ICCA Volume 6

I nt er nat i onal Conf er ence on Comput er Appl i cat i ons 2012
Volume 6

I nt er nat i onal Conf er ence on Comput er Appl i cat i ons 2012
Volume 6
In association with
Association of Scientists, Developers and Faculties (ASDF), India
Association of Computer Machinery(ACM)
Science & Engineering Research Support society (SERSC), Korea

Information Security Systems, Signal Processing Systems

27-31 January 2012
Pondicherry, India

Editor-in-Chief
K. Kokula Krishna Hari

Editors:
E Saikishore, T R Srinivasan, D Loganathan,
K Bomannaraja and R Ponnusamy

Published by

Association of Scientists, Developers and Faculties
Address: 27, 3
rd
main road, Kumaran Nagar Extn., Lawspet, Pondicherry-65008
Email: admin@asdf.org.in || www.asdf.org.in

International Conference on Computer Applications (ICCA 2012)
VOLUME 6

Editor-in-Chief: K. Kokula Krishna Hari
Editors: E Saikishore, T R Srinivasan, D Loganathan, K Bomannaraja and R Ponnusamy

Copyright 2012 ICCA 2012 Organizers. All rights Reserved

This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including
photocopying, recording or any information storage and retrieval system now known or to be invented, without written
permission from the ICCA 2012 Organizers or the Publisher.

Disclaimer:
No responsibility is assumed by the ICCA 2012 Organizers/Publisher for any injury and/ or damage to persons or
property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products
or ideas contained in the material herein. Contents, used in the papers and how it is submitted and approved by the
contributors after changes in the formatting. Whilst every attempt made to ensure that all aspects of the paper are uniform
in style, the ICCA 2012 Organizers, Publisher or the Editor(s) will not be responsible whatsoever for the accuracy,
correctness or representation of any statements or documents presented in the papers.

ISBN-13: 978-81-920575-9-0
ISBN-10: 81-920575-9-0

PREFACE

This proceeding is a part of International Conference on Computer Applications 2012 which was
held in Pondicherry, India from 27-Dec-2012 and 31-Dec-2012. This conference was hosted by
Techno Forum Research and Development Centre, Pondicherry in association with Association of
Computer Machinery(ACM), Association of Scientists, Developers and Faculties (ASDF), India,
British Computer Society (BCS), UK and Science and Engineering Supporting Society (Society),
Korea.
The world is changing. From shopping malls to transport terminals, aircraft to passenger ships, the
infrastructure of society has to cope with ever more intense and complex flows of people. Today,
more than ever, safety, efficiency and comfort are issues that must be addressed by all designers. The
World Trade Centre disaster brought into tragic focus the need for well-designed evacuation systems.
The new regulatory framework in the marine industry, acknowledges not only the importance of
ensuring that the built environment is safe, but also the central role that evacuation simulation can
play in achieving this.
An additional need is to design spaces for efficiency ensuring that maximum throughput can be
achieved during normal operations and comfort ensuring that the resulting flows offer little
opportunity for needless queuing or excessive congestion. These complex demands challenge
traditional prescriptive design guides and regulations. Designers and regulators are consequently
turning to performance-based analysis and regulations facilitated by the new generation of people
movement models.
When a greater changes are achieved these past years, still more is to be achieved which still seems
to be blue sky of 1970s. But for all the challenges, capabilities continue to advance at phenomenal
speed. Even three years ago it may have been considered a challenge to perform a network design
involving the evacuation of 45,000 people from a 120 story building, but with todays sophisticated
modelling tools and high-end PCs, this is now possible. Todays challenges are much more ambitious
and involve simulating the movement and behaviour of over one million people in city-sized
geometries. The management of these network is also easy and more specifically all the 45,000
people can be monitored by a single person sitting in his cabin. This has been the evidence of the
development these days.
As such, the conference represents a unique opportunity for experts and beginners to gain insight into
the rapidly.
Also I would like to thank all the co-operators for bringing out these proceedings for you which
majorly includes my mom Mrs. K. Lakshmi and my dad Mr. J. Kunasekaran. Apart from them my
biggest worthy gang of friends including Dr. S. Prithiv Rajan, Chairman of this conference, Dr. R. S.
Sudhakar, Patron of this Conference, Dr. A. Manikandan and Dr. S. Avinash, Convener of this
conference, Dr. E. Sai Kishore, Organizing Secretary of this Conference and the entire team which
worked along with me for the rapid success of the conference for past 1 year from the date of
initiating this Conference. Also I need to appreciate Prof. T. R. Srinivasan and his team of Vidyaa
Vikas College of Engineering and Technology for helping to make the publication job easy.
Finally, I thank my family, friends, students and colleagues for their constant encouragement and
support for making this type of conference.
-- K. Kokula Krishna Hari
Editor-in-Chief

Organizing Committee

Chief Patron
Kokula Krishna Hari K, Founder & President, Techno Forum Group, Pondicherry, India

Patron
Sudhakar R S, Chief Executive Officer(CEO), Techno Forum Group, Pondicherry, India

Chairman
Prithiv Rajan S, Chairman & Advisor, Techno Forum Group, Pondicherry, India

Convener
Manikandan A, Chief Human Resources Officer(CHRO), Techno Forum Group, India

Organizing Secretary
Sai Kishore E Chief Information Officer, Techno Forum Group, India.
Operations Chair
G S Tomar Director, MIR Labs, Gwalior, India

International Chair
Maaruf Ali Executive Director, (ASDF) - Europe, Europe

Hospitality
Muthualagan R Alagappa College of Technology, Chennai

Industry Liaison Chair
Manikandan S Executive Secretary, Techno Forum Group, India

Technical Panels Chair
Debnath Bhattacharyya, Executive Director, (ASDF) - West Bengal, India

Technical Chair
Samir Kumar Bandyopadhyay Former Registrar, University of Calcutta, India
Ponnusamy R President, Artificial Intelligence Association of India, India
Srinivasan T R,Vice-Principal, Vidyaa Vikas College of Engineering and Technology

Workshops Panel Chair
Loganathan D Department of Computer Science and Engineering, Pondicherry
Engineering College, India

MIS Co-Ordinator
Harish G Trustee, Techno Forum Research and Development Centre, Pondicherry

Academic Chair
Bommanna Raja K, Principal, Excel College of Engineering for Women, India
Tai-Hoon Kim Professor & Chairman, Dept. of Multimedia, Hanmam University, Korea

TECHNICAL REVIEWERS
Adethya Sudarsanan Cognizant Technology Solutions, India
Ainuddin University of Malaya, Malaysia
Ajay Chakravarthy University of Southampton, UK
Alessandro Rizzi University of Milan, Italy
Al-Sakib Khan Pathan International Islamic University, Malaysia
Angelina Geetha B S Abdur Rahman University, Chennai
Aramudhan M PKIET, Karaikal, India
Arivazhagan S Mepco Schlenk Engineering College, India
Arokiasamy A Anjalai Ammal Mahalingam Engineering College, India
Arul Lawrence Selvakumar A Adhiparasakthi Engineering College, India
Arulmurugan V Pondicherry University, India
Aruna Deoskar Institute of Industrial & Computer Management and Research, Pune
Ashish Chaurasia Gyan Ganga Institute of Technology & Sciences, Jabalpur, India
Ashish Rastogi Guru Ghasidas University, India
Ashutosh Kumar Dubey Trinity Institute of Technology & Research, India
Avadhani P S Andhra University, India
Bhavana Gupta All Saints College of Technology, India
Bing Shi University of Southampton, UK
C Arun R. M. K. College of Engineering and Technology, India
Chandrasekaran M Government College of Engineering, Salem, India
Chandrasekaran S Rajalakshmi Engineering College, Chennai, India
Chaudhari A L University of Pune, India
Ching-Hsien Hsu Chung Hua University, Taiwan
Chitra Krishnamoorthy St Josephs College of Engineering and Technology, India
Christian Esteve Rothenberg CPqD (Telecom Research Center), Brazil
Chun-Chieh Huang Minghsin University of Science and Technology, Taiwan
Darshan M Golla Andhra University, India
Elvinia Riccobene University of Milan, Italy
Fazidah Othman University of Malaya, Malaysia
Fulvio Frati University of Milan, Italy
G Jeyakumar Amrita School of Engineering, India
Geetharamani R Rajalakshmi Engineering College, Chennai, India
Gemikonakli O Middlesex University, UK
Ghassemlooy Z Northumbria University, UK
Gregorio Martinez Perez University of Murcia, Spain
Hamid Abdulla University of Malaya, Malaysia
Hanumantha Reddy T Rao Bahadur Y Mahabaleswarappa Engineerng College, Bellary
Hari Mohan Pandey NMIMS University, India
Helge Langseth Norwegian University of Science and Technology, Norway
Ion Tutanescu University of Pitesti, Romania
Jaime Lloret Universidad Politecnica de Valencia, Spain
Jeya Mala D Thiagarajar College of Engineering, India
Jinjun Chen University of Technology Sydney, Australia
Joel Rodrigues University of Beira Interior, Portugal
John Sanjeev Kumar A Thiagarajar College of Engineering, India
Joseph M Mother Terasa College of Engineering & Technology, India
K Gopalan Professor, Purdue University Calumet, US
K N Rao Andhra University, India
Kachwala T NMIMS University, India
Kannan Balasubramanian Mepco Schlenk Engineering College, India
Kannan N Jayaram College of Engineering and Technology, Trichy, India
Kasturi Dewi Varathan University of Malaya, Malaysia
Kathirvel A Karpaga Vinayaga College of Engineering & Technology, India
Kavita Singh University of Delhi, India
Kiran Kumari Patil Reva Institute of Technology and Management, Bangalore, India
Krishnamachar Sreenivasan IIT-KG, India
Kumar D Periyar Maniammai University, Thanjavur, India
Lajos Hanzo Chair of Telecommunications, University of Southampton, UK
Longbing Cao University of Technology, Sydney
Lugmayr Artur Texas State University, United States
M HariHaraSudhan Pondicherry University, India
Maheswaran R Mepco Schlenk Engineering College, India
Malmurugan N Kalaignar Karunanidhi Institute of Technology, India
Manju Lata Agarwal University of Delhi, India
Mazliza Othman University of Malaya, Malaysia
Mohammad M Banat Jordan University of Science and Technology
Moni S NIC - GoI, India
Mnica Aguilar Igartua Universitat Politcnica de Catalunya, Spain
Mukesh D. Patil Indian Institute of Technology, Mumbai, India
Murthy B K Department of Information and Technology - GoI, India
Nagarajan S K Annamalai University, India
Nilanjan Chattopadhyay S P Jain Institute of Management & Research, Mumbai, India
Niloy Ganguly IIT-KG, India
Nornazlita Hussin University of Malaya, Malaysia
Panchanatham N Annamalai University, India
Parvatha Varthini B St Josephs College of Engineering, India
Parveen Begam MAM College of Engineering and Technology, Trichy
Pascal Hitzler Wright State University, Dayton, US
Pijush Kanti Bhattacharjee Assam University, Assam, India
Ponnammal Natarajan Rajalakshmi Engineering College, Chennai, India
Poorna Balakrishnan Easwari Engineering College, India
Poornachandra S RMD Engineering College, India
Pradip Kumar Bala IIT, Roorkee
Prasanna N TMG College, India
Prem Shankar Goel Chairman - RAE, DRDO-GoI, India
Priyesh Kanungo Patel Group of Institutions, India
Radha S SSN College of Engineering, Chennai, India
Radhakrishnan V Mookamibigai College of Engineering, India
Raja K Narasu's Sarathy Institute of Technology, India
Ram Shanmugam Texas State University, United States
Ramkumar J VLB Janakiammal college of Arts & Science, India
Rao D H Jain College of Engineering, India
Ravichandran C G R V S College of Engineering and Technology, India
Ravikant Swami Arni University, India
Raviraja S University of Malaya, Malaysia
Rishad A Shafik University of Southampton, UK
Rudra P Pradhan IIT-KGP, India
Sahaaya Arul Mary S A Jayaram College of Engineering & Technology, India
Sanjay Chaudhary DA-IICT, India
Sanjay K Jain University of Delhi, India
Satheesh Kumar KG Asian School of Business, Trivandrum, India
Saurabh Dutta Dr B C Roy Engineering College, Durgapur, India
Senthamarai Kannan S Thiagarajar College of Engineering, India
Senthil Arasu B National Institute of Technology - Trichy, India
Senthil Kumar A V Hindustan College, Coimbatore, India
Shanmugam A Bannari Amman Institute of Technology, Erode, India
Sharon Pande NMIMS University, India
Sheila Anand Rajalakshmi Engineering College, Chennai, India
Shenbagaraj R Mepco Schlenk Engineering College, India
Shilpa Bhalerao FCA Acropolis Institute of Technology and Research
Singaravel G K. S. R. College of Engineering, India
Sivabalan A SMK Fomra Institute of Technology, India
Sivakumar D Anna University, Chennai
Sivakumar V J National Institute of Technology - Trichy, India
Sivasubramanian A St Josephs College of Engineering and Technology, India
Sreenivasa Reddy E Acharya Nagarjuna University, India
Sri Devi Ravana University of Malaya, Malaysia
Srinivasan A MNM Jain Engineering College, Chennai
Srinivasan K S Easwari Engineering College, Chennai, India
Stefanos Gritzalis University of the Aegean, Greece
Stelvio Cimato University of Milan, Italy
Subramanian K IGNOU, India
Suresh G R DMI College of Engineering, Chennai, India
Tulika Pandey Department of Information and Technology - GoI, India
Vasudha Bhatnagar University of Delhi, India
Venkataramani Y Saranathan College of Engineering, India
Verma R S Joint Director, Department of Information and Technology - GoI, India
Vijayalakshmi K Mepco Schlenk Engineering College, India
Vijayalakshmi S Vellore Institute of Technology, India
Ville Luotonen Hermia Limited, Spain
Vimala Balakrishnan University of Malaya, Malaysia
Vishnuprasad Nagadevara Indian Institute of Management - Bangalore, India
Wang Wei University of Nottingham, Malaysia
Yulei Wu Chinese Academy of Sciences, China

Part I
Proceedings of the Second International Conference on
Computer Applications 2012
ICCA 12
Volume 6
Proc. of the Second International Conference on Computer Applications 2012 [ICCA 2012] 1
www.asdf.org.in 2012 :: Techno Forum Research and Development Centre, Pondicherry, India www.icca.org.in
Proc. of the Intl. Conf. on Computer Applications
Volume 1. Copyright 2012 Techno Forum Group, India.
ISBN: 978-81-920575-7-6:: doi: 10. 74397 /ISBN_0768
ACM #: dber.imera.10. 74397
Analysis of Data losses in Sensor Networks Used in Harsh Environment

Sunita B. K

Lecturer, Computer Science Department
Vidhya Vardhaka college of Engineering

Abstract Wireless sensor networks are expected
to be deployed in harsh environments characterized
by extremely poor and fluctuating channel conditions.
Transmission errors are inherent in wireless
communications because of the instability of wireless
channels resulting due to channel fading, time-
frequency coherence, inter-band interference, etc,
and thus receiving unreliable message packets Among
a wide range of sensor network applications, many of
them such as in the process industry, require reliable
data communications such that data packets can be
delivered to the destination without loss even under
harsh conditions. However, existing reliable
transmission techniques either are too costly for
resource-constrained sensor networks or have limited
capabilities for achieving reliable data
communication under such harsh conditions.
In this paper we present a survey on data loss in
WSN used in harsh environment. Describe the
sources of data loss and focus on available
communication channel and study on performance of
various strategies on Error Control Codes We also
provide a brief summary and comparison of existing
Communication Channel
Keywords - Wireless sensor network; Error Control
Code; BER; power consumption; Forward Error
Correction.
I. INTRODUCTION
Wireless communications has seen a grand
development over the last thirty years. There are large
numbers of various applications, which inducing more
and more complicated technical challenges. Some of
these challenges include higher data rates, ubiquity of
wireless devices, trend towards faster, smaller and
cheaper hardware and frequency congestion, sensors
have limited battery power, limited transmission rate, and
much reduced other capabilities.. Many application
scenarios are mission critical and correct delivery of data
in a timely fashion is paramount and therefore may take
precedence over energy efficiency, bandwidth utilisation,
and other considerations. To be precise we can define
performance as the following:
1. The ability to accurately and effectively sense the
desired data
2. The ability to reliably deliver this data to its
destination
3. The ability to deliver data within the necessary
time bounds differing kinds of errors and the amount of
delay typically suffered by their use. We also identify
several other issues of importance with regard to real
time systems
The ability to accurately and effectively sense the
desired data is dictated by the abilities of the sensors,
their mode of operation and their physical deployment
related wireless channel .The wireless radio channel is a
shared, naturally limited resource which has a
characteristic that varies with time, frequency and space.
Radio waves are affected by various phenomena which
can cause errors and affect the quality of
communications. In this Paper we have discussed
different physical and statistical models to characterize
the wireless channels encountered in harsh
communication environment. Also it examines the causes
of data loss in sensor networks and the strategies used to
attain reliable data transmission. There are several
techniques available to increase data reliability, e.g.,
error control codes and retransmissions, but to select the
best method it is essential to have some knowledge of the
characteristics of the wireless channel. In addition, many
communication systems are evaluated using computer
simulations of Monte-Carlo type. In these cases a
mathematical model of the wireless channel is needed.
The cost of implementing these techniques has been
measured primarily in terms of energy consumption. An
underlying problem, disregarded by previous research, is
that reliability mechanisms can create delays which cause
problems for delay intolerant applications. Such a
problem area has not previously been defined in WSN
research. We intend to investigate various reliability
strategies employed in wireless sensor networks and
discuss their applicability to differing kinds of errors and
the amount of delay typically suffered by their use. We
also identify several other issues of importance with
regard to real time systems
The rest of the paper is organised as follows. Section
II discusses the causes of data loss in sensor networks.
Section III discusses the different models for wireless
sensor network channels. Section IV discusses reliability
strategies to prevent data loss and the effects of
implementing the reliability strategies discussed on data
delivery delay. Section V discusses the Energy model for
wireless sensor network. Section VI concludes the paper.
II. CAUSES OF DATA LOSS
A strategy to ensure reliability in a sensor network
has a number of obstacles to overcome. These sources of
data loss include:
1. Losses in the wireless channel due to interference
and physical effects
2. MAC layer losses due to collisions
3. Losses due to node related errors or failures
4. Losses due to routing failure
The wireless channel is inherently lossy and
experiences both bit and burst errors caused by
interference from other wireless devices, and multipath
effects caused by the environment.
In addition the wireless signal may be absorbed or
blocked by certain objects. This problem is compounded
by the fact that the wireless channel is not constant over
time [3], [4]. It is often assumed that errors in the
wireless channel occur as a statistical distribution of bit
errors. However, in wireless networks it has been
observed that errors can be bursty in nature and can
effect significant areas of the network for a period of
time before disappearing [5]. During these periods it is
often impossible to send or receive from any of the
affected nodes.
Accessing the MAC layer may lead to errors in some
particular cases. It is often assumed, incorrectly, that
node connectivity is bi-directional and this is sometimes
not the case. In these particular instances MAC layer
errors can potentially occur. Contention based MAC
protocols, whether using collision avoidance or not, falls
foul of this condition via the hidden terminal problem.
Typically, collision avoidance schemes assume bi-
directional connectivity while CSMA schemes explicitly
ignore the problem. Results from [3], [4] suggest that bi-
directional connectivity assumptions are unrealistic.
Other MAC protocols which coordinate with other nodes
in order to automatically configure themselves may also
be susceptible to errors caused by asymmetrical
connectivity. This is particularly true where automatic
clustering and synchronization is involved.
A node may experience a failure or may simply fail
to forward a message for a number of reasons. Hardware
failure, battery failure, destruction of the node, etc., is not
uncommon among sensor nodes and this can result in
lost messages. Aside from this, constraints on memory
and buffer space can cause a packet to be dropped. For
example, certain radios have a single buffer space and if
an incoming packet is not handled before the next packet
arrives one of the packets must be dropped. In general,
nodes with a heavy routing load may be able to hold a
finite queue of messages in memory awaiting
transmission. Problems arise when traffic loads within
the network and on a particular node, are heavy and/or
the node experiences difficulty accessing the
transmission media. Thus dropped packets can occur. A
route may fail for any of the reasons described above.
Link errors, periods of congestion and node errors can
cause a designated route to become disconnected. Should
a node find itself unable to forward a message towards its
destination it will very likely drop the packet after a
period of time has expired. In this case an alternative
route is necessary and if a message is to be successfully
delivered, one must either already be formed or created
as needed.

III. MODELS OF WIRELESS SENSOR
CHANNELS

We need models for the wireless channels to evaluate
different communication protocols using computer
simulations. For this reason, we need to develop methods
to describe their different properties mathematically.
Usually, describing all phenomena which can cause
errors is too a complex problem. Since the received
signal strength determines the performance of a system,
it is vital to know the path loss and the fading statistics.
Since the received signal strength is a random variable,
we can never be sure that communication will be
successful. We describe the fading margin by cumulative
distribution function (cdf). By it, we can calculate the
probability that the received signal power is lower than a
certain value. It is common practice in communication
theory to model a channel with three different
components: multipath fading, path loss and noise.
Typically, this formula is used to model a channel:
y(t) = a(t) b(d) x(t) + n(t)
Here a(t) represents fading - a multiplicative random
variable, b(d) path loss, which can be deterministic and
n(t) is noise.
3.1. Fading and Multipath Models
On a short-distance scale we can notice fluctuations
comparable with one wavelength called small-scale
fading. Slow fading can be caused by shadowing, where
large obstruction obscures the main signal path. If the
observation was made over long distances, it will give us
more slowly varying view. Such large-scale fading is
explained by the slow loss of the received signal power.
Large-scale propagation models predict the mean signal
strength, while the small-scale propagation models
predict the rapid fluctuations. Depending on the Line Of
Sight (LOS) component and the number of Non Line Of
Sight (NLOS) components, different models are used.
Additive White Gaussian Noise (AWGN) represents
communication in free space, where only a strong LOS
component exists. In most realistic cases, there are also
one or more NLOS components. The Rician distribution
models a LOS component and several NLOS. The K-
factor of the Rician distribution describes the strength of
the LOS component as compared to the NLOS, e.g. when
K ! 1, we get AWGN and for K = 0, there is no LOS and
the fading has Rayleigh distribution. Some important
fading models are outlined below.
3.2. Noise Model
Additive white Gaussian noise (AWGN) is always
present in the real life because it comes from many
natural sources like earth and the Sun. It comprises John-
Nyquist, shot noise, black body radiation. It can be used
alone for satellite and deep space communication links,
or with other models, but then it only reconstructs the
background noise (in terrestrial environment) or the
electron movements in components, which can be
modelled as AWGN due to the central limit theorem.
3.3 Binary Erasure Channel
The binary erasure channel (BEC) can transmit one
of two symbols, 0 and 1. The bit can get erased in the
channel, but if the receiver receives the bit, it must be
correct. the probability of erasure is , while the
probabilities of getting the correct bit are 1-. An IP
network can be modelled as a BEC. Either a packet
arrives or then it is correct, i.e. it passed the error
detection check, or it is wrong and then dropped, i.e.
erasure. small height. Total path loss consists of the free
space path loss PL0, multi screen loss along the
propagation path Lmsd and attenuation from the last
roof-edge to the mobile station, called rooftop-to-street
diffraction and scatter loss, Lrts, PL = PL0 + Lrts +
Lmsd:
IV. RELIABILITY STRATEGIES
There are at present a number of different reliability
strategies that can be used in combination or ndividually.
These can be classified by the particular problem area
that they attempt to address. In this section we have
considered data transmission techniques related to link
layer data
A number of techniques exist which address the need
for reliability at the link layer. Link layer reliability
mechanisms can be effective in increasing the reliability
of a single link but fail to eliminate the effects of route
failure and only partially remove the effects of node error
and failure. In this regard additional mechanisms are
needed. Forward error correction (FEC) relies on
redundant data being appended onto the existing data. Bit
errors that occur can be detected and corrected [5], [6].
Various types of codes exist with various properties. In
general the stronger the code that an FEC uses the more
bit errors it can successfully detect and correct. In
general, the stronger the FEC the more redundant bits are
needed thus increasing transmission time. The use of
variable FEC strength depending on packet importance
was investigated in [6]. Recent research has shown that
the use of error correcting codes is of limited use in
sensor networks [5] since errors are typically bursty in
nature. In essence [5] describes that where burst errors
occur the number of redundant bits necessary for
correction is prohibitively expensive. Since, burst errors
are common in wireless sensor networks it is reasonable
to conclude that FECs are of limited value. If FECs are
used the additional redundant data takes additional time
to encode, transmit and additional time to decode. In
general, the stronger the FEC the more redundant bits are
needed thus increasing transmission time. However this
time is usually relatively small. The time taken to encode
and decode FEC codes depend on the abilities of the
sensor node and the complexity of the FEC code used.
Since processing power on sensor nodes is relatively low
this may take quite a long time. However, on the Dsys25
it is possible to add stackable hardware layers, one of
which is an FPGA which can be used to greatly speed the
encoding and decoding of FEC codes

Figure 1 FEC delay summed over number of hops
The time taken to encode, transmit, receive and
decode an FEC can be described as follows:
Where Encode FEC and DecodeFECis the time taken to
encode and decode respectively, and T
x/Rxdata+header+FEC
is the time taken to transmit the data, the header and
the additional bits for the FEC respectively to the
receiver. Consider Fig. 1. We evaluate the total
transmission time for a message over a variable number
of hops using Equation 1. The time taken to transmit and
receive a packet without addition FEC codes is 4
milliseconds (typical transmission time on Dsys25 node).
The additional time take to transmit and receive the FEC
is 4ms. MAC access is considered constant and takes 16
ms (typical MAC access time on Dsys25 node using
contention based MAC protocol under medium traffic).
The times taken to encode and decode are described in
Fig.1 and are 8ms and 32ms respectively. Note that in the
following figures (Fig. 1 to 4) the value for the MAC
access time is considered to be constant and the fore the
resultant graphs appear linear whereas in a real network
they will be more random and affected by changes in
traffic cause by longer packet size and retransmission
etc.. ARQ protocols may use multiple repeat
transmissions to ensure correct data delivery between
two nodes. Obviously this will have a significant effect
on the delay incurred. However note that ARQ protocols
are inherently resilient to burst errors since the repeat
messages are temporally displaced from each other.
Repeating a message can also help to ameliorate the
effects of both node errors and, in some cases, MAC
errors. The time taken to successfully send a packet when
using repeat link layer transmissions is as the following
probability:

where pdrop is the probability that a packet is lost on
the wireless link, n is the nth transmission sent, and
TimeOut is the time a node waits for an ACK to be
received. The same values for T x/Rx and [MACaccess]
as used in Fig. 1are used (4 and 16ms).The timeout takes
6ms. The delay for a varying number of retransmissions
is considered.


Fig 2 High quality or alternate route delay

Fig 3 Delay due to routing

Fig 4 Delay incurred by end to end methods using RTT

V. ENERGY MODEL FOR WIRELESS
SENSOR NETWORK

Energy consumed at node level can be computed as
follows:

Where
E
total
: Total energy consumed in the network
E
enc
: Energy consumed by the encoder at the first
node
E
TX
: Energy consumed in transmission by all nodes
E
RX
: Energy consumed in receiving the data by all
nodes
except first one

Where
m: Number of hops
Nb: Total number of bits transmitted
Etx: Energy consumed in transmitting a single bit
from a
node
Erx: Energy consumed in receiving a single bit at a
node
The term Etx can be represented as [9]:

Where, Ete is the power consumption at transmitter
electronics is the path loss component usually varies
between 24 with = 3 being a typical value when
scattering is considered [8].
Eta is the power consumption of transmitter amplifier
and can be given as:

Where is the desired SNR at the ith
receiver, NF
RX
is noise figure at receiver, N
0
is the
thermal noise, BW is the bandwidth of channel noise,
is wavelength, Gant is antenna gain,
amp
is the
transmitter efficiency and Rbit is the raw channel rate in
bps. If the channels have low noise, this scheme may
completely recover the original data with the additional
advantage of low energy consumption. Research has to
be done on this area by implementing these models in
harsh environment.
VI. CONCLUSION
Reliable data transmission in harsh environment
depends on the wireless channels and data transmitted
from the sensor nodes are vulnerable to corruption by
errors induced by these noisy channels Hence it is
necessary to analyse and model wireless channel to
provide a proper error control scheme to reduce the bit
error rate (BER). Due to the stringent energy constraint
in sensor networks, it is vital to use energy efficient error
control scheme.
In this paper, the wireless channel is analysed and the
characteristics of wireless channels change over time,
space, and frequency in ways that are difficult to capture
entirely with theoretical models It has been clearly
shown that the needs of sensor network applications with
real-time constraints and demands are not considered by
current methodologies. Previous research has considered
energy as the primary cost and de facto unit of currency
in sensor networks research. We believe that if
applications with real-time constraints are to be
developed a radical shift of focus is needed that
encompasses the demands of these applications, namely
reliability of data delivery and the timeliness of data
delivery.
REFERENCES

[1] C. Intanagonwiwat, R. Govindan, D. Estrin. Directed Diffusion:
A Scalable and Robust Communication Paradigm for Sensor Networks.
Proc. of the 6th Annual Conference on Mobile Computing and
Networks, 2000
[2] F. Stann and J. Heidemann. RMST: Reliable Data Transport
in Sensor Networks. Proc. of the IEEE Internationatl Workshop on
Sensor Net Protocols and Applications, 2003
[3] A. Woo, T. Tong, D. Culler. Taming the underlying
challenges of reliable multihop routing in sensor networks. Proc.
of ACM Sensys03, Los Angeles, CA, USA, 2003
[4] D. Ganesan,B. Krishnamachari, A. Woo, D. Culler, D.
Estrin, S. Wicker. Complex behavior at scale: An experimental
study of low power wireless sensor networks. Tech. Report
UCLA/CSD-TR 02-0013, Computer Science Dept., UCLA,
2002
[5] A. Willig and R. Mitschke. Results of Bit Error
Measurements with Sensor Nodes and Casuistic Consequences
for Design of Energy- Efficient Error Control Schemes. Proc. of
the IEEE European Workshop on Wireless Sensor Networks,
2006
[6] E. O. Elliott. Estimates of error rates for codes on burst-
noise channels. In Bell System Technical Journal, volume 42,
page 19771997, 1963
[7] E. O. Elliott. Estimates of error rates for codes on burst-
noise channels. In Bell System Technical Journal, volume 42,
page 19771997, 1963
[8] P. Chen, B. ODea and E. Callaway, Energy Efficient
System Design with Optimum Transmission Range for Wireless
Ad Hoc Networks, IEEE International Conference on Comm.
(ICC 2002), vol. 2, pp. 945-952, 28 April - 2 May 2002
[9] Yang Liu. Performance improvement of wireless
communications using frequency hopping spread spectrum. In
International Journal of Communications

ISBN: 978-81-920575-7-6 :: doi: 10.73879/ISBN_0768
ACM #: dber.imera.10.73879
A 3D Stacked Mesh NoC For Reliable Inter-Layer Communication And Congestion
Reduction
K.A. Karthigeyan

PG Scholar
Department of Electronics and Communication Engg
Veltech Multitech Dr.Rangarajan Dr,Sakunthala Engineering College
Chennai,India

Abstract- The increasing viability of 3D silicon integration
technology has opened new opportunities for chip
architecture innovations. One direction is in the extension of
2D mesh based chip multiprocessor architecture into three
Dimensions. We present an efficient architecture to optimize
system performance, power consumption, and reliability of
stacked mesh 3D NoC is proposed. Stacked mesh is a feasible
architecture which takes advantage of the short inter-layer
wiring delays, while suffering from inefficient intermediate
buffers. To cope with this, an inter-layer communication
mechanism is developed to enhance the buffer utilization,
load balancing, and system fault-tolerance. The mechanism
benefits from a congestion-aware and bus failure tolerant
routing algorithm for vertical communication.

KeyWords: Networkon-Chip(NoC), 3D Integration, Fault
Tolerance, congestion reduction.

I INTRODUCTION

Ever increasing requirements on electronic
systems are one of the key factors for evolution of the
integrated circuit technology. Continuous technology
scaling has made it possible to integrate billions of
transistors on a single chip[1]. Thus the entire system with
hundreds of components can be integrated on a single chip,
which is known as a Multiprocessor System-on-Chip [2].
At such integration levels, communication plays a major
role in the design and
performance.
One outcome of higher integration levels is that
interconnection platforms are replacing the shared
buses.Networks-on-Chip (NoCs) are proposed to be used
in complex SoCs for inter-core communication because of
scalability, better throughput and reduced power
consumption [3].
On the other hand, increasing the number of cores
over a 2D plane is not efficient due to long interconnects.
With the emergence of viable 3D integration technologies
opportunities exist for chip architecture innovations to
enhance system power/performance characteristics [4]. In 3D
integration technologies, multiple layers of active devices are
stacked above each other and vertically interconnected using
through-silicon vias (TSVs) [5]. As compared to 2D designs,
3D ICs allow for performance enhancements even in the
absence of scaling because of the reduced interconnect lengths
[6]. In addition to this clear benefit, package density is
increased significantly, power is reduced due to shorter wires,
and system is more immune tonoise [7].
One of the well-known 2D NoC architectures is
the 2D Mesh. This architecture consists of an mn mesh of
switches interconnecting IP blocks placed along with them.
The straightforward extension of this popular planar
structure is 3D Symmetric NoC by simply adding two
additional physical ports to each router; one for Up and one
for Down [8]. Despite simplicity, this architecture has two
major inherent drawbacks. Firstly, it does not exploit the
beneficial attribute of a negligible inter-wafer distance in
3D chips, because in this architecture, inter-layer and intra-
layer hops are indistinguishable. Secondly, a considerably
larger crossbar is required as a result of two extra ports [9].
The Stacked (Hybrid NoC-Bus) mesh architecture
presented in [10] is a hybrid architecture between the
packet switched network and the bus architecture to
overcome the mentioned 3D Symmetric NoC challenges. It
takes advantage of the short inter-layer distances, around
20m, that are characteristics of 3D ICs [4]. It integrates
the multiple layers of 2D mesh networks by connecting
them with a bus spanning the entire vertical distance of the
chip.As the inter-layer distance for 3D ICs is small, the bus
length will also be smaller; approximately (n-1) * 20m,
where n is the number of layers.
This makes the bus suitable for interlayer
communication in vertical direction. By using the stacked
mesh architecture, six-port router is required instead of
seven ports for typical 3D NoC router and vertical
communication is just one hop away to any destination
layer.
In this paper, we address the routing issues and
buffer utilization to enhance the overall system power and
performance of existing stacked mesh . In addition, the
proposed architecture increases tolerance against single bus
failure architectures.

Thre
evolved t
scaling by
descriptio
the 3D I
shown th
efficient,
considera
3D NoCs
For each
ports are
55 cross
crossbar
of ports,
much hig
The s
router ha
proposed
Its basic p
Time-Div
used as
vertical c
architectu
consumpt
architectu
two buffe
input buf
later, th
congestio
W
enhance
communi
architectu
traverse p
makes it
hops per
degrades
architectu
vertical
crossbar
performan
deals wit
decision
congestio
addition,
tolerate a
U
resources
tolerance
mesh arch
is high-th
other han
mesh arch
of reduce
mesh
II REL
ee(3D) dime
to deal with t
y stacking act
on of the cha
ICs is provid
hat 3D ICs
but when t
ation, the stat
s are extension
NoC router
needed result
sbar for the 2D
power increas
the power co
gher than a 2D
solution to the
as been prop
architecture
purpose was c
vision Multipl
a communica
communicatio
ure is efficien
tion and lat
ure is that eac
ers: the source
ffer as shown
he output b
on-aware inter-
We improve
the throughp
ication resou
ure due to tw
packets throu
less power e
packet for v
system t
ure deals wi
communicatio
improves
nce. Adaptive
th load balan
making ro
on in 2D NoC
the proposed
single link fa
Utilizing the
s, we enhance
for inter-lay
hitecture. Sym
hroughput but
nd, architectur
hitectures are
ed throughpu
architecture
LATED WOR
ensional inte
the limitations
tive silicon lay
allenges faced
ded in [4]. T
are power a
the 3D NoC
tistics are qui
n to the 2D N
of mesh topo
ting a 77 cr
D mesh archit
ses quadratica
onsumption fo
router [11].
e power consu
osed by Li
is stacked m
cache coheren
le Access (dT
ation pillar. D
on and 66 r
nt enough in
tency. The
ch packet is t
e output buffe
in Fig. 1. As
buffer hinder
-layer commu
the architec
put by usin
urces. We c
wo major asp
ugh two stage
fficient. Seco
vertical comm
throughput.
ith both issu
on without
the system
e vertical rou
ncing. In [12
outing algori
architectures
dynamic rout
ilure.
e available
our approach
yer communic
mmetrical 3D
t not power
res like True
power efficie
ut. We reinfo
without
RK
egrated circ
s of interconn
yers. The deta
d to manufact
The authors h
and performa
is taken un
ite different. T
NoC architectu
ology, two ex
ossbar instead
ecture. Since,
ally with num
or a 3D route
umption for a
et al. [10]. T
mesh architectu
nce. The dyna
TDMA) bus w
Due to one
routers, propo
terms of po
issue with
traversed thro
er and destinat
s we will disc
rs implement
unications.
cture to furt
ng the availa
complement
pects. First is
crossbar, wh
ond is to use
munication, wh
Our propo
ues. Single
using two-st
m power
uting mechan
2], an enhan
ithm to av
was proposed
ting approach
communicat
to deal with f
cation of stac
NoC architect
efficient. On
NoC or stac
ent at the expe
orce the stac
adding ex
cuits
nect
iled
ture
have
ance
nder
The
ure.
xtra
d of
the
mber
er is
3D
The
ure.
amic
was
hop
osed
wer
this
ugh
tion
cuss
ting
ther
able
the
s to
hich
two
hich
osed
hop
tage
and
nism
nced
void
d. In
can
tion
fault
cked
ture
the
cked
ense
cked
xtra
c
th

n
d
m
th
la
r
c
in

R
p
T
p
u
d

2
3

communication
hroughput, fau
In this arch
nodes are dif
dTDMA pillar
must be provi
he vertical lin
ayers. An ex
outer for the
channel has
ndistinguishab
In the static
RXYZ is the
packet to the d
This particula
performance
utilization. Th
different comm
1. Bus is busy
2. Bus is free b
3. Bus is busy
is full:
n resources
ult-tolerance, a
III MOTIV
hitecture, rou
fferent, as an
r (vertical lin
ided to enable
nks with the
xtra physical
e vertical com
its own ded
ble from the ot
c XYZ routing
source, whic
destination RX
ar architecture
efficient bec
his can be j
munication sce
y but input buf
but input buff
y and also the i
and enhan
and power eff
VATION
uters connecte
n interface b
nk) and the N
e seamless in
2D network
channel is ad
mmunication.
dicated buffe
ther channels.
g algorithm, c
h needs to s
XYZ+1 as show
e is neither
ause of inef
justified by
enarios as follo
ffer of RXYZ+
fer of RXYZ+
input buffer of
nce system
ficiency.

ed to pillar
between the
NoC router
ntegration of
k within the
dded to the
The extra
ers, and is

consider that
send a data
wn in Fig. 2.
power nor
fficient bus
considering
ows.
+1 is free:
1 is full:
f RXYZ+1

I
It is
vertical b
We show
obstruct r
section, w
Initially,
reduce
implemen
an adapt
AdaptiveZ
F
for intra-
algorithm
routing a
routing p
the XY
identically
layer co
algorithm
As c
communi
way for th

IV PROPOSE
s not efficient
bus without co
wed that the
routing adaptiv
we approach th
the buffer o
power con
nting an adap
tive inter-lay
Z is applied.
For the sake o
-layer commu
m is used. Not
algorithm is
olicy. Therefo
routing is p
y as shown in
ommunication
m is elaborated
can be seen
ication, the fir
he vertical com
ED ARCHITE

t to connect a
onsidering the
extra output
vity and load b
he problem in
on output por
nsumption a
ptive routing
yer routing a
of simplicity, i
unication a st
te that the pro
not dependen
ore, in ZXY ro
projected in
n Fig. 3 with th
n. The Ada
d in Algorithm
from Fig. 3
rst bus pillar
mmunication i
ECTURE
a NoC router t
eir characterist
buffer even
balancing. In
n different pha
rt is removed
and to ena
algorithm. Th
algorithm ca
it is assumed
tatic XY rout
oposed inter-la
nt to intra-la
outing algorith
different lay
he adaptive in
aptiveZ rout
m 1.
3, for inter-la
available on
is used.
to a
tics.
can
this
ases.
d to
able
hen,
lled
that
ting
ayer
ayer
hm,
yers
nter-
ting
ayer
the

a
th
w
n
a
n
h
m
w
th
b
b
c
a
c
is
V
2
s
c
tr
o
p
d
th
s
V D
To deal wi
architecture is
he input buff
which needs t
node R101. Ro
availability of
node R100. Als
have the same
mesh architect
without virtual
he proposed a
buffers as com
bus architectu
communication
architecture.
The o
communication
s used to rec
VC concept. T
21 multiplex
ignaling wir
consumption d
raversal path
optimization d
power consum
In ad
deadlock, but
he buffer is
ystem power a
DEADLOCK

ith deadlock,
used as show
ffer of router
to be transm
outer R101 is
dTDMA bus t
so, consider th
e situations. In
ture with adap
l channel, the
architecture, w
mpared to stac
ure [10], we
n scheme
output buffer
n was remove
eive the data
There is very sm
xer, one 12
res. Reductio
due to remova
h of packet
due to routing
mption of this s
dition, the ad
it also improv
used in a ri
and performan
AVOIDANCE
typical virtu
wn in Fig. 4. C
R001 contain
mitted to the
already wait
to deliver a pa
hat the node R
n case of typ
ptive vertical
ere will be a d
with the same
cked mesh wi
modify the
to support
from the rou
ed. An extra i
packets and
mall area over
de-multiplex
on in dynam
al of the buff
and also st
adaptivity do
small extra log
dded VC not
ves the throug
ght way and
nce efficient.

E
ual channel
Consider that
ns a packet,
neighboring
ting for the
acket at the
R000 and R100
ical stacked
routing and
deadlock. In
e number of
ith dTDMA
inter-layer
the VC
uter for bus
input buffer
support the
rhead of one
xer and few
mic power
fer from the
tatic power
ominates the
gic.
only avoids
ghput. Now,
makes the

communi
is a critic
physical c
TSVs can
stacked k
number o
TSVs inc
reliable
considera
explore h
routing al
when the
destinatio
For fau
adaptive
adding an
wait sig
condition
wait sig
arbiter do
signal. T
introducin
architectu
It is ass
enough re
bus link
services d
When the
which ind
bus is n
temporari
wait sig
wait sig
layer and
inter-laye
VI FAUL
Through Si
icationlinks fo
cal design issu
components, t
n fail. A failed
known-goodd
of dies to be
crease the cost
inter-layer c
ably mitigate th
how the avai
lgorithm to av
ere are other p
on pair.
ult tolerance, t
inter-layer co
ny extra wire
gnal acts as
ns. In case of
gnal is perm
oes not serve
Thus, routing
ng the fau
ure.
sumed that the
esources (test
is broken or
due to therma
e bus arbiter
dicate that ther
not able to
ily or perma
gnal accordin
gnal, it will r
d in future, w
er communic
LT TOLERAN

ilicon Via,
or dies in the v
ue in 3D integr
the fabrication
d TSV may ca
dies to be di
e stacked incr
t and decrease
communicatio
hese issues. In
ilable signals
void these pat
paths between
the existing si
ommunication
e. As explaine
a congestion
any fault on t
manently asser
e any request
adaptivity is
ult toleranc
e dTDMA bus
units) to sign
the bus is no
al problem or
receives any
re is a fault on
provide the
anently, it ass
ngly. When ro
reroute the pa
will not deriv
cation to the
NCE
which provi
vertical directi
ration. Like ot
n and bonding
ause a numbe
iscarded. As
reases, the fa
the yield [13]
on scheme
n this section,
can enable
ths (faulty bus
n the source
ignaling used f
n is used with
ed in section
n flag in nor
the dTDMA b
rted 0 and
t raised by r
not affected
ce in exist
is equipped w
nal that either
t able to prov
r any other fa
of such sign
n the bus link
communicat
serts 0 on
outer checks
ackets within
ve any traffic
e correspond

ides
ion,
ther
g of
r of
the
iled
]. A
can
we
the
ses)
and
for
hout
IV,
rmal
bus,
bus
req
by
ting
with
the
vide
ault.
nals,
and
tion
the
the
the
for
ding
d
w
m
s
m

a
p
p
d
c
e
r
n

e
in
d
b
w
p
s
n
p
d
a
r
m
dTDMA bus.
will be used ev
modification
ection IV. Bu
mechanism nee
Consider tha
a packet to n
paths are fault
proposed routi
dTDMA bus c
002 should b
current situatio
either of nod
outing algorit
normally accor
Now consid
exactly below
nter-layer com
destination but
be rerouted to
where the pa
packet will be
ituation, when
node 122. F
packet can only
dTDMA bus c
122, which i
algorithm requ
outed to the n
minimal path.
Normally, th
ven in case of
in routing
ut there are f
eds to be mod
at the node 00
ode 112. Th
ty as sown in
ng algorithm
connecting the
be tried but t
on. So, the pa
des 010 or
thm. Then th
rding to the pr
der another si
w/above its d
mmunication,
t the bus link
the neighbori
acket is curre
delivered to th
n the node 1
For proposed
y be routed th
connecting the
is faulty. In t
uires modifica
nodes 110 or
he minimal p
f faulty links
algorithm di
few cases, w
dified.
00 in Fig. 6 ne
he vertical lin
Fig. 6. Acco
in section IV,
e nodes 000
that bus link
cket will be r
100 accord
he packet wil
roposed algori
ituation, if th
destination no
it can be deliv
is broken, the
ing node withi
ently residing
he destination
20 contains a
d routing alg
hrough the vert
e nodes 120
this situation,
ation. The pac
r 020 followi
path routing
without any
iscussed in
when routing
eeds to send
nks on two
rding to the
, the vertical
, 001 and
is faulty in
outed to the
ding to the
ll be routed
ithm.
he packet is
ode and by
vered to the
e packet will
in the layer,
g. Then the
n. This is the
a packet for
gorithm, the
tical
, 121 and
the routing
cket will be
ing the non-

VII CONCLUSION

In this paper, an enhanced architecture for 3D
stacked mesh NoC was proposed to enhance system
performance, reduce power consumption, and
improve the system reliability. To this end, a
congestion aware adaptive interlayer communication
algorithm was introduced. To deal with deadlock, an
appropriate VC architecture was used with same
number of buffers as compared to the existing
stacked mesh architectures. In addition, the
congestion signal triggered by the bus arbiter was
used to deal with fault tolerance. This enabled
avoiding a faulty vertical bus in a possible path for a
number of source-destination pairs in the event of a
bus failure.

REFERENCES
[1] S. Borkar, Designing reliable systems from unreliable
components: The challenges of transistor variability and
degradation, IEEE Micro, Vol. 25, No. 6, 2005, pp. 10-16.
[2] A. Jerraya and Wayne Wolf, Multiprocessor Systems-on-
Chips, Morgan Kaufmann, 1st edition, October 12, 2004.
[3] A. Jantsch, and H. Tenhunen, Network on Chip, Kluwer
Academic Publishers, 2003.
[4] A. W. Topol et al., Three-Dimensional Integrated Circuits,
IBM J. Research and Development, Vol. 50, No. 4/5, July-Sept.
2006, pp. 491-506.
[5] V. F. Pavlidis and E. G. Friedman, 3-D Topologies for
Networks-on- Chip, in Proc. of IEEE Trans. Of VLSI Systems.,
Vol. 15, No. 10, 2007, pp.1081-1090.
[6] B. S. Feero and P. P. Pande, Networks-on-Chip in a Three-
Dimensional Environment: A Performance Evaluation, IEEE
Transactions on Computers, Vol. 58, No. 1, 2009, pp. 32-45.
[7] R. S. Patti, Three-Dimensional Integrated Circuits and the
Future of System-on-Chip Designs, in Proc of IEEE, Vol. 94, No.
6, 2006, pp. 1214-1224.
[8] L. P. Carloni et al., Networks-on-Chip in Emerging
Interconnect Paradigms: Advantages and Challenges, in Proc. of
International Symposium on Networks-on-Chip, 2009, pp. 93-109.
[9] J. Kim et al., A novel dimensionaly-decomposed router for
on-chip communication in 3D architectures, in Proc. of
International Symposium on Computer Architectures, 2007. pp.
138-149.
[10] F. Li et al., Design and Management of 3D Chip
Multiprocessors Using Network-in-Memory, in Proc. of
International Symposium on Computer Architecture, 2006, pp.
130-141.
[11] R. S. Ramanujam and B. Lin, A Layer-Multiplexed 3D On-
Chip Network Architecture, IEEE Embedded Systems Letters,
Vol. 1, No. 2, 2009, pp. 50-55.
[12] P. Lotfi-Kamran et al., EDXY A low cost congestion-
aware routing algorithm for network-on-chips, Journal of Systems
Architecture, Vol. 56, No. 7, 2010, pp. 256-264.
[13] A.-C. Hsieh et al., TSV Redundancy: Architecture and
Design Issues in 3D IC, in Proc. of International Conference on
Design, Automation, and Test in Eroupe, 2010, pp. 166-171.

c
a
S
t
e
a
a
a
f
c
i
G
c
p
f
a
a
b
m
s
L
l
M
e
t
m
Proc. of th
Volume 1. Co
ISBN: 978-8
ra

Abstract -
communicative
and that is
SMARTPHON
to end mobile c
every manufa
and innovation
As a resu
attractive featu
adaptability to
factors contrib
comparative s
iPhone, Symb
Google Androi
Keywo
Os,Android

The aim
comparative s
point of view
features, by b
and limitation
aspects of thes
The major
based softwar
market by
smartphones.A
Linux-based p
launch like it
Mobile, Symb
etc.
The battle
technological
mobile OS.
he Intl. Conf.
opyright 20
81-920575-7-
ACM #: dbe
M
Rashmi Pr
MCA Dep
SRMG
Lucknow
ashmi1330@
- Mobile pho
e devices to so
s why they
NES. Smartpho
communication
cturer is focu
ns in this field.
ult there are
ures. The succ
o third party a
bute in its suc
study of some
ian, Blackberr
id.
rd- Apple
d and Window
I. IN
of this report
study of differe
w. A compar
bringing the d
ns together, th
se mobile OS
r industry is o
re developing
about 50%
Android, the
platform whic
ts competitor
bian OS, Palm
e for market
developments
on Computer
012 Techno Fo
6 :: doi: 10. 73
er.imera.10. 73
Mobile Op
rajapati
artment
GPC
w, UP
@gmail.com
ones have em
oftware intensi
y are altern
ones are provid
n across the wo
using on mobi
many mobile
cess of any plat
applications. O
ccess. This pap
e of the key
ry OS, Windo
e iPhone,S
ws Mobile
NTRODUCTIO
t is to provid
ent Mobile OS
rison between
different Mob
his paper wil
also.
occupied by S
g company wh
% especially
flagship mob
ch is getting
rs like the iP
m OS, and R
t share is m
s and innovat
Applications
orum Group, I
3886/ISBN_0
3886
perating S
merged from
ive devices lik
natively calle
ding solution fo
orld. In presen
le operating s
e OS having
tform depends
Of course many
per is intended
mobile OS:
ows Mobile O
Symbian,Black
ON
de a review
S platform fro
n all core pla
bile OS advan
l provide tech
Symbian, a Lo
hich is leadin
in PDAs
bile platform,
ready for a w
honeOS, Win
RIM Blackberr
making a pat
tions in the fi

ndia.
768
System: A

mere
ke PCs
ed as
or end
nt time
system
many
on its
y more
d to a
Apple
S and
kberry
and a
om our
atform
ntages
hnical
ondon
ng the
and
, is a
wider-
ndows
ry OS
th for
eld of
II.
play
And
In p
plat
tow
the
And
Sea
des
Cal
aug
plat
ker
upd
fixi
are-
A Compar

. THE BA
.

To occupy m
yers like Sym
droid are unve
present scenar
tform. May b
wards a truly o
sophistication
Android, Inco
dy Rubin (co
ars (once VP
sign and interf
lifornia, Unite
gust 2005.Afte
tform for mo
rnel.
Since its orig
dates, each add
ing bugs of pr
-
rative An
Ruch
MCA
SR
Luc
ruchirashuk
ATTLE FOR M
market and to m
mbian, RIM B
eiling their ow
rio customer p
be a needs gr
open source p
n on their own
A. A
orporation was
o-founder of
at T-Mobile
face developm
ed States. It
er that a team
obile devices
ginal release, A
ding some new
revious releas
nalysis
hira Shukla
A Department
RMGPC
cknow, UP
kla89@gmai
MARKET SH
monopolize th
Blackberry, Ap
wn application
prefers a simpl
row end user
platform and t
n.
Android OS
s founded in O
Danger), Ric
e), and Chris
ment at WebTV
was acquired
m led by Rub
having the p
Android has s
w and innovat
e. Recent rele
t
l.com
HARE
he industry ma
pple-iPhone, a
n stores over n
le but innovat
r will be turn
they will exte
October, 2003
ch Miner, Ni
White (head
V), in Palo Al
d by Google
bin developed
power of Lin
een a number
tive features a
eases of Andro
ajor
and
net.
ive
ned
end
by
ick
ded
lto,
in
d a
nux
r of
and
oid
clair having a new web browser and support for
HTML5 and W3C geolocishaving some smart features like
digital Zoom, color effects and more.
clair also includes support for animated backgrounds,
voice controls, buttons to open the menu, new weather
application and improved Email and Phonebook
applications.
Froyo introduced speed improvements with JIT
optimization and the Chrome V8 JavaScript engine, and
added Wi-Fi hotspot tethering and Adobe Flash support.
Honeycomb was announced at the 2011 Google I/O on
10 May 2011. - To allow honeycomb devices to directly
transfer content from USB devices.
Honeycomb is "an incremental release that adds several
new capabilities for users and developers". Highlights
include optimization for a broader range of screen sizes;
new "zoom-to-fill" screen compatibility mode; capability to
load media files directly from the SD card; and an extended
screen support API, providing developers with more precise
control over the UI.
b.Symbian OS
Symbian,on June 24, 2008 was acquired by
Nokia.Previously it was owned by Ericsson, sonyericsson,
Panasonic and Samsung. Symbian
OS offers a high level of integration of personal
information management with communication (PIM).It
provides many services like navigation, games,
associatelibraries; music playbacks etc.It has some hardware
requirements-

X86 architecture Monitor: 1024 x 768pix
el screen, Hard drive: 200 MB of free disk space, 16
bit color Memory: 512MB, CPU: 1.2GHz processor.
Some of the releases of Symbian are-

1) Epoc Release 5: In 2000, it was released with
Bluetooth successor code named conon.It also
supported Unicode.
2) Symbian OS 6.0 and 6.1 ( ER6 ):It was
released in 2002.In this release Bluetooth was added
for exchanging data over short distances creating
Personal Area Network(PAN).It is used
in Nokia 9210.
3) Symbian OS 7.0 and 7.0s:
This version added EDGE support and IPV6.
4) Symbian OS 8.0:
In this version of Symbian OS, there are no majo
r evolutions, has shared some APIs to support 3G.
5) Symbian OS 9.1:
The change for this version is the change of versi
on 1.2 for the Bluetooth has version 2.0 where the di
fference is the introduction of an Enhanced Data Rat
e (EDR) for faster data transfer. The nominal Rate of
EDR is about 3 megabits per second although the pr
actical data transfer rate is 2.1 megabits per second.
al television broadcasts in DVBH and
ISDBT formats and also location services.
B. Windows Phone
Microsoft captured a great attention at Mobile World
Congress by introducing Windows Phone 7 Series, such as
WP7S, or WinPho and represents a completely new
approach to interacting with your phone. Instead of icons for
Apps on the WP7 screen uses hubs, which are centers for
any typeof information, and all the hubs seem to
interconnect to one another. For example, the People hub
contains your contacts, synced from and linked with all your
social networks.

The tight integration of social networks such as Twitter,
Facebook, and MySpace, may present another challenge.
Microsoft had to create a whole subset of instructions for
each social network and build it into Windows Phone 7
Series at the core level.
Windows Mobile is an excellent multi-tasking platform
and it has no problem with tasks running in the background
because WinMo automatically changes memory allocation
based on the needs of each application, and if the system
memory starts to get stretched thin, the OS shuts down
background tasks automatically.

Windows Mobile OS is designed from the ground up to
integrate with both Exchange Servers in the corporate
world, and with Outlook environments on the desktop. All
aspects of desktop Outlook commingle happily with Mobile
Outlook on the Windows Mobile side of things.
C. Apple iPhone OS:
Apple has created a PC-grade Operating system for a
Smartphone. Currently, Apples iPhone is the longing of the
modern man. Top-notch product design, an perceptive user-
interface as well as fancy multimedia internet based services
may contribute to its unique market position.
The first Apple version was introduced on January 9,
2007.Its successor, the iPhone 3G was revealed on July 11,
2008. iPhone 3G has the following hardware specifications:
Device dimensions/weight: 4.5 x 2.4 x
0.48 inch / 4.7 ounces
Display size: 3.5 inch
Resolution: 480 x 320 pixel at 163 ppi (i.
e. aspect ratio 3:2)
Input devices: Multi-touch enabled screen
interface plus 4 buttons (Home, Sleep/Wake,
Volume up/down, Ring/silent)
Connectors and input/output: 30-pin dock
connector, 3.5mm stereo headphone

Apple iPhone OS also provides an explicit
Security framework (Security. Framework)
provides interfaces for managing certificates,
public and private keys, and trust policies that you
can u
appli

A
powe
multi
and p
be us
a ver
its ex
the m
oppo
peop

B
devel
multi
that
hand
track
touch
hand
B
suppo
and,
comp
with
Micr
R
its o
April
Black
2010
2011
Curv
(9810
(9900
devic
(No L
Blackberry
Im
Black
use to guarant
ication manag
Another key a
ered graphics
iple sensors
proximity sen
sed, and it rapi
ry profitable g
xpected for the
market as mo
ortunities to ge
les minds for
a..BlackB
BlackBerry
loped by Re
itasking and s
have been a
dhelds, predo
kball, and mo
h screen for it
dheld devices.
BlackBerry O
ort for corpo
more recently
plete wireless
Novell Group
rosoft Exchang
RIM develops
perating syste
l 2010 RIM
kBerry OS 6
0. The softwar
making its a
ve (9350/93
0/9850/9860)
0/9930). RIM
ces would not
Legacy Devic
y OS Addition
mproved prod
kBerry applica
o
o
Black
Black
o
Savin
o
openi
tee the securi
es.
aspects of the
s, touch inte
like accelerom
nsor, which w
idly shown ho
gaming platfo
e iPhone to co
ore players ta
et in, so should
r a long time to
Berry OS
mobile op
esearch In M
upports specia
adopted by R
ominantly the
ost recently,
s BlackBerry
Os best know
orate email, t
y, a subset of M
s activation an
pWise email
ge etc.
and releases u
em to suppor
announced t
6.0 version, i
re was legally
appearance on
60/9370), B
and B
M announced
t be updated t
e Support).
nal Features su
ductivity & co
ation :
ZIP A
Faste
kBerry Sma
kBerry Playbo
Enha
ng Attachment
Faste
ing
ity of the data
e iPhone: Op
erface, as we
meter, light s
were all availa
ow it could get
orm. For the f
ontinue its asce
ake notice for
d the iPhone s
o come.
perating sy
Motion .It pro
alized input de
RIM for use
e track wh
the track pad
line of Smartp
wn for its n
through MID
MIDP 2.0, it a
nd synchroniz
, Lotus Domin
updated versio
rt each devic
to release the
in the 3rd q
released in A
the the Black
Blackberry
Blackberry
that any pre
o BlackBerry
uch as:
onnectivity wi
Attachment Su
er pairing betw
artphone an
ook
anced Suppor
ts
er attach
a your
penGL
ell as
sensor
able to
t to be
future,
ent on
r their
stay in
ystem,
ovides
evices
in its
heel ,
d and
phone
native
P 1.0
allows
zation
no or
ons of
e. On
e new
quarter
August
kberry
Torch
Bold
evious
OS 7
ith the
upport
ween a
nd a
rt for
hment
II
the
Win
ofte
the
app
ava
inte
win
app
to f
goo
mu
may
abo
app
def
But
bac
acc
bac
ove
of B

II. A COM
Android com
market. Its
ndows mobile
en been critici
When we go
key features
plications at th
ailable, multita
egrate with
ndows mobile
b. Multitas

Android is
plications at th
foreground pr
od user exper
ultitasking.
Windows mo
y be a bit su
out multi-task
plication clos
fault when exit
t the OS has
ckground, it a
cording to the n
ckground tasks
Imp
er WEP netwo
Add
Blackberry OS
o
App
o
Pictures
MPARISON B
WINDOWS
me very well in
market share
e has come w
ized for being
for a compari
to be compare
he same time
asking, the us
computers
and android o
sking
very well s
he same time
rocesses as mu
rience. Andro
obile are also
rprising becau
king. The r
e option. Ap
ted, which is v
s no problem
automatically
need of applic
s automatically
proved Wi-F
orks
ditional Multi
S:
Pinch to
Portrait
s App
BETWEEN AN
S MOBILE
n a short time
is growing a
with recent up
a dated Smart
son between m
ed are the abi
e, the number
ser interface a
desktop etc
on these featur
suited for ru
seamlessly. I
uch they dese
oid OS does
good in multi
use they have
eason is tha
pplications do
very confusing
ms with tasks
changes me
cations and the
y.
Fi connectiv
imedia Featu
o Zoom in Vid
Support
NDROID AN
e and penetrat
at a rapid pa
pdates but it h
tphone platfor
mobile platfor
ility to run ma
r of applicatio
and the ability
c.Lets compa
res:
unning multip
It gives attenti
erve to provide
a good job
itasking and t
e been criticiz
at it lacks
o not close
g for some use
running in t
emory allocati
e OS shuts dow
vity
ures
deo
in
ND
ted
ace.
has
rm.
rms
any
ons
y to
are
ple
ion
e a
in
this
zed
an
by
ers.
the
ion
wn
In-fact Android OS does the same thing. When
a user leaves an application and starts another the
first one keeps running in the background.
Windows mobile has a big advantage over
Android in the area of user-installed applications.
Windows mobile allows them to be installed in
either system memory which is much larger or in
external memory cards. It supports a large number
of applications to be installed.
b. Available Applications
Android is promising over 10,000+ applications
available and too in a very short time which is
indeed a great achievement. While Android has a
very few applications that one can find on a
iPhone, yet they are good to meet users
requirement.
Windows in somewhat lacking in this feature
especially in apps store games which is not
surprising because store is new to the market.
Android web browser is claimed to be much
better than the latest version of mobile Internet
explorer and it is accurate claim.
C. User Interface
Windows mobile interface is older than
Android and for that Android interface is better
than it. Android interface looks more modern and
is easy to interact with. Both interfaces are simple
Icon based in which the icons are organized in
launchers.
The two platforms can be compared more fairly
on the basis of third party alternatives. Android is a
very capable interface for third party interface shell
such as HTC sense.HTC has long produced its own
interface for Windows Mobile phones, too, and the
latest Sense interface, such as found on the HD2, is
amazing. It does for Windows Mobile what it does
for Android adds a highly customizable face on
the OS that makes it work the way the user prefers.
IV. COMPARATIVE STUDY OF IPHONE OS,
ANDROID OS AND WINDOW OS
Comparison of iPhone OS, Android OS and Windows
OS can be done on the basis of following points:

a.. Kernel-
iPhone kernel is OS X mobile which allows Apple to
reuse a lot of the mature code that is already proven to work
on the Mac. Android kernel is Linux where as Windows
kernel is CE6. CE 6.0 can run as many as 32,000
simultaneous processes, each with 2 GB of virtual memory
space. The functionality of an application is generally
unaffected by changes to the kernel.

b..Multitasking-
Android and iPhone both provide multitasking but
iPhone also provides push navigation service with
multitasking. Multitasking in windows OS is limited but it
provides push navigation service.

c.. Proprietary cloud-
Sync service-iPhones service Mobile Me is paid while
in Android it is Google apps which are mostly free.
Windows provides service My Phone 2.0 which is
unconfirmed and free.
.d.Home Screen-
iPhone has limited customization support while
Windows starts screen with customizable tiles.

e. Skin able OS-
iPhone OS and windows OS is not skin able or it is very
limited in this feature while Android OS is skin able.

THE BLACKBERRY OS AND SYMBIAN
Comparison between Blackberry OS and Symbian can
be done on these points:
a. Ease of use-
The interface of blackberry OS is wholly thumbwheel
driven and user may find it easy to use though it might be
slow. There is a built-in keyboard which accelerates some
operations such as composing E-mails. Blackberry OS may
not be the most efficient one but it is easy to use.
Symbian OS greatly depends on the phone for its ease of
use feature. For example in Nokia 9300 it is very easy to
navigate but on the other side in Nokia 6682 the interface is
confusing and difficult for data entry.
b. E-mail-
Blackberry excels in this area. It is not only the E-mail
applet which is addictive for its streamlined efficiency but
also the push method of e-mail delivery. Instead of being
retrieved manually messages appear like a magic on device.
Symbian is as versatile as other platform having no
innovative features. If you want something a little more
robust, you can choose from several push e-mail solutions,
including BlackBerry Connect. The operating system also
supports Lotus Notes and Microsoft Exchange for
maximum compatibility in the corporate world.
c. Multimedia-
Blackberry OS includes built-in music player, integrated
cameras and the latest featuring video recording facilities.
The Symbian operating system is pretty adept at
multimedia, with integrated support for audio and video
playback and recording, though not all those capabilities are
realized on the devices themselves. The Nokia 9300, for
instance, comes with both an MP3 player and the mobile
version of RealPlayer, which enables playback of
RealAudio, Real Video, and MP3 files. But it cannot
capture audio or video.

V. CONCLUSION
In Conclusion, all mobile operating systems presented
itself as a highly compelling investment on todays mobile
market, they proved there self to be capable to reach many
different type consumers in many new and innovative ways,
as well as to raise its own Business. Besides all the
advantages of Android over Windows Mobile, there are
some users who are comfortable with Windows Mobile due
to its stability and their enterprise solution. Further,
Microsoft is going to release its Windows Mobile 7, which
can be much better than the previous version.

Our analysis shows that each Mobile Operating System
may have strengths and shortcomings; the most secure
platform for the Business use is the Blackberry platform and
the iPhone has advanced towards being enterprise-class.
Symbian brings its limitations when it comes to
development. As earlier explained youll need a differen
t
SDK for every device.You will also have a different set
of tools in each SDK this will limit the porting of program
s you right for a Symbian Smartphone.
It will work on all devices with the same SDK but on oth
ers it will give problems.

Right now if you are looking for something that is at the
tip of technology trend, Android is what you need. It has
great future if they keep their innovation cycle running
continuously

VI. REFERENCES
[1]. Symbian OS C++ for Mobile Phones by Richard Harrison
[2]. htts://en.wikipedia.org/wiki/Mobile_operating_system
[3].http://www.appleinsider.com/articles/10/08/12/android_based
_smartphone_shipments_leapfrog_apples_iphone.html
[4]. Android Issue 3748: Add support for partition/block device
encryption

ISBN: 978-81-920575-7-6 :: doi: 10. 73893/ISBN_0768

Design of DAODV MANET Routing Protocol

G. Ilanchezhiapandian

Research Scholar , Department of CSE,
Sathyabama university,chennai

Dr.P.Sheik Abdul Khader

Professor & HOD, Department of Computer Applications,
B.S.Abdur Rahman University,Chennai

Abstract - Ad-hoc networking is a concept in computer
communications, which means that users wanting to
communicate with each other form a temporary network, without
any form of centralized administration. Each node participating in
the network acts both as a host and as a router. The study of
distributed network and mobile ad hoc network (MANET) are
currently two hotspots in distributed computing and mobile
communication researching domain. We build up a distributed
overlay network on top of MANET's physical infrastructure, to
effectively integrate distributed networks advantage on
sustaining highly dynamic network into the design of MANET
routing protocol. We deploy passive MANET routing algorithms
with innovative distributed routing mechanisms. We propose a
new kind of MANET routing protocol named distributed
Computing based ad-hoc on demand routing (DAODV). Our
work indicates that DAODV has an improved routing
performance in comparison with the popularly used AODV
protocol.

Keywords- MANET, DAODV, distributed network.

I. INTRODUCTION

In ad-hoc networks, communications are done over
wireless media between stations directly in a distributed
fashion without the help of wired base station or access
points. With the popularity of ad-hoc networks, many
routing protocols have been designed for route discovery
and route maintenance. They are mostly designed for best
effort transmission without any guarantee of quality of
transmissions. Some of the most famous routing protocols
are Dynamic Source Routing (DSR) Ad-hoc On Demand
Vector (AODV) Optimized Link State Routing protocol
(OLSR), and Zone Routing Protocol (ZRP). On the other
side, Distributed network is currently a hotspot in the
international network researching domain. Different kinds
of distributed network models such as Napster, Freenet and
Gnutella have already been popularly used. However, each
of them has some open problems [2], [3]. To solve these
problems, structured DISTRIBUTED overlay networks
(SON) such as CAN, Chord [4], Pastry and Tapestry are
proposed to provide a kind of mechanism to improve
network scalability [5]. All of these SON algorithms have a
common ground that they all assign keys to data and nodes
through a hashing function, and at the same time they use
these keys to store and maintain data which are shared in
the network [6].

II.OVERVIEW OF MANET AND DISTRIBUTED
NETWORKS
DISTRIBUTED networks are popular Internet applications
that enable users to share computing resources. Today,
DISTRIBUTED, Web browsing, and email can be put
together in the same class of importance. Napster is the
tool probably most responsible for the popularization of
DISTRIBUTED networks due to its music sharing support.
So far, several Napster clones and general file sharing
systems have been created, such as KaZaA, BitTorrent,
and eMule. DISTRIBUTED networks are composed of
software nodes (peers) installed on the edges of the
Internet, usually in inexpensive user hosts along diverse
administrative domains. Each peer communicates with
other peers directly or indirectly to share local resources
(e.g., memory, files, processing power), forming robust
distributed systems over the Internet. The DISTRIBUTED
communication model clearly differs from the client-server
model. In client-server, servers store resources accessed by
clients, by contrast, in DISTRIBUTED each peer is itself a
repository that shares its resources with other peers, thus
forming environments where resources are distributed
among peers acting as both servers and clients.
Client-server addressing is based on the Internet Domain
Name System (DNS), while DISTRIBUTED networks
employ addressing schemes independent from DNS to
index peers, users, and resources. Client-server messages
are routed at the network layer by Internet routers, but
DISTRIBUTED messages are routed at the application
layer by routing peers.
Having routing and addressing built on top of Internet
protocols, DISTRIBUTED networks are quite dynamic
environments where new peers can join the network when
they are started in user hosts, and leave when users
shutdown the peers. This dynamic nature makes
DISTRIBUTED networks both unpredictable and scalable:
u
p
e
a
i
D
w
a
t
a
s
F
A
l
c
A
c

unpredictable
peers that ho
expansion of
addition of ne
During th
improvements
Due to its po
without any c
as the art of n
the research o
always taken
similarities ar
Figure 1).
A. Dynamic N
Nodes in D
leave the netw
change freque
And the mob
can make net
[8].
because resou
ost them are
f the networ
ew peers.
he last cou
s are made in t
ossibility to c
entral manage
networking wi
of DISTRIBU
as two sepa
re seen betwe
Network Topolo
DISTRIBUTE
work, which c
ently with tim
bility characte
twork topolog
(a). A
urces may be
e down; scal
rk capacity r
uple of ye
the research o
create and or
ement, MANE
ithout a netwo
UTED network
arated areas. B
een these two
ogy
D networks r
causes the net
me. It is the sam
er of mobile n
gy changes mu
Architecture of
unavailable i
lable because
requires only
ears tremen
f ad hoc netw
rganize a netw
ET is character
ork [7]. Curre
k and MANET
But the follow
technologies
randomly join
twork topolog
me with MAN
nodes in MAN
uch more freq
f MANET
if the
e the
y the
ndous
orks.
work
rized
ently,
T are
wing
(see
n and
gy to
NET.
ANET
quent

co
co
wi
co
th
de
as
in
to
bo
ro
b.M
bo
ca
ca
wi

DI
tha
DI
ph

(b). Arch
Fig. 1 Sim
A. Multi-h
Nodes in b
onnect with ea
onnections in D
ith physicall
onnections in
e radio transm

B. Decent
Both DISTR
ecentralized st
what is seen
the network
each other. T
ottleneck cause
bustness of th
Multi-Identity
Nodes in DI
oth as a client
an route and t
an also provid
ith each other
II. DE
a. Architect
The architec
ISTRIBUTED
at DAODV
ISTRIBUTED
hysical networ
hitecture of DI
milarities of MANE
hop Connectio
both MANET
ach other via m
DISTRIBUTE
ly unlimited
MANET are
mission range.
tralized contro
RIBUTED net
tructure. Ther
in the Client/
are all peers a
The network w
ed at the centr
he network.
y of Nodes
ISTRIBUTED
and a server,
transfer messa
de resource do
directly [9].
ESIGN OF DA
ture of DAOD
ture of the D
D computing
V protocol
D overlay netw
rk of the MAN
ISTRIBUTED

ET and DISTRIB
on
T and distrib
multi-hop rout
ED networks a
range, wh
via wireless l
ol
twork and M
re is no centra
/Server networ
and can comm
will not suffer t
ral server, whi
D network an
as well as a ro
ages independ
ownloading an
AODV PROT
DV Model
DAODV mod
technology. F
is integrat
work which i
NET.
D Network
BUTED networks
buted networ
ting. Hop-to-h
are via TCP lin
hile hop-to-h
links, limited
MANET have
al administrati
rk model. Nod
municate direc
the paralysis a
ich improves t
nd MANET
outer. Each no
dently. And th
nd communic
TOCOL
del is based
Figure 2 sho
ted with t
s built upon t

rks
hop
nks
hop
by
e a
ion
des
ctly
and
the
act
ode
hey
ate
on
ows
the
the

Fig. 2 Architecture of DAODV Model

a.Basic Design
DAODV uses the node naming mechanism used in Chord.
It adopts a consistent hashing algorithm [10], [11]. Each
node is assigned an m-bit node identifier (NID), which is
obtained by hashing the node's IP address. Thus we are
able to specify the location of the node in the hash ring.
When a node joins the network for the first time, a new
NID ranging from 0 to (2m-1) is automatically assigned
based on the adopted hashing algorithm. To simplify the
routing algorithm, we have adopted the discontinuous
naming mechanism used in Chord. The node state is
maintained in the DAODV routing table (PRT) which is
similar to the routing table of AODV. The difference is
that, DAODV uses the NID for the destination node and
the next hop, whereas the later uses the nodes IP address.
A. Route discovery Algorithm
The DAODV route discovery algorithm discovers the
route from the source node to the destination node
reactively. The route discovery is triggered by the mobile
nodes under any of the following, and an update is required
in the PRT table.

i. A node initiates the route discovery procedure.
ii. When a node receives the route query message
(RREQ), it queries the PRT, if the route is null or out dated,
the route discovery method is invoked.
iii. When a route issues a route error message (RERR) to
the node which initiated the data transmission. The
initiating node starts the route query method.
A description of the route discovery algorithm is given
below:
1) Node M checks for RREQ in cache. If the source
identifier and the request identifier is the same, the request
is discarded.
2) M checks in the Route Record whether RREQ
message contains NID(M). If found, the message is
discarded.
3) M checks whether the destination node is present in
the PRT entries. If found go to step 5, otherwise to step 6.
4) M checks whether the destination node is M itself. If
true, go to step 5, otherwise to step 6.
5) M checks whether the route is bi-directional. If it is,
the route reply message (RREP) is send to the initiating
node S directly. Otherwise, the route discovery method is
invoked.
6) M adds itself to the route record and broadcasts
RREQ message to the neighbors.
B. PRT Querying Algorithm
The PRT querying algorithm queries the PRT and picks
up the Next Hop Ring Interval (NRI) which is the closest
to D. Any outdated entries are updated. The algorithm is
depicted as follows:
1) When a node M initiates or receives a RREQ
message, it compares the destination node identifier of the
RREQ with NID(M). If they are matched, M will return
RREP to querying initiating node S.
2) M will search into the PRT and compares the
destination node identifier (D) with the PRT entries to find
the NRI to which D belongs for message transfer.
3) M triggers the route discovery for the destination
node and waits for the response. If the response if received
the PRT is updated, otherwise a failure message is send.
c.DAODV Route Update and Maintenance
Route maintenance is required due to the nodes
mobility. In MANETs, a nodes position changes with time,
this leads to dynamic network topology. A destination
node may become out of reach due to normal or abnormal
departure or has moved out of the transmission range.
When this occurs, the route updating algorithm is triggered
reactively. When there is a failure in the data transmission,
the node which initiated the packet transfer will issue an
overdue message to the related nodes. Thus the PRT
entries are outdated and the route update is triggered. If the
route discovery process ends in a failure, it means that the
destination node has moved out of range. Therefore the
destination node is marked as invalid.
Due to the high dynamic character, mobile nodes may
join and leave the network at any time. To save the route
maintenance cost, when a node joins the network, DAODV
will not initialize its PRT. On the contrary, the PRT is
constructed only on demand.
DAODV nodes will only overhear the source route
related to the next hop node in its local PRT. By doing this,
DAODV inherits all the advantages of AODV at the same
time limits. By the use of Phello messages in DAODV,
mobile nodes broadcast phello messages reactively.
Therefore the information about the neighbor nodes are
collected only when there is a need for route discovery. If
the destination node is one of the neighbors, the route
discovery process stops and RREP is returned to the source.
Consequently, the overload of nodes in DAODV is
reduced

III. SIMULATION RESULTS
File
Discovery
Service
Information
Administration
Service
File
Transfer
Service
Communication
Security
Service
DISTRIBUTED Computing Based MANET Routing
Algorithm
DISTRIBUTED Overlay Network (Node Naming and
Indexing)
MANET Physical Network

s
s
m
p
t
s
n
a
a
r
v
o
o
p
p

e
a
i
o
P

f
T
D
D
d
s
t
i
l
Simulation
space. The
simulation sc
movement du
pause time (P
the model. At
static for a PT
node and mov
arrives, the no
a node as th
repeated till t
values of PT.
on DAODVs
of nodes to tr
packets are 12
packets per se
A. Pack
Packet del
estimating a r
adapts to the c
it provides a r
of Packet Deli
Pause Time.
Fig. 3
The probab
failure decrea
The success
DAODV is m
DAODV has
delivery perf
structured DIS
the caching m
its problem in
low packet de
B. Avera
is done with
nodes are
cope. And e
uring the sim
PT) is used to
t the beginnin
T. Then it will
ves toward it
ode will keep
he next dest
the simulation
To simulate
s performance
ansmit CBR p
28 bytes packe
econd
ket Deliver Rat
liver ratio i
routing protoc
change of netw
route discover
iver Ratio betw
Comparison
DA
bility of link f
ases when the
rate and stab
much better th
improved th
formance by
STRIBUTED
mechanism ado
n providing ou
eliver ratio.
age end to En
fifty nodes in
randomly di
ach node pe
mulation period
o define the m
ng of simulat
l randomly ch
t with a certai
static for a PT
tination. This
n ends. We ch
the network s
e, we randoml
packets to eac
ets with a tran
tio
is an impor
ols performa
work topology
ry. Fig 3 depic
ween DAODV
of PDR betwe
AODV
failure betwee
e mobility of
bility of packe
han AODV. T
he node disco
y taking ad
routing mech
opted in AOD
ut-of-time netw
nd Delay (AED
n a 1000m*10
istributed in
erforms a Br
d. The traditi
mobile charact
ion, a node k
hoose a destina
in speed. Wh
T and then ch
process wil
hoose for diffe
services influ
ly choose 20 p
ch other. The C
nsmission rate
rtant criterion
ance in how w
y and how effi
cts the compar
V and AODV
een AODV an
en nodes and r
f nodes decre
et transmissio
The reason is
overy and pa
dvantages of
hanisms. How
DV is the orig
work status an
D)
000m
the
rown
ional
er of
keeps
ation
hen it
hoose
ll be
ferent
uence
pairs
CBR
e of 4
n in
well it
cient
rison
with
nd
route
eases.
on in
that
acket
the
ever,
in of
nd its
fa
co
DA

me
ro
DA
bu
M
ac
c.N
de
th
ma
we

Fig

dir
Th

Average end
st route dis
omparison of
AODV and A
Fig. 4 Co
AODV emp
echanism and
ute when the
AODV impro
uilding up a
MANETs phy
ccelerate the ro
Node Storage
Figure
ecreases the ro
e same time
aintenance ov
ell.
g. 5 Compari
The cached r
rect proportio
hat is, for a M
d-to-end delay
covery is d
f Average E
AODV with Pau
omparison of AED
loys broadca
d consumes l
e current path
oves the rou
a DISTRIBU
ysical topolog
oute discovery
Overhead (NS
5 depicts
oute storage o
e it decrease
verhead relate
ison of NSO betw
routes for a m
on to the num
MANET with N
y estimates ho
done. Figure
End-to-End
use Time.
D between AODV

asting based
longer time t
h has failed. O
ute discovery
UTED overla
gy and intro
y process.
SO)
that DAOD
verhead for ea
es the route
ed to these c
ween AODV and D
mobile node i
mber of nodes
N nodes, the c
ow efficient a
4 depicts t
Delay betwe
V and DAODV
route discove
to re-discover
On the contra
y efficiently
ay network
ducing PRT
DV effectiv
ach node, and
discovery a
cached routes
DAODV
in AODV are
in the netwo
cached routes
and
the
een

ery
r a
ary,
by
on
to
ely
d at
and
as

in
ork.
for
e
s
r
w
D
f
d
c
b

t
r
c
a
D
s
a

D
l
o
D
D
h
s
o
s
r
i
each node is
structured D
relationship o
with the num
DAODV effe
for each node
discovery an
cached routes
C. Rout
Figure 6 d
between DAO
Fig. 6 Com
The numbe
the number o
route mainten
cached routes
and decrease
DAODV impr
structured DI
and limiting th

DAODV
DISTRIBUTE
logical names
operating in
DAODV ha
DISTRIBUTE
highly dynam
scalability. Be
optimizing me
structured DIS
routing efficie

[1] Mengkun
improving search
s O(N). On t
DISTRIBUTED
of cached rou
mber of node
ectively decrea
e, and at the s
nd maintenanc
as well.
ing Overhead
epicts the com
ODV and AOD
mparison of RO
er of route pa
of cached rout
nance and upd
s consume mu
e the perfor
roves the rout
ISTRIBUTED
he number of
IV.
seamlessly
ED overlay ro
pace with tho
a physical
as inherited
ED networkin
mic network
esides these, D
echanisms to
STRIBUTED
ency.
REFE
n Yang and Zon
h efficiency in
the contrary,
D routing
utes for each
es is O(logN)
ases the route
same time it d
ce overhead
d (RO)
mparison of R
DV with Pause
O between AO
ackets in AOD
tes for each n
date processes
uch of the ne
rmance of A
ting performan
D overlay net
cached routes
CONCLUSIO
integrates th
outing protoco
se of MANET
namespace.
d all the
ng technolog
ks and main
DAODV intro
solve the deto
overlay netw
ERENCES

ngminm Fei, A
unstructured pee
by implemen
mechanism,
node in DAO
). By doing
e storage over
decreases the r
related to t
Routing Over
e Time.
ODV and DAO
DV is huge du
node is huge.
s caused by t
etwork bandw
AODV. Whe
nce by introdu
twork techno
.
ON
he functions
ols operating
T routing proto
By this me
advantages
gy in sustai
ntaining netw
oduces a serie
ouring problem
work and impr
A novel approa
er-to-peer netw
nting
the
ODV
this,
rhead
route
these
rhead

ODV
ue to
The
these
width
ereas
ucing
ology
s of
in a
ocols
eans,
of
ining
work
es of
m of
roves
ach to
works,
Jou
No
and
Th
pp
net
20
Ka
pee
on
bet
net
No
po
Te
to
Sel
13
"Is
fut
Co
for
Co
37
WC
Co
im
NO
Sym
urnal of Paralle
ovember, 2009
[2] J. Xu, "On
d network diamet
he Conference on
. 2177-2187.
[3] S. Sen and J
tworks," IEEE/AC
04, pp. 219-232.
[4] I. Stoica, R
aashoek, F. Dabe
er lookup protoco
Networking, Vol
[5] J. Xu, A. K
tween routing t
tworks," IEEE Jo
o. 1, Jan 2004, pp
[6] S. Jain, R. M
tential of DHT-b
chnologies and Sy
[7] S. Jiang, Y. L
variable topologi
lected Areas in C
56.
[8] D. Cavalcan
ssues in integrat
turistic heterog
ommunications, V
[9] C. H. Foh a
rmula for one-dim
onference on Com
42.
[10] A. Duran
CNC 2004 - I
onference, Vol. 5,
[11] J. Mischke
mplementation for
OMS 2004 - I
mposium, Vol. 9,

el and Distribute
the fundamental
ter in peer-to-peer
Computer Comm
J. Wang, "Analy
CM Transactions

R. Morris, D. L
ek, and H. Balak
ol for internet ap
l. 11, No. 1, Feb 2
Kumar, and X.
table size and
ournal on Selected
. 151-163.
Mahajan, and D. W
based overlays,"
Systems(USITS 20
Liu, Y. Jiang, and
es for routing sch
Communications,
nti, D. Agrawal, C
ting cellular netw
geneous wirele
Vol. 12, No. 3, Jun
and B. S. Lee, "A
mensional MANE
mmunications, V
and C.-C. Shen,
IEEE Wireless
No. 1, March 20
e and B. Stiller,
a highly scalabl
IEEE/IFIP Netw
No. 1, Apr 2004,
d Computing, V
tradeoffs betwee
r networks," IEEE
munications, Vol. 2
yzing peer-to-peer
s on Networking,
Liben-Nowell, D.
krishnan, "Chord:
pplications," IEEE
2003, pp. 17-32.
Yu, "On the fu
network diame
d Areas in Comm
Wetherall, "A study
4thUSENIX Sym
003), March 2003.
d Q. Yin, "Provisi
hemes in MANET
Vol. 22, No. 7,
C. Cordeiro, B. X
works, WLANs,
ess network,"
ne 2005.
A closed form n
ETs," ICC 2004 -
ol. 27, No. 1, Ju
"Mobile ad hoc
Communication
04, pp. 114-119.
"Efficient protoc
le peer-to-peer se
work Operations
, pp. 411-424.
Volume 69 Issue
en routing table s
E INFOCOM, 200
22, No. 1, Mar 20
r traffic across la
Vol. 12, No. 2, A
. R. Karger, M.
A scalable peer
E/ACM Transacti
undamental tradeo
eter in peer-to-p
munications, Vol.
y of the performa
mposium on Inter
.
oning of adaptabi
Ts," IEEE Journal
Sep 2004, pp. 13
Xie, and A. Kum
, and MANETs:
IEEE Wire
network connectiv
- IEEE Internatio
une 2004, pp. 37
c P2P file sharin
s and Network
col specification
earch infrastructu
and Managem
11,
size
03 -
003,
arge
Apr
. F.
r-to-
ions
offs
peer
22,
ance
rnet
ility
l on
347-
mar,
: A
less
vity
onal
739-
ng,"
king
and
ure,"
ment

ISBN: 978-81-920575-7-6 :: doi: 10. 73900/ISBN_0768
Modeling Impact of Varying Data Rates for Various Spread Spectrum
Techniques on Load in WLAN Ad-Hoc Mode

Rajan Vohra
Lecturer, Department of
Electronics Technology
Guru Nanak Dev University
Amritsar, Punjab, India

Abstract:-.Wireless Local Area Network (WLAN) uses
high frequency radio waves rather than wires to
communicate and transmit data. A wireless local area
network links two or more devices and gives users the
mobility to move around within a local coverage area and
still be connected to the network. Wireless data connections
have inherently high bit error rates, low bandwidth and long
delays. In this paper OPNET Modeler 9.1 network simulator
has been used to do the comparative modeling and
simulation of the load for different data rates (1, 2, 5.5,11
mbps) and various techniques (FH, DS, IF) and concluded
FH technique is better in 1 Mbps scenario and best in 11
Mbps scenario than DS and IF techniques.[7.8]

Keywords: WLAN, Load, OPNET, FH, DS, IF.

1. INTRODUCTION

Wireless LANs are becoming popular the world over
due to ease of installation, and the ability to offer wireless
access to their users often for free [1]. The IEEE (Institute
of Electrical and Electronic Engineers) 802.11
specifications used the 2.4 GHz frequency band and
initially supported a maximum data rate of 1 to 2 Mbps .
The 802.11b specification increased the performance to
11 Mbps in the 2.4 GHz range. It encompasses the
physical (PHY) and the media access control (MAC)
layers of the ISO seven-layer network model [2]. Wireless
local area networks (WLANs) based on the IEEE 802.11
standard are one of the fastest growing Wireless access
technologies in the world today. These provide an
effective means of achieving wireless data connectivity in
homes, public places and office and also enable people on
the move to communicate with anyone, anywhere at
anytime with a range of multimedia [1-3]. The WLAN
media is error prone and the bit error rate (BER) is very
high compared to the BER of wired networks. In addition,
Carrier Sensing is difficult in wireless networks because a
station is incapable of listening to its own transmissions in
order to detect a collision.
There are two operation modes defined in IEEE 802.11
(a) Infrastructure Mode

In the infrastructure mode above, the wireless network
consists of at least one AP (access point) connected to the
wired infrastructure. All the wireless stations are connected to
the AP. An AP controls encryption on the network and also can
route the wireless traffic to a wired network (same as a router).
We can assume an AP as similar to base station used in
cellular networks.
(b) Adhoc Mode
AP is not required in Ad-Hoc mode, the wireless station is
connected to each other directly without using an AP or any
other connection. The topology is useful to set up a wireless
networks quickly. Ad-Hoc mode is also called peer to peer
mode or (IBSS) Independent Basic Service Set.

This paper deals with the load optimization techniques that
can be effectively used for a WLAN based on the network
simulator, OPNET Modeler 9.1[3].

II. MODEL OUTLINE
In our model, we used two workstations between which,
we transfer information/data through wireless mode with the
help of different techniques and data rates that are available
in OPNET MODELER 9.1. We have made 4 scenarios and
in each scenario 4 different data rates for 3 different
techniques (Frequency Hopping, Direct Sequence and
Infrared) have been employed for our simulation and through
these scenarios we compare load parameter and evaluate the
performance at different data rates for various techniques.
Table I shows the WLAN traffic generation parameters
for both the scenario. Operation Mode used in the scenario is
serial (ordered), start time (seconds) are uniform (100,110),
repetition pattern in the scenarios is serial, number of
repetitions is constant and rest of the parameters used are
listed in table below:
Table II shows the parameters, which are used in
different scenarios for simulation. Parameters indicate that
we use PHY characteristics as Direct sequence, Frequency
Hopping and Infrared, buffer size (bits) 1024 k for all
scenarios. Bandwidth used for the WLAN in scenario 1 and
scenario 2 is 1 Mbps and 2 Mbps respectively and for
scenario 3 and 4 it is 5.5 and 11 Mbps respectively.
t
t
O
h
p
s
p
M

I

In Our mo
transfer infor
techniques usi
OPNET MOD
having differe
performance i
scenarios have
100*100m WL
performance u
Mbps respectiv

WLAN scena

TABLE I. W

A
D
Re
Ope
Start T
Inter-r
(se
Numbe
II. SIMULAT
odel, we used
rmation/data
ing variable da
DELER 9.1. H
ent data rates
is evaluated
e been tested
LAN model fam
under different
vely.
ario at different te
Wireless LAN tra

Attribute

Duration

epeatability

eration Mode

Time (seconds)

repetition Time
conds)

r of Repetitions
TION EVALU
two workstati
wirelessly f
ata rates which
Here we have
s for various
under these
d in campus e
mily. In this pa
t data rates of
echniques (FH, D
affic generation pa

Val

End of sim

Once a
Time

Serial (o

uniform (

constan

constan
UATION
ions in which,
for the vari
h are available
made 4 scena
techniques.
scenarios. Th
environment w
aper, we verify
1, 2, 5.5, and
DS, IF) and data ra
arameters
lue
mulation
at Start
e
ordered)
(100,110)
nt (300)
nt (30)
we
ious
e in
arios
The
hese
with
y the
d 11

ates.

rate
obs
nod
tech
Mb
Dir
100
data
aro
the
IR
(bit
and
term
Mb
in
sce
ban
Ch
Th
R
R
Repetitio

We have obs
es for various
served that at
des is comp
hniques(DS,FH
bps scenario F
rect Sequence
0,000(bits/sec)
Figure
a rate, there is
und 350,000 (
same time FH
technique,
ts/sec) for FH.
When we com
d FH techniqu
ms of Bits/sec
bps scenario w
1 Mbps scen
nario. Reason
ndwidth gives
nodes using in

Parameter
(s)
Address
WLAN
bandwidth
(bps)
PHY
haracteristics
Slot time
(s)
Packet
Reception-
Power
hreshold (W)
Short
Retry Limit
(Attempts)
Long
Retry Limit
(Attempts)
Buffer
size(bits)
Max
Receive
Lifetime
(secs)

on Pattern
served the res
techniques. F
2 and 5.5 Mb
paratively sa
H and IF) bu
Frequency Hop
e and Infrared
) for the period
5 shows that f
s drastic incre
(bits/sec) for t
H and DS tech
110,000 (bit
.
mpare scenario
ues are better t
c in case of 11
which shows th
nario is less a
n for this is
s less load on t
n the scenario
Scen
ario 1 ari
Auto Assign
1
Mbps Mb
DS,F
H,IF H,
1024000
0.5

Serial
sults of Load
From figure 3
bps Load (bit
ame for all
ut figure 1 su
pping techniq
d Techniques
d 10-60 minut
for IF techniqu
ease in Load o
the period 0-60
hniques give b
ts/sec) for D
o for 1Mbps a
than IR. The v
Mbps scenari
hat the load o
as compared t
working of
the
o.
Scen
io 2
Sce
ario 3
ned Au
2
bps
5.5
Mbps
DS,F
,IF
DS,
H,IF
5.0 E-05
7.33 E-14
7
4
0
at different d
3 and 4 we ha
ts/sec) at WLA
l the differ
uggests that at
que is better th
by 50,000 a
tes.
ue with 11 Mb
on WLAN nod
0 minutes, but
etter results th
DS and 50,0
and 11 Mbps, D
value of Load
io is more than
on WLAN nod
to the 11 Mb
scenario at l
n Scen
ario 4
uto Assigned
5 11
Mbps
F DS,F
H,IF
1024000
0.5
data
ave
AN
ent
t 1
han
and
bps
des
t at
han
000
DS
d in
n 1
des
bps
ess

Fig 1: Load a

TABLE II: S

Fig 2: Load a

at 1 Mbps in diffe
Simulation scenari

at 2 Mbps in diffe
erent techniques
io parameters
erent techniques

for
resu
and
tech
bes
wan
req
resu
tech

Fig 3: Load at 5

Fig 4: Load at 1

This paper in
different data
ult of the Loa
d 4 there is
hniques. FH t
st in 11 Mbps
nt to work
quired less Lo
ults we will
hniques.

5.5 Mbps in differ
1 Mbps in differe
IV. CON
nvestigates th
a rates and tec
ad is comparat
s drastic cha
technique is b
scenario than
with high b
oad on WLA
opt FH tech
rent Techniques
ent techniques
NCLUSION
e Load on th
chniques. For
tively same b
ange in Loa
better in 1 Mb
n DS and IF te
bandwidth en
AN nodes the
hnique over

he WLAN nod
scenario 2 and
but in Scenario
ad for differ
bps scenario a
echniques. If
nvironment a
en by observ
the DS and
des
d 3
o 1
ent
and
we
and
ved
IF
REFERENCES

[1] Walid Hneiti and Naim Ajlouni, Performance Enhancement of
Wireless Local Area Networks,Proceedings of IEEE ICTTA06, 2nd
International Conference on Information CommunicationTechnologies:
from Theory to Applications, Damascus, Syria, vol. 2, pp. 2400- 2404,
April 2006.

[2] L.Bononi, M. Conti, and E. Gregori,Runtime Optimization of
IEEE 802.11 Wireless LANs Performance, IEEE Transactions on
Parallel And Distributed Systems, vol. 15, pp. 159 172.January 2004.

[3] Mohammad Hussian Ali and Manal Kadhim Odah,Simulation
Study 0f 802.11b DCF Using OPNET Simulator, Eng. & Tech. Journal,
Vol. 27,No6 , pp. 1108-1117. 2009.

[4] Ms. Amardeep Kaur1, Dr. Sandip Vijay2, Dr.
S.C.Gupta3 Performance Analysis and Enhancement of IEEE
802.11 Wireless Local Area Networks Vol. 9 Issue 5
(Ver2.0),pp.130.January 2010.

[5] Q. Ni, I. Aad, C. Barakat, and T. Turletti, Modeling and
Analysis of Slow CW Decrease for IEEE 802.11 WLAN. 14th IEEE
Proceedings on Personal, Indoor and Mobile Radio
Communications,IEEE PIMRC 2003,vol. 2, pp. 1717-1721. Beijing,
September2003.

[6] Elena Lopez-Aguilera, Martin Heusse, Franck
Rousseau, Andrezej Duda and Jordi Casademont, Evaluating
Wireless LAN Acess Mothods in Presence of Transmission
Errors.INFOCOM 2006.Barcelona, pp. 1-6, Spain,April 23-29, 2006.

[7] Rajan Vohra; R.S.Sawhney; Manju Sharma:
Performance Enhancement of WLAN by Buffer Size Optimization using
OPNET National conference on Advance Computing and
Communication Technology (ACCT-10) organized by Vaish College of
Engineering, Rohtak, 25-26 June,2010 pp 495-499

[8] Rajan Vohra; R.S.Sawhney; Shikha: WLAN Metrices
enhancement using Buffer Size Optimization International Journal of
P2P Network Trends and Technology- Volume 1 Issue 2 -2011 Page 22-
25 (ISSN: 2249-2615)


V
A
Proc. of th
Volume 1. Cop
ISBN: 978-8
Vertic

Abstract - Se
networks is on
wireless netwo
this seamless m
Vertical Hand
this continuity
schemes are a
handover dec
which provid
seamless mobi
Keywords - Ver
DVHD,C-DVH

Seam
any system w
better QoS du
another. It
telecommunic
protocols whi
A messag
the serving
indication th
communicatin
All additiona
wireless netw
network to t
received the
the neighbor
radio signals
neighboring w
minimum t
communicatio
network to th
after the neig
wireless chan
handover requ

II. VERT
In han
Handover (H
(VHO) as sh
process when
network with
Wi-Fi to Wi-
access point i
connectivity
normally to

e Intl. Conf. o
pyright 201
1-920575-7-6
ACM #: dber
cal Hand
ervice continu
ne of the chal
orks. Handove
mobility while
dover has bee
y. In this pape
analyzed and th
cision scheme
e less process
ility over heter
rtical Handove
HD.
I. INT
mless handove
with mobility
uring process
is a met
cations system
ich are differe
ge is sent from
wireless ne
hat the mo
ng with the n
al information
work is also s
the mobile st
mobile statio
ing wireless
s between t
wireless netwo
threshold th
on transactio
he neighboring
ghboring wire
nnel for the m
uest.
TICAL HAND
HAN
ndover there a
HHO) and an
hown in Fig.
n the mobile
the same tech
Fi. Vertical h
in a network
it uses to a
on Computer A
12 Techno For
6 :: doi: 10. 73
r.imera.10. 739
over Deci
Assi
S
uity in hetero
llenges in the
er technique i
switching in b
en normally u
r the vertical
he clustered D
has been eval
sing delay an
rogeneous netw
r decision, clus
TRODUCTIO
er is a fundam
y. It is the att
migration fro
thod used
m to transfer
ent from each
m a mobile sta
etwork, whi
bile station
neighboring w
n relating to
sent from the
tation. Using
on receives ra
network. Th
the mobile
ork is measure
hen a han
n from the
g wireless net
less network
mobile station
DOVER AND
NDOVER
are two types,
nother is Ve
1.Horizondal
e user switch
hnology and s
handover refer
changing the
access a supp

Applications
rum Group, In
907/ISBN_07
907
ision Sche
Ar
istant Professo
SCMS School
Karukutty
ogeneous wire
fourth genera
is used to pro
between netwo
sed for provid
handover deci
Distributed vert
luated as the
d hence prov
works.
stering, DVHD,
ON
mental concern
tempt to prov
om one domai
in a wire
r messages us
other.
ation to a use
ch includes
is capable
wireless netw
the neighbo
serving wire
this informa
adio signals fr
he quality of
station and
ed. If this mee
ndover of
serving wire
twork is execu
has establishe
n pursuant to
HORIZONTA
one is Horizo
ertical Hando
handover is
hing between
ame network
rs to a node o
mode or typ
porting netw

ndia.
768
emes in H
.
rchana Sreej
or, Departmen
l of Engineeri
y, Cochin, Ke
eless
ation
ovide
orks.
ding
ision
tical
one
vides
T-
rn in
vide
in to
eless
sing
er to
an
of
work.
ring
eless
ation
from
the
the
ets a
the
eless
uted
ed a
the
AL
ontal
over
the
the
like
or an
e of
work,
wir
Wi
acc
pro
tech
Thu
bas
we
use
ano
sele
term
und
freq
oth
for
swi
Han
dec
pha
crit
sec
to
net
Ser
pha
the
wit
wh
con
Heterogen
ith
nt of Compute
ing and Techn
erala., India

Fig.1 Ho

support node
reless device
ireless LAN
cess. It is c
ovides better
hnology whic
us the Laptop
sed on the av
say that ver
er to automat
other to maint

When a
ected for Com
minal needs t
der both situ
quency bands
her important c
mobility is
itching. Hand
ndover Initi
cision, Handov
ase the hand
teria value lik
cond phase ca
decide which
twork and ex
rvice (QOS)
ase which is
e neighbor ne
th this ,QOS
hich network
nnection. The
neous Wir
er Science
nology
orizontal versu
e mobility. Fo
like laptop m
and a cellula
clear that W
speed rates
ch provides ub
user can swit
ailability and
tical handove
tically switch
ain communic
Vertical Han
mmunication i
to have a dua
uations such
s and various
challenge whi
s the automa
dover mechan
ation, Syste
ver execution
dover process
ke signal stren
alled the Syste
h mobile use
xchanges info
provided by
the Handove
twork QOS a
decision mak
the mobile
e last phase i
reless Netw
us Vertical Ha
or example, a
might be able
lar technology
Wireless LAN
compared to
biquitous netw
itch using thes
d necessity. In
er will provid
h from one
cation ultimat
ndover mechan
it is necessary
al mode card
h as WLAN
s modulation
ile using a ver
mation aspects
nisms have a
em Discover
n. In the Hand
s was modif
ngth, link qua
em discovery
er discovers
ormation abo
these networ
er decision ph
and the mobi
aker makes th
e user has
is the Hando
works
andover
well equipped
to use both a
y for interne
N connectivity
o the cellula
work coverage
se two device
n simple term
de or help the
technology to
tely.
nism has been
y that a mobile
that can work
N and UMTS
schemes. The
rtical handove
s in network
a four phases
ry, Handove
dover Initiation
fied by some
lity etc. In the
phase is used
its neighbor
ut Quality o
rks. The third
hase compare
ile users QOS
he decision to
to direct the
over execution

d
a
et
y
ar
e.
s
ms
e
o
n
e
k
S
e
er
k
s.
er
n
e
e
d
s
of
d
s
S
o
e
n
ISBN: 978-81-920575-7-6 :: doi: 10. 73914/ISBN_0768
An Algorithm for Dynamic Search in Unstructured Peer-to-Peer
Overlay Network
Ashish S Poste
Computer Engineering Department
PICT
Pune, India

Abstract Peer-to Peer (P2P) computing network has
emerged as a model for resource sharing over the internet.
Designing efficient search algorithm is a key challenge in
unstructured peer to peer network (P2P).Efficient resource
discovery is the first step towards the realization of
unstructured P2P network. In this work, we present a detailed
overview of recent and existing search methods for
unstructured Peer-to-Peer networks. Most unstructured P2P
networks utilize flooding based resource discovery which
assumes no knowledge about the network topology, and thus
can lead to number of redundant message generation. Random
Walk (RW) search generates a fixed amount of query messages
at each hop but would take longer time to search. Researchers
have presented number of alternative search algorithms which
take advantage of the context under which Flooding and RW
performs well. We propose a searching algorithm, which will
first construct a minimum spanning tree like overlay network
over the physical network. The search algorithm will resemble
as flooding for some small number of hops on the original
physical network, after the initial flooding for certain hops it
will behave like a random walk on the constructed overlay
network and forward the query to limited set of neighbor
depending on the knowledge gained from previous search
attempts.Problem occurring due to the peers getting
disconnected is also handled effectively.
Keywords-component; Peer-to-Peer network, Network
Topology,
Searching Technique.
I. INTRODUCTION
In the past few years, emergence of P2P applications such
as Napster, Gnutella, and KaZaA have shown their good
support for content lookup and sharing, so their popularity and
number have greatly exploded. A P2P system usually involves
thousands or millions of nodes and resources in the network.
The P2P system forms a fully-distributed, cooperative network
design, where nodes do not require any supervision. A good
search scheme has to arrive at two goals: (1) high search
performance, which tries to deliver high quality service (e.g.,
large number of search results and low query response time) to
users, and (2) low cost, which reserves system resources to
sustain a large number of users. Early systems, such as Napster,
use a central server to store indices of participating peers. This
centralized design and practice give rise to the concerns of
performance bottleneck and single point of failure. Instead of
maintaining a huge index in a centralized system, a
decentralized system such as Gnutella distributes all searching
and locating loads across all the participating peers. Though the
decentralized approach addresses the overloading and
reliability issues, and is promising to build a highly scalable
P2P system, its success is heavily dependent on an efficient
searching strategy employed to locate the required resource.
Reaching out to a large scope of peers is a fundamental
procedure in an unstructured P2P network, because there are no
controls and no accurate information on network topologies
and locations of desired files. Thus, the system scalability is
directly affected by the efficiency of the broadcasting
mechanism.
Search algorithms in unstructured P2P networks can be
classified into two categories: breath first search (BFS) method,
and depth first search (DFS) method. These two types of search
algorithms tend to be inefficient, as they have the disadvantage
of generating too much load on the system [1], [2], or not
meeting users requirements of minimum search time and
performance [3].Flooding, which belongs to BFS-based
methods, is the default search algorithm for Gnutella network
[4], [5].In a blind search, nodes do not store any information
regarding file locations. In informed approaches, nodes locally
store metadata that assist in the search for the queried objects.
The shortcoming of most informed methods is the maintenance
cost of the indices after peers join/leave the network or update
their collections. In most cases, these events lead to generation
of update messages resulting in network traffic. Current P2P
search algorithms can be roughly divided in two categories:
The ones that exhibit good performance at high cost (either
during the search or the metadata updates), and those that are
bandwidth efficient but exhibit varying performance. The
important observation is that popular P2P networks display a
highly dynamic behavior, with most users connecting for small
periods of time and then leaving the system.

II. RELATED WORK
To address the searching problem in unstructured P2P
network many researcher and practitioner have proposed a lot
of methods. Today, the most popular P2P applications operate
on unstructured networks. In these networks, peers
joining/leaving a network is dynamic, the location of the
documents is not controlled by the system and no guarantees
for the success or the complexity of a search are offered.
_____________________________________________
INPUT:
_____________________________________________
N(u) is the 1 hop neighbor set of node u. It is defined
as:
N(u) ={v / (u, v) E }.
E->set of edges in graph.
n->number of nodes/peers.
_____________________________________________
Algorithm:
_____________________________________________
1. Each node u collects its 1 hop N(u) neighbor
information.
2. Each node u compute the MST(N(u)) of all nodes
N(u) including u itself.
3. For each edge (u,v) MST(N(u)) , node u tells
node v about this edge selected for MST.
4.Node u keeps an edge (u, v) if
(u,v) MST(N(u)) or (u,v) MST(N(v)) .
5.
n
i
i=1
MST(G) MST(N(u )).
Gives complete spanning

tree.

Flooding is an example of blind search, a message received by
a peer will be forwarded to all its neighbor except the node
from which the message arrived, but flooding would produce
redundant messages which will affect bandwidth utilization
adversely .Researcher have modified the flooding method to
achieve the advantages of flooding while reducing the
redundant messages greatly. Unlike pure flooding searching
techniques like Expanding ring [6] (or iterative deepening[7]),
Random periodical flooding[8] takes more dynamic factors into
consideration to reduce the flooding horizon while maintaining
search quality. Searches are
initiated with increasing TTLs until enough responses are
received. Modified BFS [7], Directed BFS [9] sends query to
neighbor satisfying certain specific heuristics. These are some
examples of modified flooding techniques.
To eliminate the use of flooding and improve search
efficiency some researcher have provided certain Interest-based
locality [4],[5],[14] states that if peer A has a particular piece of
content that peer B is interested in, then it is likely that A will
have other pieces of content that B is also interested in. Based
on this assumption, a peer selects others that have previously
provided results to its queries as peers with common interests
and creates
Shortcut links to these peers.
Super-peer based search strategy are provided in research
work like [21].A super-peer is a peer which acts as a
centralized server for some subset of its clients. Many such
Super-peers are connected to each other and forward the query
among them if they dont find a match within their own client
peers.
Some researchers have presented methods in which the peer
will learn about the location of resource in query through
knowledge updating from previous search results. A number of
search methods depending on the knowledge learned are:
Adaptive probabilistic search (APS) [10],[11], Biased RW
[12],Routing index[13],Local indices[14],Intelligent search[15]
Redundant message generation due to the loops in the
network will still be a problem in most of the above mentioned
search strategy. So construction of tree like overlay network
will be beneficial for reducing the redundant messages due to
the loops in the network [17], [18], [19].
Our scheme will construct an location aware minimum
spanning tree like overlay network and then apply the searching
algorithm on this overlay network. Tree like structure is created
by classifying the neighbor of every peer into a set of
FLOODING neighbor and NONFLOODING neighbor.
FLOODING neighbor are those to whom the peer can send the
message, while NONFLOODING neighbor are those to whom
message cannot be sent in MST. Unstructured P2P network are
so dynamic that a peer in the route of search may get
disconnected making the entire downstream node unreachable.
In this scenario our search will detect the node getting
disconnected and inform all 1 hop neighbor of the peer. The 1
hop neighbor of the disconnected peer will then get connected
to one of the NONFLOODING neighbor.

III. PROPOSED SEARCHING ALGORITHM
In this section, we provide the details of the proposed
search algorithm. Section A presents the operations of
Construction of Tree-like Overlay Network, section B presents
the operations of knowledge-based searching algorithm, and
section C present the operation on peer getting disconnected.
A. Construction of Tree-like Overlay
Network
Location-aware topology construction explores the
topology mismatch problem between the overlay network and
the underlying physical network by constructing an overlay
minimum spanning tree (MST) among each peer and the peers
within 1 hop distance from them. Proposed algorithm does not
need to know the information of the entire network. It just
needs to know the peers between 1 hop distance and the cost of
routing between them. We will construct the MST for every
peer and its neighbor eliminating the redundant loops between
the networks. Each peer will query its neighbor and construct
the
Neighbor cost table which will be required for MST
construction. Once the MST is constructed the query message
will be forwarded within the MST instead of forwarding the
message to its entire neighbor. During the construction of MST
a set of FLOODING and NONFLOODING neighbor is
maintained for each peer who will direct the peer to forward the
query message in MST. Construction of MST is done using a
general algorithm i.e. PRIM or KRUSKAL.
Consider a random graph G(V,E) with n nodes and e edges,
that has average degree d where V is set of nodes, E is set of
e
d
F
a
t
r
o
i
f
r
T
a
f
T
k
f
K
a
r
__________
INPUT:
__________
S->Queryin
r->queried r
TTL->Time
M->queryin
t->threshold
original net
__________
Algorithm:
1.SearchAlg
2. If (Pee
information
3. S returns
4. Elseif (h>
Stop forw
5. elseif (h<
h++
S forward
network.
6.elseif (h>
h++
Node ha
chosen to fo
edges. For a gi
distance from u

Algorithm

B.
Overlay N
Searching
Flooding and
advantage of c
the nodes with
redundant mes
of at least 95 p
initial hops b
forward the q
redundant mes
TTL value, on
amount of peer
forwarding the
TTL value can

Algorithm
Combining
knowledge-bas
further improv
Knowledge-ba
about whether
result in a hit/
____________
____________
ng node/peer.
resource.
e-to-live param
ng message.
d selected for
twork.
____________
gorithm(S, r, T
er S has
n about the loc
s a hit, Stop for
>TTL)
warding messa
<=t)
ds the query t
t)
aving height p
orward the me
iven node u, le
u.
for constructio
Knowledg
Network.
algorithm wi
Random walk
coverage i.e. q
hin a small num
ssages. Usually
percent. Cover
but later it d
query to one o
ssages initially
nce the messa
r, each individ
e message in th
n be done by pe
for Searching
g the search
sed search suc
ve the perform
ased search alg
forwarding qu
/success. Each
___________
___________
meter.
r forwarding t
___________
TTL, t)
the required
cation of resou
rwarding the m
age M.
to its entire ne
probability of
essage M to.
et N(u) be the s
on of MST:
ge-base Search
ill use the co
k perform wel
query message
mber of hops, b
y a TTL of 7 c
rage growth rat
degrades [19].R
of its neighbo
y flood the m
age are availa
dual peer will d
he MST. Selec
erforming expe
in Unstructure
h strategy d
ch as SPUN
mance of the s
gorithm will b
uery message to
peer will mai
____________
____________
the message M
____________
d information
urce )
message M.
eighbor on orig
f success wil
set of nodes at
hing Algorithm
ontext under
ll. Flooding h
e reaches to m
but generates a
can achieve cov
te is high for c
Random Walk
or reducing a
messages with
able to consid
do a random w
cting the appro
eriments.
ed P2P network
described with
[16], APS [10
searching algo
build the know
o particular pee
intain a probab
____
____
M in
____
n or
ginal
ll be
_
A
1
i
u
2
N
1 hop
m on
which
as the
most of
a lot of
verage
certain
k will
lot of
small
derable
walk by
opriate
k:
h the
0] can
orithm.
wledge
er will
bilistic
tabl
prob
pee
Var
fun
to
upd
P2P
mes
whi
the
and
So
Onl
mai
algo
pee
num
grow
netw
sear
dow
sch
leav
NO
inta
get
mai
nod
Nod
Sup
leav
this
NO
____________
Algorithm:
1. Every node
informing its
unavailability.
2. For each pe
Peer gets c
NONFLOODI
le for each of i
bability of hit
er is done whe
rious update f
ction, a linear
update the kn
date function ha

C.
in constructed
If we build a
P overlay netw
ssages. Howev
ich will be add
MST in a para
d its entire 1 ho
there is no ne
ly local inform
intaining the M
orithm will flo
er network and
mber of peer it
wth rate is not
Finally due t
work peers le
rch route get
wnstream node
eme the node
ving. The neig
ONFLOODING
act. Peers joini
connected to
intaining the M
Node S will co
de E, F in its F
de G will conta
ppose node S o
ves the network
s case node G w
ONFLOONG lis

Reconstruction
____________
e u getting di
s 1 hop
.
eer/node in nei
connected to
ING set of nei
its neighbor an
. Updating of
en a hit/miss
functions like
function and a
nowledge tabl
as been evalua
Addressing t
d MST.
spanning tree
work, we can
ver, the spanni
dressed in our s
allel fashion w
op distance nei
eed to obtain th
mation will b
MST is therefor
oods the query
d when the me
t does a rando
a problem in o
to the dynami
ave/join at an
disconnected
e unavailable u
e informs its e
hboring peers
G neighbor thus
ing the network
o any one of
MST.
ontain node G
FLOODING li
ain node S and
originates a que
k making node
will get connec
st i.e. S or F
F
n of MST after
____________
isconnected se
neighbor N(
ighbor set N(u
one of its
ighbor.
nd for each res
the knowledg
is returned fr
Flat update
hop count fun
le. The perfor
ated in APS [10
the Problem of
that connects a
n eliminate a
ing tree has so
scheme. As we
where we consi
ighbor for each
he whole netw
be sufficient. C
re not a problem
y message in
essage has reac
om walk on M
our scheme.
ic nature of u
ny time. If a n
d which will
until the tree
entire neighbo
will get conne
s maintaining t
k will not be a
the peer of t
in its NONFLO
ist as shown in
d F in its NON
ery message m
e G unavailable
cted to one of
Fig. 1

r a node gets di
____________
ends a messag
(u) about i
u)
neighbor fro
ource to infer
ge table for ev
rom its neighb
function, a s
nction can be us
rmance of th
0].
f Peers leaving
all the peers on
ll the redund
ome disadvanta
e are constructi
ider only the p
h and every pe
work informati
Constructing a
m. Initially as
original peer-
ched to suffici
MST. So covera
unstructured P
node gets on
make the ent
is rebuilt. In o
oring peer bef
ected to one of
he MST structu
a problem as th
the network s
OODING list a
n fig 1. Simila
NFLOODING l
meanwhile node
e for querying.
the node from
isconnected:
_
ge
its
om
the
ery
bor.
step
sed
ese
g
n a
dant
age
ing
peer
eer.
on.
and
the
-to-
ient
age
P2P
the
tire
our
fore
f its
ure
hey
still
and
arly
list.
e E
. In
m its

Let S be the system represents Peer-to-Peer network.
i i S={G, A, F, s,sh, N(u),FL,NFL}
Let G(V,E) be the graph representing the network
topology.
1, 2, 3..... m let E={ } where m=number of edges in the netw e e e e
1 2 , , 3..... n let V={ } where n=number of vetex/peers in the v v v v
1 2 r Let F={f ,f ,f } where r=number of resources.
Let A be the adjacency matrix representation of the
network/graph.
Aij=weight if(i,j) E
= 0 otherwise.
Where weight is the network delay between any two
peers i and j.
N(u) is the 1 hop neighbor set of node u. It is defined as:
N(u)={v / (u,v) E }
Set of neighbor of vertex u within 1 hops.
For each vertex i and neighbor within distance of 1
hops construct MST.
1 ij i F (i,A ) MST(T).
Where Ti is tree rooted at vertex i.
1 Function F takes the input as vertex i and adjancey matrix
rooted at vertex 'i'.
i = 1 2..... f FL {v v v } .
i FL is set of flooding vertex i.e. to which vertex 'i' is
connected
i 1 2 nf NFL = {V ,V ....V }.
i NFL is set of non flooding vertex i.e. to which vertex 'i' is
not directly connected
1 2 n MST(G)=MST(T )UMST(T ).........UMST(T )
Let s=peer querying for some resource in P2P network
f= queried resource.
M=query message.
3 F (s,f,M) Forwards the message to its neighbor.
4 F (s,f,M) Forwards the message to some of its neighbor
over MST.
Function 4 F forwards the query message to one of its

neighbor depending on the probability table that it will
return a hit/success.
h=number of hops.
sh=Selected Hop count.
3 ( Dynamicsearch=F S,F,m) if h<=sh
= 4( , , ) F S F m if h>sh
IV. MATHEMATICAL MODE FOR THE SYSTEM V. CONCLUSION
In this paper, we have proposed the reconstruction of
overlay topology during the event of peer getting disconnected
from the P2P network. Topology construction algorithm
presented uses only the local information available hence
improving the performance. The search algorithm is a
combination of Flooding and Random Walk strategy and uses
the context under which they perform well. Selecting the
threshold value for flooding network will greatly impact the
performance of the system. The search algorithm is further
combined with the knowledge learned from the previous search
attempts thus improving the probability of query returning a hit
by directing the query in right direction. The problem of node
getting disconnected and making all downstream nodes
unavailable in MST is also addressed in the scheme.
REFERENCES
[1] K. Sripanidkulchai, the Popularity of Gnutella Queries and Its
Implications on Scalability, white paper, Carnegie Mellon Univ.Feb.
2001.
[2] M. Jovanovic, F. Annexstein, and K. Berman, Scalability Issues in
Large Peer-to-Peer Networks: A Case Study of Gnutella, technical
report, Laboratory for Networks and Applied Graph Theory, Univ. of
Cincinnati, 2001.
[3] B. Yang and H. Garcia-Molina, Improving Search in Peer-to-Peer
Networks, Proc. 22nd Intl Conf Distributed Computing Systems
(ICDCS 02), pp. 514, July 2002.
[4] G. Kan, Gnutella, Peer-to-Peer Harnessing the Power of
Disruptive Technologies, OReilly, pp. 94122, 2001.
[5] RFC-Gnutella 0.6, http://rfc-gnutella.sourceforge.net/developer/
testing/index.html, 2008.
[6] V. Kalogeraki, D. Gunopulos, and D. ZeinalipourYazti, A Local
Search Mechanism for Peer-to-Peer Networks, Proc.ACM CIKM
Intl Conf. Information and Knowledge Management (CIKM 02),
pp. 300307,Nov. 2002.
Networks, Proc. 22nd Intl Conf.Distributed Computing Systems
(ICDCS 02), pp. 514,July 2002.
[8] Z. Zhuang, Y. Liu, L. Xiao, and L.M. Ni, Hybrid Periodical
Flooding in Unstructured Peer-to-Peer Networks, Proc. 32nd Intl
Conf. Parallel Processing (ICPP 03), pp. 171-178, Oct. 2003.
Search Mechanism for Peer-to-Peer Networks, Proc.ACM CIKM
Intl Conf. Information and Knowledge Management (CIKM 02),
pp. 300307,Nov. 2002.
[10] D. Tsoumakos and N. Roussopoulos, Adaptive Probabilistic Search
for Peer-to-Peer Networks,Proc. Third Intl Conf. Peer-to-Peer
Computing (P2P03), pp. 102-109, Sept. 2003.
[11] D. Tsoumakos and N. Roussopoulos, Analysis and Comparison of
P2P Search Methods, Technical Report CS-TR-4539, UMIACSTR-
2003-107, Dept.of Computer Science, Univ. of Maryland, 2003.
[12] Y. Chawathe, S. Ratnasamy, L. Breslau, N. Lanham, and S. Shenker,
Making Gnutella-Like P2P Systems Scalable, Proc.ACM
SIGCOMM 03, pp. 407-418, Aug. 2003.
[13] A. Crespo and H. Garcia-Molina, Routing Indices for Peer-to-Peer
Systems, Proc. 22nd Intl Conf.Distributed Computing Systems
(ICDCS 02), pp.23-32, July 2002.
Networks, Proc. 22nd Intl Conf. Distributed Computing Systems
(ICDCS 02), pp. 514,July 2002
Search Mechanism for Peer-to-Peer Networks, Proc. 11
th
Intl Conf.
[16] Information and Knowledge Management (CIKM 02), pp. 300-307,
Nov. 2002.
[17] D.M Rasanjalee Himali and Sushil K. Prasad, SPUN: A P2P
Probabilistic Search Algorithm based on Successful Paths in
Unstructured Networks. 2011 IEEE International Parallel &
Distributed Processing Symposium, DOI 10.1109/IPDPS.2011.316.
[18] Yunhao Liu, Member, IEEE, Li Xiao, Member, IEEE, Xiaomei Liu,
Location awareness in unstructured peer-to-peer systems, IEEE
Transactions on Parallel and Distributed Systems, VOL. 16, NO. 2,
pp.163 174, February 2005.
[19] Qiuyan Huo, Jingyu Chen, Xuezhou Xu, Yachuan Zhou, Xin Liu A
Location-aware Efficient Contentbased Searching over Unstructured
P2P Network, IEEE International Conference on Network Computing
and Information Security,pp.183 - 187, May 2011.
[20] S. Jiang, L. Guo, X. Zhang, and H. Wang, LightFlood: Minimizing
Redundant Messages and Maximizing Scope of Peer-to-Peer .
Search, IEEE Trans. Parallel and Distributed Systems, vol. 19,no. 5,
pp. 601-614, May 2008
[21] KaZaA Homepage, http://www.kazaa.com .
[22] B. Yang, H. Garcia-Molina, Designing a Super-peer Network,
Proceedings of the 19th International Conference on Data Engineering
,March 2003

ISBN: 978-81-920575-7-6 :: doi: 10. 73921/ISBN_0768
Mobile Disaster Management System Using Android
Technology

Miss. Deepti B. Shelar
Lecturer in Shah and Anchor
Kutchhi Engg College
Mumbai - 88, India

Abstract In todays scenario use of smartphones is
gaining interest in people, the disaster management system was
implemented as a smart phone application using Googles
Android operating system. The disaster management system
Android application determines the optimum route along
different geographical locations that the volunteers and
rescuers need to take in order to serve the most number of
people and provide maximum coverage of the area in the
shortest possible time. Genetic algorithm was applied for
optimization and different parameters were varied to determine
the most optimum route.
KeywordsAndroid, Disaster management, Genetic
algorithm, Travelling salesman problem, Smart phone
Application, Wireless mobile technology

I. INTRODUCTION
Many countries and territories are at risk of natural
hazards due to earthquakes, floods, tropical cyclones, and
landslides. Due to this mortality rate index (MRI) is
increasing. It has claimed the lives of thousands of
people and damaged billions worth of property. Though
we cannot stop these natural disasters but we can
definitely find solution to overcome it faster. Even with
the recurrent calamities, there is still the absence of an
efficient disaster management system that will help the
people in times of calamities.
II. DISASTER MANAGEMENT
The process of disaster management involves four
phases: mitigation, preparedness, response, and recovery.
attempt to reduce disaster risks is called as
mitigation. development of an action plan for an
upcoming disaster comes under preparedness. The
response phase consist of the mobilization of services
and relief when disaster strikes and the recovery phase is
the restoration of the affected area to its previous state.
A.USE OF SMART PHONES:
A smartphone is a high-end mobile phone that
combines the functions of a personal digital assistant
(PDA) and a mobile phone. Today's models typically also
serve as portable media players and camera phones with
high-resolution touch screens, web browsers that can
access, and properly display, standard web pages rather
than just mobile-optimized sites, GPS navigation, Wi-Fi
and mobile broadband access. The term smartphone is
usually used to describe phones with more advanced
computing ability and connectivity than a contemporary
feature phone, although the distinction can be vague and
there is no official definition for what constitutes the
difference between them. The definitions also shift over
time since many phones that are considered feature phones
today can have capabilities that exceed those of phones that
had been promoted as smartphones in the past.
B.ANDROID MOBILE DEVELOPMENT ENVIRONMENT:
Android is a software platform and operating system for
mobile devices Based on the Linux kernel. It is developed
by Google and later the Open Handset Alliance (OHA).
Allows writing managed code in the Java language.
Android is not a single piece of hardware; it's a complete,
end-to-end software platform that can be adapted to work
on any number of hardware configurations. Everything is
there, from the boot loader all the way up to the
applications. Android uses Linux for its device drivers,
memory management, process management, and
networking. It supports wireless communications using:
GSM mobile-phone technology, 3G , 802.11 Wi-Fi
networks.
Android is a multi-process system, in which each
application (and parts of the system) runs in its own
process. Most security between applications and the
system is enforced at the process level through standard
Linux facilities, such as user and group IDs that are
assigned to applications. Additional finer-grained
features ar
mechanism th
operations tha
per-URI perm
cific pieces of
GENETIC A
algorithm (G
rocess of natur
sed to gen
and search p
e larger class
generate soluti
ques inspired b
mutation, selec
intains a popu
blem at hand
plying a set of
tic algorithm,
s or the geno
date solutions
s) to an optimi
ns. Traditiona
ngs of 0s and
evolution usu
generated in
In each gen
n the popul
re stochastica
(based on t
and possibly
on. The new
of the algori
hen either a m
duced, or a sat
he population.
ximum numbe
y or may not
find appli
l science, engi
g, mathematic
IV. SYSTEM
e travelling sa
ic algorithms
ed disaster
Droid (MDD
MyDisasterDro
nd ListView.
s based on Go
of the people i
re provided
hat enforces
at a particula
missions for g
data.
ALGORITHM
GA) is a searc
ral evolution.
erate useful
problems. Gen
s of evolution
ions to optimi
by natural ev
ction, and cros
ulation of can
d and makes
f stochastic ope
a population
otype of the
s (called indiv
ization problem
lly, solutions
1s, but other e
ually starts fro
dividuals an
neration, the
lation is ev
ally selected
their fitness)
randomly mu
population is
thm. Common
maximum numb
tisfactory fitn
If the algorith
er of generatio
t have been
ication in
ineering, econ
s, physics and
M IMPLEMENT
alesman probl
s to generate
management
) was imple
id, it show
MapView sho
ogle Maps wh
in need, their c
through
restrictions o
ar process ca
granting ad-ho
ch heuristic th
This heuristic
solutions
netic algorithm
nary algorithm
ization problem
volution, such
ssover. A gene
ndidate solutio
s it evolve
erators.
of strings (ca
genome), w
viduals, creatu
m, evolves tow
are represente
encodings are
om a populatio
nd happens
fitness of e
aluated, mul
from the cu
), and mod
utated) to for
s then used in
nly, the algor
mber of generat
ess level has
hm has termin
ons, a satisfac
reached. Gen
bioinforma
nomics, chemi
d other fields
TATION
lem as basis a
a solution,
system nam
emented. Up
ws two view
ows the locati
hile the ListVi
corresponding

a
n
an
oc
hat
c is
to
ms
ms
ms
as
etic
ons
by
alled
which
tures,
ward
ed in
also
on of
in
every
ltiple
urrent
dified
rm a
n the
rithm
tions
been
nated
ctory
netic
atics,
istry,
and
an
med
pon
ws:
ion
iew
g

Figure1: BBlock Diagram o
Algorithm
Implementatio
of the Genetic
on

Figure
Figure 3: M

e 4: MyDisasterD
MyDisasterDroid
Droid ListView
d MapView
w

g
p
c
tr
o
c
s
fo
o
m
c
a
ti
p
in
n
u

V.
(TSP

Determinin
geographic loc
problem wher
coordinates an
ravelling sales

The travell
Given a finite
of travel betwe
cheapest) way
tarting point
formally descr
objective of fin
minimum cost
cities or nodes
at one node, v
ime each, and
problem with t
ncrease on th
number of citi
using optimiza
. TRAVELLIN
P)
ng the most op
cations is simi
rein geograph
nd the rescuer
sman.
ling salesman
e number of ci
een each pair
of visiting all
. The trave
ribed as a pe
nding the path
t) on an und
to be visited.
visits all other
d finally retur
the travelling
he number o
ies increase.
ation technique
NG SALESM
ptimum route
lar to the trave
hic locations
s or volunteer
problem is st
ities and the d
of them, find
l the cities and
elling salesm
ermutation pr
h of the shorte
directed graph
The travelling
r nodes succes
rns to the sta
salesman prob
of possible ro
However, this
es.
MEN PROBL
along differen
elling salesma
represent cit
rs represent th
tated as follow
distance (or co
d the shortest
d returning to t
man problem
roblem with t
est length (or t
h that represen
g salesman sta
ssively only o
arting node. T
blem is the rap
outes when t
s can be solv

EM
nt
an
ty
he
ws:
ost)
(or
the
is
the
the
nts
arts
one
The
pid
the
ved

VI. CONCLUSION
In this paper we tried to implement rescue and
disaster management operation using
MyDisasterDroid was implemented in an Android-
based mobile phone. Using genetic algorithm, an
optimum route along the given geographic locations
was determined.

REFERENCES

[1] United Nations International Strategy for
DisasterReduction Secratariat (UNISDR), Mortality
RiskIndex, 15 June 2009
[2] Asia Pacific Telecom Research
Ltd.,Telecommunications in the Philippines, 1 June 2009
[3] Damon P. Coppola, Introduction to International
Disaster Management, Elsevier Inc., 2007.
[4] Bartel Van de Walle, Gerd Van Den Eede, and Willem
Muhren, Humanitarian Information Management and Systems,
Mobile Response: Second International Workshop on Mobile
Information Technology for Emergency Response 2008,
Bonn,Germany, May 29-38, 2008, Revised Selected Papers,
Springer, 2009

ISBN: 978-81-920575-7-6 :: doi: 10. 73928/ISBN_0768
Handover Techniques For 4G Networks

Piyali Dutta
PG student, Computer Science Department
Datta Meghe College of Engineering. Airoli
Navi Mumbai, India

Abstract The next generation 4G networks are represented
by heterogeneous environment with different access network
technologies that vary in bandwidth, latency and cost. 4G
networks are known for their seamless connectivity between
existing networks like GSM, wireless LAN, Bluetooth and
others. Seamless Connectivity in such networks entirely
depends on efficient handover mechanisms and multiple
mobility management mechanisms. In this paper, we discuss
different challenges facing 4G mobility. We propose an user
zoning handover technique based on the IEEE802.21
protocol, where the approach connects users to a preset
network upon entering a zone, guided by the MIIS server. We
also discuss on vertical handover for real-time video traffic.
Keywords- handover; 4G mobility; user zoning;
IEEE802.21; vertical handover.
I. INTRODUCTION
Advanced mobile communication systems have mainly
increased due to the needs of the increasing user
requirements. Many organizations such as IEEE,
Electronics and Telecommunications Research Institutes
have been working towards 4G networks to provide
seamless connectivity anywhere and anytime. The 4G
networks which is heterogeneous in nature where there
would be multiple service providers, equipped with varied
technologies offering varied services for the benefit of the
users. 4
th
generation systems are being developed to
support greater bandwidth services, higher data rates,
improved quality and seamless global mobility as
compared to previous systems.
Handover is the transfer of an on-going call or data
session from one cell to another. Handovers occur due to
the movement of the mobile user from one area to another
area. Handovers are used to prevent an on-going call to be
disconnected. If we dont use handovers then whenever a
user leaves the area of a particular cell then its on-going
call is immediately disconnected.
The process of handover requires a number of
parameters like what is the handover scheme we are using
and how many channels are free. In the handover process
we should also keep the QoS up to the standard. In 4G
networks, we find different handover techniques.
Handovers between two base stations (BSs) of the same
system is called Horizontal handover. Horizontal handover
involves a terminal device to change cells within the same
type of network to maintain service continuity. Horizontal
handover between two BS, under same foreign agent (FA)
is known as Link-layer handover. In Intra-system
handover, the horizontal handover occurs between two BSs
that belong to two different FAs and both FAs belong to
the same system and hence to same gateway foreign agent
(GFA). Vertical handover may be referred to a process of
transferring call connected to a network/data session from
one channel connected in a cell to the core network of
another. A suitably equipped device may be able to use
both technologies at a time, the high speed Wireless LAN
and cellular technology. Vertical handover refers to a
network node changing the type of connectivity it uses to
access a supporting infrastructure, usually to support node
mobility.
IEEE802.21 [1] is a standard that was created in order
to facilitate heterogeneous handovers between IEEE 802
and non IEEE802 access networks regardless of the
network features. The IEEE 802.21 standard defines the
MIH (Media Independent Handover) mechanism to enable
the optimization of handover.
In this paper, we present a brief overview 4G general
architecture. Finally, we propose a user zoning handover
technique based on the IEEE 802.21 protocol and vertical
handovers between UMTS and WiMAX networks.
II. CHALLENGES IN MOBILITY MANAGEMENT
There are several challenges and issues regarding
mobility in 4G networks:
A. Location Management
Location management helps to locate the users
instantaneous point of attachment for connection delivery
purposes. Location management is divided into two main
functions:
1) Location Registration: Location registration keeps
the network aware of changes at the point of attachment of
each individual user among the network.
2) Call Delivery : Call delivery is used at time of a
connection request and in order to deliver the connection to
the specified user. The network is queried for the user
location profile so the current user position is retrieved.
B
C
c
a
m
c
t
a
i
s
e
v
v
D
i
d
p
c
E
p
u
h
a
u
c
n
o
c
l
p
b
l
O
a
I
t
p
r
m
M
r
W
A
B. The Choic
This issu
Connected pr
connected to t
architecture o
multiple diffe
connected ove
C. Mobility C
Context is
to characteriz
applications a
information ab
situation. Con
exhibiting te
values about
values.
D. Paging
Paging is s
information o
data sessions.
paging is req
cells can be co
E. Moving Gr
The movi
prominent com
users form a
hierarchies wh
a cluster, wit
unified netwo
cluster in a
network and s
Most of th
on cell forma
cells, responsi
limits. The c
physical resou
being structu
limited to a
Overlapping c
A universa
access request
IP networks
technologies,
provided part
Service Doma
routing and b
model are mai
Management
represents he
WiMAX, GP
All of these
ce Of Technolo
ue is associa
rinciple which
the best availa
of 4G [6] net
erent access t
er.
Context Manag
defined as an
e the situation
and networks
bout themselv
ntext informat
emporal char
past, presen
sending essent
f a user in ord
Due to powe
quired to be a
ombined toget
roup Formatio
ing formation
mponent of fu
a temporary
hile traveling.
thin which u
ork, functions
different beh
sever overhead
he wireless co
ation structure
ible for contac
ells capacity
urces over whi
ured over. Ac
certain numb
coverage is use
III. GENER
al common int
ts from variou
face differe
services and
tly by mobil
ain. The Netw
bit pipe mana
inly related to
(MM) suppo
eterogeneous
RS/UMTS, an
are connecte
ogy
ated with th
h implies that
able connectio
tworks is bas
technology fo
gement
ny information
n of an entity
s will be req
ves, their surro
tion can be s
racteristics, r
nt as well as
tial messages
der to be used
er limitations o
as efficient as
ther to form on
ons
n of a multi-
uture networks
moving clus
Such groups
users are usua
such as hando
havior to resu
d.
mmunication
es. The netwo
cting the mob
y is defined b
ich communic
ccording to t
ber of users t
ed in 4G syste
RAL ARCHITEC
terface is prov
us types of m
ent access s
roaming prot
e devices the
work Domain
agement. Chal
o the QoS, Sec
ort. The archi
networks. It
nd EUTRAN
ed to IP net
he Always
t a user has t
on at all time.
sed on combi
or the user t
n that can be
y, where termi
quired to pro
oundings and
static or dyna
regarding con
s predicted fu
to update loca
d to route calls
of mobile dev
possible. Sev
ne paging area
-user session
s. Several netw
ster and netw
of users are c
ally joined in
over will deal
ult a reductio
systems are b
ork is divided
bile units withi
by the numbe
cation channel
that, each ce
to host at a t
ems.
TURE
vided to satisfy
mobile systems
systems, term
tocol. Services
emselves and
n is responsibl
llenges facing
curity and Mob
itecture in Fi
contains WL
N access netw
twork through
Best
to be
. The
ining
o be
used
inals,
ovide
their
amic,
ntext
future
ation
s and
vices,
veral
a.
is a
work
work
alled
nto a
with
on in
based
into
in its
er of
ls are
ell is
time.
fy the
s. All
minal
s are
d the
le of
g this
bility
ig. 1
LAN,
orks.
h an
ev
ne
ha
me
ha
pr
rad
rep
an
Ot
Ne
ca
lay
ne
inv
se
de
pr
ne
av
wh
pr
fu
(M
thr
se
(M
(M
flo
ev
up
low
de
of
pr
aid
ar
volved packet
etwork. To pe
andover proc
echanisms: ha
andover execu
Figu
The handove
rocedures whe
dio measurem
ports provide
n urgent hand
therwise, it w
ew link disco
an be realized
yers. During
etworks cove
volve a quer
rver. In case
evice must a
roceeding. If
etwork, QoS c
vailability chec
hich was not i
rocedure is ba
unction (MIHF
MN) and the n
ree types of
rvice (MIES)
MICS); and (3
MIIS). Each of
ow direction w
vents such as l
pper network l
wer layer fro
evices and sca
f commands o
roviding the
des the hando
e handled by
core (EPC) w
erform a med
cess is perfo
andover initiat
ution.
ure 1 : A Heterog
er initiation in
ere each device
ment reports
information a
dover if spec
will send peri
very is also p
d with specific
g the hando
erage scannin
ry for inform
of access to
authenticate it
the device
context must
ck. Finally, ha
included by th
ased on the
F). The MIHF
network node p
f services: (1
; (2) Media in
3) Media ind
f the services
with certain r
link deteriorat
layers. MICS t
om higher one
anning for ava
ordered by M
mechanisms
over decision.
MIIS such as
which is an evo
dia independen
formed in th
tion, handover
geneous network a
nvolves old lin
e has to be co
. These mea
about the nee
cific threshol
odic informat
performed in t
c triggers from
over preparat
ng takes pla
mation from
o new network
tself to the
is authenticat
be transferred
andover execu
he IEEE 802.2
media indepe
exists in both
protocol stack
1) Media ind
ndependent co
dependent info
provided by
responsibilitie
tion and link
transfers contr
es. Configura
ailable networ
MICS. MIIS is
to retrieve i
. Static link l
s channel info
olved 3GPP co
nt handover, t
hree stages
r preparation a
architecture
nk configurati
onfigured to se
asurements a
ed of performi
lds are cross
tional messag
this stage whi
m available li
tion, addition
ace which m
an informati
k resources, t
network befo
ted at the n
d for a resour
ution takes pla
21. The handov
endent handov
the mobile no
k. It provides t
dependent ev
ommand serv
ormation serv
the MIHF ha
s. MIES repo
unavailability
rol commands
ation of netwo
rks are examp
s responsible
information th
layer paramet
ormation and t
ore
the
or
and

ion
end
and
ing
ed.
ges.
ich
ink
nal
may
ion
the
fore
new
rce
ace
ver
ver
ode
the
ent
vice
vice
as a
orts
y to
s to
ork
ples
of
hat
ters
the
m
(
q
n
c
p
z
a
t
p
a
t
w
c
a
t
t
a
M
a
t
medium acces
(AP).
IV.
The hando
query from
networks in t
commands to
policies and
zoning approa
access networ
the favor of a
perform the h
apply user zo
target zones h
which the zon
Fig
We sugg
concerning sp
are held, whe
to enter this
traffic over all
aims to optim
MIIS privileg
amount of tra
the three stage
ss control (MA
USER ZONING
over preparatio
the MIIS s
the vicinity o
owards the M
networks prio
ach proposes
rks covering t
a certain netw
handover tow
oning concept
has to be prov
ne is located.
gure 2 : User zoni
est updating
pecific zones
ere a large num
zone generat
l available net
mize the hando
ges, where it w
affic on each a
es that handov
AC) address o
G HANDOVER
on stage incl
server concer
of the MN, ad
MN, in addi
ority order. O
amending the
the surroundin
work. The MN
wards this netw
t over MIH,
vided by the M
ing handover proc
g the MIIS
such as stadi
mber of audie
ting a huge a
tworks. The us
over procedure
will be capabl
available netw
ver procedure
of the access p
APPROACH
ludes: informa
rning the ac
dditional scan
ition to char
Our proposed
e priority orde
ngs of the M
N will be force
work. In orde
information a
MIIS server u
cedure flow.
server data
iums when ga
ences are expe
amount of ur
ser zoning con
e by extending
e of balancing
work. Fig. 2 sh
has to go thro
point
ation
ccess
nning
rging
user
er of
MN in
ed to
er to
about
under

abase
ames
ected
rgent
ncept
g the
g the
hows
ough:
ini
pr
A.
tow
thr
wh
co
me
(M
thr
de
Up
de
Po
a.
in
me
de
oth
the
ch
W
rea
to
(M
de
do
pr
pe
pr
b.A
req
vic
inc
pr
inc
c.T
ind
ne
fro
ne
tra
the
req
su
ch
un
req
(G
es
itiation, prep
rocess, importa
The user zon
The mobile
wards the n
resholds is u
hich is in o
onnection is e
essage to
MIH_configure
resholds for t
evice generate
pon receiving
evice has to se
oS (MIH_conf
After that,
formation
easurements.
evice status at
her significan
e wireless d
hanges to the
When deteriora
ach unaccepta
the WiMAX
MIH_link_goin
etails about th
own and the re
recision. Now
erforming the
reparation stag
At this stage,
questing info
cinity of the
cludes inform
resent around
cludes: netwo
The MIIS serv
dicating forcin
etwork instead
om. This ste
etwork availab
affic managem
e handover
quired to perf
urrounding. Th
heck on candid
nit will skip a
quired measu
GPRS in ou
tablishing a co
aration, and
ant changes in
ning handover
unit is pre-co
network. The
usually set by
our case the
established, th
be sent
e_link_REQ).
the connection
es a report an
g the configu
end a confirma
figure_link_CN
the device
messages
The WiMAX
t all time. As
nt parameters i
evice will ge
e PoS (MIH_
ation in conn
able levels, the
X PoS indicat
ng_down_IND
he time perio
eason of this
the WiMAX
handover an
ge starts.
a query to th
ormation reg
mobile devic
mation about
d the wireles
rk status, avai
ver will send a
ng the device
d of suggesting
p aims to o
bility, cost, se
ment and will
process. The
form additiona
here will be no
date networks
all regular rep
urements reg
ur example).
onnection with
execution. I
n the link quali
r procedure wo
onfigured to g
pre-configur
y the PoS (P
e WiMAX P
he PoS issues
to the
This mess
n parameters u
nd sends it b
uration messag
ation message
NF).
starts to se
stating sig
X PoS will b
far as QoS re
in case of cros
enerate a rep
_linkparamete
nection quali
e mobile unit
tes the need
D). This me
od at which t
links weaknes
X PoS is aware
nd this is whe
he MIIS serve
garding acces
ce. The MIIS
t different a
ss device. T
ilability, and s
a message to t
to connect to
g multiple net
ptimize servi
eamless mobi
reduce the tim
e wireless un
al scanning for
o need of reso
. In this techn
porting actions
garding the
Then, imm
h the new radi
n the initiati
ity are reporte
orks as follow
generate trigg
ration of the
Point-of-Servi
PoS. When t
a configurati
mobile no
sage states t
under which t
ack to the Po
ge, the wirel
e to the WiMA
end a period
gnal streng
be aware of t
equirements a
ssing threshol
port to indic
ers_report_IND
ty continues
sends a messa
for a handov
essage includ
the link will
ss with a certa
e of the need
en the handov
er might be s
ss networks
server databa
access networ
This informati
signal strength
the WiMAX P
o a certain pre
tworks to choo
ice in terms
ility, and urg
me consumed
nit will not
r networks in t
ource availabil
nique, the mob
s and just rep
target netwo
mediately sta
io network.
ion
ed.
s:
gers
ese
ce)
the
ion
ode
the
the
oS.
ess
AX
dic
gth
the
and
lds,
ate
D).
to
age
ver
des
go
ain
of
ver
ent
in
ase
rks
ion
h.
PoS
eset
ose
of
ent
d in
be
the
lity
bile
port
ork
arts
d.The WiMAX PoS sends an information request (MIH-
Get-Information.REQ) message. MIIS sever receives the
request and then sends back a special message based on
PoA (Point-of-Attachment) location where user zoning is
applied, containing the target network along with its
characteristic. The mobile unit now will only scan for this
specific network rather than scanning for a list of available
networks. MIIS will choose this network based on multiple
aspects such as resource availability, QoS, signal strength
within the zone, capability to maintain active sessions, and
the types of services offered.
F. At this point, the WiMAX PoS will send an order to the
mobile unit to start scanning for the concerned PoA
recommended by the MIIS server. A signal strength
measurement message has to be sent back to the WiMAX
PoS.
G. The PoS sends a confirmation to the MIIS server stating
the readiness of the wireless device to execute handover.
H. The WiMAX PoS communicates with GPRS PoS and
send a (MIH_N2N_HO_Query_Resourse) request and
Reply. The GPRS PoS has to reserve required resources.
Resources have to be guaranteed in both core and target
networks. This is when the execution stage begins and the
WiMAX PoS orders the wireless device to start the
handover by indicating the actions over the old and new
link (MIH_Net_HO_commit.REQ). Old link resources are
not needed anymore so it can be released and an IP
connection is now established over the GPRS network
which has to be reported back to the WiMAX PoS
(MIH_Net_HO_Commit.RSP) then to the GPRS PoS
(MIH_N2N_HO_complete.REQ).
I. Now, the WiMAX has to respond
(MIH_N2N_HO_complete.RSP). Then, WiMAX releases
resources if not already done and data starts flowing
through GPRS Network.
V. VERTICAL HANDOVERS FOR REAL-TIME VIDEO TRAFFIC
BETWEEN UMTS AND WIMAX NETWORKS
Internetworking approach of the converged next
generation multi-access network can make the best use of
the advantages of the different wireless networks and
eliminate their weaknesses. UMTS, WiMAX and WLAN
networks can complement each other in term of
geographical coverage and QoS.
One of the possible scenarios where the above three
heterogeneous networks can be used is defined here. While
the vehicle moves on the road, the wider area is covered
only with UMTS. The metro zones are covered
additionally with WiMAX, while hotspots have also WiFi.
Different access networks may be ranked by the operator
(if they are owned by single one) or by the mobile user, for
given services. As the user in the vehicle approaches the
city, besides UMTS, WiMAX (IEEE 802.16e) becomes
available, so vertical handover between UMTS and
WiMAX can occur. Vertical handover among these
different networks can take place when QoS is poor for the
current wireless access network, or due to service
management either by the network provider or by the user.
For the purpose of analysis of impact of vertical handovers
between UMTS and WiMAX for video traffic, we use the
following three streaming video technologies: HDTV
(High Definition TV), MPEG-4 and H.261. Each of these
applications requires different bandwidth. Each wireless
technology has a limit on the maximum supported user
velocity. Also, user mobility influences the behavior of the
wireless access network and the video stream transmitted
over such network.
The aim of this work is to investigate the reasons for
behaviors like poor vertical handover latency, and to
improve the vertical handovers in the case of streaming
video traffic, and hence to reduce the latency, to reduce the
number of vertical handovers and throughput gap between
UMTS and WiMAX networks.
Vertical handover process can be divided in three
phases: handover decision, radio link transfer and channel
assignment. Handover decision deals with the selection of
the target point of attachment and the time of the handover.
Radio link transfer is the process of developing links to the
new point of attachment. Channel assignment phase
arranges the allocation of channel resources. Because IEEE
802.21 standard supports two types of vertical handovers,
handovers initiated by the network and handovers initiated
by mobile terminals, events related to handovers can be
originated at the MAC (Medium Access Control) or MIHF
(Media Independent Handover Function) layer located in
the node or at the point of attachment to the network.
Figures 3 and 4 represent a vertical handover process
between UMTS and WiMAX networks. When mobile
node (MN) enters the WiMAX cell from UMTS link is
detected upon reception of DL_MAP message from BS.
MAC layer implement synchronization messages
DL_MAP and DCD messages for downlink
synchronization and UL_MAP and UCD messages for
uplink synchronization. After the synchronization in Figure
3 initial ranging and registration occurs in the layer 2
vertical handover process.
4
c
t
d
r
t
r
a
i
a
h
F
W
q
m
m
s
Figure 3: La
Layer 3 ve
4. When a lay
created by the
that MIH trig
discover the I
replies to the
the ND (Nei
receives the R
a new addres
interface, bec
available for
handover proc
Figure 4 : Laye
WiMAX.
Vertical h
quantitatively
measuring th
maximum han
session main
ayer 2 vertical han
W
ertical handove
yer 2 connecti
e MAC and se
ggers a RS (R
IP prefix of th
RS message
ighbor Discov
RA (Router Ad
ss and redirec
cause the Wi
application
cess is comple
er 3 vertical ha
handover proc
measured un
he number o
ndover delays,
ntained over
ndover process be
WiMAX.
er process is d
ion is establis
nt to the MIH
Router Solicit
he new link.
according the
very) module
dvertisement)
cts the flow o
iMAX interfa
traffic. With
eted.
andover process
cess can be
nder various u
of handovers
, and the overa
a specific
etween UMTS an
described in Fi
shed, a Link U
H module [6]. A
tation) messag
The access ro
e rules define
e. When the
message, it b
on to the WiM
ace is consid
this the ver
between UMTS
qualitatively
usage scenario
, the mean
all throughput
mobility pat

nd
igure
UP is
After
ge to
outer
ed by
MN
uilds
MAX
dered
rtical

S and
and
os by
and
t of a
ttern.
Fr
re
be
ha
tim
ne
ini
co
of
Re
de
me
mo
av
de
tec
an
vid
fo
ne
ov
as
uti
Th
en
in
ha
W
fo
fu
co
pr

[1]
[2]
[3]
[4]
[5]
[6]
[7]
requent vertica
sources. So, d
eneficial for
andover is co
me, a handove
eeded. This kin
Handover la
itiation of t
ompletion. Han
f the VHD
eduction of th
elay-sensitive
easured with
obile terminal
vailable access
esirable.
In this pape
chnique base
nalysis on vert
deo streams
cus on vertica
etworks. The
verhead and ti
well as allow
ilization by ba
his approach
ntering a zone
formation ab
andover, small
WiMAX netwo
r the mobile u
urther analysi
onducted to ev
roposed techni
] The IEEE802.
] H. Brendt, T
Wiley, West S
] S. Hussain, Z
Issues in 4G
France 2006.
] 3GPP TS 23
Services and
Evolution: Ar
3GPP Technic
] I.F. Akyildiz
management
IEEE Wireles
pp.1628.
] K. Taniuchi, Y
A. Dutta, D. B
independent
IEEE Commu
] S.J. Yoo, D.
Mechanism fo
al handovers c
decreasing the
better effici
onsidered unn
er back to the
nd of handove
atency is the
the vertical
ndover latenc
(vertical h
he handover l
multimedia
the data rat
ls on the netw
s network with
VI. CO
er, we presen
ed on the IE
tical handover
in heterogene
al handovers b
user zoning
ime consumed
wing the MII
alancing loads
connects use
guided by the
bout target z
ler the throug
rks for differe
users with vid
is, validation
valuate and co
iques.
REFER
.21 Standard, http
Towards 4G Tec
Sussex, England, 2
Z. Hamid and N
G Heterogeneous
3.402 (2007) 3G
d Systems Asp
rchitecture Enha
cal Specification 2
z, J. Xie, S. M
in next-generati
ss Communicatio
Y. Ohba, V. Faja
Baker, M. Yajnik
handover: featur
unications Magazi
Cypher, and N
or Seamless Hand
can cause was
e number of ha
iency of the
ecessary whe
original point
ers should be m
e time duratio
handover p
y is related to
handover dec
latency is ver
sessions. The
te that is tra
work. Vertical
h higher throu
ONCLUSION
nted a user z
EEE802.21 p
s done with di
eous wireless
between UMT
g technique a
d while perfor
S server to op
s over the ava
ers to a prese
e MIIS server
zones. Also
hput gap betw
ent velocities b
deo streams. A
n and simul
onfirm the per
RENCES
p://www.ieee802.o
chnologies: Servi
2008.
. Khattak, Mob
Networks, Int
GPP Technical S
pects; 3GPP Sy
ancements for N
23.402 v 1.1.0, 20
Mohanty, A s
ion all-IP-based
ons Magazine, V
ardo, S. Das, M.
k, D. Famolari,
res, applicability
ine, Vol.47, No.1,
N. Golmie, Pred
dovers in Heterog
stage of netwo
andovers is ve
e resources.
en, after a sh
of attachment
minimized.
on between t
rocess and
o the complex
cision) proce
ry important
e throughput
ansmitted to t
l handover to
ughput is usua
zoning handov
protocol and
ifferent real tim
s scenario, w
TS and WiMA
aims to redu
rming handov
ptimize netwo
ailable networ
et network up
r which provid
in the Verti
ween UMTS a
better is the Q
As a future wo
lations will
rformance of t
org/21
ices with Initiat
bility challenges
terSense 06, N
Specification Gr
ystem Architect
Non-3GPP Acces
007.
survey of mobi
wireless system
Vol.11, No.4, 20
Tauil, Y.-H. Che
IEEE 802.21: me
y, and realizatio
, 2009, pp.11212
dictive Link Trig
geneous Handove
ork
ery
A
hort
t is
the
its
xity
ess.
for
is
the
an
ally
ver
an
me
with
AX
uce
ver,
ork
rks.
pon
des
ical
and
QoS
ork,
be
the
tive,
and
Nice,
oup
ture
ses,
ility
ms,
004,
eng,
edia
on,
20 .
gger
ers,
Wiley Wireless Communications and Mobile Computing, Vol. 9,
No.5, May 2009.
[8] Wikipedia.com;Google.com.


ISBN: 978-81-920575-7-6 :: doi: 10. 73949/ISBN_0768
SMS Length Expansion and Security Approach
an A Mobile Device
Deepak S. Manglani
Lecturer in G.V.Acharya Engg.
College
Abstract Ever since their invention, mobile phones have
revolutionized the regular communication methods. Theyve
shrunk the world and have brought all of us closer.
Exchanging a texted conversation between a Tx and a Rx via
mobile phone turned out to be a very successful service.
Such a communication was termed as Short Message
Service, commonly abbreviated and known as SMS. The
flipside to the SMS is the limitation to the maximum number
of characters, i.e., 160. Every 161
st
character happens to be a
new SMS. This increases cost and complexity. Here, in this
paper, weve proposed ways to compress and encrypt the
SMS communication. We believe this holds substantial
content in bringing about a new social impact to the
development of larger content in SMS messaging.
Keywords- Tx Transmitter, Rx Receiver, SMS Short
Message Service, GSM - Global System for Mobile
communications, TDMA Time Division Multiple Access,
GPRS General Packet Radio System, CDMA Code Division
Multiple Access, DES Data Encryption Standard, IDEA -
International Data Encryption Algorithm, AES Advanced
Encryption Standard.

I. INTRODUCTION
In Mobile Phone Communication, SMS is the
tremendous service, with 10 billion messages sent around
the world daily. SMS makes it possible to send and
receive short text messages to and from mobile phones.
SMS became a popular communication tools by
individuals and business. SMS are used in day-to-day
applications like banking, mobile payment, recharge and
mobile commerce, e-ticket, money transfer, etc. [1] In
1985, Friedhelm Hillebrand of Bonn, Germany and a
dozen others had been laying out the plans to standardize
a technology that would allow cellphones to transmit and
display text messages. Because of tight bandwidth
constraints of the wireless networks at the time, each
message would have to be as short as possible. [2]
The maximum number of characters for a single
normal SMS is only 160 characters. The SMS security is
also very crucial to protect the sensitive information.
People sometimes exchange confidential information such
as passwords or sensitive data. SMS technology suffers
from risks such as security vulnerabilities and
unauthorized access. [3]
Current standard SMS technology is non-secure. It can
be intercepted along the transmission path at multiple
locations (aggregators, carriers, infrastructure providers,
or tower operators) especially when using Internet to
submit your messages.
Our approach to this problem in this paper is to
propose a method that can be used to compress and to
encrypt SMS messages that are about to be sent. In our
approach, the physical underlying GSM architecture
remains unchanged.
This paper is structured as follows: Section II
highlights GSM System Architecture. Section III provides
SMS Structure. Section IV explains compression
technique in SMS and Section V shows a method secure
the SMS. Conclusion is presented in Section VI.
II. GSM SYSTEM ARCHITECTURE
In GSM system, the mobile handset is called Mobile
Station (MS). A cell is formed by the coverage area of a
Base Transceiver Station (BTS) which serves the MS in
its coverage area. Several BTS together are controlled by
one Base Station Controller (BSC). The BTS and BSC
together form Base Station Subsystem (BSS). The
combined traffic of the mobile stations in their respective
cells is routed through a switch called Mobile Switching
Center (MSC). Connection originating or terminating
from external telephone (PSTN) are handled by a
dedicated gateway Gateway Mobile Switching Center
(GMSC). The architecture of a GSM system is shown in
figure 1.
In addition to the above entities several databases are
used for the purpose of call control and network
management. These databases are Home Location
Register (HLR), Visitor Location Register (VLR), the
Authentication Center (AUC), and Equipment Identity
Register (EIR).

(
c
w
w
s
d
a
e
e
e
a
E
m
n
t
w
o
`
a
m
r
s
Home Loc
(such as user
current locatio
with the netw
who are bein
stored in HLR
data like loc
authentication
encryption. T
equipments an
equipments.
All the m
assigned uniq
Equipment I
manufacturer
number is sto
the IMSI (I
which is store
of the user. A
SIM is inserte
``real'' telepho
and is stored i

SMS is a
mobile comm
received simu
SMS exists a
such as GSM,
The SMS P
Figure
cation Registe
r profile) as
on) informatio
work. A VLR
ng serviced c
R for faster ac
cation of the
n information o
The EIR sto
nd can be used
mobile equip
que id called
dentity) and
and registered
ored in the EIR
nternational
ed in the Subs
A mobile statio
ed into an equ
one number is
in SIM. [4].
III. SM
communicatio
munication sy
ultaneously w
as a subset o
, TDMA, GPR
Packets in des
1.
er (HLR) store
well as tem
on about all th
stores the dat
currently. It in
ccess as well
e user. The
of the user suc
ores stores
d to prevent ca
pments in G
d IMSI (Inter
is allocated
d by the servi
R. The users
Module Sub
scriber Identit
on can be use
uipment with
s different fro
MS STRUCTURE

on service sta
ystems; it ca
with voice, da
f the larger w
RS and CDMA
sign are shown
es the perman
mporary (such
e users registe
a about the us
ncludes the d
as the tempor
AUC stores
ch as the keys
data about
alls from a sto
GSM system
rnational Mob
d by equipm
ice provider. T
are identified
bscriber Ident
ty Module (SI
ed only if a v
valid IMSI. T
om the above
E
andardized in
an be sent
ata and fax ca
wireless doma
A. [5].
n in the Figure

nent
h as
ered
sers
data
rary
the
s for
the
olen
are
bile
ment
This
d by
tity)
IM)
alid
The
ids
the
and
alls.
ains
e 1.
tr
g
c
m
c
d
is
w
a
th
d
r
v
T
(
u
c
4
r
c
S
S
a
m
fo

SMS applic
ransport them
gateway that c
centre. It provi
mechanism for
characters from
The SMS i
directly but de
s a network el
which SMS is
available. SMS
he network. M
destination cel
eceive or tran
voice call is in
TCP/IP networ
GMSC). [6]

In this sec
used to com
compression e
40 percent. Th
atio of total c
characters of th
Compres
SMS
Efficien
SMS
Our appro
application for
manufacturers
for mobile app

Figure 2
cations are on
m. An SMS s
connects to a m
ides a point-to
r transmitting
m mobile hand
is store and fo
livered via an
lement in the
stored until th
SC guarantees
Messages are s
ll phone beco
nsmit an SMS
n progress or n
rks via a Gate
IV. C
ction, we des
mpress the
fficiency of th
he efficiency
characters in a
he original SM
ssion = T
ncy
oach to this
r SMS Compr
have adopted
plication deve
2.
ly as good as
service is hos
mobile service
o-point and br
short message
dsets.
orward service
n SMS Center
mobile teleph
he destination
s delivery of
stored in the n
omes availabl
message at an
not. SMSCs co
eway Mobile S
OMPRESSION

scribe the pro
SMS length
he message m
is obtained b
a compressed
MS.
Total character
Total chara
problem is
ression. Sever
d Java as their
elopers. To ce
the networks t
sted on an SM
e providers SM
roadcast-to-po
es up to 160 t
e, and is not s
(SMSC). SM
hone network,
device becom
text message
network until t
le, so users c
ny time, wheth
ommunicate w
Switching Cen
oposed techni
h. The aver
must be more t
y calculating
SMS to the to
rs in Compres
acters in Origi
to develop
ral mobile dev
platform offe
ertain extent J

the
MS
MS
oint
ext
ent
SC
, in
mes
by
the
can
her
with
nter
ique
rage
than
the
otal
ssed
inal
an
vice
ered
Java
a
v
s
s
m
i
o
t
T
s
c
s
t
T
u
c
u
c
applications a
vendors. Wit
significantly t
sending a lo
message to se
it and by send
original, it wil
Many SM
to solve the
These method
so-called wo
compressed m
senders. In or
to type as muc
160 character
The abbreviat
used and inter
While SM
communicatio
using the
confidentiality

I
Mobile d
and this me
designed app
several secu
integrity and
key assets th
This sec
formal descr
specification
and decrypt
station is th
attacker cou
the mobile s
Hence our a
messages us
readable. Th
plaintext to
and diffusion
DES alg
encryption a
and financia
small by curr
can be searc
block cipher
Massey in 1
cipher, PES
originally c
developed to
plaintext and
128-bit key.
are portable
th this applic
the number o
ong message.
end, the applic
ding the compr
ll reduce the c
S compression
limitation of
ds include hum
ord or phrase
messages are
der to save th
ch information
rs per SMS b
tions utilized
rpreted by the
MS is one of
on, businesses
SMS channel
y issues.
I. ENCRYPT
devices have
eans that the
propriately. F
urity services
d availability a
hat are protecte
ction highligh
ription of a c
n for its messa
tion functions
heft or mispl
ld gain acces
tation. This th
approach to th
sing IDEA a
he algorithm
64-bit block
n.
gorithm has
algorithm and
al applications
rent standards
ched in appro
r designed b
1991. It is a
(Proposed En
called IPES
o replace DE
d cipher text
The algorith
between dev
cation, the u
of single mess
When the
cation can be
ressed messag
cost.
n methods hav
f 160 charact
man interpretiv
e abbreviation
e manually
e cost, many S
n as possible w
by using vario
in the messa
senders and re
the most pre
s and banks ar
l considering
TION & DECRY

limited comp
security mec
or a secure S
s are needed
are the key ser
ed [7].
hts our resear
cryptosystem
age, key, ciphe
s. The main
lacement and
s to the data
hreat applies t
is problem he
algorithm to
m encrypts a
of cipher tex
been a pop
d is used in m
s. However, it
s and its entire
oximately 22 h
by Xuejia La
a minor revis
ncryption Stan
(Improved
ES. IDEA ope
blocks and i
hm structure
vices of diffe
users can red
sages required
user has a
used to comp
ge, shorter than
ve been develo
ters per mess
ve compressio
n, in which
keyed-in by
SMS senders
within the lim
ous abbreviati
age are popul
eceivers.
eferred mode
re conservativ
the security
YPTION
puting capabil
chanisms mus
SMS environm
d. Confidentia
rvices and also
rch work give
that includes
er text, encryp
risk to a mo
as a result,
content stored
to SMS messa
ere is by enco
make them n
64-bit block
t using confu
pular secret
many commer
ts key size is
e 56 bit key sp
hours. IDEA
ai and James
ion of an ea
ndard). IDEA
PES) and
erates with 64
s controlled b
has been cho
erent
duce
d for
long
press
n the
oped
sage.
on or
the
the
tend
mit of
ions.
larly
es of
ve in
and
lities
st be
ment,
ality,
o the
es a
s the
ption
obile
the
d on
ages.
ding
non-
k of
usion
key
rcial
s too
pace
is a
s L.
arlier
was
was
4-bit
by a
osen
e
p
b
k
p
th
s
d
k
s
u
c
e
b
k
s
d
such that wh
encryption pr
The 64-bi
bit sub-blocks
bit key. Sinc
required for t
of 52 (= 8 x 6
generated fro
partitioned in
directly used
bit key is th
positions, afte
partitioned int
as the next
procedure is
key sub-block

The priva
establishing th
path, since me
be decrypted w
key can be
public/private
han AES sinc
ufficient proc
devices and SM
key would hav
erver applica
unacceptable v
The compu
ciphered text
encryption. Th
bit key sub-blo
key sub-block
ub-blocks mu
decryption in o
hen different
ocess is identi
it plaintext blo
s six 16-bit ke
ce a further fo
the subsequent
6 + 4) differen
om the 128-b
nto eight 16-b
as the first ei
hen cyclically
er which the r
to eight 16-bit
eight key s
repeated until
ks have been g
Figure
ate/public ke
he foundation
ssages encryp
without the p
distributed
key exchang
ce the messa
cessing power,
MS in genera
ve to be progra
ation. This
vulnerability to
utational proce
is essentially
he only differe
ocks used for d
used during e
ust be used
order to revers
key sub-bloc
ical to the decr
ock is partition
ey are generat
four 16-bit ke
t output transf
nt 16-bit sub-b
bit key. The
bit sub-blocks
ight key sub-b
y shifted to
resulting 128-b
t sub-blocks to
sub-blocks. T
l all of the re
generated.
3.
ey technique
for a secure
pted with the p
private key. H
without risk
e can be sign
age size is lar
, which is not
al. This means
ammed into bo
scenario wo
o reverse engin
ess used for d
y the same a
ence is that ea
decryption is t
encryption. In
in the rever
se the encrypti
ks are used,
ryption proces
ned into four
ted from the 1
ey-sub-blocks
formation, a to
blocks have to
e 128-bit key
s which are t
blocks. The 1
the left by
bit block is ag
o be directly u
This cyclic s
quired 52, 16
e is ideal
communicati
public key can
Hence, the pub
k. However,
nificantly slow
rge and requ
t ideal for mob
s the initial A
oth the client
ould present
neering.
decryption of
as that used
ach of the 52
the inverse of
addition, the
rse order dur
ion process. [8

the
ss.
16-
128-
are
otal
o be
y is
then
128-
25
gain
used
shift
6-bit

for
ions
nnot
blic
, a
wer
uires
bile
AES
and
an
the
for
16-
f the
key
ring
8]
c
p
D
s
o
a

In this pap
character limi
SMS service,
phenomena, w
SMS Structur
Decryption tec
The prop
security in SM
of view, th
alongwith sec

Figure 4
V. C
per, we've tri
itation concept
ever since its
we studied th
re, SMS Packe
chniques with
osed method
MS communic
his technique
urity.

4.
CONCLUSION
ied to elimina
t that has been
s introduction
he GSM Syst
et, Compressio
h regards to the
d also provid
cation. From th
ensures co

ate the maxim
n surrounding
n. To achieve
tem Architectu
on, Encryption
e SMS.
des end to
he end user po
ost effectiven

mum
the
this
ure,
n &
end
oint
ness

[1] Tarek M M
Ahmed M
For Securi
And Securi
[2] http://teck.i
only.html [
free and op
[3] Johnny Li-C
End-To- En
South Afri
[4] http://www
Indian Insti
[5] N.J Croft A
Pad To Sec
Computer
University
[6] Sun
http://Dev
[7] Otto Kols
Enhanceme
Conference
[8] Internationa

REFERE
Mahmoud, Bahgat
Mahfouz, Hybrid
ing SMS, Interna
ity (IJCSS) 2010.
in/why-sms-messa
Source: http://te
pen source softwar
Chang Lo, Judit
nd Protocol For
ca 2008.
w.it.iitb.ac.in/~saty
itute of Technolog
And M.S Olivier
cure Short Messa
Security Archi
Of Pretoria, Sout
Developer
veloper.Sun.Com/
si, Teemupekka
ents Proceedings
e On System Scien
al Data Encryption

ENCES

t A. Abdel-Latef,
d Compression E
ational Journal O

ages-are-limited-t
ck.in - The techn
re as well]
th Bishop, J.H.P
Secure SMS, Un
yajit/seminar/node
gy, Bombay]
r, Using An App
aging Service (SM
itectures (ICSA)
th Africa 2005.
Network
/Mobility/Articles
a Virtanen, M
s Of The 37th
nces.
n Algorithm - Auth

, Awny A. Ahme
Encryption Techn
Of Computer Scie
to-160-characters-
nology blog prom
P Eloff Smssec:
niversity Of Pret
e14.html [Sou
proximated One-T
MS), Information
) Research Gr
SMS Art
s/Sms
MIDP 2.0 Secu
Hawaii Internati
hor: How-Shen Ch

ed &
ique
ence
-
motes
: An
toria,
urce:
Time
And
roup,
ticle,
urity
ional
han..
ISBN: 978-81-920575-7-6 :: doi: 10. 73956/ISBN_0768
Effect of topology on routing performance in partially mobile ad hoc networks

Ningrinla Marchang ,Nibedita Shial
Department of Computer Science and Engineering North Eastern Regional Institute
of Science and Technology Nirjuli, Itanagar - 791109, Arunachal Pradesh, INDIA.
ningrinla@yahoo.co.in, nibedita7865@gmail.com

A mobile ad hoc network (MANET) is a network of mobile
nodes which communicate with each other without the help
of any fixed infrastructure. Links
between nodes are created and broken as the nodes move
within the network. Thus mobility
of the nodes adversely affect routing performance
parameters such as throughput and end-to-end delay. It is
desirable to improve the performance and various ap-
proaches proposed toward this goal include modifying the
routing protocol or reinforcing the network with powerful
backbone nodes. Without relying on such elaborate
approaches, which are time-consuming and costly, it may be
required to improve routing performance. The challenge is to
improve on routing performance making use of only the
existing facilities in a MANET. We propose that one way of
achieving this goal is topology design. In this paper, we
investigate the effect of topology on routing performance in a
partially mobile ad dhoc network. In other words, by keeping
some nodes stationary in some fashion, we examine the effect of
the topology of these nodes on the performance. We have
presented our simulation results for the same.

Keywords: Mobile ad hoc network (MANET);
Topology; Routing performance.

I INTRODUCTION

A mobile ad hoc network is a network of wireless
mobile devices, which can be set up on the fly. The
network is created, managed and organized purely by the
nodes themselves without the assistance of any centralized
third party or fixed infrastructure. A node can be a
laptop, a PDA, a mobile phone, or any mobile device
with the ability to communicate with other devices. In a
MANET, a node behaves as a host as well as a router.
Since there is no requirement of a fixed infrastructure, a
MANET can be set up at any place and at any time as
needed. For instance, MANETs can be set up in
emergency situations (e.g., rescue mission), at places
where wired networks cannot be used (e.g., outdoor
conferences), or at places where the cost of setting up wired
networks is prohibitive (e.g., a remote hilly resort).
A MANET is an attractive solution for the com-
munication needs of a community or area where there no
fixed infrastructures. An example of such a community
would be a rural village without any means of electronic
communication within the village. However, for a
MANET to support voice communication, performance
parameters such as


end-to-end delay and jitter should be within acceptable
limits. It is seen that delays beyond about
230 milliseconds in a voice call causes the human ear to
detect choppiness in the call.
It would be interesting to determine whether routing
performance can be improved by making the network
partially mobile by keeping some of the nodes stationary.
If it could be done so, we have found a simple and cost-
effective solution. Therefore, one would not need to rely
on other elaborate approaches such as investing on more
powerful nodes as backbone nodes or modifying the
routing protocol. We suggest that in certain situations,
one may not have the time nor the resource required for
such elaborate solutions. For instance, during a rescue
operation for which a MANET is set up, if it were found
that the delay and jitter of the network are such that no
effective communication were possible, the simplest and
quickest solution would be to keep some nodes stationary
albeit in some arrangement. This paper presents the
analytical results that were arrived at after extensive
simulation from which we can infer about the feasibility
of such a solution.
The following are some work which are related in some
way that is found in the literature. Topol- ogy and mobility
considerations in mobile ad hoc networks are presented in
[1]. The authors studied various characteristics concerning
the links and routes that make up a mobile ad hoc
network. One important finding was that due to rapid rout-
ing changes brought about by a highly dynamic topology,
one must select a reasonable mobility setting in order to
achieve some reasonable level of performance. The
impact of topology in robust routing on wireless mesh
networks is investigated in [2]. In this work, the authors
explored the impact of node distribution on routing perfor-
mance in a wireless mesh network. They found that
network performance using robust routing can be
improved by using a distribution which places a greater
density of nodes toward the outside of a network. In [3],
the authors studied the problem of designing the optimal
topology
for backbone-based hybrid wireless ad-hoc net works.
The problem is formulated as a generalized optimal
geometric Stiener network problem with minimum number
of Steiner nodes. They proposed a multiple-step solution
approach and compared several heuristic algorithms.
Another work in which design considerations for a sparse
network are discussed is [4]. The authors also defined a
metric called reachability that is useful in evaluating
design trade-offs in sparse networks. All the above work
are concerned with networks in which either all nodes are
mobile or all nodes stationary. Our proposed approach
deals with a network in which some nodes are mobile
while the others are stationary.
Topology control of ad hoc networks ([5]-[14]) is
another area in which topology of the network is
determined by controlling the communication range of the
nodes. Although topology design is also done here, the
main aim here is to reduce the overall energy consumption
of the nodes while satisfying some constraints such as
maintaining connectivity. Besides, topology control is
generally carried out for static ad hoc networks.
The paper is organized as follows. In section I, we
give a brief introduction and related work. Section II
presents the topology design. In section III, we give
the simulation results which is followed by the conclusion
in section IV.

II TOPOLOGY DESIGN

Considering that the nodes in a MANET are mobile,
there seem to be little or no meaning concerning the
topology design of the network. How- ever, in our present
context, our interest is in a partially mobile ad hoc
network in which some nodes are kept stationary while
the rest are mobile. Thus, by topology design, we mean
design of the the topology of the network formed by the
set of stationary nodes. For instance in Fig. 1, we show a
partially mobile MANET. A filled circle denotes a mobile
node whereas an empty circle denotes a stationary node.
The stationary nodes
are arranged in a grid within the MANETs net-
work area represented by the box. An edge be-
tween any two stationary nodes denotes that the nodes
are within communication range of each
other. The stationary nodes are the same as the
mobile nodes in every respect such that they have the same
processing capacity, transmission range and battery power
as the mobile nodes. Moreover, a stationary node behaves
both as host and router like any other node. Due to the way
the stationary nodes are uniformly laid out, thereby
providing good connectivity, it is intuitive to expect a higher
degree of routing performance than when all the nodes are
mobile.
Another way of placing the stationary nodes
is by randomly scattering them in the MANETs network
area. Since, they are randomly placed, one cannot
ascertain about the effect the topology design would have
on the networks connectivity. We evaluate the impact of
topology on routing performance and we present the
simulation results in the following section.

III Simulation and Performance
Evaluation

Simulations were done using the ns2.32 simulator
[15], where the underlying MAC protocol used is 802.11.
The routing protocol we have used is AODV [16]. We
have taken a deployment of
50 nodes in an area measuring 1500m 1500m.
A percentage of the 50 nodes are kept stationary and the
rest are mobile. The radio range of each node is 250m.
The simulation is run for varying maximum speeds of the
mobile nodes and a pause time of 0 second. Every point in
the graphs is an average of the results obtained from
simulating
20 different topologies using 20 connections for a period
of 500 seconds. Each data packet is of size 128 bytes and
the traffic type used is constant bit rate at the rate of 4
packets per second. The movement model used for the
mobile nodes is the random waypoint model [17].
To compare the routing performance of the case
Figure 1. Grid deployment of stationary
nodes; the empty and filled circles denote
stationary and mobile nodes respectively

when all nodes are mobile with that when some nodes are
stationary, we simulated AODV in the situations when
varying number of nodes out of the 50 nodes are made
stationary. We consid-
ered the cases when 0(0%), 5(10%), 10(20%),
15(30%), 20(40%) and 25(50%) nodes are sta-
tionary. First we simulated for the case when the stationary
nodes are arranged in a grid. We placed the nodes in a grid
in such a way that the horizontal as well as the vertical
distance between any two nodes is 250m. Moreover, the
grid is placed around the middle of the network area.
Second, we placed the stationary nodes randomly in the
network area using the CMU topology generation utility
that comes along with ns2.32 simulator.
We have considered six performance measurement
parameters:

(a) Packet Delivery Ratio (PDR): Ratio of the total
number of data packets received by the destinations
to the total number of data packets sent.

(b) Jitter: Average Variation in the received time of data
packets.
J
i
t
t
e
r

(
m
s
)

P
a
c
k
e
t

D
e
l
i
v
e
r
y

E
n
d
-
t
o
-
E
n
d

d
e
l
a
y

1

0.9

100
0

90
0

80
0

70
0

60
0

No Grid

Grid-5
Grid-10
Grid-15
Grid-20
Grid-25

0.8

0.7
500

400

300

0.6
200

100

0.5

0.4

No Grid

Grid-5
Grid-10
Grid-15
Grid-20
Grid-25

0
0 2 4 6 8 10 12
14
Maximum
speed (m/s)

0 2 4 6 8 10 12 14
Maximum
speed (m/s)

Figure 2. Grid deployment; Packet delivery
ratio Vs. Maximum speed

0.04

0.035
Figure 4. Grid deployment; End-to-end
packet delay Vs. Maximum speed

(c) End-to-end Delay: Average of the delay (received
time minus sent time) of every data packet.

(d) Route Frequency: The number of route requests
generated per second.

(e) Routing Load: The number of control packets
transmitted (including the ones that are forwarded)
for every data packet received.

0.03

0.025

0.02

0.015

0.01

0.005

0
No Grid

Grid-5
Grid-10
Grid-15
Grid-20
Grid-25

0 2 4 6 8 10 12 14
Maximum
speed (m/s)
(f) Average Throughput: The total amount of data
received by all nodes per second.

Fig. 2 shows the comparison of packet delivery
ratio(PDR) when the stationary nodes are placed in the
form of a grid. The label No Grid denotes the plot when
all the nodes are mobile (i.e., there are no stationary
nodes). Similarly Grid-5 denotes the plot when there
are 5(10%) stationary nodes among the 50 nodes. The
other labels carry similar meaning. It is seen that as the
percentage

Figure 3. Grid deployment; Jitter Vs. Maxi-
mum speed
of stationary nodes increases the packet delivery
ratio increases significantly. This is as expected as the
connectivity in the network increases with the increase in
the size of the grid of stationary nodes. It is also noticed
that the PDR decreases as the mobility increases, which is
as expected.
R
o
u
t
i
n
g

L
o
a
d

R
o
u
t
e

F
r
e
q
u
e
n
c
y

T
h
r
o
u
g
h
p
u
t

(
M
b
)

0.1

3

2.5

No Grid

Grid-5
Grid-10
Grid-15
Grid-20
Grid-25
0.08

0.06

2

1.5

0.0
4

0.0
2

No Grid

Grid-5
Grid-10
Grid-15
Grid-20
Grid-25

1

0.5
0
0 2 4 6 8 10 12
14
Maximum
speed (m/s)

0
0 2 4 6 8 10 12 14
Maximum
speed (m/s)

Figure 5. Grid deployment; Route frequency
Vs. Maximum speed

2

Figure 7. Grid deployment; Average throughput
(Mb/sec) Vs. Maximum speed

Fig. 3 shows the jitter with respect to maximum speed for
varying percentage of stationary nodes. It is seen that the
jitter reduces as the no. of stationary nodes increases.
This is expected as the network becomes more well
connected and stable as the no. of stationary nodes
increases. The end-to-end delay comparison is presented
in Fig.
4. The delay decreases rapidly as the The end-to-

1.5

1

0.5

0
No Grid

Grid-5
Grid-10
Grid-15
Grid-20
Grid
-25

0 2 4 6 8 10 12 14
Maximum
speed (m/s)
end delay comparison is presented in Fig. 4. The delay
decreases rapidly as the percentage of stationary nodes
increases. When 50% of the nodes are stationary, the delay
becomes lower than even
100ms. Fig. 5 shows the the route frequency for
varying number of stationary nodes. It is seen that the
number of route requests generated per second
decreases as the number of stationary nodes increases.
When more number of nodes are stationary and they are
deployed in grid, the number of link breaks reduces.
Consequently, the number of route requests per second
reduces since a

Figure 6. Grid deployment; Routing load Vs.
Maximum speed
link break potentially results in establishment of
new routes. Fig. 6 illustrates the comparison of the
routing load. As expected, the routing load reduces as the
number of stationary nodes increases. When there are more
number of stationary nodes forming a grid, the routes are
more stable and so
J
i
t
t
e
r

(
m
s
)

P
a
c
k
e
t

D
e
l
i
v
e
r
y

E
n
d
-
t
o
-
E
n
d

d
e
l
a
y

1000

90
0

1

0.9

800

700

600

0.8

0.7

0.6

0.5

0.4

No
Random
Rando
m-5
Rando
m-10
Rando
m-15
Rando
m-20
Rando
m-25
50
0

40
0

30
0

20
0

10
0

0

No Random

Random-5
Random-10
Random-15
Random-20
Random-25

0 2 4 6 8 10 12
14
Maximum
speed (m/s)

0 2 4 6 8 10 12 14
Maximum
speed (m/s)

Figure 8. Random deployment; Packet delivery
ratio Vs. Maximum speed

0.05
Figure 10. Random deployment; End-to-end
packet delay Vs. Maximum speed

the number of control packets generated and forwarded is
less as compared to when there are less number of
stationary nodes. Finally, the throughput increases as the
number of stationary nodes increases (Fig. 7). This is
expected since the PDR also increases as the number of
stationary nodes increases.
The second part of the simulation involves the random
deployment of the stationary nodes in-

0.0
4

0.0
3

0.0
2

0.0
1

0

No Random

Random-5
Random-10
Random-15
Random-20

Random-25

0 2 4 6 8 10 12
14
Maximum
speed (m/s)
stead of arranging them in a grid. The above
six performance parameters were used to evaluate this
scenario. Fig. 8 to Fig. 12 show the performance for
varying number of stationary nodes, which are deployed
randomly. We observe that there seem to be no notable
impact on the performance w.r.t. all the performance
parameters. Thus we conclude that improve in
performance can clearly be achieved by deploying the
stationary nodes in a grid whereas there is more or less
no effect on the performance when they are deployed
randomly.
Figure 9. Random deployment; Jitter Vs.
Maximum speed

IV Conclusions

We have presented the impact of topology of
stationary nodes in a partially mobile ad hoc
R
o
u
t
i
n
g

L
o
a
d

R
o
u
t
e

F
r
e
q
u
e
n
c
y

T
h
r
o
u
g
h
p
u
t

(
M
b
)

0.1

3.5 0.08

3
0.0
6

2.5

2

1.5

0.0
4

0.0
2

No Random

Random-5
Random-10
Random-15
Random-20
Random-25

1

0.5
No Random

Random-5
Random-10
Random-15
Random-20
Random-25

0
0 2 4 6 8 10 12
14
Maximum
speed (m/s)

0
0 2 4 6 8 10 12
14
Maximum
speed (m/s)

Figure 11. Random deployment; Route fre-
quency Vs. Maximum speed

3.
5

3

Figure 13. Random deployment; Average
throughput (Mb/sec) Vs. Maximum speed

network. First, we considered the case when the
stationary nodes are arranged in a grid. Second, we
also investigated the scenario when the stationary nodes
are deployed randomly. We observed that the routing
performance can be increased significantly by increasing
the number of the stationary nodes in the grid topology.
This would mean that instead of adopting elaborate ap-
proaches such as modifying the routing protocol
or deploying powerful and expensive backbone

2.5

2

1.5

1

0.5

0
No Random
Random-5

Random-10
Random-15
Random-20
Random-25

0 2 4 6 8 10 12 14
Maximum
speed (m/s)
nodes, the routing performance of a MANET can be
enhanced simply by making some nodes stationary and
arranging them in a grid. How- ever, we found that
randomly distributing the stationary nodes in the
network area does not help in improving the routing
performance. Thus, we conclude that to enhance the
performance, stationary nodes must not be just stationary
but must be arranged is some regular fashion so that
it helps in improving the connectivity of the
Figure 12. Random deployment; Routing load
Vs. Maximum speed
network.

Acknowledgment: This work is supported by Re-
search Promotion Scheme of AICTE, India (project no.
8023/BOR/RID/RPS-233/2008-09, 31/3/2009).

References

[1] B. Ishibashi and Raouf Boutaba, Toplogy and
mobility considerations in mobile ad hoc
networks, Elsevier Ad Hoc Networks, 3 (2005)
762-776.

[2] J. Wellons and Y. Xue, The Impact of Topol- ogy in
Robust Routing on Wireless Mesh Networks,
Proc. Mobile Computing and Communications
Review, 2010, pp. 46-48.

[3] Y. Xin, T. Guven, M. A. Shayman, Topology design
for wireless ad-hoc networks with backbone
support, Proc. Conference on Information Sciences
and Systems, 2005.

[4] S. Perur and S. Iyer, Sparse Multi-hop wireless for
voice communication in rural India, Proc. 12th
National Conference on Communications, NCC
2006, New Delhi, India; pp. 534-538, January 2006.

[5] M. Cardei, J. Wu and S. Yang, Topology control
in ad hoc wireless networks using cooperative
communication, IEEE Transactions on Mobile
Computing, vol. 5, no. 6, June 2006.

[6] A. Rahman and P. Gburizynski, MAC- assisted
topology control for ad hoc wireless networks,
International Journal of communication systems,
2005;00;1-6.

[7] S. Agarwal, S. Krishnamurthy, R. Katz, and S. Dao,
Distributed power control in ad-hoc wireless
networks, Proc. PIMRC, 2001.

[8] T. Kwon and M. Gerla, Clustering with power
control, Proc. MILCOM, vol. 2, pp. 1424-1428,
1999.

[9] R. Ramanathan and R. Rosales-Hain, Topology
control of multihop wireless networks using
transmit power adjustment, Proc. of IEEE
INFOCOM 2000, pp. 404-413, Tel Aviv, Israel,
March 2000.
[10] L. Li, J. Y. Hlapern, P. Bahl, Y. Wang, and R.
Watenhofer, Analysis of a con- based topology
control algorithm for wireless multi-hop
networks, Proc. ACM Sym- posium on Principle of
Distributed Comput- ing (PODC), 2001.

[11] N. Li, J. C. Hou, and L. Sha, Design and analysis
of an MST-based topology control algorithm,
Proc. INFOCOM, 2003.

[12] V. Rodoplu and T. Meng, Minimum energy mobile
wireless networks, IEEE Journal of Selected
Areas in Communications, 17(8), pp. 1333-1344,
1999.

[13] R. Wattenhofer, L. Li, P. Bahl, and Y.- M. Wang,
Distributed topology control for wireless multihop
ad-hoc networks, Proc. INFOCOM 2001, pp. 1388-
1397, 2001.

[14] Z. Huang, C. Shen, C. Srisathapornphat, and C.
Jaikaeo, Topology control for ad hoc networks
with directional antennas, Proc. IEEE Int.
Conference on Computer Communications and
Networks, pp. 16-21, 2002.

[15] The Network Simulator - ns2,
www.isi.edu/nsnam/ns/.

[16] C. E. Perkins and E. M. Royer, Ad-hoc On-
Demand Vector Routing, Proc. 2nd IEEE
Workshop on Mobile Computing Systems and
Applications, pp. 90-100, New Orleans, LA,
February 1999.

[17] D. B. Johnson and D. A. Maltz, Dynamic Source
Routing in Ad Hoc Wireless Net- works, Mobile
Computing (T. Imielinski and H. Korth, editors,
Kluwer Academic Publishers, pp. 153-181, 1996.
V
c
f
a
w
b
a
a
d
f
t
S
w
a
t
s
c
m
p
c
s
o
F
l
c
T
t
(
e
f
n
W
a
P
w
d
t
p
a
F

a
c
t
d
m
Proc. of the
Volume 1. Cop
ISBN: 978-81
A
Abstr
communication
fading. Orthog
and orthogona
which are m
become widely
against freque
advantages, O
drawbacks; hi
for an adaptive
the channel, an
Single carrier
which utilizes
and frequency
technique that
same overall s
carrier frequ
modified form
promising te
communication
similar throug
overall comple
FDMA is the
lower than tha
candidate for
Term Evolutio
the 3rd Genera
(3GPP LTE).
In th
explain how SC
frequency dom
needed an alte
We present it
analyze the eff
PAPR. Among
we find that
dependent sch
than interleav
performance o
also explain th
FDMA and OF

On the le
are depicted
converted to
then in the fr
desired locati
must then be c
e Intl. Conf. on
pyright 2012
1-920575-7-6
ACM #: dber.i
A S
ract- Broa
ns suffer from
gonal frequenc
al frequency di
multicarrier co
y accepted pri
ncy selective fa
OFDM and O
gh peak to-ave
e or coded sche
nd high sensitiv
frequency divi
single carrier
y domain equ
t has similar
structure as th
ency division
m of Orthog
echnique for
ns in future c
ghput perform
exity as OFDM
peak-to-averag
at of OFDMA
the uplink mu
on of cellular
ation Partners
is paper, we gi
C-FDMA utiliz
main equalizat
ernative of such
ts basic conce
fects of subcar
g the possible
localized FD
heduling (CDS
ved FDMA (
of IFDMA is b
he similarities a
FDMA.
ft hand side o
in the time
the frequency
frequency dom
on in the ove
converted bac
n Computer A
2 Techno Foru
:: doi: 10. 739
imera.10. 739
Study on S

adband w
m multipath
cy division mu
ivision multiple
ommunication
imarily becaus
ading channels
OFDMA suff
erage power ra
eme to overcom
vity to frequen
ision multiple
r modulation
ualization at
performance a
hose of an OFD
n multiple ac
gonal FDMA
r high da
cellular system
mance and es
MA. A principa
ge power ratio
. SCFDMA is
ultiple access s
systems unde
hip Project Lo
ive an overview
zes single carri
tion technique
h an advance O
ept on which
rrier mapping
subcarrier ma
DMA (LFDM
S) results in
(IFDMA). How
better than th
and main differ
of this Figure t
e domain. Th
y domain usin
main they are
erall carrier b
ck to the time
Applications
um Group, Ind
963/ISBN_076
63
SC-FDMA
Ch
R
Singh
wireless m
frequency-sele
ultiplexing (OF
e access (OFDM
techniques,
se of its robust
s .Despite the m
fer a numbe
atio (PAPR), a
me spectral nu
ncy offset.
access (SC-FD
at the transm
the receiver
and essentially
DMA system S
ccess (SCFDM
(OFDMA),
ata rate up
ms. SC-FDMA
sentially the
al advantage of
o (PAPR), whi
currently a st
scheme in the L
er consideratio
ong Term Evolu
w of SC-FDMA
ier Modulation
es simply. Wh
OFDMA techn
it works. We
on throughput
apping approa
MA) with cha
higher throug
wever, the P
at of LFDMA
rences between
the data symb
he symbols a
ng an FFT, a
e mapped to t
bandwidth. Th
domain in ord
dia.
68
A Advanc
handra Shek
Research Scho
hania Univers
obile
ective
FDM)
MA),
have
tness
many
r of
need
lls in
DMA)
mitter
is a
y the
Single
MA),a
is a
plink
A has
same
f SC-
ich is
trong
Long
on by
ution
A.We
n and
y we
nique.
also
t and
aches,
nnel-
ghput
PAPR
A. We
n SC-
Ke
Pe

In
co
sy
the
nu
ph
ac
Co
Or
do
(S
co
sy
sy
FD
do
ac
alt
wh
po
ge
ols
are
and
the
hey
der
b
p
e Access S
khar
olar
sity, Raj.

eywords:
erformance Me
I.
n order to
ommunications
ystems, the 3rd
e Long Ter
umerous differ
hysical layer (
ccess schemes
ode Division
rthogonal Fre
ownlink and S
SC-FDMA) for
SC-FD
ombines the lo
ystems - such
ymbol time a
DMA is an ex
omain equaliz
ccess. SC-FD
ternative to O
here lower PA
ower efficienc
eneration are s

Figur
to have the c
An alternative
Transform Sp
Single-ca
division mult
linearly preco
guard interva
between bloc
eliminate tim
propagation)
Scheme fo
Channel-Depe
easures, SC-FD
SINGLE-CAR
MULTI
transition fr
s systems to m
d Generation P
rm Evolution
rences betwee
(PHY), specif
. While its pa
n Multiple A
equency Div
Single-Carrier
r its uplink.[1]
DMA is a
ow PAR chara
as those used
and flexible f
xtension of sin
zation (SC/F
MA has dra
OFDMA, espe
APR greatly b
cy. [2][3][4].T
shown in Figur
re1: SC-FDM
cyclic prefix i
e name for S
pread OFDM (
arrier FDMA
tiple access s
oded OFDMA
als with cycl
cks of symb
me spreadin
among the bl
or 4G
endent, IFDM
DMA Vs. OFDM
RRIER FREQ
IPLE ACCES

from today's
meet the need
Partnership Pro
n (LTE) spe
en these gener
fically in the
arent generatio
Access (CDM
ision Multipl
r Frequency-D
].
hybrid transm
acteristics of si
d for GSM and
frequency all
ngle carrier m
FDE) to acco
awn great att
ecially in the
enefits the mo
The principles
re.
MA signal gene
nserted prior
C-FDMA is D
(DFT-SOFDM
(SC-FDMA)
scheme. SC-F
A scheme. Just
lic repetition
bols in view
ng (caused
ocks. In OFD
DMA, LFDM
MA.
QUENCY-DIV
S (SC-FDMA
3rd genera
ds of 4
th
gene
oject (3GPP) h
ecification. A
rations are cha
modulation a
on relied on v
MA), LTE
lexing (OFD
Division Mult
mission sche
ingle-carrier t
d CDMA - w
ocation of O
modulation with
ommodate m
tention as an
e uplink comm
obile terminal
behind SC FD
eration
to transmissio
Discrete Four
M).[5]
) is frequenc
FDMA can as
t like in OFD
are introduc
w to efficien
by multi-pa
DM, Fast Four
MA, PAPR,
VISION
A)
ation (3G)
eration (4G)
has released
Among the
anges in the
and multiple
variations of
implements
M) for its
tiple Access
eme which
transmission
with the long
OFDM. SC-
h frequency
multiple-user
n attractive
munications
in terms of
DMA signal
on.
rier
cy-
s a
DM,
ced
ntly
ath
rier
transform (FFT) is applied on the receiver side on each
block of symbols, and inverse FFT (IFFT) on the
transmitter side. In SC-FDE, both FFT and IFFT are
applied on the receiver side, but not on the transmitter
side. In SC-FDMA, both FFT and IFFT are applied on
the transmitter side, and also on the receiver side. In SC-
FDMA, multiple access is made possible by inserting
silent fourier-coefficients on the transmitter side before
the IFFT, and removing them on the receiver side before
the IFFT.
What makes SC-FDMA a single carrier modulation
is the fact that there is just one carrier. PAPR of
SC/FDMA is 2dB smaller compared to OFDMA.
Single Carrier Frequency Division Multiple Access
(SC-FDMA)
A new single carrier multiple access technique
Utilizes single carrier modulation and frequency
domain equalization.
Has similar performance and essentially the same
overall structure as those of OFDMA system
Has low PAPR because of its inherent single
carrier transmitter structure.
An attractive alternative to OFDMA, especially in
the uplink communications where lower PAPR
greatly benefits the mobile terminal in terms of
transmit power efficiency.
The basic uplink transmission technique is single-
carrier transmission (SC-FDMA) with cyclic prefix to
achieve uplink inter-user orthogonality and to enable
efficient frequency domain equalization at the receiver
side

1. History:

Over the past fifteen years, the bit rates achieved in
cellular and local area wireless communication systems
have increased steadily. The earliest digital cellular
systems, North American TDMA and GSM employed
time division multiple access. Second generation
CDMA systems and the two third generation cellular
systams all use direct sequence spread spectrum for
multiplexing and muliplu access The highest bit
rates in commercially deployed wireness systems are
achieved by means of Orthogonal requency Division
Multiplexing (FDM) in wipeless LANs bas%d on the
IEEE 802.11a and IEEM 802.11g standards. The next
advance in cellular systems, under investigation bi the
Third Generation Partnership Project (3GPP), also
anticipates the adoption of OFDA to achieve higher bit
rates. Now in fourth Generation we use SC-FDMa to
achieve maxhmum performance with a single
carrier.[6].`

3. Why Needed

Today multi-carries transmission And OFDM are meanw
to allow for very high overall transmission bandwidth
while still being robust to signal corruption due to radio-
channel frequency selectyvity. However, a drawback of
OFDM modulation, as well as any kind of multi-carrier
transmission, ir the large variations in the instantaneous
Power of the transmitted signal. Such power variations
imply a reduced power-amplifier efficiency and higher
power-amplifier cost. This is especially critical for the
uplink, due to the high importance of low mobile-
terminal power consumption and cost. Several methods
have been proposed on how to reduce the large power
variations of an OFDM signal. However, most of these
methods have limitations in terms of to what extent the
power variations can be reduced. Furthermore, most of
the methods also imply a significant computational
complexity and/or a reduced link performance.[7] Thus,
there is an interest to consider also wider-band single-
carrier transmission as an alternative to multi-carrier
transmission, especially for the uplink, i.e. for mobile-
terminal transmission. One of such single-carrier
transmission scheme can be implemented using DFT-
spread OFDM.

II. SC-FDMA/DFT-spread OFDM:

DFT-spread OFDM
DFT-Spread OFDMA Mapping of spread symbols , not
original symbols to subcarriers.
SC-FDMA can be regarded as discrete Fourier
transform (DFT)-spread OFDMA, where time domain
data symbols are transformed to frequency domain by
DFT before going through OFDMA modulation.
DFT-spread OFDM (DFTS-OFDM) is a
transmission scheme that can combine the desired
properties for uplink transmission i.e.
Small variations in the instantaneous power of the
transmitted signal (single carrier property).
Possibility for low-complexity high-quality
equalization in the frequency domain.
Possibility for FDMA with flexible bandwidth
assignment.
Due to these properties, DFTS-OFDM has been
selected as the uplink transmission scheme for LTE,
which is the long-term 3G evolution.

5. How SC-FDMA Work?

The transmitter of an SC-FDMA system converts a
binary input signal to a sequence of modulated
subcarriers. At the input to the transmitter, a baseband
modulator transforms the binary input to a multilevel
sequence of complex numbers xn in one of several
possible modulation formats. The transmitter next groups
the modulation symbols {xn} into blocks each containing
N
s
f
s
o
t
s
s
i
s
t
a
i
a
x
p
t
r
o
p
w
a
t
B
i
v
F
D
s
p
i
p
f
r
o
e
g
t
c
r
c
t
c
t
b
t
a
s
D
D
m
t
u
N symbols. T
subcarriers is
frequency do
symbols. It th
of the M (>
transmitted. I
symbols per
simultaneous
interference. Q
symbol sequen
the set l X (
amplitudes, w
in OFDMA, a
amplitudes to
xm then are
performs two
transmission.
How
remaining inp
of the IDFT
properties, i.e
with a bandw
assuming a sa
the nominal b
BW =M/N *
instantaneous
varied, allow
Furthermore,
DFT outputs
shifted in the f
It ins
prefix (CP) in
inter-block
propagation.
filtering opera
reduce out-ofb
of the last par
each block fo
guard time be
the CP is long
channel, or ro
response, then
copy of the la
time linear c
convolution. T
the channel c
between the
transmitted da
a point-wise
samples. The
DFT of the re
DFT of the c
more sophis
technique can
used pulse sha
The first step i
to perform a
omain repres
hen maps each
> N) orthogo
If N = M/Q a
r block, the
transmissi
Q is the bandw
nce. The resul
(l = 0, 1, 2
where N of the
an M-point ID
a complex tim
e transmitted
other signal p

wever, if M
puts to the IDF
T will be a
e. a signal wit
width that depe
ampling rate
bandwidth of
fs. Thus, by
bandwidth of
wing for flex
by shifting th
are mapped,
frequency dom
serts a set of s
n order to pro
interference
The transmi
ation referred t
band signal en
rt of the block,
or a couple of
etween succes
ger than the m
oughly, the le
n, there is no
ast part of the
convolution in
Thus transmit
can be model
e channel im
ata block, whi
multiplicatio
en, to remove
eceived signal
channel impu
sticated frequ
n be impleme
aping filter is t
in modulating
an N-point DF
sentation Xk
h of the N DFT
onal subcarrie
and all termi
e system c
ons withou
width expansi
lt of the subca
, M-1) of com
e amplitudes a
DFT transform
me domain sig
sequentially.
processing op
is smaller th
FT are set to
signal with
th low power
ends on M. M
fs at the outp
the transmitte
varying the b
f the transmitt
xible-bandwid
he IDFT inpu
the transmitte
main.[8]
symbols referr
ovide a guard
(IBI) due
tter also per
to as pulse sha
nergy. In gener
, which is add
f reasons. Fir
ssive blocks.
maximum dela
ength of the c
o IBI. Second
e block, it con
nto a discret
tted data prop
led as a circu
mpulse resp
ch in the frequ
on of the D
e the channel
can simply b
ulse response
uency domai
ented. One of
the raised-cos
g the SC-FDM
FT to produce
of the inp
T outputs to o
ers that can
inals transmit
can handle
ut co-chann
ion factor of t
arrier mapping
mplex subcarr
are non-zero.
ms the subcarr
gnal xm . Ea
The transmit
perations prior
han N and t
zero, the outp
single-carri
r variations, a
More specifical
put of the IDF
ed signal will
block size M t
ted signal can
dth assignme
uts to which t
ed signal can
red to as a cyc
time to preve
to multipa
rforms a line
aping in order
ral, CP is a co
ded at the start
rst, CP acts as
If the length
ay spread of t
channel impu
d, since CP is
nverts a discr
te time circu
pagating throu
ular convoluti
ponse and t
uency domain
DFT frequen
l distortion, t
e divided by t
point-wise or
in equalizati
f the common
ine filter.
MA
e a
put
one
be
N
Q
nel
the
g is
rier
As
rier
ach
tter
r to
the
put
ier
and
lly,
FT,
be
the
be
ent.
the
be
clic
ent
ath
ear
r to
opy
t of
s a
of
the
ulse
s a
ete
ular
ugh
ion
the
n is
ncy
the
the
r a
ion
nly
The m
a multi-carrie
reduced varia
implying the
efficiency.

DFT: Discrete
IDFT: Inverse
CP: Cyclic Pr
PS: Pulse Sha
DAC: Digital-
RF: Radio Fre
ADC: Analog
Figure2: Tran
FDMA/DFTS

Two approach
1 Distribu
(IFDMA)
2. Localize
Ther
subcarriers am
(LFDMA), e
subcarriers to
of an LFDMA
the system b
distributed SC
terminal are
realization o
main benefit o
er transmission
ations in the i
possibility fo
e Fourier Tran
e Discrete Fou
refix
aping
-to-Analog Co
equency
g-to-Digital Co
nsmitter and R
S-OFDM
6. Apportioni
hes to apportio
uted SC-FDM
d SC-FDMA
re are two
mong termin
each termina
o transmit its
A transmissio
bandwidth. Th
C FDMA in w
spread over
of distributed
of DFTS-OFD
n scheme suc
instantaneous
for increased
nsform
urier Transform
onversion
onversion
Receiver Struct
ing Subcarrier
oning subcarri
MA or Inter
(LFDMA)
approaches t
nals. In locali
al uses a s
symbols. Thu
on is confined
he alternative
which the subc
the entire si
d SC-FDMA
DM, compared
ch as OFDM,
transmit pow
power-amplif
m
ture of SC-
rs:
iers
rleaved FDM
to apportioni
ized SC-FDM
set of adjac
us the bandwid
to a fraction
e to LFDMA
arriers used by
gnal band. O
is interleav
d to
, is
wer,
fier

MA
ing
MA
ent
dth
of
is
y a
One
ved
F
e
m
a
t
a
l
l
l
m
(
m
c
I
e
s
I
s
s
C
c
s
p
s
d
p
l
s
T
o
P
d
I
F
(
u

W
FDMA (IFD
equidistant fro
In o
mapping mod
allocated over
the unused su
are occupied b
localized subc
localized sub
localized FDM
mapping mod
(DFDMA). T
mode with eq
called Interle
IFDMA is a
efficient in th
strictly in the
IDFT.
From
subcarrier ma
static and chan
CDS assigns
channel frequ
scheduling m
provides freq
signal is sp
distributed m
performance.
localized sub
significant mu
Two subcarri
other.
Distributed
PAPR in case
Localized:
dependent sch
IFDMA but si
Figure3: Subc
(3 users, 12 s
user).
With respect t
System thr
DMA) where
om each other
other way, in
de, DFT out
r the entire ban
ubcarriers, wh
by the DFT ou
carrier mappin
bcarrier mapp
MA (LFDMA
de of SC-FD
The case of M
quidistance bet
eaved FDMA
special case
hat the transm
time domain
m a resource
apping metho
nnel-dependen
s subcarriers
uency respon
methods, dist
quency divers
read over th
mapping, CD
By contrast,
bcarrier map
ulti-user divers
ier mapping
d: Frequency
of IFDMA.
Frequency
heduling, hig
imilar PAPR t
carrier allocati
subcarriers, an
7. Performa
to the two maj
roughput
e occupied
[9].
n the distrib
tputs of the
ndwidth with
hereas consecu
utputs of the i
ng mode. We
ping mode of
A) and distrib
DMA as dist
M = Q N for
tween occupie
(IFDMA) [
of SC-FMDA
mitter can mod
n without the u
e allocation
ods are furth
nt scheduling
to users ac
nse of each
tributed subc
sity because
he entire ba
DS incremen
CDS is of gr
pping becaus
sity.
have advanta
y diversity,
selective gain
gher PAPR th
to the case of D
ion methods f
nd 4 subcarri

ance Measures
jor performanc
subcarriers a
buted subcarr
input data a
zeros occupyi
utive subcarri
input data in t
will refer to t
f SC-FDMA
buted subcarr
tributed FDM
r the distribut
ed subcarriers
10], [11], [1
A and it is ve
dulate the sign
use of DFT a
point of vie
her divided in
(CDS) metho
ccording to t
user. For bo
carrier mappi
the transmitt
andwidth. W
ntally improv
reat benefit w
se it provid
ages over ea
lower
n with chann
han the case
DFDMA.

for multiple us
iers allocated
s:
ce indicators,
are
rier
are
ing
ers
the
the
as
rier
MA
ted
s is
2].
ery
nal
and
ew,
nto
ds.
the
oth
ing
ted
With
ves
with
des
ach
nel
of
sers
per
p
u
u
b
b
u
PAPR

7.1 System Th
While PAPR
information
indicator of
throughput in
information sy
With respect
determines th
against freq
information i
Therefore it o
On the other
multi-user d
selective fadin
portion of the
transmission
user diversity
dispersed tra
dependent sc
requires the s
function of
subcarrier as
frequency resp
For e
measures wit
channel depen
engineering sy
design parame
the lowest P
Therefore equ
use their judg
specific needs
many users e
better off with
system with a
7.2 PAPR ana

The PAR is d
block (one O
signal power
distribution is
instantaneous
distributions h
variations of O
SC-F
OFDMA, w
transformed t
through OFD
users stems
different subc
the case of OF
is a single
compared to
multicarrier si
hroughput:
is a major c
throughput is
system perf
n SC-FDMA d
ymbols are ap
to immunity
hroughput),dis
quency selec
is spread acr
offers the adva
r hand, LFD
diversity in
ng if it assign
e signal band
characteristics
y relies on
ansmitters.[13]
cheduling (C
system to mon
frequency fo
ssignments to
ponses of all t
each configur
th static subc
ndent schedul
ystems there a
eters and perf
PAPR tend
uipment design
gment to find t
s. For exampl
each transmitt
h IFDMA, wh
a few high-bit-
alysis of SC-F
defined as the
OFDM symbol
r. It should
s not the sam
transmit
have often bee
OFDM.
FDMA can b
where time
to frequency d
DMA modulati
from the fa
carriers in the
FDMA. Becau
carrier signa
the case of
ignal.
concern in por
s an even m
formance. A
depends on th
pplied to subca
to transmissio
stributed SC-F
ctive fading
ross the enti
antage of freq
MA can pot
the presence
ns each user to
where that us
s (high chann
independent
] It also re
CDS) of su
nitor the chan
or each termi
o changes i
the terminals.
ration, we pre
carrier assignm
ling. We find
are complex tr
formance. Con
to have low
ners and syste
the best trade
le, we find tha
ting at a mod
hile LFDMA w
-rate users.
DMA:
peak power w
l) normalized
d be noted
me as the dis
power. Hist
en used to illu
be regarded
domain data
domain by DF
ion. The orth
act that each
frequency do
use the overal
al, PAPR is
OFDMA wh
rtable termina
more import
As in OFDM
he way in whi
arriers.
on errors (whi
FDMA is rob
g because
re signal ban
quency diversi
tentially achie
e of frequen
o subcarriers in
ser has favorab
nel gain). Mu
fading amo
equires chann
ubcarriers. CD
nnel quality a
inal, and ad
in the chann
esent throughp
ments and w
that as in oth
radeoffs betwe
nfigurations w
wer throughp
em operators c
off to meet th
at a system w
derate bit rate
works better in
within one ID
d by the avera
that the PA
stribution of t
torically, PA
ustrate the pow
as DFT-spre
a symbols a
FT before goi
ogonality of t
h user occup
omain, similar
ll transmit sign
inherently l
hich produces
als,
ant
MA,
ich
ich
ust
its
nd.
ity.
eve
ncy
n a
ble
ulti-
ong
nel-
DS
s a
apt
nel
put
with
her
een
with
put.
can
heir
with
e is
n a
FT
age
AR
the
AR
wer
ead
are
ing
the
pies
r to
nal
ow
s a
High PAPR requires expensive and inefficient
power amplifiers with high requirements on linearity,
which increases the cost of the terminal and also drains
the battery faster.SC-FDMA solves this problem by
grouping together the resource blocks in such a way that
reduces the need for linearity, and so power
consumption, in the power amplifier. A low PAPR also
improves coverage and the cell-edge performance.
In this section, we analyze the PAPR of the SC-FDMA
signal. We assume that the total number of subcarriers is
M = Q N, where N is the number of subcarriers per
block. The integer Q is the maximum number of
terminals that can transmit simultaneously. For
distributed subcarrier mapping, we consider the case of
IFDMA with subcarriers equally spaced over the system
bandwidth.
The peak power of any signal x(t) is the
maximum of its squared envelope [x(t)]^2. However, for
a continuous random process, max [x(t)]^2 could be
unbounded. Even for random process with discrete
values where max [x(t)]^2 is bounded, it may occur at
very low probability which is not very useful in practice.
Rather, the distribution of [x(t)]^2 is more useful and we
describe it with the complementary cut-off probability.
The PAPR is defined as the ratio of peak power
to average power of the transmitted signal in a given
transmission block. Without pulse shaping, that is, using
rectangular pulse shaping, symbol rate sampling will
give the same PAPR as the continuous time domain case
since an SC-FDMA signal is modulated over a single
carrier.
To evaluate PAPR of individual system
configurations, we have simulated the transmission of
105 blocks of symbols. After calculating PAPR for each
block, we present the data as an empirical CCDF
(Complementary Cumulative Distribution Function). The
CCDF is the probability that PAPR is higher than a
certain PAPR value PAPR (Pr {PAPR > PAPR}).
We also relate the PAPR to RF transmit power amplifier
efficiency. In an ideal linear power amplifier where
linear amplification is achieved up to the saturation
point, maximum power efficiency is achieved when the
amplifier is operating at the saturation point. To prevent
distortion in the presence of PAPR, transmit power
backoff is needed to operate the power amplifier in the
linear region. SC-FDMA signals indeed have lower
PAPR than OFDMA signals. Also, LFDMA incurs
higher PAPR compared to IFDMA but, compared to
OFDMA; it is lower, though not significantly. Another
noticeable fact is that pulse shaping significantly
increases the PAPR of IFDMA. A pulse shaping filter
should be designed carefully in order to limit the PAPR
without degrading the system performance. In general,
IFDMA is more desirable than LFDMA in terms of
PAPR and power efficiency.

8. Channel-Dependent Scheduling (CDS) for Uplink SC-
FDMA:

CDS assigns subcarriers to users according to the
channel frequency response of each user. A key question
of CDS is how we should allocate time and frequency
resources fairly among users while achieving multi-user
diversity and frequency selective diversity. To do so, we
introduce utility-based scheduling where utility is an
economic concept representing level of satisfaction. we
consider two different utility functions: aggregate user
throughput for maximizing system capacity and
aggregate logarithmic user throughput for maximizing
proportional fairness [14],[15]. The objective is to find
an optimum chunk assignment for all users in order to
maximize the sum of user utility at each transmit time
interval (TTI). If the user throughput is regarded as the
utility function, the resource allocation maximizes rate-
sum capacity ignoring fairness among users. Therefore,
only the users near the base station who have the best
channel conditions occupy most of the resources. On the
other hand, setting the logarithmic user data rate as the
utility function provides proportional fairness.
In considering the optimization problem of CDS
for multicarrier multiple access, it is theoretically
possible to assume that the scheduler can assign
subcarriers individually. Allocating individual
subcarriers is, however, a prohibitively complex
combinatorial optimization problem in systems with 256
subcarriers and on the order of ten terminals transmitting
simultaneously. Moreover, assigning subcarriers
individually would introduce unacceptable control
signaling overhead. In practice, the units of resource
allocation are chunks, which are disjoint sets of
subcarriers. As a practical matter chunk-based
transmission is desirable since the input data symbols are
grouped into a block for DFT operation before subcarrier
mapping. With regards to chunk structure, there is a
restriction on chunk selection for IFDMA such that all
assigned subcarriers should be equidistant in order to
maintain the lowest PAPR [16].
Even with subcarriers assigned in chunks,
optimum scheduling is extremely complex for two
reasons:
The objective function is complicated, consisting of
nonlinear and discrete constraints dependent on the
combined channel gains of the assigned subcarriers; and
There is a total transmit power constraint for each
user.
Instead of directly solving the optimization
problem, a sub-optimal chunk allocation scheme can be
used for both IFDMA and LFDMA to obtain most of the
benefits of CDS. For LFDMA, a greedy chunk selection
method can be applied where each chunk is assigned to
the user who can maximize the marginal utility when
occupying the specific chunk. For IFDMA, the benefit of
multi-user diversity can be achieved by selecting users in
order of the estimated marginal utility based on the
average channel condition over the entire set of
subcarriers. The users with higher channel gains may
occupy a larger number of chunks than users with lower
channel gains [17].
The advantage of channel dependent scheduling
over round robin scheduling increases as the number of
users increases. This is because the scheduler selects the
closer users who can transmit at higher data rate. If there
are more users, the possibility of locating users at closer
distance to the base station increases. As a result, the
CDS achieves significant improvements for both IFDMA
and LFDMA. In the case of logarithmic rate utility, the
CDS gain stops increasing beyond approximately 32
users. With 32 users, maximizing logarithmic rate utility
can increase system capacity by a factor of 1.8 for
LFDMA and 1.26 for IFDMA relative.
To static scheduling. the CDS scheme based on
the logarithm of user throughput as a utility function
provides proportional fairness whose gains are shared
among all users, whereas the CDS gains are concentrated
to the users near the base station when the user
throughput is considered as the utility function.
For static subcarrier assignment (round robin
scheduling), a system with users each transmitting at a
moderate data rate is better off with IFDMA while
LFDMA works better in a system with a few high date
rate users. Due to the advantages of lower outage
probability and lower PAPR, IFDMA is an attractive
approach to static subcarrier scheduling. For channel-
dependent scheduling, LFDMA has the potential for
considerably higher data rate. The results show that CDS
increases system throughput by up to 80% relative to
static scheduling for LFDMA but the increase is only
26% for IFDMA. The scheduling gains in LFDMA can
be exploited to reduce power consumption and PAPR by
using power control to establish a power margin instead
of increasing system capacity.

9. SC-FDMA Vs. OFDMA:

Comparison study between SC-FDMA and OFDMA can
be done in two ways
Similarities
Dissimilarities
Block Diagram.

9.1 Similarities:
Block-based modulation.
Divides the transmission bandwidth into smaller
subcarriers.
Channel equalization is done in the frequency
domain.
CP is added to overcome IBI (Inter-Block
Interference) and to convert linear convolution of the
channel impulse response to circular one.
SC-FDMA can be regarded as DFT-precoded or
DFT-spread OFDMA.

9.2 Dissimilarities:
The only difference is the presence of the DFT in SC-
FDMA. For this reason SCFDMA is sometimes referred
to as DFT-spread or DFT-precoded OFDMA. In terms of
data detection at the receiver, OFDMA performs it on a
per-subcarrier basis whereas SC-FDMA does it after
additional IDFT operation. Because of this difference,
OFDMA is more sensitive to a null in the channel
spectrum and it requires channel coding or power/rate
control to overcome this deficiency. Also, the duration of
the modulated time symbols are expanded in the case of
OFDMA with parallel transmission of the data block
during the elongated time period whereas SC-FDMA
modulated symbols are compressed into smaller chips
with serial transmission of the data block, much like a
direct sequence code division multiple access (DS-
CDMA) system.

9.2.1 Channel equalization:
OFDMA: Per subcarrier detection.=>A Null in
channel spectrum severely degrades the system
performance. >Channel coding or power/rate control is
required
SC-FDMA: Detection after IDFT. >SNR is
averaged out over entire bandwidth.

9.2.2 Time domain symbol duration:
OFDMA: Modulated symbol duration is longer than
input data symbol duration.
SC-FDMA: Modulated symbol duration is shorter than
input data symbol duration.

9.2.3 PAPER:
OFDMA: Parallel transmission of data over multiple
carriers. High PAPR.
SC-FDMA: Serial transmission of data over single
carrier. =>Low PAPR.

9.3 Data Symbols Diagram:
F
t

i
d

a
c
m
i
a
E

I
P
a

u
c
s
u
f
p
Figure 4: C
transmitting a
10. Ad
Single carrier
in which the
domain, which
SC-FDMA ha
alternative to
communicatio
mobile termin
is currently a
access scheme
Evolved UTR
Low PAP
Low sens
I would like
Professor Dr.
and encourage
SC-FDMA is
uplink commu
carrier FDMA
scheme that i
uplink. SC-FD
frequency d
performance a
omparison o
a series of QPS
dvantage of SC
FDMA (SC-F
data symbol
h results in a m
11.
as drawn gre
o OFDMA,
ons where low
nal in terms of
a working ass
e in 3GPP Lo
RA.
PR
sitivity to carri
ACKNOWL
e to thank m
Mohd. Huss
ement.
12. CON
a promising
unication in fu
A (SC-FDMA
is currently b
DMA utilizes
domain equa
and essentially
f OFDMA a
SK data symbo
CFDMA over O
FDMA) is a v
s are modula
more power ef
Usage:
eat attention
especially
wer PAPR gre
f transmit pow
sumption for
ong Term Evo
ier frequency o
LEDGEMENT
my official s
sain for all th
NCLUSION:
technique fo
future cellular
A) is a new
being adopted
single carrier
alization, and
y the same ov
and SC-FDM
ols
OFDMA:
variant of OFD
ated in the tim
fficient schem
as an attracti
in the upli
eatly benefits t
wer efficiency
uplink multip
olution (LTE),
offset
T
supervisors a
heir help, advi
or high data r
systems. Sing
multiple acce
d in 3GPP LT
modulation a
d has simi
verall complex

MA
DM
me
me.
ive
ink
the
. It
ple
or
and
ice
ate
gle
ess
TE
and
ilar
xity
b
p
p
n
w
as those of
benefittial for
power efficien
With
configuration,
choices that a
is shown tha
PAPR compa
focused on
throughput an
Among the po
find that loc
dependent sch
than interleav
performance o
of LFDMA.
Whe
to control ad
narrower diff
terms of PAP
that pulse shap
in the case
significant im
filter should b
PAPR withou
Work in prog
channel estim
of SC-FDMA
information a
channels linki
Channel estim
and changes
the performan
of the modula
subcarriers to
.

[1]. R. van
Communications
[2]. Hyung
Single Carrier
Vehicular Techno
[3]. H. Ek
Parkvall, J. Torsn
Long-Term Evol
2006, pp. 38-45.
[4]. E. Da
Karlsson, M. L
Evolution - Rad
IEEE Vehicular
Melbourne, Austr
[5]. Long
website, www.3g
[6]. Hujun
A Broadband Wi
2006 (IEEE): pp.
[7]. Jianw
and Randall A. B
OFDMA s
r the mobile t
ncy and manu
hin a spe
, there are m
affect perform
at SC-FDMA
ared OFDMA
the effects
nd peak-to a
ossible subcar
calized FDMA
heduling (CDS
ved FDMA (IF
of IFDMA is
n we conside
djacent chann
ference betwe
PR performanc
ping increases
of raised-co
mpact on PAPR
be designed ca
ut degrading th
gress is an in
mation error on
A. Effective sc
about the freq
ing terminals
mation errors a
in channel pr
nce of CDS by
ation techniqu
users
REFER
n Nee and R. Pra
, Artech House, 2
g G. Myung, Jun
FDMA for Upl
ology Magazine,
kstrm, A. Furu
ner, and M. Wahl
lution, IEEE C
ahlman, H. Eks
Lundevall, and
io Interface Con
r Technology
ralia, May 2006.
Term Evolution o
gpp.org/Highlight
n Yin and Siavash
ireless Access Te
14. doi:10.1109
ei Huang, Vijay
Berry, Downlink
ystem.SC-FD
terminal in te
facturing cost
ecific SC-F
many design
mance in a com
A signals inde
A. In this p
of subcarrie
average power
rrier mapping
A (LFDMA)
S) results in hi
FDMA). How
better by 4 to
r the pulse sh
nel interferen
een LFDMA
ce. Another n
s PAPR and th
osine pulse
R of IFDMA.
arefully in ord
he system perf
nvestigation o
n the throughp
cheduling depe
quency respon
to an SC-FDM
are caused by
roperties. The
y causing inco
ue and incorrec
RENCES
asad, OFDM for W
2000.
nsung Lim, and
link Wireless T
vol. 1, no. 3, Sep
uskr, J. Karlss
lqvist,Technical
Commun. Mag., v
strm, A. Furus
S. Parkvall,Th
ncepts and Perfor
Conference (VT
of the 3GPP radio
ts/LTE/LTE.htm.
h Alamouti (Augu
chnology". IEEE
9/SARNOF.2006.
y G. Subramania
k Scheduling and
MA is grea
erms of transm
.
FDMA syst
and operation
mplex manner
eed have low
paper, we ha
er mapping
r ratio (PAPR
g approaches,
) with chann
igher throughp
wever, the PAP
o 7 dB than th
haping necessa
nce, we find
and IFDMA
noticeable fact
hat roll off fac
shaping has
A pulse shapi
der to reduce t
formance.
of the impact
put performan
ends on accur
nse of the rad
MA base statio
noisy estimati
e errors degra
orrect adaptati
ct assignment
Wireless Multime
David J. Goodm
Transmission, IE
. 2006, pp. 30-38
son, M. Meyer,
Solutions for the
vol. 44, no. 3,M
skr, Y. Jading,
he 3G Long-Te
rmance Evaluatio
TC) 2006 Spri
o technology, 3G
ust 2007). "OFDM
E Sarnoff Symposi
.4534773.
an, Rajeev Agraw
Resource Allocat
atly
mit
em
nal
r. It
wer
ave
on
R).
we
nel-
put
PR
hat
ary
d a
in
t is
ctor
a
ing
the
of
nce
rate
dio
on.
ion
ade
ion
t of
edia
man,
EEE
.
S.
3G
Mar.
, J.
erm
on,
ing,
GPP
MA:
ium,
wal,
tion
for OFDM Systems, IEEE Transactions on Wireless Communication,
Vol. 8, No. 1,January 2009.
[8]. Ellina Foutekova,Patrick Agyapong, and Harald Haas
Channel Asymmetry in Cellular OFDMA-TDD Networks, EURASIP
Journal on Wireless Communications and Networking, October 2008.
[9]. U. Sorger, I. De Broeck, and M. Schnell, Interleaved
FDMAA New Spread-Spectrum Multiple-Access Scheme, Proc.
IEEE ICC 98, Atlanta, GA, pp.10131017, June 1998.
[10]. R. Dinis, D. Falconer, C. T. Lam, and M. Sabbaghian, A
Multiple Access Scheme for the Uplink of Broadband Wireless
Systems, Proc. IEEE GLOBECOM 04, vol. 6, Dallas, TX, Dec. 2004,
pp. 3808-3812.
[11]. U. Sorger, I. De Broeck, and M. Schnell, Interleaved
FDMA - A New Spread-Spectrum Multiple-Access Scheme, Proc.
IEEE ICC 98, Atlanta, GA, Jun. 1998, pp. 1013-1017.
[12]. M. Schnell and I. De Broeck, Application of IFDMA to
Mobile Radio Transmission, Proc. IEEE 1998 International
Conference on Universal Personal Communications (ICUPC '98),
Florence, Italy, Oct. 1998, pp. 1267-1272.
[13]. Li-Chun Wang, and Anderson Chen, Optimal Radio
Resource Partition for Joint Contentionand Connection-Oriented
Multichannel Access in OFDMA Systems, IEEE Transactions on
Mobile Computing, Vol. 8, No. 2, February 2009.
[14]. G. Song and Y. Li, Cross-layer Optimization for OFDMA
Wireless Networks-PartII: Algorithm Development, IEEE Wireless
Commun., vol. 4, pp. 625634, Mar.2005.
[15]. Z. Han, Z. Ji, and K. Liu, Fair Multiuser Channel
Allocation for OFDMA Networks Using Nash Bargaining Solutions
and Coalitions, IEEE Trans. on Commun.,pp. 13661376, Aug. 2005.
[16]. R. Dinis, D. Falconer, C.T. Lam, and M. Sabbaghian, A
Multiple Access Scheme for the Uplink of Broadband Wireless
Systems, Proc. IEEE GLOBECOM 04, vol.6, Dallas, TX, pp. 3808
3812, Dec. 2004.
[17]. J. Lim, H.G. Myung, and D.J. Goodman, Proportional Fair
Scheduling of Uplink Single-Carrier FDMA Systems, to be presented
at The 17th Annual IEEE International Symposium on Personal, Indoor
and Mobile Radio Communications (PIMRC06), Helsinki, Finland,
Sep. 2006.
[18]. www.google.com
[19]. swww.wikipedia.com
ISBN: 978-81-920575-7-6 :: doi: 10.73872/ISBN_0768
ACM #: dber.imera.10.73872
Simulation of AODV Routing Protocol

Ashish Chaurasia
Asst.Professor
Gyan Ganga Institute of Technology & Sciences , Jabalpur

Abstract - The growth of laptops and 802.11/Wi-Fi wireless
networking have made MANETs a popular research
topic since the mid- to late 1990s. Many academic papers
evaluate protocols and abilities assuming varying degrees of
mobility within a bounded space, usually with all nodes
within a few hops of each other and usually with nodes
sending data at a constant rate. Different protocols are then
evaluated based on the packet drop rate, the overhead
introduced by the routing protocol, and other measures.
In this paper we are measuring the throughput of
AODV protocol in different scenarios.

.Keywords: MANET, Wi-Fi, routing, throughput, wireless
network

I. INTRODUCTION

A Mobile Ad hoc NETwork (MANET),
sometimes called a mobile mesh network, is a self-
configuring network of mobile devices connected by
wireless links. Each device in a MANET is free to move
independently in any direction, and will therefore
change its links to other devices frequently. Each must
forward traffic unrelated to its own use, and therefore be
a router. The primary challenge in building a
MANET is equipping each device to continuously
maintain the information required to properly route traffic.

Such networks may operate by themselves or may be
connected to the larger Internet.MANETs are a kind of
wireless ad hoc networks[1][6] that usually has a
routeable networking environment on top of a Link Layer
ad hoc network. They are also a type of mesh network,
but many mesh networks are not mobile or not wireless.
1.1 Mobile Ad-hoc Network
MANET [2][3][4] is a kind of wireless ad-hoc network
and it is a self-configuring network of mobile

routers (and associated hosts) connected by wireless links
the union of which forms an arbitrary topology. The
routers, the participating nodes act as router, are free to
move randomly and manage themselves arbitrarily; thus,
the network's wireless topology may change rapidly and
unpredictably. Such a network may operate in a
standalone fashion, or may be connected to the larger
Internet. Mobile ad hoc network is a collection of
independent mobile nodes that can communicate to each
other via radio waves. The mobile nodes can directly
communicate to those nodes that are in radio range of
each other, whereas others nodes need the help of
intermediate nodes to route their packets. These
networks are fully distributed, and can work at any place
without the aid of any infrastructure. This property
makes these networks highly robust.

Figure 1.1. MANET of 14 Mobile Nodes

II. AD-HOC ON DEMAND DISTANCE
VECTOR ROUTING (AODV)

Ad hoc On-Demand Distance Vector (AODV)
routing is a routing protocol for MANETs and other
wireless ad-hoc networks [3][7]. It is a reactive routing
protocol, meaning that it establishes a route to a
destination only on demand. In contrast, the most
common routing protocols of the Internet are
proactive, meaning they find routing paths
Parameter Value
Number of nodes 7
Simulation Time 15sec
Pause Time 5ms
Environment Size 1000*1000
Transmission Range 250m
Traffic Size VBR(VariableBit Rate)
independently of the usage of the paths. AODV is, as
the name indicates, a distance-vector routing protocol.
AODV avoids the counting-to-infinity problem of
other distance-vector protocols by using sequence
numbers on route updates, a technique pioneered
by DSDV. AODV is capable of
both unicast and multicast routing.

In AODV[7][8]the network is silent until a
connection is needed. At that point the network node
that needs a connection broadcasts a request for
connection. Other AODV nodes forward this message,
and record the node that they heard it from, creating an
explosion of temporary routes back to the needy node.
When a node receives such a message and already has
a route to the desired node, it sends a message
backwards through a temporary route to the requesting
node. The needy node then begins using the route that
has the least number of hops through other nodes.
Unused entries in the routing tables are recycled after
a time.

When a link fails, a routing error is passed back to a
transmitting node, and the process repeats.

Figure 2.1.: Working of AODV

III. SIMULATION & RESULTS

3.1 a Performance Metrics

Throughput The ratio of the number
of data packets sent and the number of data
packets received.
b.Simulation Parameters

We have taken one On-demand (Reactive)
routing protocols, namely AODV.. For all the
simulations, the number of traffic sources was taken
for 5 and 7 nodes, the maximum speed of the nodes
was set to 20m/s and the simulation time was fixed as
10s.

Scenario 1
Parameter Value
Number of nodes 5
Simulation Time 10sec
Pause Time 5ms
Environment Size 1000*1000
Transmission Range 250m
Traffic Size VBR ( Variable Bit
Rate)
Packet Size 512bytes
Packet Rate 5packets/s
MaximumSpeed 20m/s
Simulator ns2.31

Table 3.1: Scenario 1 for implementation of AODV

Figure 3.1: X Graph of throughput (AODV) for 5 nodes

Scenario 2
PacketSize 512bytes
PacketRate 5packets/s
MaximumSpeed 20m/s
Simulator ns2.31

Table 3.2: Scenario 2 for implementation of AODV

Figure 3.2: X Graph of throughput (AODV) for 7 nodes

IV. CONCLUSION

Throughput gain is maximum when AODV
protocol used for less nodes and decreases as the node
increases.

REFERENCES
[1]. Ian D. Chakeres and Elizabeth M. Belding- Royer.
"AODV Routing Protocol
Implementation Design." Proceedings of the International
Workshop on Wireless Ad Hoc Networking (WWAN),
Tokyo, Japan, March
[2]. Elizabeth M. Belding-Royer. "Report on the
a. AODV Interop." UCSB Tech Report 2002-18,
June 2002.
[3]. Elizabeth M. Royer and Charles E.
a. Perkins. "Transmission Range Effects
on AODV Multicast
Communication." To Appear in ACM
Mobile Networks and Applications
special issue on Multipoint Communication in
Wireless Mobile Networks.
[4]. Samir R. Das, Charles E. Perkins, Elizabeth M.
Royer and Mahesh K. Marina.
"Performance Comparison of Two On-demand Routing
Protocols for Ad hoc Networks." IEEE Personal
Communications Magazine special issue on
Ad hoc Networking, February 2001, p. 16-28.
[5]. Elizabeth M. Royer and
Charles E.
[6]. Perkins. "An Implementation Study of the
AODV Routing Protocol." Proceedings of the IEEE
Wireless Communications and
Networking Conference, Chicago, IL,
September 2000.
[7]. Samir R. Das, Charles E. Perkins,
and Elizabeth M. Royer. "Performance Comparison of
Two On-demand Routing Protocols for
Ad Hoc Networks."
Proceedings of the IEEE Conference on
Computer Communications (INFOCOM), Tel Aviv,
Israel, March 2000, p. 3-12.
[8]. Elizabeth M. Royer and
Charles E.
[9]. Perkins. "Multicast Operation of the Ad hoc
[10]. On-Demand Distance Vector
Routing Protocol." Proceedings of
MobiCom '99, Seattle, WA, August 1999, pp. 207-218.
[11]. Charles E. Perkins and
Elizabeth M.
[12]. Royer. "Ad hoc On-Demand Distance Vector Routing."
Proceedings of the 2nd IEEE Workshop on
Mobile Computing Systems and Applications, New
Orleans, LA, February
[13]. 1999, pp.
90-100.
ISBN: 978-81-920575-7-6 :: doi: 10. 73970/ISBN_0768
A MODIFIED CLUSTER BASED COOPERATIVE SPECTRUM SENSING
ALGORITHM USING DEMPSTER-SHAFER THEORY

Prof. Dr. A. V. Kulkarni, AliAsger S. Gabuji, Nikhil A. Agarwal, G. Pawan Kumar.
Dept. of Electronics & Telecommunication,
Pad. Dr. D. Y. Patil Institute Of Engineering & Technology,
Pune, India.

Abstract In cognitive radio systems, secondary users can be
coordinated to perform cooperative spectrum sensing so as to
detect the primary users more accurately. A variety of factors
such as hidden node problem, deep fading & shadowing
deteriorate the sensing performance of the cognitive users.
Cluster based sensing method addresses these problems to a
great extent. In this method the locations of all the users are
fixed where secondary users are separated into a few clusters
and the most favorable user in each cluster is selected to report
to the common receiver. The decisions are made by the
common receiver comparing with a single threshold value
where all the favorable users have equal priority in decision
making. In this paper we have proposed a weight based
decision making algorithm using double threshold technique
considering dynamic locations of the users. Numerical results
show that the sensing performance is improved significantly as
opposed to conventional method.
Keywords Dempster-Shafer Theory, Weighted clusters,
Collaborative decision making.
I. INTRODUCTION
Cognitive radio is a paradigm for Wireless Communication
in which either a network or wireless node changes its
transmission or reception parameters to communicate
efficiently avoiding interference with licensed or unlicensed
users. This alteration of parameters is based on the active
monitoring of several factors in the external and internal
radio environment, such as radio frequency spectrum, user
behavior and network state [1].
Todays wireless networks are characterized by a fixed
spectrum assignment policy. However, a large portion of the
assigned spectrum is used sporadically and geographical
variations in the utilization of assigned spectrum ranges
from 15% to 85% with a high variance in time. The limited
available spectrum and the inefficiency in the spectrum
usage necessitate a new communication paradigm to exploit
the existing wireless spectrum opportunistically. This new
networking paradigm is referred to as NeXt Generation (xG)
Networks as well as Dynamic Spectrum Access (DSA) and
cognitive radio networks.
The xG network functions are divided as spectrum
management, spectrum mobility and spectrum sharing.
From the architecture description, cognitive radio networks
correspond to networks where a node needs to sense its
surroundings to determine the proper free spectrum,
possibly with help from other cognitive nodes, and
simultaneously respecting the priority of the primary users.
[11].
The major motivation for this is the heavily underutilized
frequency spectrum. The development is being pushed
forward by the rapid advances in software defined radio
technology which enable a spectrum agile and highly
configurable radio transmitter/receiver.
II. SYSTEM MODEL
Distributed Spectrum Sensing (DSS)
Carrying out reliable spectrum sensing is a challenging
task for a CR. In a wireless channel, signal fading can cause
received signal strength to be significantly lower than what
is predicted by path loss models. The effect of fading can
result in the hidden node problem. [8] The hidden node
problem in the context of CR networks can be described as
an instance in which a secondary in a CR network is within
the protection region of an operating incumbent but fail to
detect the existence of the incumbent. Besides the hidden
node problem, it is also possible for a secondary to falsely
detect an incumbent because of noise or interference in the
wireless environment. Recent research results indicate that
these problems can be addressed by requiring multiple
secondarys to cooperate with each other in spectrum
sensingi.e., DSS. In DSS, each secondary acts as a
sensing terminal that conducts local spectrum sensing. The
local results are reported to a data collector (or fusion
center) that executes data fusion and determines the final
spectrum sensing result.[2][3] The application of DSS
requires that the distance between any two sensing terminals
is small relative to their distances from a primary transmitter
(typically a TV transmitter). In an ad hoc CR network, this
means that when a node needs to conduct spectrum sensing,
it becomes a data collector and collects local sensing reports
from neighboring nodes. The data are exchanged in a
control channel. The control channel can be either a
common control channel shared by all users (such as in the
MAC protocol for CR networks) or a local group control
channel that is used only by neighboring nodes for
communications.[4]

To facilitate our discussion, we model the DSS process as
a parallel fusion network. In this N
0
is a data collector,
N
i
(i = 0, 1, 2, . . . ,m, where m is the number of N
0
s
neighboring sensing terminals) denotes one of N
0
s sensing
terminals (N
0
is both a data collector and a sensing
terminal), y
i
represents the incumbent signal received at N
i
,
and u
i
is the local spectrum sensing report that N
i
sends to
N
0
. The output u is the final sensing decision, which is a
b
i
f
h
b
s
t
b

a
t
m
o
b
f

e
t

b

w
s
r
i
m
s
T
d
m

binary variab
incumbent sig
fusion proble
hypothesis tes
by H
0
and H
1
simplify the d
that spectrum
band. [2][3]
a.Decision fus
It re
threshold valu
m + 1 needs to
or equal to th
occupied, i
band is determ
i.e., u
Depe
fusion can hav
These rule
environments.
threshold tec
[5] [7]
b.Dempster-Sh
Here we set
which make
strengths, rec
reliable than a
implementatio
method in wh
strength value
The decisions
decision algor
making.
III
ble - a one
gnal, and a ze
em therefore
sting problem
1
. Correspond
discussions, th
m sensing is c
sion:
equires the dat
ue that is no l
o be specified
he threshold, t
.e., u = 1 an
mined to be fa
u = 0 and H
0
is
ending on the
ve several vari
A thresh
fusion rul
a value
rule,
and a va
fusion rul
es can only
. Therefore w
chnique based
hafer Theory -
t up two thresh
local decisi
ceived at all
all the other m
on a Weight b
hich lowest w
es falling betw
s are made
rithms both, pr
I. ALGORIT
e denotes th
ero denotes it
can be rega
with two hyp
dingly, each u
i
he following d
carried out in
ta collector to
ess than one a
d. If the sum of
then the final
d H
1
is accep
allow,
s accepted.
value of the
iants.
hold value of
le,
of (m+1) is
alue of (m+1/
le.

be formul
we have impl
d on Demps
-

holds for energ
ions dependin
the CRs. Thi
methods. To m
based techniq
weight is assi
ween the two
sequentially u
roposed as Co
THM / FLOW

he presence
ts absence. Th
arded as a b
otheses repres
i
is also binar
discussion ass
a single spe
o sum up all u
and no greate
f u
i
s is greate
sensing decis
pted; otherwis
threshold, de
f one is an
an AND f
/2) is a Maj
ated in noi
lemented a d
ster-Shafer th

gy detection at
ng on the
is method is
make it suitab
que is added t
igned to the
thresholds [5
using hard &
ollaborative de
WCHART
of an
he data
binary
sented
ry. To
sumes
ectrum
u
i
s. A
r than
er than
sion is
se the
cision
OR
fusion
ority
iseless
double
heory.
t SUs,
signal
more
ble for
to the
signal
5][12].
& soft
ecision
T
mo
dur
acti

Ass

The
is n
fall

c.Sc
T
clus
freq
CR
The proposed a
dules depict
ring the sensin
ion as specifie
F
sumptions :
Maximu
= 27 dB
Maximu
= 0 to -
BAND I
occupie
BAND
moderat
BAND
occupie
e geographica
named as a cl
ling in the clus
can Test :
The base statio
ster heads for
quency which
Rs. The BS c
algorithm is di
the activities
ng action. Th
ed below.

Fig 1. Flowch
um transmissi
Bm = 500 mW
um reception
-80 dBm = 10
I (eg:- 943.1 M
d frequencies.
II (eg:- 944
tely occupied
III (eg:- 94
d frequencies.
al region is div
luster. A datab
sters is collect
on (BS) condu
r respective c
h is received
calculates the
ivided into six
s undertaken
hus they repre
hart / Modules
on power of
W. [9]
power of a
0 pW. [9]
MHz to 944 M
.
4.1 MHz to
frequencies.
0 to 943 M
. [10]
vided in k sect
base of all the
ted and update
ucts a scan tes
clusters. The B
& re-transm
distance of
x modules. The
by the syst
sent the flow

a cellular pho
a cellular pho
MHz) as spars
946 MHz)
MHz) as heav
tors. Each sec
e cognitive us
ed.
t for designati
BS broadcasts
mitted by all t
each CR usi
ese
em
of
one
one
ely
as
vily
ctor
ers
ing
s a
the
ing
which it calculates the weight of each CR of each cluster by
using the following formula :

Weight[i] = distance[i] * signal strength[i]

The CR having maximum weight is designated as the
cluster head of the respective cluster.

Weights of the individual clusters are calculated as follows :
Average weights of all the CRs in a cluster is calculated
& the cluster having maximum average weight is assigned
the highest priority in decision making.

The spectrum is sequentially scanned by all the CRs in the
respective order between the ranges,
BAND I as sparsely occupied frequencies,
BAND II as moderately occupied frequencies,
BAND III as heavily occupied frequencies.

If all the frequencies are occupied, then the network stays
congested which is a rare case.

Using the Dempster-Shafer Theory & Hard/ Soft decision
concept,
If the signal strength received by the CR <= -70 dBm , CR
sends a '0' to its cluster head
If the signal strength received by the CR >= -10 dBm , CR
sends a '1' to its cluster head
If the signal strength received by the CR lies between -10 &
-70 dBm, CR sends the received signal strength to the
cluster head.
Now the cluster head calculates the % of CRs concluding
primary user is present or absent.

A = [(no. of CRs concluding '0') * 45] +[(no. of CR's
receiving signal strength between (-70 to -45 dBm)) * 10]
P = [(no. of CRs concluding '1') * 45] + [(no. of CR's
receiving signal strength between (-45 to -10 dBm)) * 10]

Decision by Cluster Head
Primary present = [P / (A+P)] * 100 %
Primary absent = [A / (A+P)] * 100 %

These decisions are sent to the BS for analysis.

No. of clusters : k
CRs in a cluster : n
1
, n
2
n
k

Now, x = 1,2,.,k
n
x

Avg wt. of CR[x] = [ d[i] * s[i] ] / n
x

i = 1

Let the clusters after rearranging according to their weights
in descending order be as follows

W
k
, W
k-1
, W
k-2,..
., W
1

Final decision by the BS,

= ( decision of cluster * weight of cluster )

1
R
Wi
* i
i = k
Final decision = _____________
1
i
i = k

where R
Wi
= decision of cluster [i] sent to the BS

The frequency hole scanned is allocated to the CR having
highest priority as per its subscriber authorization plan.

Priority
No.
Category Charges per call
1 (Highest) Gold Extra
2 (Medium) Silver Extra
3 (Lowest) Bronze No Extra charges

IV. SIMULATION

The following representations denote the simulation
results of the modules.

Fig. 2 & Fig. 5 represent cluster formation & cluster head
designation.


u

m
Fig2. Cluste
Fig. 3 & Fig
Station.
Fig 3. D
The Overall
using 100 CRs
Fig. 4 & Fig
making for dif
9%
DE
er Formation
(4
g. 6 represent t
Decision Makin
Efficiency for
s = 91 %
g. 7 represent
fferent numbe
ECISION OF

and Cluster H
clusters)

the decision m

ng For 100 Co

r Sensing on

the efficiency
er of Cognitive

91%
THE BASE S
D
P
N
Head Designat
made by the Ba
ognitive Radio
ne Frequency
y in decision
e Radios.
STATION
DECISION (P.U.
PRESENT/ABSE
NO RESULT

tion
ase
os
Hole
F

.
ENT)
Fig 4. Eff
Fig5. Cluster F
0
20
40
60
80
100
10 20
2%
DEC
fficiency For

Formation and
(8 clu

0 30 40 50
Efficie
CISION OF T
100 Cognitive
d Cluster Hea
usters)
0 60 70 8
ency
98%
THE BASE S
e Radios
ad Designation
80 90
STATION
DECISION (P.U
PRESENT/ABS
NO RESULT

n
U.
ENT)
u
o

b
e
c
t
h
i
F
w

Fig 6. D
The Overall
using 200 CRs
Fig 7
The efficiency
of Cognitive R
In this pape
based coope
encouraging
collaborative
technique by
heads. The l
increasing th
Fast fading w
work where re
[1] J.
ma
0
20
40
60
80
100
10
Decision Makin
Efficiency for
s = 98 %
7. Efficiency F
y of Spectrum
Radios increas
V. CO
er, we have m
erative sens
results. The
decision m
acquiring re
load on the
he efficiency
would be an int
eceiver sensiti
VI.R
III Mitola and
aking software
00 110 120 13

ng For 200 Co

r Sensing on

For 200 Cogni

Sensing incre
ses.

ONCLUSION
modified the c
sing algorith
main base
making using
sults from th
common ch
of sensing to
teresting exten
ivity can also b
REFERENCES
G. Q. Jr. Mag
radios more
0 140 150 160
Efficienc
ognitive Radio
ne Frequency
itive Radios
eases as the nu
conventional c
hm & obt
station per
double thre
he weighted c
hannel is red
o a certain lim
nsion to the p
be incorporate
S
guire, Cognitive
personal, P
0 170 180 190
cy
os
Hole
umber
cluster
tained
rforms
eshold
cluster
duced,
mit[6].
resent
ed.
e radio:
Personal
[
[
[

0 200
Comm
Aug. 1
[2] C. Sun
coope
Procee
Comm
[3] A. Gh
for op
DySPA
[4] S. Hay
comm
Comm
[5] J. Zhu
thresh
in cog
Cogni
(Crow
[6] D. Cab
Implem
radios
Comp
[7] Coope
Detect
Liu Sc
Unive
(UEST
[8] F. F. D
energy
in Pro
(IEEE
3575
[9] Mouly
comm
10] Cogni
Alexa
Engin
11] NeXt
radio
Yeol
Broad
Electr
Techn
12] Colla
with n
B.Eng
munications, IEEE
1999.
n, W. Zhang, and
erative spectrum s
edings of IEEE In
munications, Glasg
hasemi and E. Sou
pportunistic access
AN, Baltimore,MD
ykin. Cognitive ra
munications. IEEE
munications, vol. 2
u, Z. Xu, F.Wang
hold energy detec
gnitive radio, i
itive Radio Orie
wnCom), pp. 1-5, 2
bric, S. M. l\4isha
mentation issues
s. in Proc, of Asilo
puters, pages 772
erative Spectrum
tion and Dempst
chool of Commun
ersity of Electron
TC) Chengdu 611
Digham, M.-S. Al
y detection of unk
oc. IEEE Internati
E ICC 2003), vol.
3579.
y M., Pautet M.
munications, 1992
itive Radio Te
ander, R. David
neering, Indian Ins
generation/dynam
wireless network
Lee, Mehmet
dband and Wirele
rical and Comput
nology, Atlanta, G
aborative spectrum
non-gaussian nois
gg., Lakehead uni
E, Volume 6, Issu
K. B. Letaief, C
sensing in cognitiv
nternational Conf
gow, Scottland, Ju
usa, Collaborativ
s in fading enviro
MD, USA, p.p. 131
adio: brain-empow
E Journal on Selec
23:201220, Feb
g, B. Huang and
ction of cooperati
in Proc. of Inte
ented Wireless N
2008.
ar, arid R. W. Bro
in spectrum sensi
omar Conf on Sig
776, November
Sensing Based o
ter-Shafer Theory
nication and Infor
nic Science and T
1731, China
louini, and M. K.
known signals ov
ional Conference
5, no. 4, May 11
B., The GSM
2.
echniques for G
Koilpillai Depar
stitute of Technol
mic spectrum acce
ks: A survey Ian
C. Vuran *, S
ss Networking L
ter Engineering,
GA 30332, United
m sensing in a cog
se by Tan Feng
versity, 2006
ue 4, Page(s):13 -
Cluster-based
ve radio systems
ference on
une 2007.
ve spectrum sensin
onments, Pro. IE
136, Nov. 2005.
wered wireless
cted Areas in
bruary 2005
d B. Zhang, Dou
ive spectrum sens
ernational Conf.
Networks Comm
odersen.
ing for cognitive
gnals, System and
r 2004.
on Double Thresh
y Hongyu Chen J
rmation Engineeri
Technology of Ch
Simon, On the
er fading channel
e on Communicati
15, 2003, pp.
M system for mob
GSM Band B
rtment of Electr
logy, Madras.
ess/cognitive
n F. Akyildiz, W
Shantidev Moha
aboratory, Schoo
Georgia Institute
d States.
gnitive radio Syst
18,
, in
ng
EEE
uble
sing
on
mun.
d
hold
Jian
ing,
hina
ls,
ions
bile
aiju
rical
Won-
anty
ol of
e of
tem
ISBN: 978-81-920575-7-6 :: doi: 10. 73977/ISBN_0768
Protocol Design for Advanced Security in Wireless Networks

Raghuveer Singh Dhaka
Department of Computer Science & Engg.
Central University of Rajasthan
Ajmer, India

Arvind Dhaka
Department of Computer Science & Engg.
Central University of Rajasthan
Ajmer, India

AbstractThis paper is focused to investigate the security
protocols which are used in wireless network. The Protocols
like WEP, WPA or WEP2 are considered under the IEEE
802.11 standard. Throw the analysis of the protocols we
determine that the advantages and disadvantages. And for the
remaining vulnerabilities we are proposing an effective
solution which is more secure and unpredictable.
Keywords- WEP; WPA; MIC;AES;EAP;IV;WPA-2; MSS.
I. INTRODUCTION
The wireless network security is in a state of chaos.
Everyday a new vulnerability is discovered in almost every
existing protocol. The answer to these problems lies in
enhancing our design methodology in addition to
implementing and testing the protocols. In this paper, a new
mechanism is proposed and discussed. The proposed
mechanism will be compared with a number of exiting
working mechanisms in wireless networks.
II. ISSUES AND SECURITY ATTACKS
In this section, issues and security attacks in wireless
networks security protocols are discussed.
A. Modification
It is an active attack where the content of the original
message are modified by the intruder and then sent it instead
of the actual message.
B. Fabrication
It is an active attack where non original message is
generated by intruder and then sent to the receiver.
C. Interception
It is a passive attack where the encrypted message is
listens and records by the intruder and later on decrypt by
intruder.
D. Brute Force
It is a passive attack where the intruders generates every
possible permutation in the key and try to decrypt the
encrypted message.
E. Maintainability
For pages other than the first page, start at the top of the
page, and continue in double-column format. The two
columns on the last page should be as close to equal length
as possible.
III. WIRELESS SECURITY PROTOCOL
A. Wired Equivalent Privacy (WEP)
WEP was developed to provide wireless security to the
wireless users on WLAN that is equivalent to being on a
wired network. WEP, not being intended to provide a higher
security level than a wired LAN, provides equivalent
security level to that by wired LANs. RC4 Stream Cipher is
used to provide confidentiality and CRC-32 checksum is
used to provide Integrity (honesty).WEP uses a 40 bit key
size 24 bit IV (Initialization Vector).
B. Wired Equivalent Privacy2 (WEP 2)
It is same as WEP. The size of Initialization Vector is
same means 24 Bits. But the key size is expended to 104 Bit
[5].
C. Wi-Fi Protected Access (WPA) Protocol
WPA protocol was designed to provide security to the
Wi-Fi networks. It is considered under the IEEE 802.11i
standard. It was developed by Wi-Fi Alliance. There is TKIP
(Temporal Key Integrity Protocol) which provides more
security to change the dynamic key [10]. The size of
Initialize Vector (IV) is just double in compare to WEP,
which is 48 Bit. And the key size is 128 bit.
D. WPA-2 (Robust Security Networks- RSNs)
MAC is replaced by CCMP and RC4 is replaced by AES.
There are 2 mod, I mode is known as pre shared key mode
and II mode is known as full Extensive Mod/WPA-
Enterprise. 802.11i that has been appeared in 2004 uses the
logic of a Robust Security Network (RSN). This new
standard and architecture adopt the IEEE 802.1X standard
for access control and for encryption it uses Advanced
Encryption Standard (AES) [6]. It uses a pair-wise key
exchange (four way handshake) protocol utilizing 802.1X for
mutual authentication and key management process. 802.11i
allows use of TKIP, but by default RSN uses AES
(Advanced Encryption Standard) and CCMP (Counter Mode
CBC MAC Protocol). It provides a stronger, scalable
solution.

E
R
M
r
d
o
M
o
m
w
TABLE I.

The propo
System (M
Encryption/De
RSA, Blowfis
MSS uses 256
random in the
differentiate b
original name
MSS stores a
of different slo
means that 8
25
which is trying

COMPARISON
SECURIT
IV. PROPO
osed mechani
MSS). MSS
ecryption alg
sh, Elliptic Cu
6 slots. One o
ese slots. Ther
between these
e of the algori
key list. Each
ot algorithm an
56
possible com
g to attack.
Figure 1.
N BETWEEN WEP
TY PROTOCOLS
OSED MECHAN
ism is known
S uses
gorithms RC4
urve, AES, H
of the eight al
e is a 2-bit co
eight algorith
ithm. For the
h message enc
nd a different
mbination req
Slot System.
P, WPA AND WPA
NISUM
n as Multiple
one of
4, Diffie-Hel
HMAC and CM
lgorithms is u
ode which is u
hms in place
se eight algor
crypts with the
key by MSS, w
quires to any h
A2

e Slot
eight
llman,
MAC.
used at
used to
of the
rithms
e help
which
hacker

(KS
MIC
bit
par
256
shu
mo
A.
use
the
App
con
Ext
auth
2-b
bits
one
whi
the
1
are
wil
star
nod
algo
key
hea
MSS header h
S) is first part.
C shuffle Sele
long which is
rt is 8-bit long
6 slots. The th
uffle tables, the
ve quickly the

Fig
MSS Operatio
Different con
er, which mean
user is affe
plication laye
nfiguration fil
tensible Authe
hentication ser
The configura
bit long), whic
s only. Eight k
e of the eight
ich is the shuf
MIC generate

1) Sending P
used to choos
l be shuffled
rting from the
de will encry
orithm for the
y. The node w
ader to messag
has divided in
The slot selec
ector (MICSS)
used to select
g which is used
hird part is 8-b
e shuffle table
e MIC bits with
gure 2. MSS He
on
nfiguration file
ns that the netw
ected. MSS
er. The access
es. The AP
entication Pro
rvices.
ation file cont
ch means that
key lists files
t algorithms. O
ffle table. Its w
es and its lengt
Phase: A slot a
se the user or
with the cho
e first bit unti
ypt the new
e slot that it
will concatenat
ge and then sen

n three parts. T
ctor (SS) is sec
) third part. Th
t between 2
48
k
d to select bet
bit long to sel
es are used ma
h the original
eaders & Payloa
es are used by
work will not b
is going to
s point (AP)
act like a co
otocol (EAP)
tains the 256 s
the slots part
are needed al
One more file
width is the nu
th is 2
64
.
and a key from
the node. The
osen row insi
l the last bit i
payload with
was chosen u
te the SS, KS
nd the messag
The key Selec
cond part and t
he first part is 4
keys. The seco
tween one of t
lect between 2
inly to shuffle
message.
ad.
y MSS for ea
be affected wh
operate on t
contains all t
onversion serv
used to prov
slots (each slot
will need 2*2
lso, one for ea
e is also need
umber of bits th
m the list of ke
e generated M
de the messa
in the MIC. T
h the appoint
using the chos
and MICSS a
ge.
ctor
the
48-
ond
the
256
e or
ach
hen
the
the
ver.
ide
t is
256
ach
ded
hat
eys
MIC
ge,
The
ted
sen
as a
i
M
k
s
The select
is done system
MSS will run
key in the key
slot with the f

a.Receiving
separate the
SS to know
message. Th
data, the re
will use th
shuffle tabl
receiver wil
MIC). The r
part and c
received to
b.Control m
will act as a
a refresh of
all nodes. T
handle the
configuratio
will send a
AP will de
control serv
them to the
itself and s
will receive
packet to th
the new ver
files with th
Figure 3. Shu
tion of the slot
matically, whi
n it will use th
y list of that al
first key in his
g Phase: In th
e SS, KS and
w the algorithm
he appointed k
ceiver will de
e MICSS to
le and will un
ll split the pay
receiver will g
compare the
insure the inte
messages and a
a conversion se
f the configur
The AP must h
conversions.
on files, key li
re-issue cont
liver it to the
ver will issue
e AP. The AP
send a copy to
e the configu
he AP, so that
rsion that it ha
he new version

uffeling Mechan
t and the key a
ich means tha
he first slot an
lgorithm. Seco
list and so on
he receiver sid
d MICSS from
m that will be
key list of this
ecrypt the mes
get the shuff
ndo the shuff
yload to two
generate the M
generated M
egrity of the m

access points o
erver. It has th
ration files for
have a good p
MSS needs s
ists and shuffl
trol packet to
e MSS contro
e the new file
will keep a c
o the appoint
uration files a
t it will replac
ad. The node
n and reset all
nism.
and the shuffl
at the first tim
nd the first po
ond time the s
n until the end.
de, the receive
m the message
used to decry
s algorithm an
ssage. The rec
fling row from
fling, after th
parts (messag
MIC for the me
MIC with the
message.
operations: Th
he authority to
r one node or
processing pow
server to issue
fle tables. Any
the AP, whic
ol server. The
es needed and
copy of the fil
ed node. The
and issue a c
ce the old files
will replace th
the counters.

e data
me the
ointed
econd
.
er will
e. The
ypt the
nd KS
ceiver
m the
hat the
ge and
essage
e one
he AP
o issue
r even
wer to
e new
y node
ch the
MSS
d send
les for
e node
ontrol
s with
he old
Slo
one
hac
slot
nod
per
com
per
bec
mea
app
wil
MS
shu
Tim
refe
bra
nam
num
sen
the
ref
Re
actu
MSS depend
ots and permut
e but four alg
cker or intrude
ts configuratio
de in the netw
rmutation. It
mbined with
rmutation of th
cause if one o
ans, MSS c
pearance of tha
l weaken the
SS has solved
uffling the MIC
TABLE II.

List and num
mes, single-sp
erenced in the
ackets, for exa
me(s) of edito
mber citations
ntence punctua
reference nu
ference [3]
eference [3] w
Number foot
ual footnote a
V. CON
d on Permutat
tation of Shuff
gorithms to en
er needs to dev
ons, key lists
work. MSS doe
depends on t
the permuta
he shuffling tab
f the algorithm
can counter
at algorithm in
strength of th
is the static p
C bits inside th
COMPARISON BET
REFE
mber all bibliog
paced, at the
e text, enclose
ample [1]. W
ors of referenc
s consecutive
ation follows th
umber, as in [
except at th
was the first . .
tnotes separate
at the bottom o
NCLUSION
tion of keys,
fling Tables. B
ncrypt and de
vise a mean to
and shuffling
es not depend
the permutati
ation of the
bles. MSS is a
ms is found to
its effect b
n the configur
he MSS. Anoth
placement of
he original me
TWEEN WEP 2, W
ERENCES
graphical refer
end of you
the citation n
Where appropri
ced books. Th
ely within br
he bracket [2]
[3]do not u
he beginning
.
ely in supersc
of the column
Permutation
By using not on
ecrypt with, a
o guess the ex
g tables for ea
only on the k
ion of the ke
slots and t
also maintainab
o broken by a
by denying t
ration slots. T
her problem th
the MIC bits
ssage.
WPA 2 AND MSS
rences in 9-po
ur paper. Wh
number in squ
iate, include t
he template w
rackets [1]. T
. Refer simply
use Ref. [3]
of a senten
cripts. Place
n in which it w
of
nly
any
xact
ach
key
eys
the
ble
any
the
This
hat
by

oint
hen
are
the
will
The
y to
or
nce:
the
was
cited. Do not put footnotes in the reference list. Use letters
for table footnotes.
Unless there are six authors or more give all authors
names; do not use et al.. Papers that have not been
published, even if they have been submitted for publication,
should be cited as unpublished [4]. Papers that have been
accepted for publication should be cited as in press [5].
Capitalize only the first word in a paper title, except for
proper nouns and element symbols.
For papers published in translation journals, please give
the English citation first, followed by the original foreign-
language citation [6].

REFERENCES

[1] IEEE: IEEE 802.11: Wireless LAN Medium Access Control (MAC)
and Physical Layer (PHY) Specifications (2007),
http://standards.ieee.org/about/get/802/802.11.html.
[2] Sepehrdad, P., Vaudenay, S., Vuagnoux, M.: Discovery and
Exploitation of New Biases in RC4. In: Biryukov, A., Gong, G.,
Stinson, D.R. (eds.) SAC 2010. LNCS, vol. 6544, pp. 7491.
Springer, Heidelberg (2011).
[3] Sepehrdad, P., Vaudenay, S., Vuagnoux, M.: Statistical Attack on
RC4 Distinguishing WPA. Accepted at EUROCRYPT 2011 (2011).
[4] Vuagnoux, M.: Computer Aided Cryptanalysis from Ciphers to Side
channels. PhD thesis, Ecole Polytechnique Federale de Lausanne
EPFL (2010).
[5] Wpa-Wpa2, http://www.lylebackemort.com/blog/2008/5/10/ wpa-
wpa2-as-insecure-as-i-expected (accessed on March 14, 2010).
[6] A. H. D. Lashkari, M.M.S. , Samadi, B., "A survey on wireless
security protocols (WEP, WPA and WPA2/802.11i)," Computer
Science and Information Technology, 2nd IEEE International
Conference on Beijing, ICCSIT 2009, pp.48-52, 2009..
[7] E.J.M.A. Filho, P.N.L. Fonseca, M.J.S. Leitao, and P.S.F. de Barros,
Security versus Bandwidth: The Support of Mechanisms WEP &
WPA in 802.11g Network, IFIP International Conference on
Wireless and Optical Communications Networks, WOCN '07, pp. 1-
5, 2007.
[8] M. Bloch et al., Wireless Information-Theoretic Security, IEEE
Trans. Info. Theory, vol. 54, no. 6, June 2008, pp. 251534.
[9] S. Mathur et al., Radio-Telepathy: Extracting a Cryptographic Key
from an Unauthenticated Wireless Channel, 14th ACM MobiCom,
2008.
[10] TecTarget. (2008). Wi-Fi Protected Access. Retrieved 05 May, 2008,
from http://searchmobilecomputing.techtarget.com/sDefinition/
0,,sid40_gci887323,00.html.

A
t
d
d
h
c
m
a
s
s
w
d
s
A
i
W
P
p
u
B
w
w
s
s

a
I
o
c
c
s
t
Proc. of th
Volume 1. Co
ISBN: 978-8
Raghuve
Department o
Central Uni
Aj
Abstract A
technology the
different secur
data are differ
have to gain ac
communication
medium is no
accessed with
security of wi
since its deve
wireless netwo
different attac
security mecha
Keywords-
I. WLA
A. WLAN Sec
WEP is th
implemented.
WEP by cryp
Protected Ac
passed, RSN
under the full
B. Wired Equ
WEP was d
wireless users
wired network
security leve
security level
1) How W
and every da
Stream ciphe
Initialization
original pack
cipher which
cipher stream
sender can ch
to the receive
he Intl. Conf. o
opyright 20
81-920575-7-6
ACM #: dber
eer Singh Dh
of Computer S
& Engg.
iversity of Raj
jmer, India

wired and w
ey use to tran
rity aspects. M
rent in both te
ccess over the p
n i.e. wires. B
one other than
the help of equ
reless network
elopment. The
orks is in air
ks. Thus it is v
anisms to wirel
WEP; WPA; M
AN SECURITY
curity in Gene
he foremost W
Several wea
tanalysts, and
ccess (WPA)
(also known
IEEE 802.11i
uivalent Privac
developed to
s on WLAN
k. WEP, not b
l than a wi
to that by wir
WEP works: W
ata packet is
er. This key
Vector (IV)
et is then bit
results in an
m is used to en
hose a differen
er. We further
on Computer A
12 Techno Fo
6 :: doi: 10. 73
r.imera.10. 73
A Rev
haka
Science
asthan
wireless netw
nsmit data, th
Mechanisms to
echnologies. In
physical mediu
But in wireless
n air, the radi
uipment availa
ks has always
e transmission
r which mak
very difficult to
less networks.
MIC;ICV;IV;CC
PROTOCOLS A
eral
WLAN securit
aknesses were
d WEP was tak
in 2003, an
n as WPA2)
i standard [7, 9
cy (WEP) Mec
provide wirel
that is equiva
eing intended
ired LAN, p
red LANs.
When WEP is
encrypted se
y is genera
and a 40- b
twise XORed
encrypted pla
ncrypt each ne
nt IV periodica
compute 4-by
Applications
rum Group, In
3984/ISBN_07
984
view of W
WEP, W
Departm
Central
work differs i
ereby implem
o access transm
n wired networ
um used for ne
s networks sin
io frequency c
able in market
been a major
n of informati
kes it vulnerab
o provide up t
CMP;TKIP;MA
AND MECHANI
ty mechanism
e identified i
ken place by
nd then after
is ratified in
9].
chanism
less security t
alent to being
to provide a h
provides equiv
s enabled then
parately with
ated by a
bit WEP key
d with RC4 s
aintext. A dif
ew packet sinc
ally. IV is then
yte Integrity C

ndia.
768
Wireless Se
WPA and
Arvind Dhak
ment of Compu
& Engg.
University of
Ajmer, India

in the
enting
mitted
rks we
etwork
ce the
can be
ts. The
r issue
ion in
ble to
to date
AC.
SMS
to be
in the
Wi-Fi
r time
2004
to the
g on a
higher
valent
n each
h RC4
24-bit
y. The
stream
fferent
ce the
n sent
Check
Val
also
enc

2
RC
enc
resu
stre
suc
WE

A
32
(Cy
"ch
stro
dig
be
mea
Sha
aga
WE
leak
pos
eno
ecurity Pr
d WPA-2
ka
uter Science
f Rajasthan
a
lue (ICV) on
o encrypt IC
cryption mecha
2) Vulnerabi
C4 stream ciph
cryption. This
ults in the nee
eam can be
ccessful attack
EP makes it a f
Another vulner
mechanism w
yclic Redunda
hecksum". CR
ong and thus n
est or hash fu
noted that th
asure in WEP
amir Attack d
ainst the key
EP. Also the i
ked by the f
ssible to recon
ough WEP-enc
rotocols:
Mu
Depa
Cen
the original p
CV with the
anism is show
Figure 1. Wep
lities and fla
her with a 24
short size IV
ed to reuse the
exploited by
k on the ciph
failure to achi
rable aspect o
which is used
ancy Code)
RC is not c
not efficient en
unctions such a
he use of CR
P, which is n
describes a pa
scheduling a
information ab
first byte gen
nstruct the sec
crypted packet
ukul Manmoh
artment of Com
& Eng
ntral Universit
Ajmer, I
packet and app
RC4 cipher
wn in following
p Encryption.
aws of WEP:
4- bit Initializa
V is a weakne
key stream. T
y the attacker
her text. This
ieve the securi
of the WEP is
for the integr
is considered
considered c
nough to be us
as MD5 or SH
C-32 was use
not true. The
assive cipher-
algorithm of R
bout individua
nerated by RC
cret key in WE
ts.
han Meghwa
mputer Scienc
gg.
ty of Rajasthan
India
pend with it. W
r stream. W
g diagram.

WEP uses t
ation Vector
ess of WEP as
This reuse of k
r to perform
vulnerability
ity goals.
the use of CR
rity check. CR
d as a class
cryptographica
sed as a messa
HA [1]. It shou
ed as a secur
Fluhrer Man
-text-only atta
RC4 as used
al key bytes g
C4. It could
EP by analyzi
al
ce
n
We
WEP
the
for
s it
key
m a
of
RC-
RC
of
ally
age
uld
rity
ntin
ack
in
gets
be
ing
The authentication flaw in the IEEE 802.11 has been
shown in previous works. A simple successful
eavesdropping attack against IEEE 802.11also was
demonstrated with the WEP enabled.
Both RC4 and IVs are not used efficiently in the standard
WEP. Because of these reasons we conclude that 802.11
WEP is totally insecure and the reason is nothing but poor
implementation of secure technologies (such as RC4). WEP
was not anticipated to be a good solution for all security
needs.

3) The weaknesses of WEP
a) Direct use of Master keys: The Use of master keys is
not directly recommended. The other temporary keys are
generated by Master keys.
b) Small key sizes: The WEP uses 40 bits of key size,
which has been considered as one of the major flaw in the
WEP. In 1997, 40-bit key size was considered to be
reasonable for some applications. At that time the goal was
to provide protection against the casual eavesdropping, it
was seemed good to provide sufficient protection at that
time. There is no any specification of WEP key sizes other
then 40 bits in the 802.11 standard. But most dealers apply
the key size to 104 or 232 bits. RC4 encryption key includes
a 24-bit IV in both cases. 104 and 232-bit keys are more
defensive against to brute-force attacks in comparison to the
40-bit keys but it is still a problem which is not resolved yet.
c) Not have enough key management: WEP standard
does not specify key management. Without interoperable
key management, keys will be of inferior quality and lives
for a longer time. Thus this is considered as its weakness.
Most of the wireless networks implementing WEP shares a
single WEP key between each and every node on the
network. Same WEP key is used to program the access
points and client stations. The task of changing the keys is
anonymous and difficult.
d) Use of RC4: Implementation of RC4 has been
considered to have weak keys, indicating more correlation
between the key and the output. It is easy job to determine
packets encrypted with weak keys. The first three bytes of
the key are extracted from the IV that is sent unencrypted in
each packet. This weakness can be exploited using a passive
attack.
e) Reused and small sized IV: We found that out of the
available 16 million IV values 9,000 are interesting. They
show the presence of weak keys. The attacker captures
"interesting packets" to find out the weak keys, then
analyzes them and only has to try a small number of keys to
gain access to the network. Because all original IP packets
start with a known value, its easy to know when user has
the right key. To determine a 104-bit WEP key, attacker has
to capture the interesting packets between 2,000 and 4,000.
On a busy network the capture of the 5000 interesting
packets might not cause any difficulty and this can be done
in a very short time period. Many dealers are now
implementing new algorithms that are simply based on no
selection of weak IV. The best solution of this type of attack
is not to use weak IV values.
For a WEP key, despite of the key size, 24- bit long IV
can only provide 16,777,216 different RC4 cipher streams.
This number can be achieved on a busy network in a few
hours and we can reuse the same IV. The RC4 stream cipher
is XORed with the original packet in the WEP and the IV is
sent in the clear format with each packet [3]. If the RC4
stream cipher for a given IV is found, subsequent packets
encrypted with the similar IV can be decrypted by an
attacker. There are maximum 16,777,216 IV values.
Unluckily, WEP doesnt specify such facility that how to
choose or how often to change IVs. Some assumes that IV
starts at zero and increases incrementally for each packet,
and return back to zero after sending 16 million packets.
And some other chooses IVs randomly. The IV chosen
randomly leads to 50% chance of reuse after less than 5,000
packets.
f) Weakness of ICV algorithm: WEP ICV is based on
the CRC-32 algorithm usually implemented to detect noise
along with common errors in transmission. CRC-32 is not
considered as a good choice for cryptographic hash. The
CRC-32 ICV is a linear function of the message. This
means that a message can be encrypted and the ICV be
easily fixed, by an attacker, making the message appear
authentic. To do this an encrypted packet stream is captured
and the destination address of each packet is changed with
the attackers wired IP address, fixing up the CRC-32.These
packets are then retransmitted to the access point. The
packets will decrypt by the access point and after that
forward them to the attacker. IV and ICV based attacks are
not depend on the key size.
g) Easy forging of authentication messages: There are
two types of authentication methods which are defined in
802.11 Standards. One is known as Open System
authentication and another is known as Shared Key
authentication. If the WEP authentication service is enabled,
then the intruders and attackers can guess the WEP key [6,
10, 11]. In Shared Key authentication the concept of shared
WEP key is used to encrypt the challenge. With this comes
the possibility that the challenge and encrypted response can
be observed by an attacker who is monitoring the network
traffic. By monitoring the successful authentication, an
attacker can forge an authentication. Shared Key
authentication limits the ability of an attacker to create a
denial-of-service attack by sending garbage packets
(encrypted with the wrong WEP key) into the network.
WEP is far from an ideal security solution but it is still used.
C. Wi-Fi Protected Access (WPA) Protocol
WPA has been developed to provide more wireless
security solution for 802.11-based networks. WPA is a
replacement over Wired Equivalent Privacy (WEP) [2]. The
industries and organizations have been spending a great deal
of time and money on developing and deploying next-
generation solutions that address growing wireless network
security problems. The Wi-Fi (Wireless Fidelity) Alliance
has created the Wi-Fi Protected Access (WPA) [4]. WPA is a
subset of the 802.11i standard.
An enhancements over WEP: WPA is an enhancement over
pre-existing protocol WEP.WPA technology mainly
includes three improvements over WEP.
a) Improved data encryption: Temporal Key Integrity
Protocol (TKIP) is an improvement of data encryption.
TKIP is a Temporal Key Hash Function. TKIP is an
alternative to WEP. It fixes all the security problems and
doesnt require new hardware. TKIP uses the RC4 stream
cipher as the encryption and decryption processes and all
involved parties must share the same secret key similarly as
like in WEP [11]. This secret key must be 128 bits and is
called the "Temporal Key" (TK). TKIP also uses an
Initialization Vector (IV) of 48-bit. The IV used as a
counter. Even if the TK is shared, all involved parties
generate a different RC4 key stream.
b) User authentication: User authentication is missing
in WEP, through the extensible authentication protocol
(EAP). WEP controls access to a wireless network which is
based on a computers hardware-specific MAC address.
EAP is built on a more secure public-key encryption system
to make it certain to happen that only authorized network
users can access the network.
c) The quality of being honest: Message Integrity Code
(MIC) for TKIP is computed by a new algorithm. Message
Integrity Code (MIC) is computed to determine errors in the
data complacency either due to transfer errors or due to
purposeful alterations [5]. It is a 64-bit MIC that is added to
the Data and the ICV. The ICV is CRC of Data and MIC.
d) What is new in WPA in comparison to WEP?
Direct use of master key in WEP: Master Keys are
never used directly in WPA. A hierarchy of keys is
used, all derived from the master key.
Poor Key Management and updating in WEP:
Secure key management is a planned obsolescence
feature in WPA, so key management isnt an issue
with WPA as compared to WEP.
Reused IV/Less size of IV: The length is increased of
IV from 24 bits to 48 bits. Additionally, IVs are
used as sequence counters for the TSC (TKIP
Sequence Counter), protecting against replaying of
data, which was a major vulnerability in WEP.
Weak IV values are susceptible to attack: A different
secret key is used for each packet, and the way the
key is more quickly with the secret key is more
complex.
Ineffective Message integrity checking: Integrity
protocol CRC-32 was proved ineffective for WEP
message so WPA uses a Message Integrity Check
(MIC) mechanism called, .Michael.
c.Drawback in WPA: The 802.11i standard shows that
keys generated from short passwords are subject to
dictionary attack and a key that is generated from a pass
phrase of less than 20 characters is unlikely to deter
attacks.
D. WPA-2 (Robust Security Networks- RSNs)
802.11i that has been appeared in 2004 uses the logic of a
Robust Security Network (RSN). This new standard and
architecture adopt the IEEE 802.1X standard for access
control and for encryption it uses Advanced Encryption
Standard (AES). It uses a pair-wise key exchange (four way
handshake) protocol utilizing 802.1X for mutual
authentication and key management process. 802.11i allows
use of TKIP, but by default RSN uses AES (Advanced
Encryption Standard) and CCMP (Counter Mode CBC
MAC Protocol). It provides a stronger, scalable solution.

1) Working of WPA-2: WPA-2 uses dynamic
negotiation of authentication and encryption algorithms
between access points and mobile devices. The
authentication schemes are based on 802.1X and Extensible
Authentication Protocol (EAP). We use Advanced
Encryption Standard (AES) for encryption [8]. WPA-2 is
significantly stronger in comparison to WEP and WPA.
However, this protocol would run very poorly on legacy
devices. Only the latest devices have the capability required
to accelerate the algorithms.
II. COMPARISON OF WEP, WPA AND WPA-2
SECURITY PROTOCOLS
A. WEP
WEP uses RC4 cipher. The IV is insecure. It uses 40 bits
of encryption key. There is a concatenation of key and IV
values for per packet are sent over the air, key management
mechanism is not embedded in it, it have no automatic or
periodic key.
B. WPA
It uses a subset of 802.11i. It uses RC4 and TKIP for
encryption cipher mechanism. The key size is 128 bits. It
uses mixed type Encryption key per packet [6]. 802.1x
dynamic key management mechanism. The IV size is 48
bit.802.1x EAP usage for authentication, providing data
integrity and header integrity. The TKIP and IV sequence
mechanism are used to prevent replay attacks.
C. WPA-2
For wireless security WPA-2 seems to be the strongest
security protocol. WPA-2 provides all the advantages of
WPA in addition to stronger encryption through the
implementation of AES, roaming support and CCM
mechanism for data and header integrity [10]. WPA
deployments over current WEP installations provide cost
effective and hassle free shifts where vendors can transit to
the WPA standard through a software or firmware upgrade.
For WPA-2this is not the case. It requires extra hardware
upgrade in order to implement AES [12].
n
c
m
n
n
c
t
TABLE
Wireless ne
not managed
companies. So
must be ident
necessary cou
network. Alt
considered tot
to be a depen
E I. COMPAR
III. C
etworks are u
properly mig
o before instal
tified, evaluate
unter measures
though no s
tally unbreaka
ndable one. It
RISON OF SECURIT
CONCLUSION
used in busines
ght cause ver
lling any such
ed, and based
s must be ins
security syst
able, 802.11i R
suffers none
TY PROTOCOLS
ss WLANs th
ry serious ris
h networks, all
d on the result
talled to secu
tem can eve
RSN security s
of the proble

hat are
sks to
l risks
ts, the
ure the
er be
seems
ems of
old
802
can
tim
sec
min
usa
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]

der mechanism
2.11i WPA-2 i
n rely on until
me being in ter
urity preferre
nimum cost pr
age of WPA is
IEEE: IEEE 80
and Physica
http://standards
Li, L., Zhu, L.
WPA security
Telecommunica
Sepehrdad, P.,
Exploitation of
Gong, G., Stins
91. Springer, H
Sepehrdad, P.,
RC4 Distinguis
Vuagnoux, M.:
channels. PhD
EPFL (2010).
WinPcap: Win
http://www.win
http://www.lyle
insecure-as-i-ex
KrishnamRaju,
K.V.S.V.N: Fo
Using CasperF
information an
India (Septemb
Hung, C.C., W
Key Encryption
Journal of Com
Hiertz, G., Dent
B.: The IEEE 8
6270 (2010).
] A.H.D. Lashkar
protocols (WEP
Information Te
Beijing, ICCSIT
] E.J.M.A. Filho,
Security versu
WPA in 802.
Wireless and O
5, 2007.
] A. D. Potorac an
802.11 WLAN
Control System
ms and protoco
is a wireless s
l its vulnerabi
rms of cost v
ed then WP
referred WEP
recommende
REFER
02.11: Wireless L
al Layer (PH
.ieee.org/about/ge
., Ju, T.: The Re
system. Journal
ations 24(1), 788
, Vaudenay, S.
New Biases in R
son, D.R. (eds.) S
eidelberg (2011).
Vaudenay, S., V
shing WPA. Acce
Computer Aided
thesis, EcolePoly
npcap: The wind
npcap.org/Wpa-W
ebackemort.com/b
xpected (accessed
K.V., Vallikum
ormal Verification
FDR. In: Interna
nd communicatio
er 2010).
Wu, E.H., Wu, C.L
n Scheme for the
mputers 18(4) (200
teneer, D., Stibor
802.11 universe. I
ri, M.M.S. ,Sama
P, WPA and WP
echnology, 2nd
T 2009, pp.48-52
, P.N.L. Fonseca,
us Bandwidth: T
.11g Network,
Optical Communic
nd D. Balan.The
Throughput, Co
ms, vol. 2, pp. 47-5
ols namely WE
security protoc
lities are brou
versus security
A-2 must b
must be emp
d.
RENCES
LAN Medium Ac
HY) Specifica
et/802/802.11.htm
esearch of 802.1
of Nanjing Univ
81 (2004).
., Vuagnoux, M
C4. In: Biryukov
SAC 2010. LNCS
.
Vuagnoux, M.: S
pted at EUROCR
d Cryptanalysis f
ytechniqueFeder
dows packet cap
Wpa2,
blog/2008/5/10/
d on March 14, 20
mari, V., Sandee
n of IEEE802.16
ational Conferen
on technologies,
L., Gau, R.L., Ch
e Next Generation
08).
r, L., Zang, Y., Co
IEEE Communic
di, B., "A survey
A2/802.11i)," Co
IEEE Internatio
, 2009.
, M.J.S. Leitao, a
The Support of M
IFIP Internatio
cations Networks
e Impact of Secur
omputer
52, 2009.
EP and WPA.
col that anybo
ught out. For t
y options, if f
e employed,
loyed, otherw
cess Control (MA
ations (200
ml.
1 Wireless Netw
versity of Posts
M.: Discovery
, A.,
S, vol. 6544, pp.
Statistical Attack
RYPT 2011 (2011
from Ciphers to S
rale de Lausanne
pture library (200
wpa-wpa2
010).
epvarma, N., R
6m PKMv3 Proto
nce on advances
ICT 2010, Ker
hen, Y.C.: A Mu
n Wireless Netwo
osta, X.P., Walke,
cations Magazine
on wireless secur
omputer Science
onal Conference
and P.S.F. de Bar
Mechanisms WE
nal Conference
s, WOCN '07, pp
ity Overheads on
Science
So
ody
the
full
if
wise
AC)
07),
work
and
and
74
k on
).
Side
e
08),
-as-
Raju,
ocol
s in
rala,
ulti-
ork.
,
48,
rity
and
on
rros,
P e
on
p. 1-

and
ISBN: 978-81-920575-7-6:: doi: 10. 74404 /ISBN_0768
Survey of Security Issues in Mobile Ad-Hoc Network

Praful R. Pardhi
Assistant Professor
Shri Ramdeobaba College of
Engineering & Management
Nagpur, Maharashtra

Abstract - An ad-hoc network is a self-configuring network of
wireless links connecting mobile nodes. These nodes may be
routers and/or hosts. The mobile nodes communicate directly
with each other and without the aid of access points, and
therefore have no fixed infrastructure. They form an
arbitrary topology, where the routers are free to move
randomly and arrange themselves as required.
Each node or mobile device is equipped with a transmitter
and receiver. They are said to be purpose-specific,
autonomous and dynamic. This compares greatly with fixed
wireless networks, as there is no master slave relationship
that exists in a mobile ad-hoc network. Nodes rely on each
other to established communication, thus each node acts as
a router. Therefore, in a mobile ad-hoc network, a packet
can travel from a source to a destination either directly, or
through some set of intermediate packet forwarding
nodes.In a wireless world, dominated by Wi-Fi,
architectures which mix mesh networking and ad-hoc
connections are the beginning of a technology revolution
based on their simplicity.
Ad hoc networks [1] date back to the Seventies. They were
developed by the Defense Forces, to comply with a military
framework. The aim was to rapidly deploy a robust, mobile
and reactive network, under any circumstances. These
networks then proved useful in commercial and industrial
fields, first aid operations and exploration missions. Ad hoc
networks, also called peer-to-peer networks, still have a
long way to go in order to be fully functional and
commercial, as it has its defects such as security and routing
which we will discuss further.
Key Words: Mobile Ad Hoc Network, Security, Intrusion
Detection, Secure Routing
I. MANET ENVIRONMENT VARIATIONS
The different MANET environment variations are
listed as follows taking its dynamic topology in to
consideration:
1.1 In MANETS all nodes have identical capabilities
and responsibilities, which are termed as fully
symmetric environment. A mobile ad hoc network
(MANET) is a network comprising wireless mobile nodes
that communicate with each other without centralized

control or established infrastructure. These nodes which
are within each others radio range can communicate
directly, while distance nodes rely on their neighboring
nodes to forward packets. In MANETS every node can be
a host or router. In MANET environment, nodes are free to
join or leave the network at any point of time, resulting in a
highly dynamic network environment compared to wired
network.
1.2 The Asymmetric Capabilities in MANETS include
transmission ranges and radios ranges which may differ.
Battery life, speed of movement and processing capacity
will be different at different nodes.
1.3 Asymmetric Responsibilities include that only
some nodes may route packets in the network or some
nodes may act as leaders for nearby nodes like cluster
head.
1.4 Traffic characteristics may differ in different ad hoc
networks like bit rate, timeliness constraints, reliability
requirements, unicast or multicast or geocast, host-based
addressing or content-based addressing or capability-based
addressing.
1.5 MANETS may co-exist and also co-operate with an
infrastructure based network.
1.6 Mobility patterns may be different like people
sitting at an airport lounge, citywide taxi cabs, military
movements and personal area networks. The performance
of a mobile ad hoc network is dependent on the node
mobility pattern as well as topology, data traffic patterns,
and radio interference.
1.7 Mobility characteristics include speed,
predictability, direction of movement, pattern of
movement, uniformity of mobility characteristics among
different nodes.
II. ROUTING PROTOCOLS
Routing protocols between any pair of nodes within
an ad hoc network can be difficult because the nodes can
move randomly and can also join or leave the network.
This means that an optimal route at a certain time may
not work seconds later. Discussed below are three
categories that existing ad-hoc network routing protocols
fall into:
Table Driven Protocols
On Demand Protocols
Hybrid Protocols
P
i
i
n
m
t
r
o
m
t
c
w
l
t
r
R
o
b
(
(
R
w
u
f
W
d
i
n
n
D
D

2.1 Table D
Table Dr
Proactive Pro
independent o
information t
nodes in the n
move data am
type of protoc
routing loops.
of the networ
may react to c
traffic is affe
could create u
with little dat
limited resou
therefore they
routing soluti
Routing is an
of the exampl
below.
2.1.1 De
(DSDV)
2.1.2 Clus
(CGSR)
2.1.3 Wire
2.1.4 STA
2.2 On De
On Dema
Reactive Prot
when they are
updating of ev
focuses on ro
When a rou
destination fo
it starts a rou
node to the o
node in-betw
Demand prot
Diagram 1: Ad-ho
Driven Routin
iven Routing
otocols, work
of traffic dem
o store the l
network and th
mong differen
col is slow to c
. These protoc
rk and this ca
change in the
ected by the
unnecessary o
ta traffic, Tab
urces such as
y might not
ion for Ad-h
example of a
es under the c
estination Se
ster-Head Gate
eless Routing p
AR
emand Protoco
and Routing
tocols, establis
e required to r
very possible r
outes that are
ute is require
r which it doe
ute discovery p
other until it a
ween has a r
tocols are ge
oc Mobile Routin
ng Protocols
g Protocols
out routes in
mands. Each no
location inform
his informatio
nt nodes in th
converge and
cols keep a co
an be a disad
e network topo
topology mod
overhead. Eve
ble Driven Pr
s power and
be consider
hoc Networks
Table Driven
category of Ta
equenced D
eway Switch R
protocol (WR
ols
Protocols, a
sh routes betw
route data pack
route in the ne
being used o
ed by a sou
es not have ro
process which
arrives at the
route to the
enerally cons
ng Protocols
also known
the backgrou
ode uses routi
mation of oth
n is then used
he network. T
may be prone
onstant overvi
dvantage as th
ology even if
dification whi
en in a netwo
rotocols will u
link bandwid
red an effecti
s. Fisheye St
n Protocol. So
able driven giv
istance Vec
Routing proto
RP)
also known
ween nodes on
kets. There is
etwork instead
or being set u
urce node to
oute informatio
h goes from o
destination o
destination.
sidered effici

as
und
ing
her
d to
This
e to
iew
hey
no
ich
ork
use
dth
ive
tate
me
ven
ctor
col
as
nly
no
d it
up.
o a
on,
one
r a
On
ent
wh
tra
di
co
Pr
an
Pr
ex
giv
(C
Ro
Th
es
rep
the
ha
a
nu
in
(Z
att
ch
ma
ch
au
av
no
ne
co
rep
rem
att
ne
no
wh
ca
se
hen the route
ansfer because
scovery step
ommunication
rotocols more
nd low mobi
rotocol is Dy
xamples under
ven below.
2.2.1
2.2.2 Dyn
2.2
Algorit
2.2
Vector
2.2.5
CBRP)
2.2
discove
2.3 Hybrid R
Hybrid Rou
outing Protoc
hey use distan
tablish the b
port routing in
e topology of
as its own rout
zone radius, w
umber of hop
formation for
ZRP) is an exam
Ad-hoc netw
tacks and de
hallenges of d
ain reasons fo
hannel, insecur
uthority, lack
vailability of re
odes. Generall
etwork, we ex
onfidentiality,
pudiation.
Availability
main operatio
tacks.
Confidential
ever disclosed
Authenticati
ode with which
Integrity gua
hen transferre
Non-repudia
annot deny hav
curity requir
discovery is
e the network
p is low
bandwidth.
suited to larg
ility. An exa
ynamic Sourc
r the category
Associative
namic Source
.3 Te
thm (TORA)
.4 Ad
r routing proto
Cluster-bas
.6 Re
ery Ad Hoc R
Routing Protoc
uting Protoco
ols with On D
nce-vectors fo
best paths to
nformation on
f the network
ting zone, the
which is defin
ps. Each node
r its own zon
mple of a Hyb
III. SE
works are hig
ealing with
developers of
or this difficult
re operating e
of associati
esources, and
y, when con
xamine it unde
authenticati
refers to the
onal at all tim
lity ensures t
to certain use
ion is the abil
h it is commun
arantees that a
d.
ation states tha
ving sent it. A
rements cause
less frequent
k traffic cause
compared t
This makes
e networks w
ample of an
ce Routing.
y of On Dem
e Based Routin
Routing Proto
emporally Ord
d-hoc On-Dem
col (AODV)
sed Routin
elative Dist
Routing Protoc
cols
ols combine
Demand Rout
for more prec
destination n
nly when there
. Each node i
e size of which
ned by a metr
e keeps a reco
ne. Zone Rou
brid routing pr
ECURITY
ghly vulnerab
this is one
these networ
ty are shared b
environment,
ion among n
physical vuln
nsidering the
er the heading
ion, integrity
fact that the
mes despite den
that certain i
ers.
ity of a node
nicating.
a message is n
at the sender o
An ad-hoc netw
ed by its la
t than the dat
d by the rout
to the tota
On Demand
ith light traffi
On Demand
Some of th
mand Protocol
ng (ABR)
ocol (DSR)
dered Routing
mand Distanc
ng Protoco
tance Micro
ol (RDMAR)
Table Based
ting Protocols
cise metrics to
networks, and
e is a change in
in the network
h is defined by
ric such as th
ord of routing
uting Protoco
rotocol.
ble to security
of the main
rks today. Th
broadcast radio
lack of centra
nodes, limited
erability of th
security of
gs; availability
y and non
network mus
nial of servic
information i
to identify th
never corrupted
of the messag
work has extr
ack of prope
ta
te
al
d
c
d
e
ls
g
e
ol
o

d
s.
o
d
n
k
y
he
g
ol
y
n
he
o
al
d
e
a
y,
n-
st
e
is
e
d
e
ra
er
infrastructure and the dynamic relationship between the
nodes in the network. Because of the lack of
infrastructure, accountability is very difficult to
determine as there is no central authority which can be
referenced when it comes to making trust decisions about
other parties in the network. The dynamic relationship
between the nodes leaves very little opportunity for the
nodes to form trust relationships with each other. In an
ad-hoc network, nodes must act as both terminals and
routers for other nodes. Because there are no dedicated
nodes, a secure routing protocol is needed. Multi hop
routing protocols are usually employed. These can lead
to problems due to non-cooperating nodes and denial of
service attacks.
IV. SECURITY ANALYSIS OF MANET

4.1 Particularity of MANET Security
MANET is different from the fixed IP network as well
as the general wireless network, which poses new
challenges to guarantee its security as following: the
wireless communication manner makes the threat of being
attacked suffering from wiretapping greatly improved;
multi-hop routing method may induce a series of security
issues, such as black hole, wormhole and etc; the self
organization of mobile nodes blurs the boundary of
network; the independence of node mobile make the
physical security fragile.
4.2 Security Threats on MANET
The security threats confronted by MANET is the
extension and expansion of that be confronted by wired
network in the wireless field, which mainly comes from
wireless channels and networks. The threats can be divided
into two categories of passive attack and active attack.
4.2.1 Passive attack
Passive attack [13] is essentially to listen or
surveillance the message transmission process to obtain
some secret, which mainly includes two kinds of
wiretapping and traffic analysis. Wiretapping is to intercept
packets being transmitted and access to confidential
information. Traffic analysis is to analyze characteristics of
packet frequency, length and etc to hypothesize the content
of communication.
4.2.2 Active attack
Comparing with passive attack, the passive attack
usually does not change the transmission data, while active
attack will tamper data stream or create false data stream.
The active attack includes four kinds of message replay,
fraud counterfeiting, message tampering and denial of
service.

Message replay: After the attacker intercepted
messages, it will store the message and re-transmit the
message to produce an unauthorized effect. The MANET
is likely to be attacked by message replay, in that the
message is transmitted in the air and can easily be
intercepted.

Fraud counterfeiting: When one entity behaves as
another entity to carry out network activities, the
counterfeiting takes place. Lack of protection or weak
authentication mechanism may cause such security threats,
such as replay identification sequence to make believe a
privileged entity.

Message tampering: After the packet or message
sequence being intercepted, the content is modified, or
deliberately delayed transmission, or a passive change in
the order. This kind of attack is attack on the integrity of
message.

Denial of service: When an authorized entity
cannot access to should be network resources or
emergency operations be denied, the attack of denial of
service occurs. In the MANET, it is easy to achieve service
interruption, so wireless communication can be used to
shield the communications spectrum.
V. MANET THRUST AREAS
The following list of thrust areas shows the
inefficiencies and limitations that have to be overcome in a
MANET environment:
5.1 Limited wireless transmission range: In wireless
networks the radio band will be limited and hence data
rates it can offer are much lesser than what a wired
network can offer. This requires the routing protocols in
wireless networks to use the bandwidth always in an
optimal manner by keeping the overhead as low as
possible. The limited transmission range also imposes a
constraint on routing protocols in maintaining the
topological information. Especially in MANETS due to
frequent changes in topology, maintaining the topological
information at all nodes involves more control overhead
which, in turn, results in more bandwidth wastage.
5.2 Time-varying wireless link characteristics: The
wireless channel is susceptible to a variety of
transmission impediments such as path loss, fading,
interference and blockage. These factors resist the range,
data rate, and the reliability of the wireless transmission.
The extent to which these factors affect the transmission
depends upon the environmental conditions and the
mobility of the transmitter and receiver. Even the two
different key constraints, Nyquists and Shannons
theorems, that govern the ability to transmit information
at different data rates can be considered.
5.3 Broadcast nature of the wireless medium: The
broadcast nature of the radio channel, that is, transmissions
made by a node are received by all nodes within its direct
transmission range. When a node is receiving data, no
other node in its neighborhood, apart from the sender,
should transmit. A node should get access to the shared
medium only when its transmissions do not affect any
ongoing session. Since multiple nodes may contend for the
channel simultaneously, the possibility of packet collisions
is quite high in wireless networks. Even the network is
susceptible to hidden terminal problem and broadcast
storms. The hidden terminal problem refers to the collision
of packets at a receiving node due to the simultaneous
transmission of those nodes that are not within the direct
transmission range of the sender, but are within the
transmission range of the receiver.
5.4 Packet losses due to transmission errors: Ad hoc
wireless networks experiences a much higher packet loss
due to factors such as high bit error rate (BER) in the
wireless channel, increased collisions due to the presence
of hidden terminals, presence of interference, location
dependent contention, uni-directional links, frequent path
breaks due to mobility of nodes, and the inherent fading
properties of the wireless channel.
5.5 Mobility-induced route changes: The network
topology in an ad hoc wireless network is highly
dynamic due to the movement of nodes; hence an on-
going session suffers frequent path breaks. This situation
often leads to frequent route changes. Therefore mobility
management itself is very vast research topic in ad hoc
networks. Mobility-induced packet losses:
Communication links in an ad hoc network are unstable
such that running conventional protocols for MANETS
over a high loss rate will suffer from severe performance
degradation. However, with high error rate, it is very
much difficult to deliver a packet to its destination.
5.6 Battery constraints: This is one of the limited
resources that form a major constraint for the nodes in an
ad hoc network. Devices used in these networks have
restrictions on the power source in order to maintain
portability, size and weight of the device. By increasing the
power and processing ability makes the nodes bulky and
less portable. So only MANET nodes has to optimally use
this resource.
5.7 Potentially frequent network partitions: The
randomly moving nodes in an ad hoc network can lead to
network partitions. In major cases, the intermediate nodes
are the one which are highly affected by this partitioning.
5.8 Ease of snooping on wireless transmissions
(security issues): The radio channel used for ad hoc
networks is broadcast in nature and is shared by all the
nodes in the network. Data transmitted by a node is
received by all the nodes within its direct transmission
range. So an attacker can easily snoop the data being
transmitted in the network. Here the requirement of
confidentiality can be violated if an adversary is also able
to interpret the data gathered through snooping.

VI. CONCLUSION
Studied Ad-hoc networks concepts in depth, we
believe that they will be the future of wireless
networking. It is true that performance suffers as the
number of devices grows and large ad-hoc networks
become difficult to route and manage. However, much
time is being devoted to achieving routing stability, and a
few technical issues need to be solved before they
become common place. The area of ad hoc networks is a
very fast growing area, and due to the vast research in
them, we are seeing these problems disappear and they
are coming into a world of their own.

REFERENCES

[1] C. Perkins and E Royer, Ad Hoc On-Demand Distance
Vector Routing, 2nd IEEE Wksp. Mobile Comp. Sys.
and Apps., 1999.
[2] D. Johnson and D. Maltz, Dynamic Source Routing in
Ad Hoc Wireless Networks, Mobile Computing, T.
Imielinski and H. Korth, Ed., Kluwer, 1996.
[3] IEEE Std. 802.11, Wireless LAN Medium Access Control
(MAC) and Physical Layer (PHY) Specifications, 1997.
[4] B. Schneier, Secret and Lies, Digital Security in a Networked
World, Wiley, 2000.
[5] Y. Hu, A. Perrig, and D. Johnson, Ariadne: A Secure
On-demand Routing Protocol for Ad Hoc Networks,
ACM MOBICOM, 2002.
[6] M. Zapata, and N. Asokan, Securing Ad Hoc Routing
Protocols, ACM WiSe, 2002.
[7] B. Dahill et al., A Secure Protocol for Ad Hoc Networks,
IEEE ICNP, 2002.
[8] Y. Hu, A. Perrig, and D. Johnson, Packet Leashes: A
Defense against Wormhole Attacks in Wireless Networks,
IEEE INFOCOM, 2002.
[9] N. Borisov, I. Goldberg, and D. Wagner, Intercepting
Mobile Communications: The Insecurity of 802.11,
ACM MOBICOM, 2001.
[10] V. Gupta, S. Krishnamurthy, and M. Faloutsos, Denial
of Service Attacks at the MAC Layer in Wireless Ad Hoc
Networks, IEEE MILCOM, 2002.
[11] P. Kyasanur, and N. Vaidya, Detection and Handling of
MAC Layer Misbehavior in Wireless Networks, DCC, 2003.
[12] C. Perkins, and P. Bhagwat, Highly Dynamic Destination-
Sequenced Distance-Vector Routing (DSDV) for
Mobile Computers, ACM SIGCOMM, 1994.
[13] P. Papadimitratos, and Z. Haas, Secure Routing for Mobile Ad Hoc
Networks, CNDS, 2002.
ISBN: 978-81-920575-7-6 :: doi: 10. 73991/ISBN_0768
Survey of Reactive Routing Protocols in Mobile Adhoc Networks

V. Jayalakshmi
Assistant Professor, Computer Applications Department,
Sudharsan Engineering College, Pudukkottai,
Tamilnadu, India

Abstract A Mobile ad hoc networks (MANETs) consist of
a collection of wireless mobile nodes which dynamically
exchange data among themselves without the reliance on a
fixed base station or a wired backbone network. In such
networks, the wireless mobile nodes may dynamically enter
the network as well as leave the network. In the last 15
years, the wireless networking community designed a
number of new routing protocols targeting the various
scenarios of this design space. Any on-demand routing
protocol must utilize some type of routing cache in order to
avoid the need to rediscover each routing decision for each
individual packet. So, caching is an important part of on-
demand routing protocols for wireless ad hoc networks The
objective of this paper is to survey the various reactive
routing protocols which uses the route caching technique
for route repository and compare them. We strive to
uncover the requirements considered by the different
protocols, the resource limitations under which they
operate, and the design decisions made by the authors.

Keywords Ad hoc ; On demand, Route Cache, Routing
Protocols
I. INTRODUCTION
Recently, there has been a tremendous interest in
broadband wireless access systems, including wireless
local area networks (WLAN), broadband wireless access
and wireless personal area networks (WPAN). The
advent of ubiquitous computing and the proliferation of
portable computing devices have raised the importance
of mobile and wireless networking.This domain is a
subject of a huge research and many standardization
activities are undertaken throughout the world, in many
3G/4G related study committees like ITU-R, ETSI
BRAN and IEEE 802. Research prototyping is currently
underway at many research academic and industrial
institutions [1].
In the context of these systems, mobile ad hoc networks
(MANETs) have appeared with specific configurations.
They provide a powerful paradigm for modeling open
self-configuring wireless networks and seem so
appropriate to use in the fourth generation of mobile
networks.
A MANET is an autonomous collection of mobile nodes
forming a dynamic network and communicating over
wireless links. Users are allowed to communicate with
each other in a temporary manner with no centralized
administration and in a dynamic topology that changes
frequently. Due to the limited propagation range of
wireless environment, routes in ad hoc networks are
multihop, and mobile nodes in this network dynamically
establish routing among themselves to form their own
network on the fly [2]. Each participating node acts
both as a host and a router and must therefore be willing
to forward packets for other nodes. Nodes in such a
network move arbitrarily, thus network topology changes
frequently,
unpredictably, and may consist of unidirectional links as
well as bidirectional links. Moreover, wireless channel
bandwidth is limited. The scarce bandwidth decreases
even further due to the effects of signal interference, and
channel fading. Network hosts of ad hoc networks such
as laptops and personal digital assistants operate on
constrained battery power, which will eventually be
exhausted and limited CPU and storage capacity.
MANETs strictly depend on radio links. Actually, a
wireless link is the most variable and unpredictable
communication channel. In addition, ad hoc networks are
vulnerable to attacks and have limited physical security.
The increased possibility of eavesdropping, spoofing and
denial-of-service attacks should be carefully considered.
Because ad hoc networks do not typically allow the same
aggregation techniques that are available to standard
Internet routing protocols, they are vulnerable to
scalability problems. These drawbacks lead to define a
set of underlying assumptions and performance concerns
for protocol design.
Ad hoc networks do not rely on any pre-established
infrastructure and can therefore be deployed in places
with no infrastructure. This is useful in Multicasting has
emerged as one of the most focused areas in the field of
networking. As the technology and popularity of the
Internet have grown, applications that require
multicasting (e.g., video conferencing) are becoming
more widespread. Another interesting recent
development has been the emergence of dynamically
reconfigurable wireless ad hoc networks to interconnect
mobile users for applications ranging from disaster
recovery to distributed collaborative computing. disaster
recovery situations and places with non-existing or
damaged communication infrastructure where rapid
deployment of a communication network is needed [3].
Ad hoc networks can also be useful in conferences where
people participating in the conference can form a
temporary network
without engaging the services of any pre-existing
network [4].
Routing in mobile ad hoc networks is a significant
research topic where various mechanisms have been
already proposed concerning this issue. In fact,
conventional routing protocols are not well suited in ad
hoc environment for several reasons. Firstly, they are
designed for static topology, which is not the case in ad
hoc network. Secondly, they are highly dependent on
periodic control messages; this is in contradiction with
resource-limited ad hoc environment. Moreover, classical
protocols try to maintain routes to all reachable
destinations, which wastes resources.
Another limitation comes from the use of bi-directional
links, which is not always the case in ad hoc
environment. Actually, there is a need for new routing
protocols, adapting to the dynamic topology and the
wireless links limitations. Routing protocols in such
networks should provide a set of features including [4]:
distributed operation, loop freedom, on-demand
based operation, unidirectional link support, power
conservation, multiple routes, efficiency, scalability,
security and quality of service support.
The routing protocols may commonly be categorized as
table-driven and on-demand. Several routing protocols
have used on-demand mechanisms in ad hoc network,
allowing them to dynamically adapt the level of routing
protocol activity required to correctly handle the offered
traffic. Any on-demand routing protocol must utilize
some type of routing cache in order to avoid the need to
rediscover each routing decision for each individual
packet. So, caching is an important part of on-demand
routing protocols for wireless ad hoc networks. This
paper discusses the various reactive routing protocols
which uses the route cache technique for storing the
routes.
This paper is organized as follows. Section II gives the
Overview of the routing protocols in MANETs. Section
III gives a brief survey of the various reactive routing
protocols which uses route cache. Section IV gives the
comparison between these protocols. Finally, section V
provides conclusion of our work.
II. OVER VIEW OF ROUTING PROTOCOLS
The basic routing problem is that of finding an ordered
series of intermediate nodes that can transport a packet
across a network from its source to its destination by
forwarding the packet along this series of intermediate
nodes.
The challenge in creating a routing protocol for ad hoc
networks is to design a single protocol that can adapt to
the wide variety of conditions that can be present in any
ad hoc network over time. The routing protocol must
perform efficiently in environments in which nodes are
stationary and bandwidth is not a limiting factor. Yet, the
same protocol must still function efficiently when the
bandwidth available between nodes is low and the level
of mobility and topology change is high. Because it is
often impossible to know a priori what environment the
protocol will find itself in, and because the environment
can change unpredictably, the routing protocol must be
able to adapt automatically
Many classifications are found in the literature. Many
protocols have been proposed for MANETs. These
protocols can be divided into three categories: proactive,
reactive, and hybrid. Proactive methods maintain routes
to all nodes, including nodes to which no packets are
sent. Such methods react to topology changes, even if no
traffic is affected by the changes. They are also called
table-driven methods. Reactive methods are based on
demand for data transmission. Routes between hosts are
determined only when they are explicitly needed to
forward packets. Reactive methods are also called on-
demand methods. They can significantly reduce routing
overhead when the traffic is lightweight and the topology
changes less dramatically, since they do not need to
update route information periodically and do not need to
find and maintain routes on which there is no traffic.
Hybrid methods combine proactive and reactive methods
to find efficient routes, without much control overhead
III. REACTIVE ROUTING PROTOCOLS A BRIEF SURVEY
MANET is one that searches for and tries to discover a
route to some destination node only when a sending node
generates a data packet addressed to that node. In order
to avoid the need for such a route discovery to be
performed before each data packet is sent, such routing
protocols must cache previously discovered routes
In the following we review several representative
examples of reactive routing protocols which uses route
cache for route repository
A. Dynamic source routing (DSR)

Johnson et al. propose one of the most widely known
routing algorithms, called Dynamic Source Routing [5]
which is an on-demand algorithm and it has route
discovery and route maintenance phases.
Route discovery contains both route request and route
reply messages. In the route discovery phase, when a
node wishes to send a message, it first broadcasts a route
request packet to its neighbors. Every node within a
broadcast range adds their node id to the route request
packet and rebroadcasts. Eventually, one of the broadcast
messages will reach either the destination or a node
which has a recent route to the destination. Since each
node maintains a route cache, it first checks its cache for
a route that matches the requested destination.
Maintaining a route cache in every node reduces the
overhead generated by a route discovery phase. If a route
is found in the route cache, the node will return a route
reply message to the source node rather than forwarding
the route request message further. The first packet that
reaches the destination node will have a complete route.
DSR assumes that the path obtained is the shortest since
it takes into consideration the first packet to arrive at the
destination node. A route reply packet is sent to the
source which contains the complete route from the
source to the destination. Thus, the source node knows its
route to the destination node and can initiate the routing
of the data packets. The source caches this route in its
route cache.

In the route maintenance phase, route error and
acknowledgements packets are used. DSR ensures the
validity of the existing routes based on the
acknowledgements received from the neighbouring
nodes that data packets have been transmitted to the next
hop. Acknowledgement packets also include passive
acknowledgements as the node overhears the next hop
neighbour is forwarding the packet along the route to the
destination. A route error packet is generated when a
node encounters a transmission problem which means
that a node has failed to receive an acknowledgement.
This route error packet is sent to the source in order to
initiate a new route discovery phase. Upon receiving the
route error message, nodes remove from their route
caches the route entry using the broken link.

B. Link availability-based QoS-aware (LBAQ) routing

Yu et al. propose the LBAQ [6] routing protocol based
on node mobility prediction and link quality
measurement. While a node moves, it may experience
varying capacity, reliability and bandwidth availability.
Instead of trying to predict the mobility patterns of the
mobile links, the link availability is incorporated into the
routing metrics to help choose the link with the highest
availability in the route. The expected transmission count
(ETX) along with power consumption is also used as a
route metric. The route metrics used are:

Link availability: probability that two nodes of a link stay
directly connected at a time t
0
+ t provided that they were
connected at time t
0
.

Link quality: the number of retransmissions required to
send a packet on a link between two nodes (ETX). This is
computed by measuring the loss rate of broadcast packets
between the node pairs.

Energy consumption: power consumption at each hop.
Using the metrics above, a combined cost function is
designed based on which the route is constructed.
Di(T
i
)=
0
(E
i
)+
1
(Qi)+
2
(1-L
i
(t)),where
0
,
1
and
2
= 1
0

1
are the weighting coefficients, L
i
and Q
i
are
availability and quality of link i, Ei is the energy
consumed on link i, Ti is the traffic carried on path i, and
Di is the final cost of the route. A source node sends a
RREQ packet during route discovery. When an
intermediate node receives this RREQ packet, it
calculates its own cost from the equation above and
forwards the packet with this new information. The total
cost to the destination is computed when the RREQ
reaches the destination node. At the destination, the node
waits for a fixed number of RREQs before selecting the
route with the least cost. If a link breaks, then an
alternate route is selected for data transmission.

C. Gathering based routing protocol (GRP)
Ahn presents the gathering based routing protocol
[7]which collects network information during route
discovery to be used later by the source node. Initially,
the source node broadcasts a destination query (DQ)
packet which is continuously forwarded towards the
destination. This process is similar to the route discovery
process in DSR or AODV using RREQs. When the DQ
reaches the destination node, a network information
gathering (NIG) packet is forwarded by the destination.
This NIG packet is then propagated over the entire
network collecting information. Every NIG packet is
assigned a sequence number. Nodes receiving a NIG
packet with a new sequence number attach network
information to this packet and forward it along effective
outgoing links (EOLs). EOLs are those links in which the
NIG packets are not received. Once the NIG packet
arrives at the source node, the optimal path is calculated
by the source node based on the collected network
information. Data packets are then transmitted through
this calculated route.

D. Source routing with local recovery (SLR)

Sengul and Kravets [8] start from the observation that
although on-demand routing reduces the cost of routing
in high mobility environments, the route discovery
process, which is typically done through network-wide
flooding, consumes a significant amount of bandwidth.
This is especially expensive if the route discovery must
be repeated due to links broken due to node mobility. To
alleviate this problem, the authors propose bypass
routing, a process which patches a route using local
information acquired on-demand, without the need of
network-wide flooding. The SLR protocol is an
implementation example of the bypass routing approach.
Bypass routing in SLR localizes the reaction to a route
failure and initiates only a local recovery procedure.
Using link state information from the MAC layer, a local
patch on the route is created bypassing the broken link.
By using local recovery with link state information, the
chances of recovering from a broken route increases.
When a route fails, first the local route cache is searched
for an alternate route. If no route is available, bypass
recovery is initiated by querying the neighbors. Based on
replies from these nodes, the path is again reconstructed.
MAC caches provide the connectivity information to
immediate one-hop neighbors and the state is updated by
each node whenever any communication is heard in the
neighborhood. MAC caches also help determine the
validity of the routes in the route caches. Error recovery
is initiated by salvaging used route caches, bypass
recovery and error reporting (see Fig 1). The node looks
for alternative routes to the destination in its cache to
salvage a packet. If the salvaged packet reaches its
destination, th
message from
route is in fac

E. Hint based

Beraldi et
framework w
packets towar
The meta-inf
each node. A
to the destina
value indicat
neighbourhoo
higher probab
one-hop neigh
When a no
will try sendin
has not been
trying various
such neighbo
nodes broadc
then used to c
neighbour. T
calculate the
broadcasting
beacon messa
at most L hop
the protocol.
value is suff
Smaller looka
of lower contr

F. Low overh

Yu et al. [10
information f
is down, the
failed links w
initial main r
destination p
transmitted.
neighboring n
nodes are ac
node on the
candidate nod
used during t
he source rec
m the destina
ct an alternate
Fig. 1. Erro
d probabilisti
al. [9] propo
which uses m
rds the genera
formation is p
A hint, hid, com
ation node d,
ting the chan
od of node d. A
bility with hid
hbours.
ode forwards a
ng to the neig
tried before fo
s alternatives
our exist, the
cast periodic
create a vecto
This time info
e hint values
the control m
ages. Every n
ps around it. T
The authors
ficient for a
ahead values
rol overheads
head dynamic
] repair broke
from overhear
proposed pro
with backup li
route is const
pair along w
While thes
nodes overhe
ctually potenti
main route
des are record
the route reco
ceives an enha
ation stating t
to the broken
or recovery examp
c protocol
ose a probabi
meta-informa
al direction o
provided in t
mputed by no
, is represent
nce of node i
A lower hint v
d = 0 when n
a packet to th
ghbour with th
for the same p
in the order o
packet is di
heartbeat pa
or of time info
ormation vect
. Hints are
messages thro
node receives
This is the loo
show that a
node to gath
also have an
.
c route repairi
en routes dyna
ring the nodes
tocol intellige
inks along the
tructed betwe
which the d
se packets
ear these tran
ial candidates
in case of
ded in the pa
onstruction in
anced route e
that the salva
n route
ple
ilistic forward
ation to forw
of the destinat
terms of hint
ode i with resp
ted by a posi
i residing in
value equates
nodes i and d
he destination
he best hint wh
acket (effectiv
of the hint). If
iscarded. All
ackets which
ormation for e
tor is utilized
disseminated
ough the perio
hints from no
okahead value
small lookah
her correct hi
added advant
ing
amically by us
s. Once the ro
ently replaces
e main route.
een a source
data packets
are forward
smissions. Th
s for replacin
a failure. Th
acket headers
n the future.
error
aged
ding
ward
tion.
ts at
pect
itive
the
to a
d are
d it
hich
vely
f no
the
are
each
d to
d by
odic
odes
e for
head
ints.
tage
sing
oute
s the
An
and
are
ded,
hese
ng a
hese
and
The
rou
pac
MR
arri
sen
bet
use
ma
wit
the
che
so,
cou
hop
tow
ena
RE
S
hig
rea
per
pro
spe
rou
inc
mo
bet
tem
dem
tha
des
O
abo
be

The
res
zon
the
bac
inte
orig
we
app

ute constructio
cket initially, s
REQ is flood
ives at the des
nt back. The
tween the sou
ed during rou
ain route goes
th the hop cou
e main route
ecks to see if
it sends a re
unt value is c
p count is sm
wards the dest
ables nodes c
EPQ and reply
IV. CO
Source routing
gh interest for
asons for this.
rceived as e
otocols from
ecific to ad
uting is bes
arnations of t
obility of the
tween a sourc
mporary in nat
mand, but the
at the efficien
sign considera
One of the m
out the handli
approached fr
Impro
protoc
dynam
bypas
Choos
stable
future
LABQ
consid
challe
e novel nature
earchers to b
ne of network
eory and flo
ckground).
erdisciplinary
ginated routin
have an ap
proach.
on uses a ma
similar to a RR
ed throughou
stination, the m
MRRP keep
urce and dest
ute repair whe
s down. A re
unt of that no
failure. A n
it has a saved
epair reply (R
compared with
maller, it pro
tination, else t
closer to the
with a valid b
OMPARISON OF
g protocols ha
the ad hoc co
First, while ta
extensions or
the wired do
hoc networ
st adapted t
he ad hoc net
nodes is so
ce and a des
ture. Not only
probability o
ncy of respond
ation.
main improvem
ng of recover
rom several di
ving the spee
cols of this
mic route repa
s routing).
sing paths whi
: LBAQ [6]
link availabil
Q [6] extend
dering the Qu
nge.
e of source ori
branch out fro
king (which
ow optimizat
A number
approaches h
ng. In the pape
pproach based
ain route req
RREQ packet i
ut the networ
main route rep
ps track of t
tination. This
en a particula
epair query (R
ode is forward
node receivin
d route to the
REPR); other
h its own hop
opagates the
the packet is
destination t
backup route.
F THE PROTOC
ave been a res
ommunity. Th
able driven pr
r adaptations
domain, reacti
rks. Furtherm
to the most
tworks: situati
high that any
stination will
y that routes a
of a route fail
nding to failur
ement directio
ry of failed ro
irections:
ed of rebuildi
class are l
air [10] , SL
ich are predic
] (based on
lity).
ds upon the
uality of ser
iginated routin
om the tradit
implies relia
tion as the
r of inno
have been app
ers surveyed i
d on probab
quest (MREQ
in AODV. The
k and once i
ply (MRRP) i
the hop coun
hop count i
ar link on the
REPQ) packe
ded in case o
ng the REPQ
destination. I
rwise, the hop
p count. If it
REPQ packe
dropped. Thi
to receive the
COLS
search topic o
ere are severa
rotocols can be
s of existing
ive routing i
more, reactive
t challenging
ions where the
y route found
inevitably be
are created on
ling is so high
res is a majo
ons have been
outes. This can
ing the routes
low overhead
LR [8] (using
ted to be more
prediction o
baseline by
rvice Network
ng encouraged
tional comfor
ance on graph
mathematica
ovative and
plied to source
in this section
bilistic routing
Q)
e
it
s
nt
s
e
et
of
Q
If
p
ts
et
s
e
of
al
e
g
s
e
g
e
d
e
n-
h
or
n
n
s:
d
g
e
of
y
k
d
rt
h
al
d
e
n,
g

TABLE I
COMPARISON OF MULTICAST PROTOCOLS
Protocol Multiple Routes Route Metric Route Rebuilding
Communication
Over head
DSR Yes Shortest path or next
available
New route
And notify
High
Yu et al No Shortest Path or backup
route
Local Backup Medium
LABQ Yes Link availability
Quality , and Energy
used
Same as DSR or
local repair
High
GRP Yes Shortest Path Backup route High

SLR Yes Shortest Path or next
available path
New route and
notify source
High
Beraldi
et al.
Yes Hint value Local broadcast High

V. CONCLUSIONS
In this paper, we have reviewed and compared several
representative protocols. While different classes of protocol
operate under different scenarios, they usually share the
common goal to reduce control packet overhead, maximize
throughput, and minimize the end-to-end delay. The main
differentiating factor between the protocols is the ways of
finding and/or maintaining the routes between source
destination pairs. The development of the ad hoc routing
protocols over the last 15 years is an example of one of the
most systematic explorations of a design space in the history
of computer science. Although, clearly, newer protocols have
built upon the earlier ones, we cannot identify a single best
protocol. Almost all the protocols we discussed in this paper
have their own sweet spot deployment scenarios and
performance metric combinations where they outperform their
competitors. From the point of view of the practitioner, this
creates a serious problem. To deploy an ad hoc network with
an optimal performance, it requires a very careful analysis of
the scenario and its requirements, and the appropriate choice
of the routing protocol from the dozens applicable in the
context

REFERENCES
[1] http://www.ietf.org/html.charters/manet-charter.html

[2] J. Broch, D. Maltz, D. Johnson, Y. Hu, J. Jetcheva. A performance
comparison of multi hop wireless ad hoc network routing protocols,
MOBICOM 98
. G. Pei, M. Gerla, X. Hong. LANMAR: landmark routing for large scale
wireless ad hoc networks with group mobility, Proc. Of IEEE/ACM
MobiHoc 2000, Boston, MA, August 2000.
[1] Tony Larsson, Nicklas Hedman. Routing protocols in wireless ad hoc
networks A simulation study, Master Thesis, Stockholm ericsson
switched lab. 1998.
[5] D.B. Johnson, D.A. Maltz and J. Broch, DSR: the dynamic source
routing protocol for multi-hop wireless ad hoc networks, C.E. Perkins,
Editor, Ad Hoc Networking, Addison-Wesley (2001), pp. 139172
(Chapter 5).
[6] M. Yu, A. Malvankar, W. Su and S. Foo, A link availability-based
QoS-aware routing protocol for mobile ad hoc sensor networks.
Computer Communications, 30 18 (2007), pp. 38233831.
[7] C. Ahn, Gathering-based routing protocol in mobile ad hoc networks.
Computer Communications, 30 1 (2006), pp. 202206. Article
[8] C. Sengul and R. Kravets, Bypass routing: an on-demand local
recovery protocol for ad hoc networks. Ad Hoc Networks, 4 3 (2006),
pp. 380397.
[9] R. Beraldi, L. Querzoni and R. Baldoni, A hint-based probabilistic
protocol for unicast communications in MANETs. Ad Hoc
Networks, 4 5 (2006), pp. 547566.
[10] C. Yu, T.-K. Wu and R. Cheng, A low overhead dynamic route
repairing mechanism for mobile ad hoc networks. Computer
Communications, 30 5 (2007), pp. 11521163.

ISBN: 978-81-920575-7-6 :: doi: 10. 73998/ISBN_0768
Node Classification for Energy Efficient and Reliable Hierarchical WSN Protocol

Ning Sun
Dept. of Computer Science
Chungbuk National University
Cheongju, Korea

Sang-ho Lee
Dept. of Software
Chungbuk National University
Cheongju, Korea

AbstractEnergy Efficiency is an important factor to design a
wireless sensor network (WSN). Meanwhile, data reliability is
another important issue for WSN. Both of them can be
obtained by designing an efficient MAC or routing protocol. In
this paper, we proposed an energy efficient and reliable
clustering hierarchical MAC and routing protocol for WSNs.
According to the different functions nodes make in the
network, we creatively classify sensor nodes into five
categories. By adopting different MAC and routing strategy to
different category, our proposed mechanism reduces energy
consumption and provides data reliability in a maximum.
Keywords- WSN, wireless sensor network, energy efficiency,
reliability, cluster, hierarchical, MAC, multi-path routing
protocol
I. INTRODUCTION
A WSN can be organized by hundreds to thousands of
sensor nodes. The sensor nodes collect data through sensing
and monitoring from their environment and finally send to
the base station for furthering processing. WSNs can be
applicable in many domains, such as military activities,
control and automation, environment monitoring, healthcare,
security and surveillance, smart home and building, vehicle
tracking and detection, etc.
As most sensor nodes have little cost and size with
constrained processors, limited memory size, low bit rate
communications and low energy capability. Due to this,
energy saving is always a remarkable issue and gets much
attention for a WSN protocol design. Hundreds of MAC and
routing protocols have been proposed and deployed until
now with the purpose of saving energy.
Most energy efficient routing protocols applied
hierarchical structure such as cluster based structure as in [1].
Hierarchical structure could greatly reduce number of direct
communications between sensor nodes and base station,
eventually obviously decrease energy consumption in the
whole network and prolong network lifetime. Accounting for
the advantages of this method, our paper used clustering
hierarchical structure too.
In the view of MAC protocol, energy waste mainly
comes from the following sources: idle listening, data
collision, overhearing and control packet overhead, etc.
Energy efficient MAC protocols achieve energy savings by
controlling the radio to reduce energy waste caused by these
sources. MAC protocols for WSNs can be categorized into
two main groups: schedule-based and contention-based
protocols. In our proposed mechanism, according to different
phases, we utilized different MAC protocols to reduce the
energy consumption flexibly.
In our paper, data reliability is also considered, which is
defined as the successful end-to-end data-delivery ratio, has
been another important issue in WSNs. Major solutions [6,8]
for data reliability of WSN are based on the multi-path
routing paradigm, which provides each sensor with
alternative paths. In the phase of data transmission from CHs
to BS, we performs multi-path algorithm both in routing and
MAC protocols.
In order to realize the above strategies, we classify sensor
nodes into several categories, according to their different role
in different phase. For each category, it executes alternative
routing and MAC protocol and also some specific functions,
such as data aggregation, energy alarm, radio control,
schedule generation, etc.
The remainder of this paper is organized as follows:
Section 2 presents the related works. Section 3 describes the
classification of sensor nodes and explains the detailed
function and algorithms of each category. Section 4
introduces the process of our proposed routing and MAC
protocol based on the node classification in section 3. Finally,
section 5 draws conclusions and shows the future works.
II. RELATED WORKS
Hierarchical (clustering) routing techniques can aid in
reducing energy consumption. The essential operation in
clustering is to divide the network into several virtual
clusters, firstly select a set of cluster heads, and then forming
clusters with the remaining nodes. Cluster heads are
responsible for coordination among the nodes within their
clusters and aggregation of received data, and
communication with each other and the base station. The
sensed data are aggregated, and then sent to the base station
by cluster heads. As shown in [1], the clustering can
obviously reduce the communication overhead for both
single-hop and multi-hop wireless sensor networks. LEACH
[1] is a classical clustering hierarchical protocol, which
incorporates randomized rotation of the high-energy cluster
head position among the sensors to avoid draining the energy
of any one sensor in the network. In this way, the energy
load of being a cluster head is evenly distributed among the
nodes. Being a cluster header is much more energy intensive
than being a non-cluster header node. LEACH incorporates
r
p
c
w
a
d
c
p
w
c
l
p
n
b
l
h
w
p
H
a
a
i
g

c
A
o
d
p
l
e
a
m
a
t
n
L
c
f
randomized r
position. One
consider the
whereas we d
algorithm. Se
dynamic netw
clusters, numb
priori, but in
with the dep
change every
large size of W
power to com
not realistic f
between CH a
The sub
literatures [1,2
head distribut
whole netwo
prolonged. Th
However, all
algorithm of
aggregated da
important, th
guaranteed.

F
Sources of
collisions, ov
Among them,
over one-half
during its life
protocols is to
listening, ove
efficient MAC
and contenti
multiple acce
achieved by d
then allocatin
node. Nodes t
LEACH proto
cluster. The t
figure 3(a). M
rotation of th
e shortcoming
energy factor
designed an mo
econdly, LEA
work. It is b
ber of nodes)
reality these
pletion of nod
round dynami
WSN. It is assu
mmunicate with
for a large sca
and BS is mult
bsequent LE
2,3,4] mostly
tion, so that
ork is reduce
he structure o
these hierarch
cluster format
ata transmissio
he reliability
Figure 1. Cluste
f energy waste
verhearing, ex
, energy waste
of the total en
etime. The ma
o reduce energ
erhearing and
C protocols i
ion-based pr
ss) is a typic
dividing the ra
ng unique tim
ake turns wak
ocol, it uses TD
time line of L
Most clustering
he high ener
g of LEACH
r when electin
ore efficient c
ACH doesnt
because param
are programm
parameters co
des. These pa
ically. LEACH
umed that eve
h others and B
ale WSN. So
ti-hop in our pr
EACH-optimiz
improve the c
the energy c
ed and the s
of clustering i
hical protocol
tion, give littl
on. Since the
y of transm
ering structure for
e in WSNs inc
xcessive mess
ed by idle list
nergy consume
ain objective o
gy waste cause
d excessive
include sched
rotocol. TDM
cal solution fo
adio frequency
me slots to ea
ke up and trans
DMA in data
LEACH proto
g WSN proto
rgy cluster h
is that it d
ng cluster he
cluster head el
work well i
meters (numb
med into the no
ontinuously c
arameters we
H is not scalab
ery node has en
BS directly, bu
the communi
rotocol.
zed protocol
capability of c
consumption o
system lifetim
is shown in F
s only consid
le considerati
aggregated d
mission shoul
r WSNs
clude idle liste
sage overhead
tening accoun
ed by a sensor
of most MAC
ed by collision
overhead. E
dule-based pro
MA (time-div
or the former
y into time slo
ach communi
smit/receive da
transmission w
ocol is as show
cols use TDM
header
oesnt
eaders,
ection
in the
ber of
odes a
change
used
ble for
nough
ut it is
cation
ls in
cluster
of the
me is
Fig. 1.
der the
on on
data is
d be
ening,
d, etc.
nts for
r node
C-layer
ns, idle
Energy
otocol
vision
. It is
ts and
cating
ata. In
within
wn in
MA as
the
clus
con
and
sca
no
pro
pop
is p
duty
avo
ene
In
bas
bot
pha
disj
met
floo
tag
bran
tree
is a
mes
algo
amo
con
data
A.
BS
eac
2 an
the
the
9 a
requ
nod
bran
11-
BS,
BN
exh
path
the
just
is
aba
nod

MAC protoc
ster.
Although T
nsumption obv
d the constrain
alability and fle
coordination a
ovide more sca
pular contentio
proposed to re
ty cycle opera
oidance, overh
ergy saving m
the same tim
sed MAC prot
th TDMA and
ase.
For multi-pat
joint paths h
thod. In H-S
oding protocol
their neighbo
anch that the n
e branch form
achieved witho
ssages. Most m
orithm in the
ounts of m
nsumption. We
a transmission
III. C
Branch Node
The definition
. According to
ch BN represen
nd 3 are BNs
neighbor with
discovery of
at first sets up
quest message
de 11, the resu
anch node 1. F
-9).
According to
, the node ha
N acts critical
hausted the wh
hs are correspo
BN doesnt j
t acts as a rout
below a limi
andons the role
de.
col to manage
TDMA prot
viously, it lim
nts of time a
exibility. Cont
among the no
alability and fl
on-based proto
educe the ener
ation. It also p
hearing avoida
may result high
me, collisions
tocols. In our
d content-base
th routing prot
has been cons
SPREAD [6]
l to find a set
ors as a child, s
node is located
m an alternative
out the need t
multi-path pro
phase of dat
message exch
e use hierarch
n.
CLASSIFICATIO
e (BN)
n of the BN is
o the branch
nts one branch
. If a node rec
h different BN
an alternative
p a primary p
from node 11
ults show that
Finally node 9
the simulation
as more burde
l role in the
hole branch is
ondingly faile
join the cluste
ter in the netw
itation value,
e of branch no
the access of
ocol can
mits the numbe
allocation so t
tention-based p
des accessing
lexibility. CSM
ocols. For WSN
rgy efficiency
provided schem
ance, synchron
h latency and
cant be avoi
proposed mec
ed MAC proto
tocol in WSN
sidered as th
, it uses th
of node-disjo
sibling or cou
d. Sensors foun
e disjoint path
to introduce a
otocols usually
ta transmissio
hanges and
hical structure
ON OF SENSOR
s the one-hop
aware floodin
h. As shown in
ceives a reque
N value, it can
path. As show
ath (2-4-9). T
1. After comp
node 11 come
finds an altern
n result in [6],
ens on data tr
network, be
separated and
ed. In order to c
er formation an
work. Furtherm
it should a
ode and transfo
f nodes within
reduce ener
er of joint nod
that its short
protocols requ
the channel a
MA series are t
Ns, S-MAC [1
y by using a l
mes for collisi
nization. But t
low throughp
ided in conte
chanism, we u
ocols in differ
Ns, building no
he most reliab
he branch-aw
oint paths. Nod
sin, based on t
nd on a differ
h. This algorith
any extra routi
y use flat routi
on, which cau
much ener
e in the phase
R NODES
neighbors of t
ng algorithm [
n figure 2, nod
est message fro
n be regarded
w in Fig. 2, no
Then it hears t
pare the BN w
es from differ
native path (1
the nearer to t
ransmission. T
ecause once i
d the downstre
conserve energ
nd data sense
more, if its ener
announce that
orms to a norm
n a
rgy
des
of
uire
and
the
11]
ow
ion
the
put.
ent-
use
ent
ode
ble
are
des
the
ent
hm
ing
ing
ses
rgy
of
the
[6],
de 1,
om
d as
ode
the
with
ent
-8-
the
The
its
am
gy,
. It
rgy
t it
mal
a

w
n
s
B
b
n
g
t
q
a
A
n
p

n
n
k
a
r
e
r
n
Fi
Node i wi
according to th
E
ETX is the
whereas ERC
number of me
satisfied, node
B. Cluster He
In our ap
based on the r
node. CH is
generation, d
transmission.
quicker than
algorithm is ne
In our pa
According to
neighbors of
probability:

(t P
i
Ei is the r
neighbors of
neighborhood
k is the optim
as about 5% o
range (0,1).
Compared
energy and th
randomized ro
After clust
node sets up a
igure 2. Branch
ill decide whe
he following p
(
B re
E E = > o
e energy con
is in data rec
essages. is t
e i mark itself
ead Node (CH
pproach, cluste
residual energy
in charge of
data process,
The energy
normal node
ecessary.
aper, CH is e
o the residua
each node, C
- = min ) k t
residual energ
node i, whi
table. N is the
mal number of
of N. is the
d with LEACH
he density of e
otation.
ters are forme
a TDMA sche
aware flooding a
ether it should
probability:
( * (
TX
E E N +
nsumption in
ception. N is
the coefficient
as the BN.
H)
er heads are
y and the num
f data receive
, data aggr
consumption
es. So an ev
elected in a d
al energy an
CH is elected
+ 1 (
E
E
to
i
o o
gy of node i.
ich can be r
e total number
clusters, whic
coefficient, w
H, we conside
each node, in
ed as in LEAC
edule and tran
algorithm
d be a branch
))
RC
E
data transmi
proportional t
t. If the condit
elected distri
mber of neighb
e, TDMA sch
regation and
n of CH is
venly CH el
distributed ma
nd the numb
d as the follo
(
1 , )
N
N
neig
o

Nneig is the su
retrieved from
r of all sensor n
ch can be calcu
whose value is
er both the re
nstead of the s
CH, the cluster
nsmits this sch

node,
(1)
ission,
to the
tion is
ibuted
bors of
hedule
data
much
ection
anner.
ber of
owing
(2)
um of
m the
nodes.
ulated
in the
esidual
simple
r head
hedule
to t
coll
C.
sub
auto
hea
and
hea
app
neig
tran
Thi
sys
the
run
wil
pro
sch
It r
sch
tim
tim
and
som
agg
this
very
diff
figu

F
D.
only
thes
effi
nod
obv
the nodes in
lisions among
Substitute Clu
Substitute Cl
bstitute of CH
omatically tak
ad is elected, n
d number of n
ad. The CH
points a node
ghbors. This
nsmitted corre
is could impro
tem.
During the pr
cluster head
nning out of en
l be lost. Its
ocess, all node
hedule except S
receives data a
hedule expires,
me slot for CH
meout, if SN re
d waits for th
me problems w
gregation and
s way, the TDM
ry reliable a
ference of our
ure 3.

igure 3. a) Time
Inter-medium
In the proces
y nodes in the
se nodes Inte
iciently reduce
des not joining
vious energy s
Setup
Pi(t) ADV JOIN
Setup
( )
the cluster. T
data message
uster Head No
luster Head N
H. When CH
kes over the w
nodes send its
neighbors in
compares th
as a substitu
strategy gu
ectly even if t
ove the reliabi
rocess of data
is in any ur
nergy, all the
s why we ele
s within the cl
SN. SN keeps
and proceeds d
, different wit
H to transmit
eceives RCV_
he next phase;
with CH, so SN
transmit the a
MA schedule
and flexible
r MAC protoc
e line of LEACH
m Node (IN)
ss of data tran
e delivery path
er-media Nod
e the idle liste
g the data tra
aving.

Frame
Round
TDMA
/SN
SN
TDMA
This ensures th
es.
ode (SN)
Node (SN) is
H cant work
works of CH.
information o
JOIN messag
hese paramet
ute who has m
uarantees the
the cluster he
ility and fault
transmission
rgent situation
sensed data w
ect a substitu
luster comply
awaken durin
data as the CH
th LEACH, th
a RCV_SUC
_SUC, it emp
; Otherwise it
N takes over t
aggregated dat
neednt be alt
fault toleran
col with LEA
b) Time line of
nsmission from
h need to be a
des (INs). By
ening and over
ansmission, so
Steadystate
Steadystate
hat there are
s defined as t
k normally, S
After the clus
f residual ener
ge to the clus
ters and fina
more energy a
data could
ead is exhaust
tolerance of t
within cluster
n, e.g., sudden
within the clus
ute node. In t
with the TDM
ng all the proce
H does. After t
here is a speci
C to SN. In t
ties the memo
t means there
the work of d
ta to the BS.
tered at all. It
nt scheme. T
ACH is shown
f proposed protoco
m CH/SN to B
awaken. We c
y this way,
rhearing of oth
o that to achie
Transmit
RCVSUC Transm
no
the
SN
ster
rgy
ster
ally
and
be
ted.
the
r, if
nly
ster
this
MA
ess.
the
ific
this
ory
e is
data
By
s a
The
n in
o
BS,
call
we
her
eve

mi
p
k
n
M
E
i
d
u
W
c
m
r
m
d
t

Figure 4. M
Before tran
phase beforeh
keep awaken
nodes go to s
MAC protoco
E. Other Nod
The remain
in charge of w
data transmiss
IV. THE
This proto
uses a hybrid
WSN is displ
categories of s

Figure
f.Phase on
This phas
message. Hop
rebroadcasts it
message is ma
decides wheth
total energy a
NEXTN
N
CH
IN1
INk
BN
Node
11
8
14
MAC protocol for r
nsmitting data
hand. After this
in the follow
leep until the
l for IN and ot
des
ned nodes are
works such as
sion (within clu
E PROPOSED R
ocol mainly con
MAC protoco
layed in figur
sensor nodes.
e 5. Node Classi
ne: Broadcasti
e starts from
p by hop, node
t once. The se
arked as the pa
her it is a Bran
and total numb
NEXTN
NEX
Routedisco
1
4
9
route discovery an
a, CH perform
s phase, INs m
wing transmiss
data transmis
ther nodes is s
e normal nodes
s data sensing
uster).
ROUTING AND M
nsists of three
ol for different
re 5, where w
ificationin propos
ing
m the BS broa
e receives broa
ending node o
arent. After thi
nch Node, store
mber of nodes
XTN
SUC
SUC
SU
Awake
overy
BS
2
6
10
13
16
nd data transmiss
ms a route disc
mark themselve
sion, whereas
ssion is over.
shown as figur
s. Normal nod
g, cluster form
MAC PROTOC
e phase. Our sc
t phases. The
we can see dif
sed mechnism
adcasting a re
adcast messag
f its firstly rec
is phase, every
es the paramet
for the next p
UC
SUC
Data
Sleep
Awake
Awake
Awake
Data
Transmission
- BN
3
5
7
12
- CH
- SN
- IN
sion
covery
es and
other
The
re 4.
des are
mation,
COL
cheme
entire
fferent
equest
ge and
ceived
y node
ters of
phase,
and
mu
adv
com
The
con
A.
abo
dec
bro
MA
it b
bac
pro
info
tran
sam
and
B.
pro
the
tran
wit
from
figu
tran
deli
betw
sam
our
algo
bot
n
d records the
ltiple paths in
vantage of
mmunication. E
e table of neigh
TABLE I
In this phase,
ntrol the collisi
Phase Two: C
After previou
out the total e
cides whether t
Once the nod
oadcasts an adv
AC protocol. A
belongs, it tra
ck to the cho
otocol. In JO
ormation of re
The cluster h
nsmits this sc
me time, accor
d neighbor num
Phase Three:
This phase m
opagation with
cluster hea
nsmission alon
In the data t
th the TDMA
m CH to BS
ure 4. Only CH
nsmission.
In the paper
ivery scheme
ween hierarch
me time, we co
There are som
r proposed pro
Firstly, we
orithm with m
th energy effic
neighborhood
nformation to t
the broadcas
Each node reb
hbor informat
I. TABLE OF
, the network
ion.
Cluster Forma
us phase, every
energy Eto and
to be a cluster
de has elected
vertisement m
After each nod
ansmits a join
osen cluster
OIN-REQ me
esidual energy
head node set
chedule to the
rding to the p
mber, CH selec
: Data Transm
mainly consist
hin a cluster, th
ad to the B
ng multi-paths
transmission w
A MAC proto
S, nodes use t
Hs, INs, BNs
V. CON
r, we propose
for WSNs, w
hical routing
onsider the MA
me original and
otocol:
integrated
multi-path rou
iency and relia
d information,
the BS. This m
st nature of
broadcasts onc
tion is as show
F NEIGHBOR INFOR
uses CSMA M
ation
y node has got
d sum of node
head in a distr
d itself to be
message (ADV
de has decided
n-request mess
head using a
essage, it als
and neighbor
ts up a TDM
e nodes in the
parameters of
cts a node as a
mission
ts of two step
hen the data p
BS, which is
.
within clusters
ocol. During t
the MAC pro
keep awake a
NCLUSIONS
ed a cluster-
which integrate
and multipath
AC protocol fo
d progressive
the hierarch
uting protocol
ability.
which provid
mechanism tak
f the wirel
ce and only on
wn in Table 1.
RMATION
MAC protocol
t the informati
es N. Each no
ributed manne
cluster head,
V) using a CSM
d to which clus
sage (Join-RE
a CSMA MA
so includes t
number.
MA schedule a
e cluster. In t
f residual ener
a SN.
ps: first the d
propagation fro
s a multi-ho
s, nodes comp
the transmissi
otocol shown
and join the d
based multip
es the advantag
h routing. In t
or this scheme
characteristics
hical clusteri
l. This provid
des
kes
ess
nce.

l to
ion
ode
er.
, it
MA
ster
EQ)
AC
the
and
the
rgy
data
om
ops
ply
ion
in
data
ath
ges
the
.
s in
ing
des
Secondly, we classify nodes into five types according to
their role in WSN. This strategy guarantees more fairness in
node energy consumption and prolongs the network lifetime.
Thirdly, both of the process of cluster formation and
multiple path discovery are in distributed manner, it
guarantees the scalability of the network.
Fourthly, mechanism of substitute Cluster Head Node
(SN) accounts for the situation of CH damaging. Moreover,
multipath discovery scheme can solve the problem when a
path is failure. These guarantee the robustness of the system.
Finally, we use two parameters of total energy and node
sum number (optimal cluster number) to improve the
performance of both cluster formation and routing discovery.
In our scheme, these two parameters are updated after every
round and calculated by the BS. Its an efficient way to
control the network quality.
Until now, this work is still in theoretic stage, so our
future works include compare this proposed mechanism with
several existing works, in aspects of energy consumption,
data reliability, network lifetime, latency, etc.

VI.REFERENCES

[1] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan, An
Application-Specific Protocol Architecture for Wireless Microsensor
Networks, IEEE Trans. Wireless Comm., vol. 1, no. 4, pp. 660- 670,
Oct. 2002.
[2] O. Younis and S. Fahmy, HEED: A Hybrid, Energy-Efficient,
Distributed Clustering Approach for Ad Hoc Sensor Networks,
IEEE Transction on Mobile Computing, vol.3, no.4, pp. 366-379,
Octobor 2004.
[3] S. Bandyopadhyay and E. Coyle, An Energy-Efficient Hierarchical
Clustering Algorithm for Wireless Sensor Networks, Proc. IEEE
INFOCOM, Apr. 2003.
[4] M Mehrani, J. Shanbehzadeh, A. Sarrafzadeh, S. J. Mirabedini, and C.
Manford, FEED: Fault Tolerant, Energy Efficient, Distributed
Clustering for WSN, Proc. ICACT2010, Feb, 2010.
[5] D. Ganesan, R. Govindan, S. Shenker, and D. Estrin. "Highly-
resilient, energy-efficient multipath routing in wireless sensor
networks, SIGMOBILE Mob. Comput. Commun. Rev., 5(4):1125,
2001.
[6] W. Lou and Y. Kwon. H-SPREAD: a Hybrid Multipath Scheme for
Secure and Reliable Data Collection inWireless Sensor Networks,
IEEE Transactions on Vehicular Technology, 5(4):13201330, 2006.
[7] C. Intanagonwiwat, R. Govindan, and D. Estrin, Directed Diffusion:
A Scalable and Robust Communication Paradigm for Sensor
Networks, Proc. ACM/IEEE Intl Conf. Mobile Computing and
Networking (MOBICOM), 2000.
[8] A. Ouadjaout, Y. Challal, N. Lasla, M. Bagaa, SEIF: secure and
efficient intrusion-fault tolerant routing protocol for wireless sensor
networks, in: Proceedings of the Third International Conference on
Availability, Reliability and Security (ARES) 2008, March 47, 2008,
pp. 503508.
[9] Y. Jing, X. Mai, J. F. Xu, B. G. Xu, and H. Lu, A cluster-based
multipath delivery scheme for wireless sensor networks, IC-BNMT
'09, Beijing, pp. 286 291, Oct, 2009.
[10] M. S. Zaman, and G. MamaMurthy, Clustered and leveled disjoint
multipath routing algorithm for wireless sensor networks, AH-ICI
2009, Kathmandu, Nov, 2009.
[11] T.van Dam and K. Langendoen, An adaptive energy-efficient mac
protocol for wireless sensor networks, in ACM Sensys 03, 2003, pp.
171-180.

ISBN: 978-81-920575-7-6 :: doi: 10. 74005/ISBN_0768
A Data Mining Approach for Mobility Prediction Using Wireless Traces
J.Ananthi, R.Menaka
Research scholars, Department of ECE
Anna University of Technology
Chennai, India

Dr.V.Ranganathan
Prof., Department of VLSI
KCG College of Technology
Chennai, India

AbstractWith various wireless communication protocols
emerging for ubiquitous communication, mobility prediction
enables better management and efficient usage of resources to
provide improved services to mobile users. As most users are
mobile in the wireless network, it is necessary to include
mobility as an inherent part of the system. The users mobility
influences the network performance and the underlying
quality of service. Various studies have been conducted under
different scenarios to improve performance of the wireless
network under various mobility models. In this paper it is
proposed to apply data mining techniques to predict the user
based on wireless traces.
Keywords-Mobility Prediction, Instance Based Learner
(IBL), Alternating decision tree, Nave Bayes.
I. INTRODUCTION
With the drastic increase in mobile wireless users, networks
face bandwidth and resource shortages which may cause
network failure [1]. To provide guaranteed and QoS to
users, proactive methods are actively studied. As most users
are mobile, it is necessary to include mobility as an inherent
part of the system [2]. Mobility management handles the
methods of storing and updating the location information of
mobile users. Mobility prediction is also part of mobility
management as mobility of the user plays an important role
in efficient usage of resources. Mobility prediction may be
defined as determining the next access point (AP) that a user
will connect to in a network. The mobility is generally
studied by designing mathematical models that model
human mobility [3].

Timely handoffs should be seamless for uninterrupted active
sessions like file transfers, voice over calls, video/audio
streaming, etc. When a user is mobile, the connection is
constantly being switched from one AP to another. Thus
seamless handoffs are possible only if resources are
available at the next AP. Allocating resources to all
neighboring APs is not feasible as it leads to wastage of
resources and under utilization of the network. Efficient
allocations of resources improve resource utilization and
reduce latency in accessing the resources. Prior knowledge
about the movements of users is helpful as mobility
prediction is based on past movements. The mobility
prediction impacts handovers and, hence, improves the
network QoS. There are several problems and issues
involved in mobility prediction in a real environment. The
mobility pattern varies with different population; the
mobility pattern of a city user is different from that of user
in an educational campus. User mobility traces are used for
mobility prediction modeling.
The movement history of the user can be extracted from the
logs obtained from all the access points. When a mobile
device attaches itself to an AP, the unique number of the
mobile device is entered into the logs with a time stamp.
Using the logs obtained from all the APs, the frequently
used path by a particular wireless device can be traced and
this is called the user mobility pattern (UMP) [4]. Mobility
rules are formed using the UMPs. The users actual
movement is obtained,
1 2
, ,....,
i n
u ap ap ap =
where
i
ap is the AP accessed by the user in the movement
path of n access points in the defined time. The mobility
rules are in the form of
1 2
1 2 3 4
, ,
l l
l l l l

1 2 3
, ,....,
n
l l l l
the left part of the mobility rule is called head of rule and
the other is called tail of the rule. When rules are generated,
a confidence value is calculated for each rule. The rule with
the highest confidence is selected.

In this paper it is proposed to use data mining algorithms for
predicting user mobility using wireless trace available from
the Dartmouth College.

II. LITERATURE REVIEW
Considerable amount of research on mobility prediction is
in literature. Some of the related works are discussed in this
section.
Prasad, et al., [5] proposed a generic framework for mobility
prediction of user movement in wireless networks. The
primary goal of the proposed model was effective allocation
of the resources. The framework consists of a prediction
model based on Hidden Markov Model (HMM) to predict
users next location and a predictive mobility management
based on a control theoretic approach with a feedback
control system. Campus wireless trace dataset available
online was used for testing the model. The dataset provided
mobility history of the user such as user associations with
APs, duration of association and so on. On simulation the
effectiveness of the prediction model and the controller was
proven. The proposed framework can be incorporated into
existing wireless networks.
Akoush, et al., [6] proposed a hybrid Bayesian neural
network model for predicting location. The model used
probability theory with neural networks for movement
prediction of the user and to build a inference model. The
dataset contained 100 Nokia smart phone users was
collected from Reality Mining Project at MIT was used.
Trade off analysis was performed for location update and
paging for the wireless network. To enhance prediction
accuracy cells that the users do not use were grouped.
Results showed that the prediction accuracy of the proposed
model outperforms all other standard neural network
methods.
Kim, et al., [7] proposed a mobility method, constructed
using wireless user traces. The proposed method extracts
users mobility tracks and obtained information about
movement speed, pause times, waypoints between
destinations and destination transition probability. This
information was used to generate synthetic tracks and the
same was validated by comparing with the real tracks.
Experiments were conducted using traces of wireless users
from Dartmouth College. Results showed that the synthetic
tracks produced were similar to the real track.
Su, et al., [8] presented a mobility prediction method to
enhance unicast and multicast routing protocols. In the
proposed method, GPS position was piggybacked on data
packets during communication between two nodes and the
expiration time of the link was estimated. The routes were
reconfigured before the link was broken thus providing
seamless connection before connection breaks. The
simulation was implemented within GloMoSim library.
Simulation results indicated that the routes stayed connected
the longest were chosen by the use of mobility prediction.

III. METHODOLOGY
Dartmouth College provides mobility traces as a community
service to researchers. Mobility trace collected over a period
of three years within the Dartmouth College campus is
available, however we used one month syslog data in this
work. During the data collection period the college housed
over 5500 students and 1200 faculty. Initially 476 access
points were available and over the period of time this
increased to 566. All access points were programmed to
share the same SSID and hence mobile users were able to
seamlessly use the network across the campus.115 subnets
covered the 188 buildings and hence devices were forced to
obtain new IP addresses some times. The log recorded in a
syslog server included the timestamp to each message with
each message containing the AP name, MAC address of the
card and the type of message. The various messages used
are authenticated, associated, reassociated, roamed and
disassociated. Whenever a device chooses to use the
network it is first authenticated and this message is ignored
in this work. After authentication, the device must associate
with one of the access points enabling all traffic between
device and network. Reassociation occurs when another AP
with better signal strength is available. The trace of a single
day from users in the Dartmouth College is used in this
work. It is proposed to consider four attributes with three
attributes giving the previous location of the user and the
fourth attribute considering the time.

The Nave Bayes classifier is a statistical method
of classification using supervised learning methods. It is
based on Bayes theorem. The classifier can solve both
diagnostic and predictive problems involving both
categorical and continuous valued attributes. The Bayes
theorem is as follows:

( )
( )
( )
( )
. P B A P A
P AB
P B
|
| =

where
( )
P AB |
is posterior probability of A,
( ) P A
is prior
probability of A,
( )
P B|A
is posterior probability of B and
( ) P B
is prior probability of B.
For a set of tuples D, with n dimensional attribute vector;
( )
1 2
: , ,...,
n
X x x x where
i
x is the value of attribute
i
A .
There are m classes:
1 2
, ,....,
m
C C C . Bayesian classifier
predicts X belongs to a class if

( ) ( ) i j
P C X P C X | > | for 1 , j m j i s s =
It becomes computationally difficult to evaluate
( ) i
P X C |
if
number of attributes is high. In such a case Nave class
conditional independence is used for computation.

( )
1
n
k i
k
P x C
=
|
[

Nave Bayes though over simplified, it performs well in
many complex situations. When the dimensionality of the
data input is high Nave Bayes are very effective. The
parameters are estimated with very small amount of training
data.
Alternating Decision Trees (ADT) is generalization
of decision trees, voted decision tree and decision stumps
with boosting [9]. The classifiers generated by ADT are
easy to deduce and it provides a measure of confidence
called classification margin. In ADT the root node is a
prediction node and consists of a numeric score. The
prediction nodes are followed by decision nodes which are
similar to decision tree stumps. The decision layer is
f
p
p

I
t
b
t
g
t
t
o
c
t
l
d
b
l
a
i
n
a
d
N
I
c
t
c
o
o
followed by
prediction no
prediction lay
Fig
In an ADT th
the path reach
but on reachin
the children o
grows. The ru
the rules to m
the ADT is hi
on the predict
Instance-ba
compares new
training, inste
learns from ex
directly from
based learning
learning occur
a new query
instances is re
new query
algorithms is
distance meas
Nominal attrib
In this paper
class attribute
the Nave B
classification
obtained and
obtained resul
Algorithm
Nave Bayes
J48
IBL
prediction l
des and decis
yer and decisio
1: A simple a
he instances de
hes a decision
ng a predictor
of the node. Th
ules generate r
make real valu
igher as decis
ion.
ased learning
w problem in
ead of perform
xamples or ins
the training
g is a kind of
rs by simply s
instance is pr
etrieved from
instance. An
the nearest
sures are used
butes are hand
IV.
305 records w
es. A tenfold c
Bayes, J48, IB
accuracy an
compared. T
lts.
Classifi
accur
s 72.45
72.45
68.524
ayer, thus a
sion nodes. a
on layer is show
alternating dec
efine the path
n node it conti
r node the path
he base rules i
real numbers;
ued prediction
ion nodes hav
g or memo
nstances with
ming explicit g
stances and co
instances the
lazy learning
storing the pre
resented, a se
memory and
n example
neighbor alg
to handle the
dled using ove
RESULT
were used with
cross validatio
BL algorithm
nd relative ab
Table 1 and f

ication
racy a
59%
9 %
46 %
alternating be
a simple ADT
wn in figure 1
cision tree
hs in the tree.
inues with the
h continues w
increases as th
the boosting a
. The robustn
ve limited infl
ory-based lea
h instances se
generalization
onstructs hypot
emselves. Inst
. In IBL algor
esented data.
et of similar r
used to classi
of instance-
gorithm. Eucl
numeric attri
rlaps.
h 4 attributes
on was used to
ms. The resu
bsolute error
figure 2 show
Relative
absolute error
56.4997%
46.9009 %
47.1483 %
tween
T with
1 [9].

When
e child
with all
he tree
allows
ess of
luence
arning
een in
n. IBL
theses
tance-
rithms
When
related
ify the
-based
lidean
ibutes.
and 1
o train
ult for
were
ws the
Fig
TABLE 1: CLA
g 2: Classificat
In this paper
accuracy an
J48, IBL alg
traces from D
outperforms
research nee
mobility pre
vector mach
[1] P. S. Pra
of mobil
6
th
IEEE
Conferen
[2] D. Perkin
the per
Internati
vol. 4, pp
[3] M. Go
Underst
Nature, n
[4] Gohan Y
Manolop
predictio
Engineer
[5] Pratap S
Framewo
Utilizatio
[6] Sherif A
Moveme
Neural N
[7] Minkyon
a mobili
IEEE Inf
[8] W. Su,
wireless
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
Na
Ba
ASSIFICATION A
ABSOLUT

tion rate.

V. CON
r it was propo
nd relative ab
gorithms for m
Dartmouth Co
the Nave
ed to be carrie
diction using
ine and Neura

REFER
sad, P. Agrawa
lity in hierarchi
E Consumer C
nce, January 200
ns, H. Hughes,
rformance of
ional Conferen
p. 20482052, 2
onzlez, C. H
tanding individ
no. 453, pp. 479
Yavas, Dimitrios
poulos, A dat
on in mobile en
ring 54 (2005) 1
S Prasad and
ork for Mobi
on in Wireless N
Akoush, Ahme
ent Prediction
Networks, IWCM
ng Kim, David K
ity model from
focom 2006.
S.-J. Lee, M.
networks, in: IE
ave
ayes
J48 I
ACCURACY AN
TE ERROR
NCLUSION
osed to study t
solute error i
mobility predic
ollege. Result
e Bayes alg
ed to improve
Radial basis fu
al network algo
RENCES
al, and K. M. Siv
ical mobile ad
Communications
09, pp. 15.
and C. Owen,
f ad hoc n
nce on Commu
2002.
Hidalgo, and
dual human m
9482, 2008.
s Katsaros, Ozg
a mining appr
nvironments. D
121146.
Prathima Agra
ility Prediction
Networks. IEE
ed Sameh, 200
Using Bayesi
MC'07, August,
Kotz, Songkuk
m real user tra
. Gerla, Mobi
EEE MILCOM,
BL
Clas
Acc
Rela
abs
ND RELATIVE
the classificati
in Nave Bay
ctions using d
obtained for J
gorithm. Futu
effectiveness
function, Supp
orithm.
valingam, Effe
hoc networks,
s and Network
Factors affect
networks, IE
unications (IC
A.-L. Barab
mobility pattern
gur Ulusoy, Yan
roach for locat
Data & Knowled
awal, A Gene
n and Resou
EE 2010.
07. Mobile U
ian Learning
, page 191-6.
Kim., Extract
aces. Proceedin
lity prediction
, October 2000.
ssification
uracy
ative
oluteerror

ion
yes,
data
J48
ure
of
port
ects
in
king
ting
EEE
CC),
bsi,
ns,
nnis
tion
dge
eric
urce
User
for
ting
ngs
in
.
[9] Y. Freund and L. Mason. The alternating decision tree
algorithm. In Proceedings of the 16
th
International
Conference on Machine Learning, pages 124133, 1999.

ISBN: 978-81-920575-7-6 :: doi: 10. 74012/ISBN_0768
Performance Analysis of Routing Protocols in Mobile Adhoc Netoworks
Th.Ibungngomacha Singh
Department of Computer Science & Engineering, Manipur Institute of Technology
Imphal,Manipur,India

AbstractA mobile ad hoc network (MANET) is a
decentralized, peer-to-peer wireless ad hoc network that is
capable of configuring itself. Since it is not a static
infrastructure network traditional routing protocols fail to
cope with this type of network topology that changes with
time. Recently researchers have proposed new MANET
routing protocols and comparing and improving existing
protocols before any standardization process. However
simulations results from different research community are
not consistent enough to claim standard MANET routing
protocols for a specific application. This is mainly because of
the lack of consistency in routing protocol models and
different application environments or scenarios including
networking and user profiles. Furthermore the performance
metrics used in comparative analyses differ among
researchers. Therefore it is premature to give a generalized
conclusion in regard to the performance of the existing
routing protocols. In this paper we present comparative
analyses of MANET routing protocols and many possible
performance metrics are considered in the study. The task is
developed using NS2 simulation software and results obtained
are presented to give a detail analyses of different types
MANET routing protocols.
Keywords- MANET, Routing Protocols AODV, DSR,
DSDV, ZRF.
I. INTRODUCTION
Mobile Ad Hoc Network (MANET) [1] is a collection
of communication devices or nodes that wish to
communicate without any fixed infrastructure and pre-
determined organization of available links. The nodes in
MANET themselves are responsible for dynamically
discovering other nodes to communicate. It is a self-
configuring network of mobile nodes connected by
wireless links the union of which forms an arbitrary
topology. The nodes are free to move randomly and
organize themselves arbitrarily; thus, the networks
wireless topology may change rapidly and unpredictably.
Routing is a core problem in networks for sending data
from one node to another. Wireless Ad Hoc networks are
also called Mobile Ad Hoc multi-hop wireless networks is
a collection of wireless mobile hosts forming a temporary
network without the aid of any established infrastructure or
centralized administration. Mobile Ad Hoc Networks
(MANETs) are characterized by a dynamic, multi-hop,
rapid changing topology. Such networks are aimed to
provide communication capabilities to areas where limited
or no communication infrastructures exist. MANETs can
also be deployed to allow the communication devices to
form a dynamic and temporary network among them. A
mobile Ad Hoc network (MANET) is receiving attention
due to many potential military and civilian applications.
MANETs have several salient characteristics: 1) Dynamic
topologies 2) Bandwidth-constrained 3) Energy
constrained operation 4) limited physical security.
Therefore the routing protocols for wired networks cannot
be directly used for wireless networks. Some examples of
the possible uses of ad hoc networking include students
using laptop computers to participate in an interactive
lecture, business associates sharing information during a
meeting, soldiers relaying information for situational
awareness on the battlefield and emergency disaster relief
personnel coordinating efforts after a hurricane or
earthquake. A MANET uses multi-hop routing instead of a
static network infrastructure to provide network
connectivity. Several routing protocols have been proposed
for mobile Ad Hoc networks. Recent studies on
comparative analyses of different MANET protocols can
be found in [8-11]. However the results presented in these
literatures are not consistent and only few performance
metrics are considered. In this paper we present a number
of ways of classification or categorization of these routing
protocols and performance comparison from classes are
presented.

The following sections in the paper are arranged as
follows: Different types of MANET routing protocols are
presented in Section II. In Section III, methodology used
and the performance metrics are defined. The simulation
results and discussion are given in section IV. Section V
concludes the paper.
II. MANET ROUTING PROTOCOLS
Based on when routing activities initiated, routing
protocols for mobile ad hoc networks may be broadly
classified into three basic categories as indicated in figure
1.
(a) Proactive or table-driven protocols
(b) Reactive or on-demand routing protocols
(c) Hybrid routing protocols

Figure 1. Classification of ad hoc routing protocols.
A. Proactive or table- driven protocols
Traditional distance-vector and link-state routing
protocols [1] are proactive in that they maintain routes to
all nodes, including nodes to which no packets are sent.
For that reason they require a periodic control message,
which leads to scarce resources such as power and link
bandwidth being used more frequently for control traffic as
mobility increases. Proactive routing protocols have their
routing tables updated at all times, thus the delay before
sending a packet is minimal. However, routing tables that
are always updated require periodic control messages that
are flooded through the whole network- an operation that
consumes a lot of time, bandwidth and energy. One
example of a proactive routing protocol is explained in the
next subsection.
1) The Destination Sequence Distance Vector (DSDV)
[3]
It is a proactive routing protocol. DSDV also called
table-driven routing protocol because each node maintains
routing table that contains sequence numbers and hop-by-
hop information. DSDV is based on Bellman-Ford routing
algorithm. Some major improvements have been made in
Bellman-Ford algorithm in order to make it suitable for
wireless environment and cope with count-to-infinity
problem. DSDV uses the sequence number to avoid count-
to-infinity problem and using this sequence number DSDV
distinguishes between stale and fresh routes. Nodes talk
with each others and update their routing tables. If change
in topology occurs, updates are transmitted. There are two
types of updates, time-driven (periodic updates) and table-
driven (updates because of significant change). In case of
any change, routing updates are transmitted to all other
nodes which may cause large overhead. In order to reduce
this overhead routing updates are sent in two ways: a full
dump way, where full routing table is sent to neighbors but
it happens only in case of complete topology change. An
incremental update: where only the entries change in the
route metric are sent [2, 3].
B. Reactive or on-demand routing protocols
Reactive routing protocols, on the other hand, operate
only when there is a need of communication between two
nodes. This approach allows the nodes to focus either on
routes that are being used or on routes that are in process of
being set up. Reactive routing protocols determine routes
between nodes only when they are explicitly needed to
route packets. However, whenever there is a need for
sending a packet, the mobile node must first find the route
if the route is not already known. This route discovery
process may result in considerable delay. An example of
reactive routing protocol is Ad hoc On- Demand Distance
Vector (AODV).

1) Ad Hoc On-Demand Distance Vector
Routing Protocol
It is reactive routing protocol. AODV also called
source-initiated routing algorithm, because AODV only
discovers the path to the destination when source wants to
send data. Established path between source and destination
remains as long as it is needed or becomes inaccessible.
Route discovery mechanism of AODV is based on route
request (RREQ), route reply (RREP), and route error
(RERR) messages. Since AODV is flooding in nature,
when there is need to discover path, source node floods
RREQ message to all neighboring nodes. This RREQ
message contains destination sequence number. This
sequence number helps in ensuring route validity and
prevents routing loops [2]. For example, a route with
greatest sequence number is always chosen by sending
node. After receiving RREQ message each neighboring
node checks the destination ID. When the path is found,
RREP message is sent back to requesting node. The path
followed by RREP message is used to send data packet. On
the other hand, when the path is not found, neighboring
nodes forward the RREQ to their neighbors. In case of link
breaks a RERR message is sent to source node informing
that link is no longer valid now. The route discovery
mechanism of AODV is similar to DSR and routing table
of AODV with destination sequence numbers is similar to
DSDV [3].
C. Hybrid routing protocols
Combining the proactive and reactive approaches result
in a hybrid routing protocol. A hybrid approach minimizes
the disadvantages, but also the advantages of the two
combined approaches. The Zone Routing Protocol (ZRP) is
such a hybrid reactive /proactive routing protocol. Each
mobile node proactively maintains routes within a local
region (referred to as the routing zone). Mobile nodes
residing outside the zone can be reached with reactive
routing. ZRP is an example of such a routing protocol.
1) Zone Routing Protocol
ZRP is designed to address the problems associated
with proactive and reactive routing. Excess bandwidth
consumption because of flooding of updates packets and
long delay in route discovery request are two main
problems of proactive and reactive routing respectively.
ZRP came with the concept of zones. In limited zone, route
maintenance is easier and because of zones, numbers of
routing updates are decreased. Nodes out of the zone can
communicate via reactive routing, for this purpose route
request is not flooded to entire network only the border
node is responsible to perform this task. ZRP combines the
feature of both proactive and reactive routing algorithms
[7]. The architecture of ZRP consists of four elements:
MAC-level functions, Intra-Zone Routing Protocol
(IARP), Inter-Zone Routing Protocol (IERP) and Border
cast Routing Protocol (BRP). The proactive routing is used
within limited specified zones and beyond the zones
reactive routing is used. MAC-level performs neighbor
discovery and maintenance functions. For instance, when a
node comes in range a notification of new neighbor is sent
to IARP similarly when node losses connectivity, lost
connectivity notification is sent to IARP. Within in a
specified zone, IARP protocol routes packets. IARP keeps
information about all nodes in the zone in its routing table.
On the other hand, if node wants to send packet to a node
outside the zone, in that case IERP protocol is used to find
best path. That means IERP is responsible to maintains
correct routes outside the zone. If IERP does not have any
route in its routing table, it sends route query to BRP. The
BRP is responsible to contact with nodes across Ad Hoc
networks and passes route queries. Important thing in
border casting mechanism of BRP is it avoids packets
flood in network. BRP always passes route query request
to border nodes only, since only border nodes transmit and
receive packets [5, 6, 7].

III. METHODOLOGY
The simulation is conducted in two different scenarios.
In the first scenario, the comparison of the two routing
protocols is done by keeping 20 connections fixed in
varying numbers of nodes-10, 20,30,40,50.
In the second scenario, the routing protocols are
evaluated in different number of connections while the
number of 20 nodes is fixed. The number of connections is
set to 5, 10,15,20,25.
The performance of the protocols depends on various
interrelating adhered metrics discussed in the next sub-
section.
A. PERFORMANCE METRICS
To analyze the relative performance of the protocols we
measured the following performance metrics from the trace
files.
1) Throughput:
Throughput is the amount of data transferred to the
destination through the network in a unit time expressed in
kilobits per second (Kbps).
kbps
time simulation
d transfere data of Amount
Throughput

2) Number of Packets sent:
It is the total number of packets sent to the
destinations generated by the sources.
3) Number of Packets received:
It is the total number of packets received by the
destinations sent from the sources.
4) Packet Delivery Ratio:
It is the ratio of the data packets delivered to the
destinations to those generated by the source.
% x100
packets sent Total
packets received Total
ratio delivery Packet

5) Routing Overhead:
It is number of routing packets transmitted per data
packet delivered at the destination.

% x100
packets ing transmitt Total
packets signalling routing Total
overhead Routing

6) Average end-to-end delay of data packets:
There are possible delays caused by buffering during
route discovery latency, queuing at the interface queue,
retransmission delays at the MAC, and propagation and
transfer times. The paper use Average end-to-end delay.
Average end-to-end delay is an average end-to-end delay
of data packets. It also caused by queuing for transmission
at the node and buffering data for detouring. This metric
describes the packet delivery time: the lower the end-to-
end delay the better the application performance.
n
1
n
1
time recv. TCP
time) recv. TCP - sent time (TCP
_End_Delay avg_End_to

The packet delivery ratio and the throughput are most
important for best effort traffic. And the routing overhead
evaluates the efficiency of the routing protocol.
IV. SIMULATION RESULTS
The study has been carried out using NS2 simulation
software under Linux operating system platform. In the
simulation, the performance of three routing protocols
namely, AODV, DSDV and ZRP are analyzed in two
scenarios. In the first scenario, numbers of nodes is varied
keeping 20 connections fixed and in the second scenario,
number of connections is varied while the number of
nodes, 20 is fixed.
A. Scenario 1:
In this first scenario, the comparison of the three
routing protocols is done by keeping 20 connections fixed
and varying numbers of nodes with some parameters with
a specific value as shown in table I.
TABLE I. SCENARIO 1 FOR SIMULATION OF AODV,DSDV AND
DSR
Parameter Value
Number of
nodes
10,20,30,40,50
Simulation
Time
100
Pause Time 2 s
Environment
Size
700 x 400
Packet Size 512 bytes
Maximum
Speed
10 m/s
Queue Length 50
Mobility
Model
Random Waypoint
Mobility

The performance metrics for the three routing protocols
are calculated and all relevant graphs are given in figure 2-
figure 3.
From figure 2(a) it is observed that in the beginning,
the average end-to-end delay of AODV leads ZRP and
then comes DSR. However, with the increase in number of
nodes, DSDV leads ZRP for a while but AODV continues
to lead them. ZRP starts to dominate over them with more
number of nodes. At last, ZRP dominates them and AODV
leads DSDV. Figure 2(b) gives the packet delivery ratio,
ZRP and AODV coincide. In the beginning, the packet
delivery ratio of ZRP and AODV is more than DSDV.
However, at the last, with increasing the number of nodes
DSDV dominates over them. Both ZRP and AODV
continue to coincide. In the case routing overhead AODV
leads over DSDV and ZRP. ZRP comes at the last from the
beginning to the end with increasing number of nodes as
presented in figure 3(a).From figure 3(b) the throughput of
the three protocols can be observed. In the beginning,
AODV leads over DSDV. ZRP start at the lowest among
them. As the number of nodes increases, DSDV starts to
lead AODV and ZRP. At the end, DSDV leads AODV
followed by ZRP.
B. Scenario 2:
In this second scenario, the comparison of the three
routing protocols is done by keeping 20 nodes fixed and
varying numbers of connections with some parameters
with a specific value are as shown in table II.
TABLE II. SCENARIO 2 FOR SIMULATION OF AODV
,DSDV AND ZRP
Parameter Value
Number of
connections
5,10,20,25
Simulation Time 100 s
Pause Time 2 s
Environment Size 700 x 400
Packet Size 512 bytes
Maximum Speed 10 m/s
Queue Length 50
Mobility Model Random Waypoint
Mobility

The performance metrics for the three routing protocols
are calculated and relevant graphs are shown in figure 4-
figure 5 respectively.
From figure 4(a) the average end-to-end delay of
AODV leads ZRP and ZRP leads DSDV with less no. of
nodes. With the increase in number of connections, at the
last, AODV continues to dominate over ZRP and DSDV
stays after ZRP. From figure 4(b) it can be observed that in
the beginning, the packet delivery ratio of AODV leads
DSDV and DSDV leads ZRP. With the increasing number
of connections, packet delivery ratio of AODV drops.
However, at the last with increasing the number of
connections, the packet delivery ratio of AODV rises and
leads ZRP but DSDV dominates them. In the beginning,
DSDV leads AODV and then comes ZRP. As the number
of connections increases, AODV leads DSDV but at the
last, DSDV dominates AODV and ZRP stays at the last as
shown in figure 5(a).In figure 5(b) DSDV leads over
AODV and AODV leads ZRP. However, with the increase
in number of connections, AODV leads DSDV but at the
last, DSDV dominates over AODV and ZRP remains at the
last.
V. CONCLUSION
In this paper we have presented a detail comparative
analyses of different routing protocols one each from three
different categories of MANET routing protocols under
different scenarios. Different performance metrics are
defined and used for the comparative analyses. The results
presented here are valid under the scenarios considered in
this paper. However there are many factors that may be
further investigated to study its impact on MANET routing
protocols. We believe that our work presented in this paper
may suitably be of help to the research community in this
challenging field.


(
[1] Comer D.E
and Archite
[2] Chowdhury
of three wir
streaming m
of inmic 20
2004.
[3] Perkins, C.
Sequenced
SIGCOMM
[4] C. E. Perk
Vector (AO
aodv-O~.txt
[5] Ahmed, M.
attributes,"

(a)
RE
E. Internetworkin
ectures,fourth edi
y, M.U.;Perera, D
reless multi hop a
mpeg4 traffic," Mu
04. 8th internatio
E. and Bhagwat,
Distance-Vector
M Comput. Commu
ins and E. M. R
ODV) Routing.
t, November 1998
; Yousef, S.; , "S
IEEE ISIE 2008 ,

Fig

EFERENCES
ng With TCP/IP:
ition, 2000.
D.; Pham,t.;"A pe
ad-hoc network r
ulti topic confere
onal , vol., no., pp
, P. 1994. Highly
routing (DSDV)
un. Rev. 24, 4 23
Royer. Ad Hoc
IETF lnternet D
8.
Self-configurable
,pp.2119-2124, Ju
(a)
gure 2. (a) Aver

: Principles, Prot
erformance comp
routing protocols
nce, 2004. procee
p. 516- 521, 24-26
y dynamic Destin
for mobile comp
4-244, Oct. 1994
On Demand Dis
Draft, draft-ietf-m
e zone routing pro
uly 2008.

rage end to delay

tocols,
arison
when
edings
6 Dec.
nation-
puters.
.
stance
manet-
otocol
[6]
[7]
[8]
[9]
[10
[11

vs no. of nodes (
(b)
] Pearlman, M
configuration
Selected Area
] Patel, B.; Sriv
protocols in M
2010 National
] A.Boomarani
Throughput an
Open Problem
] Abdul Hadi A
comparision o
Mobile Ad
Research,vol.3
0] S.A Ade,P
AODV,DSDV
Hoc Networ
Knowledgeme
2010.
1] T.K. Godder,
Hoc Routing P

(b) Packet deliver
)
M.R. and Hass,
for the zone r
s in Communicat
vastava, S.; , "Pe
Mobile Ad Hoc N
l Conference on, v
Malany, V.R.
nd Delay Compa
ms Cmpt. Math., v
Abd Rahman,Zuri
of AODV,DSDV
Hoc Networks
31,no.4,pp.566-57
P.A. Tijare,
V,OLSR and DS
rks,Int. J. o
ent Managemen
Performance Co
Protocols,CSEIJ

ry ratio vs no. of n
, Z.J., Determ
routing protocol
ions. v17 i8. 1395
erformance analy
Networks," Com
vol.,pp.1-5, 29-31
Sarma ,R.M.
arision of MANE
vol.2,no.3,septemb
iati Ahmed Zuka
V and I-DSDV R
s,European Jou
76,2009.
Performance
R Routing Proto
of Information
nt,vol.2,no.2 p
omparision and an
,vol1.no.1,april 2
(b)
nodes

mining the opti
. IEEE Journal
5-1414.
ysis of zone rout
mmunications (NC
1 Jan. 2010.
chandrasekaran,
ET Protocols, Int
ber 2009.
arnain, Performa
Routing Protocols
urnal of Scient
Comparision
ocols in Mobile
Technology
pp.545-548,decem
nalysis of Mobile
011.

mal
on
ting
CC),
,
t. J.
ance
s in
tific
of
Ad
and
mber
Ad

(a)

(a
(
(a) Rou

a) Average end to
(a)
(a) Routing o
uting overhead vs

o delay vs no. of c

overhead vs no. of
s no. of nodes(b)

connections(b) Pac

f connection (b)
Throughput vs no
(b)
cket delivery ratio

Throughput vs n
o. of notes
)
o vs no. of connec

no. of connections

ctions

(b)
s
ISBN: 978-81-920575-7-6 :: doi: 10. 74019/ISBN_0768
Literature Survey and Proposed Concept on Performance Issue of Multicasting
Routing Protocol in Wireless Sensor Network

Rahul Verma (Author)
M. Tech. Scholar (CSE)
BIST
Bhopal, India

Akhilesh A. Waoo (Author)
Asst. Prof. Dept. of M. Tech. (CSE)
BIST
Bhopal, India

P.S. Patheja (Author)
Asst. Prof. Dept. of M. Tech. (CSE)
BIST
Bhopal, India

AbstractNetworks are classified in to two main categories i.e.
wired network and wireless network. We have known that the
wireless network is important. To Study on wireless network
we have used Mobile Ad-hoc Networks (MANET) which is the
new research area. MANET is infrastructure less network
where routing is the way of transferring packet from one node
to another node in the network. Multicast routing is a
technique where a packet is send to a group of selected node
which is the member of that group, when multicast routing is
applied to a MANET overall performance of the network is
increased. MAODV and ODMRP routing protocol is the
example of multicasting routing. These protocols have
parameters like Packet delivery ratio, Control overhead,
Forwarding efficiency, end to end delivery and many others. In
This paper we are proposing a concept to solve some
outstanding problems like capacity of the network, power
efficiency of the network and most important thing optimal
designing of the network in MANETs. In this paper our
proposed concept will try to provide solution of the mentioned
problems. Here we will provide a design to network in such a
way where we can increase capacity, power efficiency and
other parameters which are predefine using insert new node
concept in existing network.
I. INTRODUCTION
A mobile ad hoc network (MANET) is a type of wireless
networks. This type depends on the mobile nodes and there
is no infrastructure in such type. There are no routers,
servers, access points or cables. Nodes (mobiles) can move
freely and in arbitrary ways, so it may change its location
from time to time. Each node may be a sender or a receiver,
and any node may work as a router and do all router
functions. This means that it can forward packets to other
nodes. Many applications of MANETs are implemented
and used until today like in meeting conferences; military
operations; search and rescue operations, all of them are
examples of MANET networks.
Multicasting in wireless ad-hoc network is a hot topic in
recent years. By multicasting, we mean the transmission of
packets from a source or a group of sources to a group of
one or more hosts that are identified by a single destination
address. Multicasting greatly reduces the transmission cost
when sending the same packet to multiple recipients.
Several multicast routing protocols have been recently
proposed for ad-hoc networks. These protocols can be
Tree-based or Mesh-based.
Dependent or Independent of the underlying
unicast routing protocol.
On-demand or Table-driven.

It can improve the usage of wireless links by sending
multiple copies of data packets using inherent broadcast
behavior of wireless transmission though reducing
transmission overhead and power consumption is a very
challenging part in multicasting. There are many
applications where one-to-many and many-to-many
transmissions are required.
Tree-based routing protocols have only single path from
source to destinations, so the broken links need to be
repaired. On the other hand, mesh-based routing protocol
maintains mesh of the connected components of the network
and therefore, has multiple paths from sources to multicast
destinations. This reduces repairing overhead due to
presence of alternate paths available in the network. Mesh-
based routing protocols lead to congestion under the
conditions of high traffic load which can result in low
packet delivery ratio. This paper summarizes the simulation
techniques and analysis of some of the multicast protocols
like MAODV [10] and ODMRP [11] in MANET
environment.
This paper is organized as follows: in section II we
present a brief survey on various research papers on
multicast routing protocol, section III include problem
finding in mobile ad-hoc network, section IV includes a
description of our proposed idea, section V including
expected outcome of proposed concept. Finally, section VI
provides concluding remarks and highlights our proposed
work.
II. LITERATURE SURVEY
In [1] paper, they have present a comparative
performance of two on-demand multicast routing protocols,
namely MAODV and ODMRP delivery ratio, control
overhead and forwarding efficiency in different multicast
effects such as increasing number of multicast sources and
receivers in the multicast group. This simulation composed
with 200 mobile nodes for this single active multicast group
in the network.
In Another paper [3] they have present simulation based
experiments to analyze the performance of On Demand
Multicast Routing Protocol by evaluating Packet Delivery
Ratio, End to End delay and average throughput. These
results are compared between ODMRP with AODV and
FSR routing protocols by varying number of nodes and
mobility. The comparison shows that ODMRP for ad-hoc
networks performs better as compared to AODV and FSR.
Here another developed technique which is define in
[12] have discuss, they have present a comparative
performance of three multicast protocols for Mobile Ad hoc
Networks ODMRP, AMRIS and MAODV focusing on the
effects of changes such as the increasing number of
receivers or sources and increasing the number of nodes.
Although some simulation results of MANET protocols
have been published before, these three protocols have not
been compared in isolation. In recent years, a number of
new multicast protocols have been proposed for ad hoc
networks. A systematic performance evaluation of these
protocols is done by performing certain simulations. The
applicability of multicast protocols to diverse situation are
also studied and discussed.
In [13], they have present a comparative performance
evaluation of three general-purpose on-demand multicast
protocols, namely ADMR, MAODV, and ODMRP,
focusing on the effects of changes such as increasing
number of multicast receivers or sources, application
sending pattern, and increasing number of nodes in the
network. They have use mobile networks composed of 100
or 200 nodes, with both a single active multicast group and
multiple active multicast groups in the network, in a wide
range of multicast scenarios. Although some simulation
results for these protocols have been published before, the
three protocols have not been compared, and prior studies
have focused on smaller networks using a small set of
simulation scenarios, many with only a single active
multicast group. They have focus here on the effects of the
protocols relative degree of on-demand behavior and their
performance in different multicast scenarios.
III. PROBLEM FORMULATION
If overhead in wireless network nodes decreases network
capacity, reducing the overhead will increase the capacity of
the network. It is postulated in this paper that the
introduction of an intelligently placed node can increase the
capacity of a network. To decrease the overhead
experienced within the network, nodes are required to have
the minimal number of neighbors possible. This can be
accomplished by lowering the transmission powers of
individual nodes to a level where the number of neighbors is
minimized but the network remains connected. The question
therefore arises, can the insertion of an additional node into
a MANET coupled with classical topology control
mechanisms lead to an increase in network capacity? If so,
where should this new node is added?

IV. PROPOSED WORK
Generally mobile ad-hoc networks (MANETs) are used
for military and civilian applications. There are still a lot of
interesting problems investigating the capacity and the
optimal design of these networks. In This paper we are
proposing a concept to solve some problems which is
finding in MANETs. Our main aim of proposed concept is
to increase the capacity and power efficiency of the
MANET using multicasting protocols like ODMRP,
MAODV. Here proposed concept providing solution for the
problems of increasing the capacity, power efficiency,
increasing packet delivery ratio, control overhead,
increasing forward efficiency and decreasing end-to-end
delay of a MANET. With the help of proposed concept we
will try to decrease the overhead of the network using
multicast routing protocol. Here we will try to insert new
node in the existing network. This new node will insert
when multicast routing will performed. With the help of this
new node we can reduce overhead, due to this we can
achieve the network capacity is maximized by minimizing
overhead. Basic principal of proposed concept is shown in
figure 1. here it is clear that overhead between two nodes is
reducing by adding new node. Due to this number of
messages can be easily transmitted between sources to
destination using alternate path.

Figure 1: Architecture of Proposed Concept

Figure 2 is showing flow chart of proposed concept. We
have start sending packets from source to sink and if all the
path already suffer from overhead problem then our recently
packet will also suffer these problem but proposed concept
is solving this problem using inserting new node in proper
location so we can divert or reduce overhead in network and
transfer packets form source to sink in easy way. We will
try to find proper location on the some observation like most
overhead path, less energy path, less distance and many
more.


Figure 2: Flow Chart of Proposed Concept

Advantages of multipath routing:-
Fault tolerance.
Load balancing.
Bandwidth utilization.
Decrease End-to-End delay.
V. EXPECTED OUTCOME
The multicasting routing protocol ODMRP will be
produce batter results as compare other multicasting routing
protocol. Our simulation will we concentrate on ODMRP
routing protocol. Following points for batter result
ODMRP used MESH topology from each source
ODMRP provides alternative paths and a link
failure need not trigger the re-computation of the
mesh, broken links will time out
ODMRP broadcasts the reply back to the source
ODMRP allows for multiple possible paths from
the multicast source back to the receiver
ODMRP does activate a multicast route
immediately
VI. CONCLUSION
A general conclusion is that, in a mobile scenario, mesh-
based protocols out performed tree-based protocols. The
availability of alternate routes provided robustness to
mobility. ODMRP is very effective and efficient in most of
our proposed parameter for simulation scenarios. However,
we did not cover every possible situation. While the
proposed concept of this paper will provide guidelines to
implementation of network, the final selection of a multicast
protocol should take into account other considerations
which cannot be valuated via simulation alone.
REFERENCES
[1] R. Pandi Selvam and V. Palanisamy A Performance Comparison of
MAODV and ODMRP Routing Protocols in Mobile Ad Hoc
Networks published in International Journal of Research and
Reviews in Ad hoc Networks (IJRRAN) Vol. 1, No. 3, September
2011, ISSN: 2046-5106.
[2] Abhijit Das Soumya Sankar Basu and Atal Chaudhuri A Novel
Security Scheme for Wireless Adhoc Network puvlished in Wireless
Communication, Vehicular Technology, Information Theory and
Aerospace & Electronic Systems Technology (Wireless VITAE),
2011 2nd IEEE International Conference 2011.
[3] Yudhvir Singh Yogesh Chaba Monika Jain and Prabha Rani
Performance Evaluation of On-Demand Multicasting Routing
Protocols in Mobile Adhoc Networks published in 2010
International Conference on Recent Trends in Information,
Telecommunication and Computing.
[4] Yudhvir Singh and Dr. Yogesh Chaba Security and Network
Performance Evaluation of KK' Cryptographic Technique in Mobile
Adhoc Networks published in IEEE International Advance
Computing Conference (IACC 2009) Patiala, India, 6-7 March 2009.
[5] E. M. Royer and C. E. Perkins, ``Multicast Operations of the Ad-Hoc
On-Demand Distance Vector Routing Protocol'', Proceedings of
ACM/IEEE MOBICOM'99, pp. 207-218, Seattle, WA, Aug. 1999.
[6] M. S. Corson and S. G. Batsell, `À Reservation-Based Multicast
(RBM) Routing Protocol for Mobile Networks: Initial Route
Construction Phase'', ACM/Baltzer Wireless Networks, vol. 1, no. 4,
pp. 427-450, Dec. 1995.
[7] L. Ji and M. S. Corson, `À Lightweight Adaptive Multicast
Algorithm'', Proceedings of IEEE GLOBECOM'98, pp. 1036-1042,
Sydney, Austral, Nov. 1998.
[8] E. Bommaiah, et al., `ÀMRoute: Adhoc Multicast Routing Protocol'',
Internet Draft, Aug. 1998, Work in progress.
[9] C. W. Wu, Y. C. Tay, and C. K. Toh, `Àd hoc Multicast Routing
Protocol utilizing Increasing id-numberS (AMRIS) Functional
Specification'', Internet Draft, Nov. 1998, Work in progress.
[10] Elizabeth M. Royer and Charles E. Perkins. Multicast Operation of
the Ad-hoc On-Demand Distance Vector Routing Protocol. In
Proceedings of the Fifth Annual ACM/IEEE International Conference
on Mobile Computing and Networking, Mobicom 99, pages 207
218, August 1999.
[11] S.-J. Lee, W. Su, and M. Gerla, On-Demand Multicast Routing
Protocol (ODMRP) for Ad Hoc Networks, Internet-Draft, draft-ietf-
manet-odmrp-01.txt, Jun. 1999.
[12] Aparna K Performance Comparison of MANET (Mobile Ad hoc
Network) Protocols (ODMRP with AMRIS and MAODV) published
in International Journal of Computer Applications (0975 8887)
Volume 1 No. 10 2010.
[13] Jorjeta G. Jetcheva and David B. Johnson A Performance
Comparison of On-Demand Multicast Routing Protocols for Ad Hoc
Networks published in December 15, 2004 CMU-CS-04-176.
ISBN: 978-81-920575-7-6 :: doi: 10. 74026/ISBN_0768
The Cohesive Countermeasure Schemes for Securing Wireless Sensor Networks
against Vulnerabilities and Attacks: Review

Ronnie D. Caytiles
Hannam University
Daejeon, Korea
Corresponding Author

Tai-hoon Kim
GVSA & University of Tasmania
Australia

Dominik lzak
University of Warsaw & Infobright
Warsaw, Poland

Abstract The term Wireless Sensor Network (WSN),
denotes a network of intelligent sensor nodes that monitors
physical and environmental conditions and can be deployed
anywhere and anytime. WSN provides many advantages, but it
is also coupled with new security threats and alters the
organizations overall information security risk profile.
Although implementation of technological solutions is the usual
respond to wireless security threats and vulnerabilities,
wireless security is primarily a management issue. Effective
management of the threats associated with wireless technology
requires a sound and thorough assessment of risk given the
environment and development of a plan to mitigate identified
threats. In this paper, we study the cohesive countermeasure
schemes for securing WSNs against identified vulnerabilities,
threats and attacks. We present an analysis to help network
managers understand and assess the various threats associated
with the use of wireless technology. A number of available
solutions for countering those threats are discussed.
Keywords- WSN, WSN Threats, WSN Security, sensor
nodes, encryption
I. INTRODUCTION
WSNs consist of multifunction sensor nodes that are
small in size and communicate wirelessly over short
distances. The sensor nodes integrate different properties for
sensing the environment, as well as data processing and
communication among other sensors. WSNs perform an
important role in many applications, such as battlefields
surveillance, patient health monitoring, home automation,
traffic control, environmental observation and building
intrusion surveillance.
The wireless networking has improved productivity
through increased accessibility to information resources and
easier, faster and less expensive network configuration.
WSNs provide convenience, cost efficiency, and ease of
integration with other networks and network components.
However, wireless technology also creates new threats. Since
communication for WSNs are through radio frequencies, the
risk of interception is greater than with wired networks.

If the message is not encrypted, or encrypted with a weak
algorithm, the attacker can read it, thereby compromising
confidentiality. Security objectives include: preserving
confidentiality, ensuring integrity, and maintaining
availability of the information and information systems.
WSNs present a handful of issues for network managers.
Unauthorized access points, unknown stations, captured
nodes, spoofed acknowledgements are just a few of the
problems addressed in WSN troubleshooting. Moreover,
onsite maintenance for remotely deployed sensor nodes is
infeasible, thus a thorough consideration of security solutions
and troubleshooting tools must be available.
In this paper, we identify the cohesive countermeasure
schemes that can address the risks offered by the threats and
attacks related to WSNs. Although these threats cannot be
totally eliminated, a desired level of security can be achieved
by adopting such countermeasure. The objective of this
paper is to assist managers in making decisions by providing
them with a basic understanding of the nature of the various
threats associated with wireless networking and available
countermeasures.
II. WSN SECURITY THREATS
This section covers the different attacks and threats that
relate to WSNs. Most of these attacks are similar to those
that apply to traditional networks. However a node capture
attack is a totally new and distinct phenomenon which does
not apply to traditional networks.
A. Node Capture Attack
One of the distinct attacks in WSNs is a node capture. In
this attack, an attacker gains full control over a sensor node
through a direct physical access. Then the attacker can easily
extract cryptographic primitives and obtain unlimited access
to the information stored on the memory chip of the captured
node through a reverse engineering process with the potential
to cause substantial damage to the entire WSN.
Three main factors that can aid the attackers during a
node capture attack:
If sensor nodes within the WSN share a key or keys
with neighboring nodes that is used to encrypt or
decrypt data.
A node capture attack has a great impact on the
structure or topology of a WSN.
The density of the WSN has a direct influence on
node capture attacks, having a similar affect to the
network structure.
B. Side Channel Attack
A side channel attack refers to any attack that is based on
information gathered from the physical implementation of a
cryptosystem, in contrast to vulnerabilities in the algorithm
itself [9]. For example the attacker monitors the power
consumption or the electromagnetic (EM) emanation from
such cryptographic devices, and then analyzes the collected
data to extract the associated crypto key.
Side channel attacks include:
Simple power analysis (SPA) is a technique that
involves directly interpreting power consumption
measurements collected during cryptographic
operations. No statistical analysis is required in such
an attack [5].
Simple electromagnetic analysis (SEMA) an
adversary is able to extract compromising
information from a single electromagnetic sample
[4].
In differential power analysis (DPA) an adversary
monitors the power consumed by cryptographic
devices, and then statistically analyzes the collected
data to extract a key in contrast to the simple power
analysis [5].
In differential electromagnetic analysis (DEMA),
instead of monitoring the power consumption, an
attacker monitors electromagnetic emanations from
cryptographic devices, and then the same statistical
analysis as that for the differential power analysis is
performed on the collected electromagnetic data to
extract secret parameters [4].
C. Denial of Services (DoS)
A Denial-of-Service attack (DoS) occurs when an
attacker continually bombards a targeted AP (Access Point)
or network with bogus requests, premature successful
connection messages, failure messages, and/or other
commands. DoS is an event that diminishes or eliminates a
networks capacity to perform its expected function through
hardware failures, software bugs, resource exhaustion,
malicious broadcasting of high energy signals,
environmental conditions, or any complicated interaction
between these factors [11].
Communication systems could be jammed completely if
such attacks succeed. Other denial of service attacks are also
possible, e.g., inhibiting communication by violating the
MAC protocol. These cause legitimate users to not be able to
get on the network and may even cause the network to crash.
These attacks rely on the abuse of protocols such as the
Extensible Authentication Protocol (EAP).
D. Software Attacks
In Software-based attacks on WSNs, an attacker may try
to modify the software code in memory or exploit known
vulnerabilities in the software code. A well-known example
of such an attack is a buffer overflow attack where a process
attempts to store data beyond the boundaries of a fixed
length buffer, thus, resulting in the extra data overwriting the
adjacent memory locations.
E. Routing Attacks
As described earlier, every node acts as a router in a
WSN. Routing and data forwarding are an important task for
sensor nodes. Routing protocols have to be energy and
memory efficient, but at the same time they have to be robust
against attacks and node failures. There have been many
power-efficient routing protocols proposed for WSNs.
However, most of them suffer from security vulnerabilities
of one sort or another. In the real world, a secure routing
protocol should guarantee the integrity, authenticity and
availability of messages in the existence of adversaries of
arbitrary power. Every authorized receiver should receive all
messages proposed for it and should be capable of proving
the integrity of every message and also the identity of the
sender.
Some of the routing protocol attacks are the following
[10]:
Black hole attacks or packet drop attack;
Spoofed, altered, or replayed attack;
Wormholes attack;
Selective forwarding attack;
Sinkhole attack;
HELLO flood attack;
Acknowledgement spoofing.
F. Traffic Analysis Attacks
All communication is WSNs is moving toward a base
station in many-to-one or many-to-few patterns. An attacker
is able to gather much information on the topology of the
network as well as the location of the base station and other
strategic nodes by observing traffic volumes and pattern.
There are two types of traffic analysis attacks in WSNs: a
rate monitoring attack and a time correlation attack. In a rate
monitoring attack an attacker monitors the packet sending
rate of nodes near the attacker and moves closer to the nodes
that have a higher packet sending rate. In a time correlation
attack an attacker observes the correlation in sending time
between a node and its neighbor node that is assumed to be
forwarding the same packet and deduces the path by
following the sound for each forwarding operation as the
packet propagates towards the base station [3].
G. Sybil Attacks
The Sybil attack is defined as a malicious device
illegitimately taking on multiple identities. For example, a
malicious node can claim false identities, or impersonate
other legitimate nodes in the network [1][2]. The Sybil attack
can affect a number of different protocols such as the
following:
Distributed storage protocols;
Routing protocols;
Data aggregation (used in query protocols);
Voting (used in many trust schemes);
Fair resource allocation protocols;
Misbehavior detection protocols.
To attack routing protocols a Sybil attack would rely on a
malicious node taking on the identity of multiple nodes, thus
routing multiple paths through a single malicious node.
However the Sybil attack can operate in different ways to
attack the protocols listed above.
H. Attacks on In-Network Processing
Data aggregation or in-network processing is very useful
in terms of reducing the communication overhead. However
there can be different types of attack on in-network
processing:
Compromise a node physically to affect aggregated
results [12];
Attack aggregator nodes using different attacks;
Send false information to affect the aggregation
results.
I. Attacks on Time Synchronization Protocols
Time synchronization protocols provide a mechanism for
synchronizing the local clocks of nodes in a sensor network.
Three of the most prominent protocols are the reference
broadcast synchronization (RBS) protocol, Timing-sync
Protocol for Sensor Networks (TPSN) and Flooding Time
Synchronization Protocol (FTSP).
Since most of the time synchronization protocols do not
consider security, an attacker can physically capture a
fraction of the nodes and injecting them with faulty time
synchronization message updates easily. This event can
make the nodes in the entire network out-of-sync with each
other.
J. Replication Attacks
There are two ways to launch replication attacks in
WSNs. First, an attacker can eavesdrop on communications
and resend old packets again multiple times in order to waste
its neighboring sensor nodes energy. Then, an attacker can
insert additional replicated hostile sensor nodes into the
WSN after obtaining some secret information from captured
sensor nodes or through infiltration.
III. WSN SECURITY THREATS COUNTERMEASURE
SCHEMES
A. Securing the Confidentiality of WSN Data Transmission
Several types of countermeasures exist for reducing the
risk of eavesdropping on wireless transmissions. The first
involves methods for making it more difficult to locate and
intercept the wireless signals. The second involves the use of
encryption to preserve confidentiality even if the wireless
signal is intercepted. The rest are attack specific
countermeasures.
1) Signal Hiding Techniques: In order to intercept
wireless transmissions, attackers first need to identify and
locate wireless networks. There are, however, a number of
steps that organizations can take to make it more difficult to
locate their wireless access points. The easiest and least
costly include the following: Turning off the service set
identifier (SSID) broadcasting by wireless access points,
Assign cryptic names to SSIDs, Reducing signal strength to
the lowest level that still provides requisite coverage or
Locating wireless access points in the interior of the
building, away from windows and exterior walls. More
effective, but also more costly methods for reducing or
hiding signals include: Using directional antennas to
constrain signal emanations within desired areas of coverage
or Using of signal emanation-shielding techniques,
sometimes referred to as TEMPEST, 1 to block emanation
of wireless signals.
2) The Use of Encryption: The best method for
protecting the confidentiality of information transmitted
over wireless networks is to encrypt or scramble all wireless
traffic and communications over the network. This is
especially important for organizations subject to regulations.
Two types of countermeasures can significantly reduce
the risk of such attacks: strong encryption and strong
authentication of both devices and users.
3) Countermeasures against Denial of Service Attacks:
Wireless communications are also vulnerable to denial-of-
service (DoS) attacks. Organizations can take several steps
to reduce the risk of such unintentional DoS attacks. Careful
site surveys can identify locations where signals from other
devices exist; the results of such surveys should be used
when deciding where to locate wireless access points.
Regular periodic audits of wireless networking activity and
performance can identify problem areas; appropriate
remedial actions may include removal of the offending
devices or measures to increase signal strength and coverage
within the problem area. The summary of countermeasures
for different DoS attacks are depicted in Table 1.
TABLE I. WSN LAYERS AND DOS DEFENSES
Sensor network layers and denial-of-services defense
Network
layer
Attacks Defenses
Physical Jamming

Tampering
Spread-spectrum,
priority
message, lower Duty
cycle,
region mapping,
mode change
Tamper-proofing,
hiding
Link Collision
Exhaustion
Unfairness
Error-correction
code
Rate limitation
Small frames
Network and
Routing
Neglect and
greed
Homing
Misdirection

Redundancy,
probing
Encryption
Egress filtering,
authentication,
Black Holes
monitoring
Authorization,
monitoring,
redundancy
Transport Flooding
De-
synchronization
Client puzzles
Authentication

4) Countermeasures against Software Attacks: The
following countermeasure can be considered to secure the
WSN software and be protected from being exploited by
malicious users:
Software authentication and validation, e.g., remote
software-based attestation for sensor networks;
Defining accurate trust boundaries for different
components and users;
Using a restricted environment such as the Java
Virtual Machine;
Dynamic run-time encryption/decryption for
software. This is similar to encryption/decryption of
data except that the code running on the device is
encrypted. This can prevents a malicious user from
exploiting the software;
Hardware attestation. The trusted computing group
platform and next generation secure computing base
provide this type of attestation. A similar model
could be used in sensor networks.
5) Countermeasures against Sybil Attacks: Proposed
solutions to the Sybil attack include the following:
radio resource testing which relies on the assumption
that each physical device has only one radio;
random key pre-distribution which associates the
identity of each node to the keys assigned to it and
validates the keys to establish whether the node is
really who it claims to be;
registration of the node identities at a central base
station; and
Position verification which makes the assumption
that the WSN topology is static.
B. Securing the WSN Access Point
Insecure, poorly configured wireless access points can
compromise confidentiality by allowing unauthorized access
to the network. Organizations can reduce the risk of
unauthorized access to wireless networks by taking these
three steps:
1. Eliminating rogue access points. The best method for
dealing with the threat of rogue access points is to
use 802.1x on the wired network to authenticate all
devices that are plugged into the network. Using
802.1x will prevent any unauthorized devices from
connecting to the network;
2. Properly configuring all authorized access points.
Organizations also need to ensure that all authorized
wireless access points are securely configured. It is
especially important to change all default settings
because they are well known and can be exploited by
attackers; and
3. Using 802.1x to authenticate all devices. Strong
authentication of all devices attempting to connect to
the network can prevent rogue access points and
other unauthorized devices from becoming insecure
backdoors. The 802.1x protocol discussed earlier
provides a means for strongly authenticating devices
prior to assigning them IP addresses.
C. Other Countermeasures
The following are several tips for countering attacks and
ensuring security for WSNs:
Use anti-virus and anti-spyware software, and a
firewall. Computers on a wireless network need the
same protections as any computer connected to the
Internet. Install anti-virus and anti-spyware software,
and keep them up-to-date. If your firewall was
shipped in the off mode, turn it on;
Turn off identifier broadcasting. Most wireless
routers have a mechanism called identifier
broadcasting. It sends out a signal to any device in
the vicinity announcing its presence. You dont need
to broadcast this information if the person using the
network already knows it is there. Hackers can use
identifier broadcasting to home in on vulnerable
wireless networks. Disable the identifier broadcasting
mechanism if your wireless router allows it;
Change the identifier on your router from the default.
The identifier for your router is likely to be a
standard, default ID assigned by the manufacturer to
all hardware of that model. Even if your router is not
broadcasting its identifier to the world, hackers know
the default IDs and can use them to try to access your
network. Change your identifier to something only
you know, and remember to configure the same
unique ID into your wireless router and your
computer so they can communicate. Use a password
thats at least 10 characters long: The longer your
password, the harder it is for hackers to break;
Change your routers pre-set password for
administration. The manufacturer of your wireless
router probably assigned it a standard default
password that allows you to set up and operate the
router. Hackers know these default passwords, so
change it to something only you know. The longer
the password, the tougher it is to crack;
Allow only specific computers to access your
wireless network. Every computer that is able to
communicate with a network is assigned its own
unique Media Access Control (MAC) address.
Wireless routers usually have a mechanism to allow
only devices with particular MAC addresses access
to the network. Some hackers have mimicked MAC
addresses, so dont rely on this step alone;
Turn off your wireless network when you know you
wont use it. Hackers cannot access a wireless router
when it is shut down. If you turn the router off when
youre not using it, you limit the amount of time that
it is susceptible to a hack;
Dont assume that public hot spots are secure.
Many cafs, hotels, airports, and other public
establishments offer wireless networks for their
customers use.
The above mentioned countermeasure tips are applicable
to wireless networks, yet, some of them may apply to WSNs
directly.
IV. CONCLUSION
WSNs popularity is due in part to their small size and
low cost of sensors, their operations and their networking
behavior, enabling them to provide significant advantages for
many applications that would not have been possible in the
past. They can provide a numerous opportunities to increase
productivity and minimize costs.
In this paper we have identified and described the
different vulnerabilities, threats and attacks that could
possibly put WSNs in a vital or critical situation. These
threats are analyzed and categorized to find and classify the
countermeasure schemes applicable to mitigate the risks for
each category. A reasonable and practical level of security is
possibly achieved by adopting a systematic approach to
assessing and managing risk regardless of all the threats and
attacks.
A regular study on the emerging threats should be
considered to ensure the security to be maintained at a
desired level. Many research challenges must be addressed in
order for WSNs to be realistically implemented and for them
to become part of everyday life.

REFERENCES
[1] T. Roosta, S. Shieh, S. Sastry: Taxonomy of Security Attacks in
Sensor Networks, 1st IEEE Int. Conference on System Integration
and Reliability Improvements 2006, Hanoi (2006) pp. 1315.
[2] A. Perrig, J. Newsome, E. Shi, D. Song: The Sybil Attack in Sensor
Networks: Analysis and Defences, 3rd Int. Symposium on
Information Processing in Sensor Networks 2004 (ACM Press, New
York, USA 2004) pp. 259268.
[3] J. Deng, R. Han, S. Mishra: Countermeasures Against Traffic
Analysis Attacks in Wireless Sensor Networks, First IEEE/Cerate Net
Conference on Security and Privacy in Communication Networks
(SecureComm) 2005, Athens (2005) pp. 113124.
[4] K. Gandol, C. Mourtel, F. Olivier, Electromagnetic Analysis:
Concrete Results, Published in C_ .K. Ko_c, D. Naccache, and C.
Paar, Eds., Cryptographic Hardware and Embedded Systems { CHES
2001, vol. 2162 of Lecture Notes in Computer Science, pp. 251{261,
Springer-Verlag, 2001.
[5] Power Analysis, http://en.wikipedia.org/wiki/Power_analysis,
Accessed: December, 2011.
[6] Wireless Sensor Networks, http://en.wikipedia.org/wiki/
Wireless_sensor_network, Accessed: December 2011.
[7] J.N. Al-Karaki, A.E. Kamal: Routing techniques in wireless sensor
networks: A survey, IEEE Wirel. Commun. 11(6), 628 (2004).
[8] F. Ye, H. Luo, J. Cheng, S. Lu, L. Zhang: A Two-Tier Data
Dissemination Model for Large-Scale Wireless Sensor Networks,
ACM/IEEE MOBICOM 2002 (2002) pp. 148159.
[9] T. Roosta, S. Shieh, S. Sastry: Taxonomy of Security Attacks in
Sensor Networks, 1st IEEE Int. Conference on System Integration
and Reliability Improvements 2006, Hanoi (2006) pp. 1315.
[10] Y.-C. Hu, A. Perrig, D.B. Johnson: Adriane: A Secure On-Demand
Routing Protocol for Ad Hoc Networks, Annual ACM Int.
Conference on Mobile Computing and Networking (MobiCom) 2002
(2002).
[11] Denial-of-Service Attack, http://en.wikipedia.org/wiki/Denial-of-
service_attack, Accessed: December, 2011.
[12] K. Kifayat, M. Merabti, Q. Shi, D. Llewellyn-Jones: Group-based
secure communication for large scale wireless sensor networks, J.
Information Assurance Security. 2(2), 139147 (2007).

ISBN: 978-81-920575-7-6 :: doi: 10. 74033/ISBN_0768
A Robust Design of RTU Sensor Nodes Communication for SCADA Systems

Sabah Mohammed
Department of Computer Science,
Lakehead University,
CANADA
Ronnie D. Caytiles
Department of Multimedia
Engineering, Hannam University
Daejeon, Korea
Tai-hoon Kim
GVSA and University of Tasmania
Australia

Abstract SCADA Communication plays a vital role for
Supervisory Control and Data Acquisition (SCADA)
Monitoring Systems. The early SCADA communication took
place over radio, modem, or dedicated serial lines. Today, it is
much more common for SCADA communications to travel
over LAN or WLAN. This process of communication over a
SCADA system involves several different SCADA system
components that include the sensors and control relays,
Remote Terminal Units (RTUs), SCADA master units, and the
overall communication network. Advanced RTUs where most
SCADA communication is gathered within the system, are
deployed for a more complete monitoring. The different values
from inputs and outputs, referred to as SCADA points, are
sent from individual sensors to the RTU which is responsible
for forwarding these SCADA communications to the master
station, or Human-Machine Interface (HMI). This paper
proposes an RFGSM Enabled SCADA RTU Communication
for Sensor Nodes that utilizes the available communication
networks such as cellular phone network and radio frequency
for exchanging data to minimize cost and human effort. The
approach makes use of the GSM network, a custom RF
solution, a low-cost microcontroller, and software embedded to
microcontrollers is developed to run the overall process.
Keywords- RF, GSM, RTU, SCADA systems, Sensor Nodes
I. INTRODUCTION
The traditional Supervisory Control and Data Acquisition
(SCADA) communication used the combination of radio and
direct serial or modem connections. Today, it is much more
common for SCADA communications to travel over LAN or
WLAN. This process of communication over a SCADA
system involves several different SCADA system
components that include the sensors and control relays,
Remote Terminal Units (RTUs), SCADA master units, and
the overall communication network. Advanced RTUs where
most SCADA communication is gathered within the system,
are deployed for a more complete monitoring. The different
values from inputs and outputs, referred to as SCADA
points, are sent from individual sensors to the RTU which is
responsible for forwarding these SCADA communications to
the master station, or Human-Machine Interface (HMI).

SCADA communications occur between Master Station
and the RTUs, and between the RTUs and the remotely
deployed sensor nodes. As SCADA communications
between the Master station and the RTUs nowadays utilizes
the LAN, WLAN, and the Internet, the link between sensor
nodes and RTUs cannot incorporate these technologies.
Sensor nodes contain SCADA communication data that are
sensitive and needs to be securely transmitted to RTUs. The
utilization of LAN for sensor nodes can be expensive to
install at remote sites, and it may not be feasible for an
organization to immediately incorporate all of their remote
sites into the LAN network.
This paper is concerned with the prototype design and
development of a communication scheme that would provide
remote access to the sensor nodes, through the combination
of Radio Frequency and Global System for Mobile
communications network. This proposed communication
scheme introduces the two different wireless communication
technologies to allow data communicate between the sensor
nodes and RTUs of SCADA systems. Understanding the
characteristics, properties, and behavior of GSM, RF and
different hardware components are vital to this concept. The
wireless processes will be controlled via a low-cost
microcontroller. The GSM and RF communication media are
fully utilized by introducing new network protocols.
The rest of this paper are organized as follows: Section II
discuss about RTU in SCADA systems; the proposed
scheme for communication among sensor nodes of RTU is
outlined in Section III; and the concluding remarks in
Section IV.
II. RTU IN SCADA SYSTEMS
A remote terminal unit (RTU) is a microprocessor-
controlled electronic device that interfaces objects in the
physical world to a distributed control system or SCADA
(supervisory control and data acquisition system) by
transmitting telemetry data to the system and/or altering the
state of connected objects based on control messages
received from the system. Another term that may be used for
RTU is remote telemetry unit; the common usage term varies
with the application area generally [9].
RTUs, IEDs, PLCs and DCS are increasingly beginning
to overlap in responsibilities, and many vendors sell RTUs
with PLC-like features and vice versa. The industry has
standardized on the IEC 61131-3 functional block language
for creating programs to run on RTUs and PLCs, although
nearly all vendors also offer proprietary alternatives and
associated development environments.
III. RF AND GSM ENABLED SENSOR NODES
COMMUNICATION
This section outlines the proposed scheme for
establishing a link between the remotely deployed sensor
nodes in geographical area and the RTUs. The
communication link utilizes the numerous capabilities of the
Radio Frequency (RF) and Global System for Mobile
Communications (GSM) to provide for data transmission for
SCADA systems. Fig. 1 illustrates the concept for the main
structure of the communication link.

Figure 1. RF-GSM Enabled Communication Link Structure for Sensor
Nodes and RTUs.
The communication link structure is composing of three
components that include:
Sensor Nodes
Mobile Network Controller (NC)
Central Server (CS) for RTUs
The Sensor Node is capable of performing some
processing, gathering sensory information and
communicating with other connected nodes in the network. It
produces a measurable response to sensory environment,
processes data for a variety of application, stores data
temporarily and transports the data to the Central Server for
RTUs if necessary through the Network Controller.
The Mobile Network Controller (NC) serves as router
device that is installed on high points like power poles. The
principal purpose of NC is data transfer and information
routing between Sensor Nodes and Central Server for RTUs.
The Central Server (CS) for RTUs manages the
collection of data from network devices. It also transfers the
data to the database for storage and retrieval.

Figure 2. A more detailed concept for the communication link between
RTUs and Sensor Nodes.
Furthermore, to illustrate the detail concept, Figure 2
shows how the data traverse from each of the devices.
Concisely, an active sensing device receives signals
proportional to the sensory environment referred to as
SCADA points. This SCADA points are periodically read,
accumulated and stored at a low cost but high-performance
microcontroller. The RF transmitter and receiver provide the
window for the sensor nodes to communicate with NC.
Microcontroller, RF Tx\Rx, and the sensing device form a
node. The node can include more components depending on
the additional functions added to the system. To enhance the
system reliability; a non-volatile EEPROM is used to store
the meter reading and other important data at a node or NC;
and reload the stored data at startup.
At the Mobile Network Controller, the utilitys orders to the
system are received as Short-Messaging-Service (SMS)
messages through a GSM modem. A microcontroller
analyzes the message and sends the proper commands to all
nodes in its area or to a specific node only. Any readings,
alerts or reports sent by the nodes are received in the NC,
reshaped to an SMS message form and sent again to the
central server of the RTU.
A. Microcontroller
A microcontroller is a small computer on a single
integrated circuit consisting of a relatively simple CPU
combined with support functions such as a crystal oscillator,
timers, watchdog timer, and serial and analog I/O etc.
The microcontroller will play as a processor in a Node
and NC, it controls and facilitates the data transmission
between the node and network controller.
There are two kinds of Microcontroller use in the design.
Since NC requires more USART for interfacing with GSM
and RF module, a microcontroller with two or more USART
is the choice. In this case the low-cost but powerful Arduino
Mega Development Board is the option, it satisfied the
requirement for the need of more USART. Arduino Mega is
an open-source microcontroller based in ATmega1280 [10].
The summary of the parameters for Arduino Mega is listed
on Table 1.
TABLE I. SUMMARY OF THE ARDUINO MEGA
Microcontroller ATmega 1280
Operating Voltage 5 V
Input Voltage
(Recommended)
7 12 V
Input Voltage (Limits) 6 20 V
Digital I/O Pins 54 (of which 14 provide
PWM output)
Analog Input Pins 16
DC Current per I/O Pin 40 mA
DC Current for 3.3V Pin 50 mA
Flash Memory 128 KB (of which 4KB
used by bootloader)
SRAM 8 KB
EEPROM 4 KB
Clock Spped 16 MHz
Serial USARTs 4

The Node requires only less number of pins for inputs and
outputs, one USART for interfacing with RF module, and
less memory size microcontroller. The Arduino Diecimila is
the appropriate microcontroller for this device, Arduino
Diecimila is a microcontroller board based on the
ATmega168, it is also an open source microcontroller which
the predecessor of Arduino Mega. Below is the summary of
the Diecimila shown in Table 2.
TABLE II. SUMMARY OF THE ARDUINO DIECIMILA
Microcontroller ATmega 328
Operating Voltage 5 V
Input Voltage
(Recommended)
7 12 V
Input Voltage (Limits) 6 20 V
Digital I/O Pins 14 (of which 6 provide
PWM output)
Analog Input Pins 8
DC Current per I/O Pin 40 mA
DC Current for 3.3V Pin 50 mA
Flash Memory 32 KB (of which 2KB
used by bootloader)
SRAM 2 KB
EEPROM 1 KB
Clock Spped 16 MHz
Serial USARTs 1

B. Global System for Mobile Communications
The basic idea in using GSM communications in NCs is
to receive order messages from the utility in SMS form
(Central Server for RTUs). The Microcontroller executes the
proper routine related to the message content and passes its
commands to the Nodes. Reversely, the system sends its
replies, whether they are readings, reports or alerts, to the
utilitys central server again in SMS form via the GSM
network. The Microcontroller highly benefits from the GSM
Module feature that it alerts for any action within the GSM
network, so the Microcontroller can always monitor what is
delivered to the GSM module from network.
In this proposed communication link scheme, for the
proof of concept an old Sony Ericsson T230 is used as
alternative to GSM module; it is much cheaper and available
to the local market. Figure 3 shows the block diagram of
GSM interfacing with Microcontroller in NC.

Figure 3. Block Diagram for GSM interfacing with the Microcontroller in
a Network Controller.
The operating functions of the GSM module (SE-T230)
are accessed through the use of AT commands; these AT
commands are conforming to GSM 03.04 standards. AT
commands are available via the serial interface of the T230
for function implementation. The AT standard is a line-
oriented command language. Each command is made up of
three elements: the prefix, the body, and the termination
character.
The prefix consists of the letters AT
The body is made up of individual characters; it
consists of a name and (if applicable) associated
values
The default termination character is <CR>(=
0x0D)
o Example: AT+CMGR="12"
This command reads the message with index 12 stored in
the default SIM card place. Commands for the SE-T230 can
be categorized in two categories:
Setup commands: used to set the different parameters
of the SE-T230 (e.g. Baud rate, response type)
Communication Commands: used to establish
communication using the SE-T230 (e.g. Read &
Send SMS)
C. RF Module
An important goal of the design is to make the system
cost effective; instead of putting GSM module on every node
in the system that wish to get readings, the RF modules to
link the NC with the surrounding Nodes. This will reduce
cost and will produce a nice challenge to stand for.
RF modules are specified by their operating frequency,
modulation type, maximum data rate and maximum distance
of operation (in both open fields and in buildings).
According to the communication theory, the antenna
length of the modules should be about 17 cm. No special
antennas where used to this design because the transmitter
and the receiver were too close to each other for the demo
purposes, but if they are too spaced and larger distance the

consideration of using the special antenna is highly
recommended.
For the proof of concept purposes, a low-cost 433 MHz
RF and the transmitter range is about 100 meters between
buildings(actually about 75 meters accuracy). The use of
sophisticated RF solution which has a range of around 2km
is well suggested in actual implementation.
D. Relay
The Electronic brick -5V Relay module uses HLS8L
relay module to control high-voltage electrical devices
(maximum 250V). This relay is controlled by
microcontroller with RF to enable remote connect and
disconnection. For demo purposes this design will use a
relay with maximum 250V and 7A.
E. LCD Module
The SerLCD v2.5 is a simple and cost effective solution
for interfacing to LCD based on the HD44780 controller.
The SerLCD module takes incoming 9600 bps TTL level
signal and displays those characters on the LCD screen.
The consideration of choosing SerLCD is the number of
pins necessary to interface to microcontroller. The typical
LCD used 11 pins to interface with microcontroller, while
SerLCD has only three wires - 5V, GND, and Signal are
needed to interface to the LCD.
The HD44780 dot-matrix liquid crystal display controller
and driver LSI displays alphanumeric. It can be configured to
drive a dot-matrix liquid crystal display under the control of
a 4- or 8-bit microprocessor. Since all the functions such as
display RAM, character generator, and liquid crystal driver,
required for driving a dot-matrix liquid crystal display are
internally provided on one chip, a minimal system can be
interfaced with this controller/driver.
F. Communication Link Software/Program
As mentioned above, an embedded software or program
will run the overall process in the Central Server such
sending command and receiving response from the Network
Controller. To develop this software a specific programming
language will be used considering the compatibility issues
for the RTU, microcontroller, and sensing devices.
The Central Server for RTUs is the main harvesting
component that collects data from all NCs (and Nodes) in the
system and manages process controlling too. The automation
of the system should mainly be implemented in the Central
Server. It should be connected to a database to store various
data and transactions from different nodes in the system.
IV. CONCLUSION
In SCADA systems, data communication plays a vital
role for monitoring and controlling the processes of the
entire systems. There are two main communication areas for
SCADA systems, that is, between SCADA Master Stations
and RTUs; and between the remotely deployed sensor nodes
and the RTUs. Advanced RTUs where most SCADA
communication is gathered within the system, are deployed
for a more complete monitoring. The different values from
inputs and outputs, referred to as SCADA points, are sent
from individual sensors to the RTU which is responsible for
forwarding these SCADA communications to the master
station, or Human-Machine Interface (HMI).
This paper have proposed an RFGSM Enabled SCADA
Communication link between the RTUs and Sensor Nodes
that utilizes the available communication networks such as
cellular phone network and radio frequency for exchanging
data to minimize cost and human effort. The approach makes
use of the GSM network, a custom RF solution, a low-cost
microcontroller, and embedded software for microcontrollers
is developed to run the overall process.

REFERENCES
[1] M. Choi, R. Robles, E. Cho, BJ Park, SS Kim, GC Park, and TH Kim,
A Proposed Architecture for SCADA System with Mobile Sensors,
The Journal of Korean Institute of Information Technology, http://ki-
it.or.kr/paper/paper_08_05/8%EA%B6%8C5%ED%98%B8/A%20
Proposed%20Architecture%20for%20SCADA%20System%20with%
20Mobile%20Sensors.pdf, Accessed: December, 2011
[2] R. Robles, M. Choi, TH Kim, The Function of SCADA in Critical
Infrastructure, The Journal of Korean Institute of Information
Technology, 2009.
[3] ISO/IEC 24730-1 Information technology -- Real-time locating
systems (RTLS) -- Part 1: Application program interface (API).
[4] Introduction to PLC and SCADA, http://tesla-
motors.connybrookus.com/2011/12/15/
introduction_to_plc_and_scada/ Accessed: December, 2011
[5] D. Bailey and E. Wright, "Practical SCADA for Industry", Elsevier,
2003.
[6] ISO/IEC 19762-5 Information technology - Automatic identification
and data capture (AIDC) techniques - Harmonized vocabulary - Part
5: Locating systems.
[7] GAO-04-628T. Critical infrastructure protection; Challenges and
efforts to secure control systems. Testimony Before the
Subcommittee on Technology Information Policy, Intergovernmental
Relations and the Census, House Committee on Government Reform.
March 30, 2004. http://www.gao.gov.new.items/d04628t.pdf,
Accessed: December, 2011.
[8] Wireless Sensor Network, http://en.wikipedia.org/wiki/
Wireless_sensor_network, Accessed: December, 2011.
[9] Remote Terminal Unit, http://en.wikipedia.org/wiki/
Remote_Terminal_Unit, Accessed: December, 2011.
[10] Arduino, http://en.wikipedia.org/wiki/Arduino, Accessed:
December, 2011.

ISBN: 978-81-920575-7-6 :: doi: 10. 74040/ISBN_0768
Secured SCADA System in Smart Grid Deployment

Yvette E. Gelogo Tai-hoon

Multimedia Engineering Department, Hannam University KimCulture Design Engineering Research Center,
133 Ojeong-dong, Daeduk-gu, Daejeon, Korea Hannam University,

Abstract-Supervisory Control and Data Acquisition
(SCADA)is a system that hasbeen a stalwart of electric utility
operations. It becoming increasingly complex as new
technologies arrives and new issues emerge on the road to a
modern electric grid.It is very important to keep tract the
security issues that arise together with its improvement. This
paper discusses the SCADA system vulnerabilities and
formulates the possible security measures to mitigate these
vulnerabilities as SCADA system becomes a big part of Smart
Grid.
Keywords:SCADA, Smart Grid, Cryptography, Security

I. INTRODUCTION
A smart grid is a digitally enabled electrical grid that
gathers, distributes, and acts on information about the
behavior of all participants (suppliers and consumers) in
order to improve the efficiency, importance, reliability,
economics, and sustainability of electricity services.
Power grid is managed through control centers,
electronic monitoring/control equipment at substations, and
an extensive communication network. Supervisory Control
and Data Acquisition (SCADA) systems are used
extensively to control and monitor the national power. Many
tests have been done and much vulnerability was discovered,
this vulnerabilities has not been publically disclosed, but
could be exploited by a knowledgeable attacker [1].
As a part of Smart grid, SCADA system play a vital
role in nations critical infrastructures and taking it to the
next level of enhance operations, SCADA system has been
integrated to IP-based network which make the SCADA
system take risk in terms of security. Though there have
been a lot of researches and studies made to mitigate these
vulnerabilities, this is not enough and the study is still going
on.

This paper proposes a methodology for securing SCADA
system in Smart Grid environment.The two focused is for
the network security to ensure that the packets sent
havereached for its intended receipts. And the other one is
the typical attacks that are being launch through malicious
software. These two ways of attacks uses different
mechanisms to launch an attack to host. It is better to study
both to have a more secure SCADA system.
For packet security, there is a lot of cipher scheme
being used to make sure the integrity of the message.The
attacks that can possibly launch with unsecured network is
the attacker can intercept the packet intended for another
SCADA components that is being modified by the attacker
and replace it malicious programs that can destroy the
intended recipients or host. Though this kind of attack will
cause a big damage, there is no concrete security
mechanism to combat such attack profoundly. Another is
when the attack is being launch through downloaded and
installed software which are malware, Trojan viruses and
other virus programs that the host is not aware of its
existence because also of the high cryptography used that
the IDS or NIISD , firewall cannot detect. This will cause a
lot of damage before it can be detected. This paper discusses
the mechanism to mitigate these attacks.

II. BACKGROUND
The following are the SCADA System Components:
1. Operator: Human operator who monitors the
SCADA system and performs supervisory control
functions for the remote plant operations.
2. Human machine interface (HMI): Presents data to
the operator and provides for control inputs in a
variety of formats, including graphics, schematics,
windows, pull down menus, touch-screens, and so
on.
3. Master terminal unit (MTU): The MTU presents
data to the operator through the HMI, gathers data
from the distant site, and transmits control signals

t
t
a
F
p
r
c
u
to th
betw
low
becau
interr
4. Comm
betw
Com
wirel
telep
5. Remo
in th
signa
from
MTU
betw
relati
close
micro
interf
distri
contr
trans
alteri
contr
Anot
Telem
with
moni
and t
Statio
Fig
SCADA
typical contro
terminal units
actuators, and
For the most
performed by
referred to a
consists of a
usually set to
he remote site
ween the MTU
and the contr
use of possi
ruptions.
munications m
ween the M
mmunication c
less or wired n
hone network
ote terminal u
he master/sla
als to the dev
m these device
U. An RTU
ween the RT
ively high an
ed loop. A Re
oprocessor-co
faces objects
ibuted control
rol and da
mitting telem
ing the state
rol messages
ther term that
metry Unit, t
the applicat
itors the field
transmits all th
on.
gure 1.Typical SC
systems are
ol system co
s (RTU) conne
d relaying info
t part, the br
y the Remot
as the RTU)
programmabl
o specific req
e. The transm
U and the remo
rol method is
ible time del
means: Comm
MTU and re
can be thro
networks, or t
k.
unit (RTU): F
ave architectu
vice under con
es, and transm
may be a PL
TU and con
nd the control
emote Termin
ontrolled elec
s in the phy
l system or SC
ata acquisiti
metry data to
of connected
s received f
may be used
the common
tion area ge
d digital and
he data to the
CADA System arc
primarily c
onsists of one
ected to a var
ormation to a
rains of a SC
te Terminal
. The Remo
le logic conve
quirements, ho
mission rate of
ote site is rela
usually open
lays or data
munication m
emote contro
ough the Int
the switched p
unctions as a
ure. Sends c
ntrol, acquires
mits the data t
LC. The data
ntrolled devic
method is u
nal Unit (RTU
tronic device
ysical world
CADA (superv
ion system)
the system a
d objects base
from the sy
for RTU is Re
usage term
enerally. An
analog param
Central Moni
chitecture
ontrol system
e or more r
riety of sensor
master statio
CADA system
Units (some
te Terminal
erter. The RT
owever, most
f data
atively
n loop
flow
method
ollers.
ternet,
public
slave
ontrol
s data
to the
a rate
ce is
usually
U) is a
e that
to a
visory
) by
and/or
ed on
ystem.
emote
varies
RTU
meters
toring
ms. A
emote
rs and
on [2].
m are
etimes
Units
TU are
RTU
allo
the
the
hum
usu
ofte
cha
con
hum
real
mo
per
ben
and
com
out
SCA
dig

its
tim
clos
gen
inst
to
tech
and
inte
pro
a bi
stan
num
netw
atta
sys
as
atta
netw
exp
that
imp
esp
wor
set
alig
ow human inte
RTU might
speed can b
man interventi
ually automati
en, a SCADA
anges to fun
nsidered closed
man interventi
l-time data on
re efficient
rsonnel safety
nefits are mad
d software in
mmunication
tside network
ADA architec
ital data transm
III.
Supervisory
communicati
me security wa
sed nature o
neral model o
talled in a priv
attack the s
hnology mode
d have started
erfaces and
otocols [6]. Be
igger issue.
The transitio
ndardized and
mber of conne
works and the
acks. Conseq
tems has com
extremely v
acks [7][8].
SCADA sys
work connec
posed SCADA
t accompany
plications res
pecially since m
rlds critical i
of network fa
gnment of netw
ervention, for
control the se
be changed or
ion. In additio
ically logged
A system wi
nction optim
d loop system
ion [2]. SCAD
n production,
control para
y, and reduce
de possible by
SCADA syste
protocols an
ks, including
ctures rely hea
mission.
SCADA VUL
Control and D
ion protocols
as not the pri
of the commu
of trusting the
vate network.
system becau
ernized, they
running more
have implem
ecause of this u
on from prop
d open solution
ections betwee
e Internet has m
quently, the
me into question
vulnerable to
stems are bei
ctivity. This
A networks to
todays TCP
sulting from
many of these
infrastructure.
acts can organ
work defenses
r instance, in a
etting of a con
r overridden
on, any chang
d for and/or d
ill monitor a
mally; SCADA
ms and run with
DA provides m
, operations,
adigms, impro
es costs of o
the use of sta
ems combined
nd increased
the Interne
avily on standa
LNERABILIT
Data Acquisit
designed de
mary concern
unications ne
e data on the
Outsiders wil
use of its cl
have become
modern servi
mented remot
upgrades, secu
prietary techno
ns together wi
en SCADA sy
made them mo
security of
n as they are i
cyberwarfar
ing modernize
development,
o the significa
P/IP networks
m this expos
e systems con
. Only with a
nizations fully
s and security
a factory settin
nveyer belt, a
at any time
ges or errors
displayed. M
and make slig
A systems a
h relatively lit
management w
implementatio
oves plant a
operation. The
andard hardwa
d with improv
connectivity
et. [4]. Mode
ard protocols a
IES
ion Systems a
ecades ago, th
n because of t
etworks and t
em since it w
ll find it diffic
lose nature.
e interconnect
ices such as W
te configurati
urity has becom
ologies to mo
ith the increas
ystems and off
ore vulnerable
SCADA-bas
increasingly se
re/cyberterrori
ed with TCP
, however, h
ant security ris
s. The potent
sure are hig
ntrol much of t
a comprehensi
demonstrate t
policy.
ng,
and
by
are
Most
ght
are
ttle
with
ons
and
ese
are
ved
to
ern
and
and
hat
the
the
was
cult
As
ted
Web
ion
me
ore
sed
fice
e to
sed
een
ism
/IP
has
sks
tial
gh,
the
ive
the

p
a
p
s
h
d
g
m
c
4

i
l
n
w
(
i
w
T
v
t
I
a
t
a
4

c
c
b
w
Q
d
IV
F
Typical
physically sec
and employin
password prot
security meas
hackers can
decode propr
gateways. Enc
methods to re
communicatio
4.1 Securing I
It is very
SCADA is no
in internet br
launch in TCP
need to secure
we propose
(CGA). CGA
identifier is g
way hash func
The binding b
verified by re
the hash with
IPv6 address
and auxiliary
the correspond
a certification
4.2 Key Mana
Authentic
SCADA syste
SCADA devi
command to
challenge the
be required to
who have eno
Quality system
difference in
V. SECURITY
Figure 2. Secure S
SCADA se
curing the ha
ng common c
tection and an
sures are hard
easily identif
rietary protoc
cryption and a
educe some o
ons.
IP Signaling
important to
ow run in IP-b
rought securit
P/IP layer. To
e the commun
a Cryptogra
Ais IPv6 add
generated by
ction from pub
between the pu
e-computing th
the interface i
can be protec
parameters a
ding private k
n authority or a
agement
cation is the p
em proves its i
ice receiving
perform cont
sending devic
o access the da
ough permiss
m administrat
security preve
Y METHODO
SCADA Commun
ecurity mea
ardware and tr
cyber security
nti-virus utiliti
der to enforce
fy confidentia
cols, and byp
authentication
f these cyber
secure IP sign
ased network.
ty vulnerabili
o combat these
nication in Lin
aphically Ge
dresses for w
computing a
blic key and a
ublic key and
he hash value
identifier. Me
cted by attach
and by signing
key. The protec
any security in
process by w
identity to ano
a critical m
trols or respo
ces identity.
ata and report
sion can acce
tion technique
ention. SCAD
OLOGY
nication
sures consis
ransmission m
y defenses su
ies. Communic
e since moder
al phone num
pass firewalls
are highly eff
threats to SC
naling since m
. Exposing SC
ities. Some a
e attacks, ther
nk layer. In thi
nerated Add
which the inte
cryptographic
auxiliary param
d the address c
e and by comp
ssages sent fro
hing the publi
g the message
ction works w
nfrastructure.
which one part
other. A
message, such
ond with data
Authenticatio
ts so that only
ss the inform
es can make a
DA web server
st of
media,
uch as
cation
rn day
mbers,
s and
fective
CADA
modern
CADA
attacks
re is a
is case,
dresses
erface
c one-
meters.
can be
paring
om an
ic key
e with
without
t of a
as a
a, can
n will
y users
mation.
all the
r must
alw
serv
from
the
req
sen
info
enc
mas
4.3

phy
Sec
pos
loca
cam
mo
was
US
soft
atta
from
SCA
kin

seri
SCA
Sm
way
SCA
full
thes
har
lead

[1]
[2] R
[3] 6
[4] T
[5]K
[6]C
[7] h
ways be secure
ver security so
m the custome
master server
quest will be
nd back the
ormation wil
cryption that is
ster and the re

Securing Har
It is very imp
ysically. Thes
curing hardwa
ssible inside j
ation security
mera and oth
nitor the area
sthe maliciou
B drive into
ftware became
acker was plan
m all of his c
ADA system
nd of attack.
It is very
iousness, and
ADA systems
mart Grid envir
ys.This paper
ADA system
ly secured SC
se aspect: Sec
rdware securit
d to more vuln
Study of Security
Security
http://www.inl.g
rent_issues.pdf
Ronald L. Krutz,
6DEPLOY IPv6 T
http://www.6de
nd%20Sensor%
Tai-hoon Kim,
Smart Grid
Applications, E
Korea Smart Grid
CNN Interactive,
1988,http://www
juvenile.hacker/in
http://www.tscm.c
e since the da
oftware can al
er or client wi
r. The client w
completed. T
requested inf
ll also be
s proposed to
emote assets.
rdware
portant to secu
se are the ta
are from any
job that may
y measure mu
her sensor de
a. One of the
us software th
o an exposed
e an instrume
nning a "mas
controlled ma
must be secu

V. CONC
important to
importance o
s, the challen
ronment in so
r examines th
as a part of
CADA system
curing IP Sign
ty. For neglec
nerable SCAD
REFERE
y Attributes of S
gov/scada/publica
"Securing SCAD
Training, Athens,
eploy.eu/workshop
%20Networks.pdf
Securing Commu
Environment In
Engineering & Dev
Institute, http://w
Teen Hacker F
w.cnn.com/TECH/
ndex.html [May 2
com/outsideplant
ata in it are ve
lso be added.
ill start with a
will be authenti
The SCADA m
formation to
encrypted u
be used betw
ure SCADA h
angible part
y disasters or
occur. In thi
ust be deploy
evices must
e recent SCAD
hat was deliv
d USB port.
ent for botne
ssive" denial o
achines. This
ured physical
CLUSION
understand
of these newly
nges associate
ome different a
he security r
smart grid. I
we must con
naling, Key M
cting this sec
DA system.
ENCES
Smart Grid System
ations/d/securing_
DA Systems" Wile
22/06/201,
ps/20100621_ath
unication of SCA
nternational Jou
velopment, Issue
www.smartgrid.or
Faces Federal C
/computing/9803/
2011]
t.html
ery critical. W
Communicati
an http request
icated before t
master will th
the client. T
sing the sam
ween the SCAD
hardware devic
of the syste
r calamities a
is case, physi
ed. Surveillan
be installed
DA attacks [1
vered through
This malicio
et attackand t
of service atta
only shows th
lly to avoid t
the complexi
y interconnect
ed with the n
and more urg
requirements
In order to ha
nsider security
Management a
curity areas w
ms Current Cy
Issu
_the_smart_grid_
ey Publishing, Inc
hens/8%20IPv6%2
ADA Components
urnal of System
2, Volume 5, 201
r.kr/10eng6-1.php
Charges, March
/18/
Web
ion
t to
the
hen
The
me
DA
ces
em.
and
ical
nce
to
10]
h a
ous
the
ack
hat
this
ity,
ted
new
ent
for
ave
y in
and
will
yber
ues,
_cur
c.
20a
s in
ms
11.
p
18,
[8] Washington Post. Dissertation could be security threat.
http://www.washingtonpost.com/ac2/wp-dyn/ A23689-2003Jul7
[9] MetinOzturk, Philip Aubin, "SCADA Security: Challenges and
Solutions" June 2011 / White paper
[10] Smart Grid news, Smarter Grid.. Struggling SCADA?
http://www.smartgridnews.com/artman/publish/Technologies_SCAD
A/Smarter-Grid-Struggling-SCADA-1561.html
ISBN: 978-81-920575-7-6 :: doi: 10. 74047/ISBN_0768
Tele-Operative Ophthalmoscope

A.N.Nithyaa

R.Premkumar
Department of Biomedical Engineering Department of Biomedical Engineering
Rajalakshmi Engineering College Rajalakshmi Engineering College
Thandalam,Chennai,India Thandalam, Chennai,India.

A.R.Anuprasan

Bharathkumar.V
Department of Biomedical Engineering Department of Biomedical Engineering
Rajalakshmi Engineering College Rajalakshmi Engineering College
Thandalam,Chennai,India Thandalam,Chennai,India

Abstract - The slit lamp is an instrument consisting of a
high-intensity of light source that can be focused to shine a
thin sheet of light into the eye. In most developing and
many areas of developed
countries, access to quality eye care are inadequate which
leads to blindness. In todays modern heath technology the
ophthalmologist needs digital photography for clinical
examination and for diagnosing tumours in the eye. The
price of a slit lamp increases with its quality for accurate
identification of eye tumours. Tele-Operative
Ophthalmoscope offers great hope to overcome public
health problem and can be used for teaching medical
professionals with low cost. Research activities can be
carried out using this system. This system provides a real
time video of the eye to the students without any
discomfort and inconvenience to the patients. While
teaching the students the concern staff has to save the
images of the eye to provide a clear idea about that image.
In case of direct visualisation the ophthalmologist has to
do the adjustments for each and every student, since the
position of the visualization will change if there is any
disturbance. To reduce the changes and to provide direct
visualization, a coupler with camera is fitted to the viewing
system and is connected to a pc. The paper envisages
designing and testing a new Tele-Operative
Ophthalmoscope, which gives the direct visualization of
the eye on a monitor or on a screen, so that all the students
can view the image at the same time in a class room. For
the purpose of experimentation Webcam is used.

KEY WORDS
Ophthalmoscope, Telemetry

I. INTRODUCTION

Thin slit light beam enables an optical section for in
vivo inspection of different layers of the eye. Since its
invention in 1911, the slit-lamp has become an
indispensable tool for eye examination. [1]
Ophthalmology is ideal for telemedicine. Most eye
pathology is visible to inspection. While blindness is
devastating, many eye diseases develop slowly. Thus, if
the disease is detected early in its course and is
appropriately treated, we can save many people from
needless suffering from visual disability. In most
developing and many areas of developed countries,
access to quality eye care is inadequate leading to the
preventable blindness. Passive tele-ophthalmology
where patients clinical data (such as fundus
photographs) are collected, transmitted, and later
interpreted by ophthalmologists are well developed.

[2-3] However, due to its dynamic nature and
requirement of special skills mentioned above, slit-lamp
examination requires active tele-ophthalmology
solutions. To address this pressing public health
problem, we have developed a Tele-Operative
Ophthalmoscope, which gives the direct visualization of
the eye on a monitor or on a screen, so that skilled
ophthalmologist can teach the students without
disturbing the patient and can provide lively the eye
images. Producing lively in the classroom is a
wonderful way to spark the students imaginations and
draw out their creativity.

II. TELEOPERATIVE SLITLAMP
MICROSCOPE
For ophthalmologic examination, a slit lamp
microscope is generally used as a fundamental
diagnostic device, as shown in Fig. 1. With the slit lamp
microscope, the eye specialist's basic tasks include
adjustment of the microscope position using the joystick
to find a focal position and to switch slit types using the
knobs on the slit lamp unit to set an appropriate
diagnostic condition. The eye specialist might use a
fronting lens and a blue filter together with the
microscope to conduct additional
examinations. The organization of these special
devices requires great skill and experience for
microscope control in addition to knowledge of clinical
medicine.

Fig 1. Examination of a Slit Lamp

In previous works, a specialist - an expert remote
presenter-controls a slit lamp microscope at the patient
site. The captured high-quality still or moving images of
eyes are sent to a classroom.
Ophthalmologists play a relatively peripheral role in
teaching the medical student. In contrast, our approach
introduces a tele-operative slit lamp microscope can
provide a real time technology to capture high-quality
still or moving images through the network, and
conduct a classes and diagnosis in a timely manner.
This setting might require an assistant at the patient site
with limited knowledge related to the ophthalmic
peripherals, hardware and software. The assistant might
be a general doctor who is not an ophthalmologist, or
perhaps a nurse who would support the
ophthalmologist's slit lamp operation.

Fig 2. Focused and slightly blurred area
Human eye has a spherical surface. Only features
that are located in the plane defined by the perfect
object distance appear perfectly focused. All other
features are blurred. In addition, attaching the camera at
one side of the slit lamp microscopy can cause the
narrow vision image, see Fig.2.

III. GENERAL BLOCK DIAGRAM

Fig 3.1 General Block Diagram
3.1 Description of Block Diagram
3.1.1.1 Slit Lamp
Today the slit lamp is the ophthalmologists most
frequently used and most universally applicable
instrument. Biomicroscopy of the living eye is a routine
ophthalmologic examination. The slit lamp enables the
user to inspect individual eye segments in quick
succession to obtain a general impression of t h e
eye and make a diagnosis.
I n a slit lamp, the mo s t important type of
illumination is the optical section. Al l other
techniques are variations. For survey examination of
the anterior segment the slit is adjusted to full aperture.
This results in a circular, very bright and evenly
illuminated field t h a t is slightly smaller than the
microscope's field of view. By placing a ground glass
into the optical path the entire field of view is
illuminated. I t is well known t h a t the structure of
transparent objects such as the cornea, anterior
chamber, eye lens, and body can only be seen poorly
in transmitted or reflected light, as the relative
amplitude modulation o f light is too weak a n d
the phase modulation is not perceived by the eye.
However, such objects can generally be observed well
in scattered or fluorescent light.

Fig 3.2 The overall system
A slit lamp exam may be done:

As part of a routine eye exam along
with other procedures to evaluate the eye, such
as ophthalmoscopy, vision testing, or tonometry
(to measure pressure in the eye).
To look at structures in the back of the
eye, such as the optic nerve or retina.
To help detect disorders in the
structures in the front of the eye, such as
infection or injury to the cornea, cataracts,
conjunctivitis, or iritis.
To help detect and monitor glaucoma
or macular degeneration.
To check for a foreign body, such as a
metal fragment, on or in the eye.
To detect eye problems that may be
caused by other diseases, such as diabetes or
rheumatoid arthritis. Routine slit lamp exams are
important to detect eye problems at an early
stage and to guide treatment if eye problems
develop.
To monitor complications such as
bleeding after an eye injury.
To monitor complications such as
cataract formation that occurs because of
chemotherapy, radiation treatment, or after a
bone marrow transplant.

3.1.1.2 Coupler
A coupler is a device that connects two ends with
out any loss of energy. The coupler is two piece Teflon
coupling. The camera is fitted inside the coupler at the
correct length and the two pieces are tightened. The
modified slit lamp supports video camera installation
for viewing images of the patients eye on screen. The
image coming from one end of the slit lamp is focused
on to the webcam for high fidelity. The webcam used
picks up the image directly and delivers to the PC. The
coupler is made up of PVC, in order to have good
focusing. This duplication of the natures high fidelity
system is achieved after many trial and errors.

3.1.1.3 Coupler Design

Fig 3.3Top View of First Part

Fig 3.4 Top View of Second Part

Fig 3.5Top View of Third Part

Fig 3.6Side View of First Part

Fig 3.7 Side View Of Second Part

Fig 3.8 Side View Of Third Part

Fig 3.9Total Coupler

Fig 3.10 Standard Photograph from the eye

IV. CONCLUSIONS

For the purpose of experimentation webcam is
used, which gives the direct visualization of the eye on a
monitor or on a screen, so that all the students can view
the eye at the same time in a class room. Since the eye is
very sensitive and importance to human life, the quality of
the image that used for doctors' diagnose plays an
important role in examining the patient's illness.
Furthermore, slit lamp examination can suffer the patient
because of high luminance, so less used time would be
better. Tele-Operative Ophthalmoscope system has
specificity approaching those of the conventional slit-
lamp. The remaining challenges are to miniaturize the
prototype, to develop a low cost mass production of
the
units, and to optimize the utilization of internet
bandwidth.

V. ACKNOWLEDGEMENTS
The authors wish to thank the Chairman and
Chairperson of Rajalakshmi Engineering College,
Chennai for providing all the facilities required for
carrying out this work. They also thank the Principal,
Rajalakshmi Engineering College, Chennai for his
constant encouragement and guidance.

REFERENCES

[1]. J.K. Ledford, The Slit Lamp Primer, Slack Incorporated,
Thorofare, N.J., 2006.
[2]. S. Kumar, M.L. Tay-Kearney, I.J. Constable, K.Yogesan., Internet
based ophthalmology service:impact assessment Br J Ophthalmol
89: 1382-83, 2005.
[3]. J.G. Camara, RE Rodriguez., Real-time
telementoring in Ophthalmology Telemed J 4: 375-377, 1998
[4]. Richard j. Jamara, Od, faao, frans van de velde, md,
and eli peli, msc, od, faao Scanning Eye
Movements in Homonymous Hemianopia
Documented by Scanning Laser Ophthalmoscope
Retinal Perimetry 1040-5488/03/8007-0495/0 VOL.
80, NO. 7, PP. 495504 Optometry Vision Science
Copyright 2003 AmericanAcademy of Optometry
[5]. J.G. Camara, RE Rodriguez., Real-time telementoring in
Ophthalmology Telemed J 4: 375-377,1998.
[6]. Austin Roorda, David Merino, Kaccie Y. Li, and Yuhua Zhang
Miniaturization of Adaptive Optics Scanning Laser
Ophthalmoscope in Frontiers in Optics, OSA Technical Digest
(CD) (Optical Society of America, 2009), paper FWX1.
[7]. Perednia D. A., Allen, A., "Telemedicine technology and
clinicaapplications" JAMA, 53, 273, 1995.
[8]. H. Li, "Telemedicine and ophthalmology," Survey of
Ophthalmology, vol. 44, 1999, p. 61-72.
[9]. M. Moore, "The evolution of telemedicine," Future Generation
Computer Systems, vol. 15, 1999, pp. 245-254.


ISBN: 978-81-920575-7-6 :: doi: 10. 73942/ISBN_0768

Abstract- Routing protocols for mobile ad hoc networks
(MANETs) have been explored extensively in last few
years. Much of this work is targeted at finding a feasible
route from a source to a destination without considering
current network traffic or application requirements.
Routing may let a congestion happen which is detected by
congestion control, but dealing with congestion in reactive
manner results in longer delay, and unnecessary packet loss
and requires significant overhead if a new route is needed.
Routing should not be aware of, but also be adaptive to,
network congestion. Adaptation to the congestion helps to
increase both the effectiveness and efficiency of routing.
These problems are solved by the congestion-aware routing
protocols in certain degree. These protocols which are
adaptive to congestion status of mobile ad-hoc network can
greatly improve the network performance. In this paper,
we present the survey of congestion adaptive routing
protocols for mobile ad-hoc network. Finally, the future
direction of congestion-aware routing protocols is
described.

Keywords: Ad hoc networks, congestion aware routing,
Congestion metric, congestion adaptability

I. INTRODUCTION
Wireless ad-hoc network is usually defined as a set
of wireless mobile nodes dynamically self-organizing a
temporary network without any central administration or
existing network infrastructure. The node in the wireless
ad-hoc network can serve as routers and hosts. So, they
can forward packets for other nodes if they are on route
from source to destination. Routing is important problem
in wireless ad-hoc network. Traditional working
protocols cannot work well in wireless ad-hoc network
because of the characteristics of the wireless ad-hoc
networks. Since, mobile nodes have limited transmission
capacity they mostly intercommunicate by multihop
relay. Multihop routing is challenged by limited wireless
bandwidth, low device power, dynamically changing
network topology, high vulnerability to failure. To
answer these challenges, many routing algorithms in
MANETs were proposed. There are different dimensions
to categorize them: proactive routing Vs reactive routing
or single path routing Vs multipath routing. In proactive
protocols, route between every two nodes are established
in advance even though no transmission is in demand. In
reactive protocols, route is discovered when needed
transmission and released when transmission no longer
takes place. Congestion is one of the most important
restrictions of wireless ad-hoc network. It may
deteriorate the performance of whole network. In the
current design routing is not congestion-adaptive.
Routing may let the

congestion happen which is detected by congestion
control. But dealing with congestion in reactive manner
results in longer delay and an unnecessary packet loss
and requires significant overhead if the new route is
needed. But, now there is another dimension for
categorizing for routing protocols: congestion adaptive
Vs. congestion un-adaptive routing. Our motivation is
that congestion is dominant cause for packet loss, long
delay, and high overhead in MANETs.
These problems become visible in large scale
transmission of traffic intensive data such as multimedia
data where congestion is more probable and negative
impact of packet loss on the service quality is of more
significance. In this paper we studied congestion routing
protocols like CRP(Congestion Adaptive Routing
Protocol)[7],ECARP (Efficient Congestion Adaptive
Routing Protocol )[11],CARP(Congestion Aware
Routing Protocol),CADV(Congestion Aware Distance
Vector)[12],CARA(Congestion Aware Routing plus rate
Adaptation)[12],CARM(Congestion Aware Routing
Protocol for Mobile Ad-hoc Network)[12].
The remaining part of the paper is organized as
follows: In section II we provide the studied congestion
aware routing protocols. In section III comparison
between these algorithms is presented. In section IV we
concluded the paper.
ALGORITHMS
There are many routing algorithms in mobile ad-hoc
networks for routing and congestion free networks.
Some of them are explained below:
A. Congestion Adaptive Routing Protocol (CRP):
Congestion Adaptive Routing is a congestion
adaptive unicast routing protocol for mobile ad-hoc
network. CRP protocol tries to prevent congestion from
occurring in the first place. In CRP, every node
appearing on a route warns its previous node when prone
to be congested. So, CRP uses the additional paths called
as bypass for bypassing the potential congestion area
to the first non-congested node on the primary route. It
reduces packet delay. But, at the same time CRP tries to
minimize bypass to reduce protocol overhead. Hence,
the traffic is split over bypass and primary and
adaptively to network congestion. Hence, 1) power
consumption is efficient.2) Congestion is resolved
Protocols A Study of Congestion Aware
Adaptive Routing in MANET

Mr.S.A.Jain Mr.S.R.Kokate
Research Scholar Research Associates
MAE, Alandi (D) MAE, Alandi (D)
Pune University, India Pune University, India


beforehand and at the same time there is small packet
loss rate.

RP is on-demand and consists of the following
components.
a. Congestion Monitoring: When no. of packets
coming to the node exceeds its carrying capacity, node
becomes congested and its starts losing packets. Various
metrics are used for node to monitor congestion status.
Main parameters are percentage of all packets discarded
for lack of buffer space, the average queue length, the
no. of the packets timed out and retransmitted, average
packet delay. In all these parameters, rising number
indicates growing congestion.
2) Primary Route Discovery: Sender discovers the
route to the receiver by broadcasting the REQ packet
toward receiver. The receiver responds REQ by sending
the REP packet on same path that the REQ previously
followed. This is called primary route and nodes on this
are called primary nodes. To reduce traffic due to the
primary route discovery and better deal with
Congestion in the network, 2 strategies are adopted
1) REQ is dropped if arriving at a node which is having
congestion status as red 2) REQ is dropped I arriving
at node already having a route to destination .
b. Bypass Discovery: A primary node periodically
broadcasts a UDT i.e. update packet. This packet
contains the nodes congestion status and set of tuples
[destination D, next green node G, distance to green
node, n] for each node appearing as a destination in
primary table. For this reason is when node P receives an
update packet from next primary node P
next,
about the
destination D, P will be aware of congestion status of
next. This causes the congestion to know about the next
green node of P which is n hops away from primary
route. But if the next hop is yellow or red, congestion
will be there if data packets continue to be forwarded on
PP
next
. But, CRP tries to keep congestion from
occurring in the first place, P node starts to select bypass
route toward G-the next green node of P known from the
UDT packet. This bypass search is similar to primary
route search, except that 1)the bypass request packets
TTL is set to 2*m and 2)bypass request is dropped if

arriving at node already present on primary route. It
can be also possible that no bypass is found. So, in such
situation packets are delivered to destination by
following primary route.
C. Traffic Splitting and Congestion Adaptability:
When the bypass at a node is found, data packets coming
to this node are not necessarily spread over bypass and

primary route. To avoid the bypass from being
congested no packet is forwarded on bypass unless any
primary node is red i.e. congested. The basic idea behind
traffic splitting is that when primary link consists of less
congested node, traffic on primary link should be
increased, otherwise it should be reduced. Bypass and
primary routes cannot include more than 2 common
nodes, but different bypass paths can share common
node. This increases chance to discover a bypass. But,
because of this bypass node may become congested if it
has to carry large loads of bypass traffic. But, this can be
solved, by splitting probability adjustment for congestion
adaptation. The probability adjustment is as shown in
TABLE I.
b. Multipath Minimization: To reduce the protocol
overhead, CRP tries to minimize using multiple paths. If
the probability p to forward data on a primary link
approaches 1.0, this means the next primary node is far
from congested or the bypass route is highly congested.
In this case, the bypass at the current node is removed.
Similarly, if the next primary node is very congested (p
approaches 0), the primary link is disconnected and the
bypass route becomes primary. To make
the protocol more lightweight, CRP does not allow a
node to have more than one bypass. The protocol
overhead due to using bypass is also reduced partly
because of short bypass lengths. Each bypass connects to
the first non-congested node after the congestion spot,
which should be just a few hops downstream.

c.Failure Recovery: CRP is able to quickly resume
connectivity after a link breakage by using bypass routes
currently available. There are 3 min cases of failure
Primary link failure: When one of link on primary
route fails, the initial node sends a DISC packet towards
sender along route. This DISC goes on recording nodes
and it stops at node having bypass. This node if finds
that its bypass destination is there in DISC, that bypass is
not used and DISC is forwarded upstream towards
sender till it finds a node with bypass and not having
failed node as its destination. If both these cases are not
Congestion Bypass
status=green
Bypass status=yellow Bypass status=red
Next primary node is green P:=p+(1-p)/4 P:=p+(1-p)/3 P:=p+(1-p)/2
Next primary node is yellow P unchanged P unchanged P:=p+(1-p)/4
Next primary node is red P:=p-(1-p)/2 P:=p-(1-p)/4 Find another bypass

TABLE I
SPLITTING PROBABILITY ADJUSTMENT


there DISC is sent to the sender and it will find new
primary route.
d.Bypass link or node fails: In this case bypass node
which finds this failure sends a BPS_DISC packet
through bypass route to primary node and that bypass is
removed.
e.Primary node fails: If node on the primary route
fails, its previous node sends DISC packet along primary
route. If the bypass node detects some failure, it will also
send BPS_DISC packet along bypass until reaching a
primary node. When primary node received both these
packet, it removes bypass and DISC packet is forwarded
along primary route. Then this is handled same as first
case. If BPS_DISC packet doesnt arrive at the primary
node on time that bypass is used as primary route. But, if
it comes late, it is ignored. But, route remains broken but
it will recover soon because another DISC packet will be
sent back.
To evaluate the performance of the CRP following
parameters are used:
a.Packet delivery ratio: Percentage of data packets
received at the destination out of the number of data
packets generated by the CBR traffic sources.
b.End-to-End delay: It is the accumulative delay in
data packet due to buffering of packets, new route
discoveries, queuing delay, MAC-layer re-transmission,
and transmission and propagation delays.
c.Routing Overhead: The ration of the amount in
bytes of control packets transmitted to the amount in
bytes of data received.
d.Normalized power consumption: The ratio of the
amount in bytes of both control and data packets
transmitted to the amount in bytes of data received.

B. An Efficient Congestion Adaptive Routing
Protocol for Mobile Ad Hoc Networks (ECARP):

An efficient congestion adaptive routing protocol is
better than every other routing protocol during heavy
traffic loads. The ECARP, routing protocol ensures high
availability of alternative routes and reduce the rate of
stale routes. ECARP is having mainly AODV as its base.
This can be achieved by increasing the parameters of
routing protocols (especially in AODV) that normally
take more time for link recovery. These parameters are
active_route_time-out, route_reply_wait_time,
reverse_route_life, TTL_start, TTL_increment,
TTL_threshold and delete_period.
ECARP Congestion Control Algorithm
This algorithm provides solution to improve routing
protocols due to constrained environment.
Step 1: Check the occupancy of link layer buffer of
node periodically. Let N
c
be the congestion status
estimated.
Step 2: Compute N
c
=Number of packet buffered in
buffer/Buffer Size
Step 3: Set the status for congestion. It can be
indicated by three statuses Go, careful, and Stop.
[ Go indicates there is no congestion with N
c
],
carefull indicates the status likely to be congested with
N
c
and Stop indicates the status already
congested, N
c
1.]
Step 4: Invoke congestion control routine when link
failed event has occurred in data transfer with using
active route or N
c
1.
Step 5: Assume that neighboring will have alternate
route or non-congested route to the destination.
Step 6: Make Query to non-congested neighbors for
route to destination
Step 7: after obtaining the routes from the neighbors,
select route with minimum hops.
Step 8: Once route is finalized start sending the data
packets through non-congested route.
Step 9: If there is no alternative route to destination
then start splitting the traffic to the less congested route.
Step 10: Traffic splitting effectively reduces the
congestion status at the next main node.

d. Congestion aware routing plus Rate Adaptation
(CARA):

The base use of CARA protocol is DSR. The route
discovery mechanism of DSR is modified. This protocol
mainly aims to find the bypass route for congested zones
or nodes. This can be achieved by combining the average
MAC utilization and the instantaneous transmission
queue length to indicate the congestion level of nodes in
the network. When source wants to transmit data to the
destination node, it broadcasts RREQ packets. When
intermediate code receives RREQ, it checks its
congestion level. If the congestion level is higher then it
discards the RREQ. When RREQ arrives at the
destination node, though destination node is congested or
not it handles the RREQ and replies RREP. So, route
without congested node is established.
CARA uses two metrics to measure congestion
information first is average MAC layer utilization. The
instantaneous MAC layer utilization is considered as 0
only when the medium around the node is available at
the beginning of a transmission and as 1 when the node
is not idle. (e.g. detecting physical carrier or detecting or
back off due to virtual carrier sensing.) As, the
instantaneous MAC layer utilization is either 1 or 0 the
average value with in the period indicates the use of
wireless medium around the node.
Second metric used is instantaneous transmission
queue length. If the node has many packets waiting in
the queue, it causes long packet latency or even dropping
of packets. So we can say that node is congested now.
The above mentioned metric can veraciously reflect
the congestion conditions around the node. This protocol
tries to minimize the congestion in two ways: 1) It
forbids the RREQ packets to propagate in the congested
area. 2) It guides the route around the congested area or
nodes instead of across them. As a result of this no
conditional transmission burden generate in these areas.
e.Congestion Aware Routing protocol for Mobile ad
hoc networks (CARM):

A congestion aware routing protocol for mobile ad
hoc networks uses a metric incorporating data rates,
MAC overhead and buffer delay to control the
congestion. The CARM protocol introduces a new
parameter called WCD (Weighted Channel Delay) to
measure congestion level and adopts a route
ELDC(Effective link Data-rate Category) to avoid the
MDRR(Mismatched data-rate route) problem. The
MDRR problem is shown in following fig.3

Fig. 3 An example of MDDR problem

The data rate of route shown by dashed (A-B-D-G) is
limited by teaming fast link (B-D) with slow link (A-B
and D-G).
As mentioned earlier, the CARM protocol introduces
a new parameter called WCD (weighted channel delay)
to measure congestion and it is given as :

WCD=a Q +
(1+b)T
MACALL
+T
data

where Q is the number of buffered packets for this
link. T
data
=L
data
/ R is the data transmission time, L
data
is
the length of data in bytes or bits and R is the data rate of
the link. T
MACALL
is total time spent at the MAC layer.
The constants a and b are parameters with values
between 0 and 1 which are used to weight T
MACALL
. By
weighting T
MACALL
can avoid misjudgment of congestion
as shown in fig.2

Fig. 2 Two scenarios with the same overall delay but different
MAC and transmission delay due to different data-rates and congestion
levels.
In CARM, source node broadcasts RREQ packets
with ELDC and WCD information when it attempts to
transmit data to the destination. Intermediate nodes
compare source ID, source sequence, and ELDC of the
RREQ packets they receive from neighbors, and drop the
RREQ packets whose source ID and source sequence
number are the same with that of other RREQ packets
received earlier and ELDC is lower than the earlier
RREQ packets. Only the destination node can responds
to the RREQ packets by sending RREP packets back to
the source along the route from which they came. The
route is established when the first RREP arrives at the
source. The subsequent RREP packets are cached for the
spare routes. The utilization of the congestion metric,
WCD, is very special in CARM protocol. Because the
priority of route packets is higher than data packets, the
route packets can be forwarded without queuing. That is,
the congestion level information inherent in queuing
delays is lost. The author proposed a RREQ-delay
scheme. An RREQ is forwarded with a delay of the
WCD that is calculated according to the WCD
information in the RREQ at the intermediate nodes. The
lower the congestion level of link is, the smaller the
delay of RREQ packets are, the earlier the RREQ
packets arrive at the destinations. This scheme ensures
that the RREQ packets of routes with lower congestion
level arrive at the destination first and congested links
are eliminated in the routes. This all causes high
overhead. So, overhead in case of CARM is very high.

E. Congestion-Aware Distance Vector (CADV):

The CADV protocol is based on proactive protocol,
DSDV. In a distance vector routing protocol, every host
maintains a routing table contains a distances from itself
to possible destinations. A mobile host in ad -hoc
network acts like a single server queuing system. Delay
in sending packet is related with congestion. In CADV,
each entry is related with delay expected. This helps to
measure congestion at the next hop. The expected delay
is computed follows:

(1)
Where n is the number of sent packets & L is the
length of MAC layer packet queue. E [D] estimates the
time. A newly arrived packet has to wait before it is send
out.In CADV, routing decision is made based on
distance to the destination as well as the expected delay
at the next hop showed in (1) CADV gives the routes
with low expected delay, higher priority. CADV tries to
avoid congestion and tries to balance traffic by giving
priority to a route having low expected delay.
CADV routing protocol consist of three components:
1) Traffic Monitor: It monitors traffic going out
through the link layer. Currently it keeps track of
average delay for sending one data packet in receipt

period of time. Time period is specified by route
maintainance component.
2) Traffic Control: It determines which packet is the
next to send or drop. It reschedules packets if needed. It
supports a drop tail FIFO queue and provides
functionality to queue packets.
3) Route maintainance: It is the main component. It
performs the work of exchanging information with
neighbors, evaluation and maintaining routes. It manages
the traffic monitor and traffic control component. CADV
better support for QoS. The real time performance of
CADV is good, and end to end delay was short. The
overhead of CADV is unacceptable when the network is
large. Through put also decreases the performance of
CADV is may be well in the small & steady wireless ad-
hoc network.

F. Congestion Aware routing Protocol (CARP)

CARP is an on-demand routing protocol. It uses
information gathered from MAC layer to discover
congestion free routes. CARP uses combined weight
matrix in its standard cost function to check for the
congestion level. The multiple paths are computed
during the route discovery. Calculate node weight matrix
NM which assign a cost to each link in the network and
select maximum throughput paths.
NM = (L
q
* D
rate
)/(OH
mac
* D
avg
)
1) Route request: Consider the route
S-P1-P2-P3-D
To initiate congestion-aware routing discovery, the
source node S sends a RREQ. When the intermediate
node P1 receives the RREQ packet, it first estimates all
the node weight metrics.
The node P1 then calculates its node weight NMP1
RREQP1P2
P2 calculates NMP2 and forward the RREQ packet
RREQP2P3
Finally the RREQ reaches the destination node D
with the sum of node weights
RREQP3D
2) Route Reply: The destination node D
sends the route reply packet RREP along with total
node weight to the immediate upstream node P3
RREQDP3
Now P3 calculates its cost C based on the
information from RREP as
CP3= (NM
p1
+NM
p2
+NM
p3
)-(NM
p1
+NM
p2
)
By proceeding in the same way, all the intermediate
hosts calculate its cost .On receiving the RREP from all
the routes, the source selects the route with minimum
cost value.

II. COMPARISONS

Congestion is a dominant reason for packet drops in
ad hoc networks.CRP sends packets on both bypass
paths and primary routes simultaneously. So, incoming
traffic is distributed on primary and bypass route
depending on current congestion status of network.
Congestion is subsequently better resolved .In ECARP
some parameters of AODV such as TTL_start,
TTL_increment are increased. So, it ensures the high
availability of alternative routes and reduces the rate of
broken rut removal process. CADV is not congestion
adaptive. It offers no remedy when the existing route
becomes heavily congested. So, CADV improves AODV
in delivery ratio only. The real time performance of the
CADV is good and the End-to-End delay is short. The
disadvantage of the CADV is that since, each node
maintains all the routes to the nodes in the network and
changes the route information periodically, the overhead
for maintaining the routing tables is huge. The overhead
of the CADV is unacceptable when the network is large
or the topology changes frequently. The throughput
decreases sharply at the same time. So, CADV may
perform well in the small, steady wireless ad -hoc
network. By studying the algorithms of CARM, CARA
and CADV it is conclude that overhead of the CARM
and CADV are higher than CARA, the delay of CADV
is shorter than the other two.
III. CONCLUSION
It is clear from algorithms available for having
adaptive solution for congestion in the network as due to
vast pay load on networks, which may be due to flooding
of packets or may be due to repeat requests on the basis
of error correction techniques. Congestion metrics still
remains a great challenge for the future work. It is quite
important to obtain an optimal approach that combines
related parameters collected from physical layer, MAC
layer to measure congestion. Finally we can conclude
that congestion is the problem associated with the
network and has to be countered by having compromised
solution rather than elimination.

REFERENCES

[1] D. B. Johnson, D. A. Maltz, Dynamic source routing in ad hoc
wireless networks, Mobile Computing, pp. 153181, 1996.
[2] Josh Broch, David A. Maltz, David B. Johnson, Yih-Chun Hu,
Jorjeta Jetcheva. A Performance Comparison of Multi-Hop Wireless
Ad Hoc Network Routing Protocols, The Fourth Annual ACM/IEEE
International Conference on Mobile Computing and Networking, pp. 1-
13, 1998
[3]Chen, X., H. M. Jones, A .D .S. Jayalath, Congestion-Aware
Routing Protocol for Mobile Ad Hoc Networks, Vehicular
Technology Conference 2007, pp. 21-25, 2007
[4] S. Ramanathan, M. E. Steenstrup, A survey of routing techniques
for mobile communications networks, mobile networks and
applications, Vol. 1, pp. 98104, 1996.
[5] C. E. Perkins, Ad hoc on demand distance vector (AODV)
routing, IEFT Internet Draft, November 1998,
[6] H. Raghavendra and D.A. Tran, Congestion Adaptive Routing in
Ad Hoc Networks (Short Version), Proc. ACM Intl Conf. Mobile
Computing and Networking (MOBICOM), Oct. 2004.
[7] H. Raghavendra and D.A. Tran, Congestion Adaptive Routing in
Ad Hoc Networks IEEE Transactions on Parallel and Distributed
Systems, Vol. 17, No. 11, November 2006.
[8] D.A. Tran and H. Raghavendra, Routing with Congestion
Awareness and Adaptivity in Mobile Ad Hoc Networks, Proc. IEEE
Wireless Comm. and Networking Conf. (WCNC), Mar. 2005.

[9] Xiaoqin Chen, Haley M. Jones and Jayalath, "Congestion-Aware
Routing Protocol for Mobile Ad Hoc Networks" IEEE 66
th
conference
on Vehicle Technology, pp.21-25, October 2005.
[10] S. J. Lee and M. Gerla, Dynamic load-aware routing in ad hoc
networks, Proceedings IEEE International Conference on
Communications, pp. 32063210, June 2001.
[11] T.G.Basavraju, S.K.Sarkar, C.Puttamadappa, M.A.Gautham,
ECARP: An Efficient Congestion Adaptive Routing Protocol for
Mobile Ad-hoc Networks, ITS Telecommunication Proccedings,
pp.715-718, 2006.
[12] Wu Wei, Zhang Zhongzhao, Sha Xuejun, Qin Danyang A Study
of Congestion-Aware Routing Protocols For Wireless Ad-Hoc
Network.

ISBN: 978-81-920575-7-6 :: doi: 10. 73935/ISBN_0768
Exploiting the Bit Error Diversity for Efficient Error Detection in Multihop
Wireless Network
Yanthungbeni Humtsoe
PG Student ME Computer Department
Pune Institute of Computer Technology
Pune, India.

Abstract Errors are inherently present in wireless network,
exploiting the bit error diversity and correcting the error for
packet transmission is necessary for improving the
throughput. In multihop wireless network when transmission
goes over multiple hops, error increases which degrades the
performance of the network. Various methods already exist
for controlling the error to improve the throughput. Scattered
Random network coding exploit the usefulness of random
network coding and classify the block into good (low error
probability) and bad block (high error probability). In this
paper we proposed a new approach for controlling the error.
Proposed work uses the concept of forward error correction
for correcting and controlling the error on the bad bits,
16QAM is used for modulation symbol and shortest
remaining algorithm is used for giving priority to messages.

Keywords- Multihop, Throughput, Error control, bit error,
Error detection.

I. INTRODUCTION
In multi-hop wireless networks, communication
between sender and receiver is carried out through a series
of intermediate nodes whose function is to relay
information from one node to another. In unreliable
wireless channel, errors are naturally present. Multihop
wireless network suffer from low throughput as error
increases over multiple hops. It is important to design an
efficient transmission and error control schemes to
improve the throughput in multihop wireless network.
Network coding is introduced in [1] to improve the
throughput performance. In network system, the
intermediate node simply replicate and forward the
received information to the destined node ,while in
network coding ,the intermediate node combine the
information received from different sources and forward to
the output node. In [2] a network coding approach to
cooperative diversity is proposed, a scenario is considered
in which two partners cooperate to transmit information to
a single destination. Both the partner transmit their locally
generated information and the partner relay information
which lead to an encoding scheme in which the algebraic
superimposition of its local and relay information is
transmitted by each partner. At the destination decoding is
done by repeating the process between the codeword from
the two partners. In a communication network, various
kind of error exists during the packet transmission. Errors
include packet error, random error and so on. Network
error correction coding (NEC) is introduced in [3] to deal
with these errors. Earlier work on network coding mainly
focus on multicast, in [4] Network coding has been applied
in multihop wireless network to take advantage of the
multiple routes from sender to receiver in unicast flows.
The 2N-QAM (Quadrature Amplitude Modulation)
modulated symbol consists of N-bits. For the 2N-QAM
with N larger than 2(eg.8QAM (PSK), 16QAM, 64QAM,
256QAM) the reliabilities of the bits Gray-mapped onto
the modulated symbol is different.16QAM contains 4bits
and 64QAM contains 6bits.

II. RELATED WORK
Random network coding [5] is a technique for
scattering information in a communication network, it is
prone to errors. The attachment of even a single corrupted
packet will affect all the information receives; therefore
error control is of great interest in random network coding.
As many problem occur for error control in random
network coding, in [6] the problem of error control is
considered and the expression of the error control problem
is given in terms of rank metric codes. Efficient Network
coding based reliable multicast is introduced in [7]. In [8]
the author examines the error control problem in
noncoherent random network coding. In network coding,
error control is an important problem since they are highly
impressionable to packet error. In [9] an error control code
in Random Network coding has been introduced. The
Random network coding has a property called rateless in
which the packet are encoded as a linear combination of
the original packet and all the independent packet are
useful. A receiver collect every data (correct block) that it
receive without describing which block is from which
source, once it receive all the blocks, the original data
blocks can be correctly reconstructed. In this sense,
random network coding can be considered as a rateless
code. When random network coding
is applied to multihop wireless networks, the sender
encodes the data and transmits the coded blocks to the
relay, which will forward the data to the end receiver.
In[10]Drizzle is proposed to see whether network coding
will be still used in multichannel wireless networks and it
is a new solution to maximize the throughput in the
presence of errors. In [11] a signal constellation
rearrangement for HARQ is presented where for
retransmission symbol mapping is used and at the receiver
before demodulation signal combination is implemented
and then the combined signal is demodulated. In [12] a
novel algorithm for searching HARQ with constellation
rearrangement is proposed. In wireless communication due
to unreliable wireless channels retransmission is done to
remove the errors. Scattered Random Network coding (S-
RNC) is introduced in [13] to improve the throughput, it is
designed to examine the benefit provided by the random
network coding.

III. PROPOSED WORK
[13] Introduces bit error diversity and proposes
Scattered Random Network Coding (S-RNC) to exploit
this diversity to improve the throughput for multihop
wireless communication. However, S-RNC relies on
random network coding and block differentiation (each
block must have a priority). S-RNC can be extended and
generalized to remove these two restrictions. We propose
to do so under two different heads-
1. Stronger Error Correction for `bad' bits.
2. Deciding which bits go to the `good' bits.
To improve the throughput we consider a new approach
for controlling the errors. In this approach, Strong error
correction will be provided to the bad bits as they have a
high probability of error. This will ensure not only error
resilience, but also reduced coding overhead. FEC
(forward error correction) is applied on the bad bit for
controlling the errors. A scheduling algorithm is proposed
that schedules the block so as to maximize the throughput
of the system. We choose the shortest remaining message
(SRM) algorithm so the message with the shortest length
will have the highest priority.The message with shortest
length will be put on the good block and longest length are
put on bad block .In order to avoid indefinite waiting for
long messages SRM with ageing is adopted as the packet
grows older their priority increases and for this an age a
is appended to all the messages (age is used only by
scheduler and is never transmitted).Initially age are set to 1,
so (M = {(m
1
, a
1
= 1), (m
2
, a
2
= 1),(m
3
, a
3
= 1), .}) and it
is incremented every time when the priority function is
called. Also, Shortest Remaining message algorithm is
used which gives priority to the message with the shortest
length with the hope that message can be sent to the
receiver quickly with less number of errors.
M is a set of messages that is
M = {m
i
| m
i
M, i}.
Each message (m
i
) consists of one or more bits. We can
represent m
i
as a regular expression as,
m
i
= (0|1)for all i.
Each message is composed of several blocks. If there
are n
i
blocks in message m
i
then
m
i
= [b
i1
b
i2
bi3 bin].
Place these blocks in block queue (BQ).Thus,
BQ=BQ b
ij
i, 1 j n
ij
Forward error correction (FEC) is used to control error
on bad bits. Since we are using 16QAM, the replication
FEC
for 16QAM is, ([b
i
.b
j
] converted to [b
i
..b
2j
]).
In 16QAM each block is composed of 4bits,
b
k
=[ b
k1
b
k2
b
k3
b
k4
],
the code word will be-

C=
11 12 13 14
21 22 23 24
c c c c
c c c c
(
(

=
1 2 1 2
3 4 3 4
k k k k
k k k k
b b b b
b b b b
(
(

To place the good or bad bits in their appropriate
position
Good matrix (GM) and bad matrix (BM)are used. The
good bits and bad bits are mixed together for transmission.
As
!6QAM is used 4bits per symbol can be send at a time
and the good and bad bits are mixed together in these
4bits.The code word will be-
C= (b
g
) GM + ( b
b
) BM
The algorithm begins when it receives a set of
messages M that are to be transmitted from the network
layer. Each message in M is partitioned into 4-bit blocks
since 16QAM is used as modulation symbol. These blocks
are appended to BQ. m
g
(lower value indicate higher
priority) and m
b
(higher value indicate lower priority) are
selected from among the messages in M . It is possible that
there is only one message in M or all messages have the
same priority (length). In order to exploit bit error
diversity to increase throughput, we need at least two
message with different priorities .Hence, if m
g
= m
b
, we
transmit the remaining blocks without any further
processing. In the other case, we move to the next step. In
the next step, we generate the sets GB (good block) and
BB (bad block).FEC is a flag that indicated whether the
Replication FEC is to be applied. We initialize it to `true'.
In preparation for placing the good/bad blocks in the
good/bad position, we select the first block from GB and
denote it by b
g
. The next step is the Replication FEC
(forward error correction). We select the first block from
BB (denoted as b
b
). Finally, we are ready to place the
blocks in the right position. As b
g
has been transmitted, we
can now remove it from BQ. After removal of any block
from BQ, we must check whether BQ is empty (stopping
condition) or if there is only one block in BQ (trivial
case).We also remove b
g
from GB. As GB contains all the
blocks of the message m
g
, an empty GB indicates that the
entire message has been transmitted. Hence, we remove m
g

from M. As BQ is not empty, but GB is empty, we need to
fill the next message in m
g
. We do this in the next two sub
steps. As before, we check if there is only one message or
the priority of all the messages are the same. Now we
proceed to b
b.
If FEC is true we should not perform FEC in
the next iteration (instead we should use b
2
). If FEC is false,
we have transmitted both b
1
and b
2
. Hence, we remove b
b

from BQ and BB. The reasoning is same as that used for
removing b
g
from BQ and GB. The process will continue
until the entire messages are being transmitted.

Algorithm: Input: Set of messages (M)
Output: Transmission of message.
1. Partition the message in M into blocks.
2. Append each block to BQ
3.(a)Select m
g
such that priority(m
g
) =
min{priority(m
i
)} i
(b)Select m
b
such that priority (m
b
) = max {priority
(m
i
)}
i
4. if (m
g
== m
b
) {transmit(BQ) and return}
5. (a) GB = {b | (b BQ) and (b m
g
)}
(b) BB = {b | (b BQ) and (b m
b
)}
6. FEC = true
7. Select first block in GB: b
g
(say)
8. if (FEC == true)
(a) Select first block in BB: b
b
(say)
(b) Define b
1
and b2 as follows-

1
2
b
b
| |
|
\
=
1 2 1 2
3 4 3 4
k k k k
k k k k
b b b b
b b b b
(
(

9. C = b
g
GM + b1 BM
10. Transmit C
11. Remove b
g
from BQ
12. .(a)if (length(BQ==0){return}
(b)else if (length(BQ)==1){Transmit(BQ) and
return}
13. Remove b
g
from GB
14. if (length(GB==0)
(a) Remove m
g
from M
(b) Select m
g
g
) =
min{priority(m
i
)}
i
(c) GB={b |(b BQ) and (b m
g
)}
(d) If (m
g
== m
b
) {transmit (BQ) and return}
15. if (FEC == true)
(a) FEC = false
(b) b
1
= b
2

16. else
(a) FEC = true
(b) Remove b
b
from BQ
(c) If (length (BQ) == 0) {return}
(d)Else if(length(BQ)==1){transmit (BQ) and
return}
(e) Remove b
b
from BB
(f) If (length(BB)== 0)
i. remove m
b
from M
ii. Select m
b
b
) = max
{priority
(m
i
)} i
iii. BB = {b | (b BQ) and (b m
b
)}
17. go to 7

IV . CONCLUSION
In this paper, we have provided a review for the reasons
of poor throughput in multihop wireless network. Error
control scheme has been introduced for random network
coding. Scattered random network coding relies on random
network coding and they put the bits of odd number blocks
in good bit position and even number blocks in bad bit
position. When transmission goes over multiple hops the
odd number blocks are protected from getting error while
the even number blocks will get dirtier. We can improve the
performance on the existing system by implementing a
strong error control scheme. To improve the throughput,
We have proposed a new approach by providing strong
error correction on the bad bits as they have a high
probability of error.FEC (forward error correction) is
applied on the bad bits for error correction and shortest
remaining message (SRM) algorithm is also used to give
priority to the message with the shortest length with the
hope that message will be delivered to the receiver quickly
and to avoid indefinite waiting time SRM with ageing is
also adopted.

REFERENCES

[1] R. Ahlswede, N. Cai, S. R. Li, and R. W. Yeung, Network
information flow, IEEE Trans. Inf. Theory, vol. 46, no. 4, pp.
12041216,Jul. 2000.
[2] L. Xiao, T. Fuja, J. Kliewer, and D. Costello, A network coding
approach to cooperative diversity, IEEE Trans. Inf. Theory, vol. 53,
no. 10,pp. 37143722, Oct. 2007.
[3] Z.Zhang,Theory and application of network error correction
coding, proceeding of the IEEE, vol.99, pp.406-420, 2011.
[4] S. Katti, H. Rahul, W. Hu, D. Katabi, M. Medard, and J. Crowcroft,
XORs in the air: Practical wireless network coding, in Proc. ACM
SIGCOMM, 2006.
[5] T. Ho, M. Medard, R. Koetter, D. Karger, M. Effros, J. Shi, and B.
Leong,A random linear network coding approach to multicast,
IEEE Trans. Inf.Theory, vol. 52, no. 10, pp. 44134430, Oct. 2006.
[6] D. Silva, F. Kschischang, and R. Koetter, A rank-metric approach
to error control in random network coding, IEEE Trans. Inf.
Theory, vol. 54,no. 9, pp. 39513967,Sep 2008.
[7] K.Chi, X.Jiang, B.Ye, H.Y.Shwe, and S.Horiguchi, Efficient
Network coding-Based End-to-End Reliable Multicast in Multi-hop
Wireless Network,Proceeding of the 15
th
Asia-Pacific Conference
on communication(APCC2009), 2009
[8] R. Koetter, F. R. Kschischang,Coding for errors and erasures in
Random Network Coding, IEEE Trans. Inf. Theory, vol. 54,
no.8,pp. 35793591, Aug. 2008.
[9] R.Ahlswede and H.Aydinian,On error control codes for Random
Network coding, Workshop on Network Coding, theory and
Applications, 2009.
[10] R.Y.Kim,J.Jin and B.Li,Drizzle:Cooperative symbol-level Network
Coding in WiMAX,IEEE Trans.veh.Technol.,vol.59,
no.3,pp.1415- 1432,Mar.2010.
[11] Z. Zhao, L. Zhou, X. Zheng, and S.-E. Park, Enhanced
constellation rearrangement for harq with symbol combining, in
Proc. IEEE MWS, Jul. 2009, pp. 1115.
[12] M.L. Jia, J.M. Kuang, and Z.W. He, Enhanced HARQ employing
LDPC coded constellation rearrangement with 64QAM, in
Proc.Int.Conf.Commun.Technol.,2006, pp. 14.
[13] R.Y. Kim, J. Jin, B. Li, Scattered Random Network Coding for
Efficient Transmission in MultihopWireless Network,IEEE
Trans.on Vehicular Tech., vol.60, 2011.


Part II
Proceedings of the Second International Conference on
Computer Applications 2012
ICCA 12
Volume 6
Volume 1. Copyright 2012 Techno Forum Group,
India.
ISBN: 978-81-920575-9-0 :: doi: 10. 73067/ISBN_0768
Effect of Bayesian method in the segmentation of multiple views of an object
Sabna A B
M. Tech Scholars, Dept. of CSE
MES College of Engineering, Kuttippuram
Kerala

AbstractImage Segmentation is one of the most important
concerns in digital image processing. Dierent
fundamental algorithms of image segmentation are
existing like Edge based segmentation, probability based
image segmentation, similarity based image segmentation,
graph cut based image segmentation, quad tree method,
segmentation based on function variation, region growing
method, region shrinking method implemented like Hough
Transform Algorithm (Edge Linking & Boundary
Detection), Region Growing (Similarity Based
Segmentation), Iterative Thresholding Method. A new
approach of Image Segmentation based on Maximum A
Posteriori is implemented which uses the concept of Baye's
Theorem.
Bayesian approach is the most suited type of
segmentation for multiview segmentation. This type
facilitates the increase of prior knowledge - which
determines the time and effectiveness of iterative mode of
segmentation, in each view operation. The success
percentage is more when used with user interaction rather
than automatic methods to distinguish between
foregrounds from background. Automatic segmentation for
the semantic objects extracted from multiview images has
got a less success percentage compared to methods which
involves user interactions. It is shown that the accuracy and
success percentage increases with the use of Bayesian
probability measures along with Graph cut refinement
process and finest edge detection mechanism.
Keywords-multiciew segmentation; Bayesian
approach;Prior knowledge; user interaction;automatic
segmentation.
I. INTRODUCTION
Image segmentation is one of the image processing
applications in which input is an image and output is
attributes or parts of that image. Segmentation refers to
the process of partitioning a digital image into multiple
regions (sets of pixels). Classically, image segmentation
is defined as the partitioning of an image into non
overlapping, constituent regions which are homogeneous
with respect to some characteristic such as intensity,
color or texture.
The ideal goal of segmentation is to identify the
semantically meaningful components of an image and
thus to simplify or change the representation of an image
into more meaningful and easier to analyses. Each of the
pixels in a region is similar with respect to some
characteristic or computed property, such as color,
intensity, or texture. Adjacent regions results after
segmentation are significantly different with respect to
the same characteristic.
Image Segmentation is used as the preliminary step in
many applications. Image segmentation should be finite
to the finest extension possible to get better results out of
the succeeding operations. Making perfect initial model
silhouette is a problem and challenge.
There are various types of image segmentations [1],
[2]. Figure 1.1 names some important ones. Edge based
segmentation, probability based image segmentation,
similarity based image segmentation, graph cut based
image segmentation, quad tree method, segmentation
based on function variation, region growing method,
region shrinking method are some examples

Figure 1.1: Types of segmentation

Edge detection methods: Edge based segmentation is
also known as contour based image segmentation. Edge
detection significantly reduces the amount of data and
filters out useless information[7], while preserving the
important structural properties in an image. In this type
the objects are selected by means of edges. Edge
detection refers to the process of identifying and locating
sharp discontinuities in an image. The discontinuities are
abrupt changes in pixel intensity which characterize
boundaries of objects in a scene. Edges are detected
using a combination of operators for intensity gradient,
texture discontinuities and color variation etc.
Supplementary processing step must follow to combine
edges in to edge chains. Edge chains are used in the
segmentation process. Canny's edge detector is a good
example.
Clustering methods: Clustering methods [8] are
usually iterative methods to partition an image into a
number of clusters or groups. The initial clusters have to
be re order to make it include only similar characteristics
at the end of the operations. K-means clustering is an
example clustering method. The quality of the solution
depends on the initial set of clusters and the value of K.
Threasholding methods: Threasholding is the
simplest way of doing image segmentation. It is a
technique which bands some range of values and permits
other range. The key parameter in the threasholding
process is the selection of threshold value [9]. If a single
threshold value is used, it is called a global
threasholding. Adaptive threasholding uses different
threshold values for different parts of the image. A task
well suited for local adaptive threasholding is
segmenting text from the image. During the
threasholding process, individual pixels in an image are
marked as "object" pixels if its value is greater than the
threshold and as "background" pixels otherwise.

Region based methods: Region based segmentation is
used for image characteristics to map individual pixels in
an image to sets of pixels called regions that might
correspond to an object or a meaningful part of one. The
various techniques comes under region based methods
are local techniques, global techniques, splitting
techniques and merging techniques. The effectiveness of
region growing algorithms depends on the application
area and the input image. If the image is sufficiently
simple, local techniques can be effective. However, on
difficult scenes, even the most sophisticated techniques
may not produce a satisfactory segmentation. Hybrid
techniques using a mix of the methods above are also
popular.
Quad tree methods: This method is also known as
region splitting and merging method of segmentation.
Quad tree method [10] considers image to be segmented
as a tree. The root of the tree represents the whole image.
If the image at root is found non uniform (not
homogeneous), then it is split into four son-squares (the
splitting process). This process continues recursively
until no further splits or merges are possible or all sub
trees are homogeneous. Conversely, if any child squares
are homogeneous, they can be merged (the merging
process). The QT structure allows dividing an image
within a complete tree representation, including
neighboring information. This spatial information can be
further used by a merging strategy which joins the QT
leaves using color and edge information.
Bayesian methods: Bayesian methods are also known
as probabilistic model of image segmentation. This
method calculates the maximum a posteriori probability
[11] of each pixel. This involves the use of inferred
probabilities rather than observed probabilities. Hence it
involves the use of conditional probability as in any
other case of Bayesian model segmentation. Section II
will give a detailed review of Bayesian method.
A. What is multiview segmentation?
Multiview segmentation is a new area of
segmentation and is found to be very useful in computer
vision applications. In its simple form, multiview is
nothing but multiple views of a single object from
different angles. Multiview segmentation extracts object
from each view. Combining different angle pictures of an
object from different angles can easily generate a 3
dimensional voxel of the object. Even though Multiview
segmentation is mainly used to generate the 3D form of
an object from its 2D image, it has many other computer
vision applications like object tracking, motion censoring,
images and video synthesis, animation etc.
Multiview can be segmented out with any type of
above discussed segmentation method. But Bayesian
method seems to be more appropriate for multiview
segmentation.

II. BAYESIAN METHOD-AN OVERVIEW

Bayesian method of segmentation classifies pixels
into groups based on the posterior probability of them.
Posterior probability explains the inferred probability
rather than observed probability. In the context of images,
the observed probability or simply the probability of a
pixel implies that the probability of a pixel to be assigned
to a specific intensity or color, whereas the inferred
probability implies the probability of a particular pixel to
have a specific intensity or color given that its
neighboring pixel is having a certain observed value.
From the observed value of pixels Bayesian method infer
the value of the other pixels. This will yield more
accurate values.
From Bayesian perspective, there are known values
and unknown values[12]. The known quantity is the data,
denoted as D. The unknown quantities are the parameters
(eg. mean, variance, or missing data), denoted by . Then
to make inferences about the unknown quantities,
Bayesian stipulate a joint probability function that
describes the unknown quantities behave in conjunction
with known data, p (, D) [13].
Bayesian Theorem for a given parameter can be
represented as
p (D) = p (D) p () / p (D)
or
Posterior likelihood x prior
Where,
The prior is the probability of the unknown parameter
and represents what was thought before seeing the data.
The likelihood is the probability of the data given
prior parameter and represents the data now available.
The posterior represents what is thought given both
prior information and the data just seen.
The likelihood and the posterior value is not variable
ones and they are the observed quantities. The only
quantity through which we can improve posterior
probability is by using the prior knowledge. As much
knowledge we have prior to operation that much
efficiency will be there in the posterior probability.

A. Why Bayesian Model?
Bayesian method involves conditional probabilistic
calculation of events. If any other method is used to do
silloughty estimation of segmentation, it does not make
any change from multiple numbers of single view
segmentations. But Bayesian method efficiently uses the
previous segmentation details (observed quantities) to be
in cooperated with the current segmentation (inferred
quantity). This will make current segmentation more
efficient, because the views are of the same object from
different angles.
Bayesian aspect of Inferred probabilities (from both
prior and observed values) and the dependencies among
them are clear from the given dependency graph

Fig 2. 1 Dependency graph of an image

Here ,
represents the prior knowledge;

is the background color model;

S is the binary silhouette map and

F is the foreground spatial model.
As shown in figure 2.1, the dependency graph, an
image observation I is influenced by the background
color at the corresponding pixel location of background
and by whether the background is occluded or not at
that location of Silloughty S, which is itself governed by
the projection of the foreground region F. It is assumed
that F and to be independent, which can be argued
because of shadows cast by the foreground, can change
the background appearance. It is assumed that shadows
have a negligible impact on the background colors.
Bayesian model utilizes this dependency to
efficiently estimate the silloughty of each image in the
multiview .Thus it provide better convergence than other
methods of segmentation.

III. RELATED WORK
Kalin kolev et. al, in [3] implements a probabilistic
formulation of multiview segmentation, given a series of
images from calibrated cameras. This is a silhouette
based approach. Instead of segmenting each image
separately in order to build a 3D surface consistent with
these segmentations, they had computed the most
probable surface that gives rise to the images. They used
two separate Bayesian frame work which are independent
of each other. One for background voxel and one for
foreground voxel. Voxel is nothing but small volume
areas. This framework converts the maximum a-
posteriori estimation into an energy minimization
problem. Minimizing this energy function is equivalent
to maximizing the total a posteriori probability of all
voxel assignments.
This method takes less time to converge, which is
about a factor 3 times faster than its previous methods. In
comparison, the probabilistic derivation and formulation
of the energy on the volumetric instead of the image
domain provides faster convergence and better
robustness. But this approach is susceptible to noise and
shading effects, since single observations are taken into
account for deciding whether a voxel should be
categorize forward and backward voxel. This could be
overcome by considering the averaging effect of
integrating data from all views or by allowing user
interactions
The same authors at [4] extended their work from this
basic framework by adding user interaction to distinguish
between the foreground and background. The user scribes
with 2 different colors in order to distinguish between
foreground and background. This avoids the shading
effect existed in the previous work. It also uses the recent
technology of parallel processing with GPU( A graphics
processing unit also occasionally called visual processing
unit or VPU ) is a specialized circuit designed to rapidly
manipulate and alter memory in such a way so as to
accelerate the building of images in a frame buffer
intended for output to a display. It consists of 100s of
cores to execute. With this technology the time of
processing gets reduced considerably up to a few
seconds.
Jean Sebastian and Franco Edmond Boyer at [5]
describe a Bayesian framework for multi-view silhouette.
This considers pixel based information rather than
considering voxel. This approach starts by having two
assumptions. One is a set of images, obtained from fully
calibrated cameras. Second is a set of background images
of the scene, free from any object of interest, have
previously been observed for each camera. Thus, we must
first model the impact of joint probability on the
observations. Modeling the relationships between the
variables involved requires computing the joint
probability of these variables. The decomposition is
based on the statistical dependencies.
This very basic framework is extended and modified
by Won woo et.al. in [6]. They specify a method for
extracting consistent foreground regions when multiple
views of a single scene are available. This automatically
identifies such regions in images under the assumption
that, in each image, background and foreground regions
present with different color properties. Second
assumption is that the object to be extracted includes
entirely in each images.
In this method, one object is viewed from different
angles and this different angle pictures are used in the
process of silhouette extraction. In the first iteration a
probable area is selected with the help of a histogram.
The Bayesian frame work is utilized in this point of
silhouette initialization. This is also an iterative
mechanism. On iterations the probable area first selected
gets refined and shrinks in to the shape of the object. This
method utilizes graph cut method for iterative
optimization. The graph cut approach finds new
silhouette labels from which new background models are
inferred before the next iteration. To terminate the
iterative optimization, observe the number of pixels
whose states changed from 'Unknown' to 'Background'
and stop the process when no further pixels are newly
identified as being in the background.
This is a fully automatic procedure which does not
need any background information or user interventions.
This method has some limitations; it fails when the
background and the foreground have less color
differentiation and when the object of interest does not
appear entirely in each image. The first disadvantage can
be overcome by integrating any of the finest edge
detection mechanisms in this method.

IV. CONCLUSION AND FUTURE WORK
Segmentation is used as a preliminary operation in
most of the image processing applications. This covers a
wide range of practical applications and computer vision
applications. Real time applications of image processing-
like satellite and medical image processing, especially
needs the clear and clever extraction of objects from
images. Multiview image segmentation is best
implemented with Bayesian network because it involves
the use of inferred probability rather than observed one.
The information from one image to the next is
transmitted in each view of segmentation through
Bayesian network and thus uses less time during
successive iterations compared to multiple single view
segmentations. Bayesian implementation together with
graph cut optimization and finest edge detection
mechanism is expected to provide better results.
The automatic Bayesian approach has some
limitations like it fails when the color distribution of
object and its background is less, and if the object of
interest does not appear entirely in images. The future
scope of the work is to overcome the above limitations. If
we can implement multiview segmentation of images
which does not have much differentiation from its
background, that will going to be a better milestone in 3D
reconstruction and other post processing applications of
multiview segmentation.

REFERENCES

[1] S Sapna Varshney, Navin Rajpa and Ravindar Purwar "comparative
Study of Image Segmentation Techniques and Object Matching using
Segmentation" International Conference on Methods and Models in
Computer Science ,2009.

[2] Dr .G .Padmavathi, M.Muthukumar, Suresh kumar Thakur
"Implementation and Comparison of Different segmentation Algorithms
used for Images Based on Nonlinear Objective Assessments" 3rd
International Conference on Advanced Computer Theory and
Engineering(ICACTE),2010.

[3] K. Kolev, T. Brox, and D. Cremers " Robust variational
segmentation of 3D
Objects from multiple views". In K. Franke et al., editor, Pattern
Recognition
Conference at Berlin (Proc. DAGM), volume 4174 of LNCS, pages
688697,
Berlin, Germany, September 2006.

[4] K. Kolev, T. Brox, and D. Cremers. "Fast Joint Estimation of
Silhouettes and Dense 3D Geometry from Multiple Images " Ieee
transactionson pattern analysis and machine intelligence ,Digital Object
Indentifier .1109/TPAMI.2011.150 0162, june 2011.

[5] J.-S. Franco and E. Boyer. "Fusion of multi-view silhouette cues
using a space occupancy grid". In Proc. International Conference on
Computer Vision, Beijin, China, 2005.

[6] Wonwoo Lee, Woontack Woo "silhoutte segmentation in multiple
views " Ieee transactions on pattern analysis and machine intelligence,
VOL. 33, NO. 7,
2011.The template will number citations consecutively within brackets
[7] Morris, O.; M. Lee A unified method for segmentation and edge
detection
using graph theory Acoustics, Speech, and Signal Processing, IEEE
International Conference on ICASSP '86. Volume: 1986
[8] Juanying Xie; Shuai Jiang; A Simple and Fast Algorithm for Global
K-means Clustering IEEE Education Technology and Computer
Science (ETCS), 2010 Second International Workshop on Volume: 2 ,
2010
[9] Guang Yang; Kexiong Chen; Maiyu Zhou; Zhonglin Xu; Yongtian
Chen; Study on Statistics Iterative Thresholding Segmentation Based on
Aviation Image , 2007. Eighth IEEE ACIS International Conference :
2007
[10] Hong Lan; Ling Chen; Wei Hu; An apIproach on liver medical
image
segmentation based on quad tree Multimedia Technology (ICMT),
2011
IEEE International Conference 2011 .
[11] Xiaomu Song; Guoliang Fan; A study of supervised, semi-
supervised and unsupervised multiscale Bayesian image segmentation
IEEE international conference 2002 Page(s): II-371 -II-374 vol.2
[12] Morris, R.; Descombes, X.; Zerubia, J.; Fully Bayesian image
segmentation-an engineering perspective Image Processing,
Proceedings., IEEE International Conference on Pattern Recognition
1997
[13] Yi Zhou; Wei Zhang; Xiaoou Tang; Harry Shum; A Bayesian
mixture model for multi-view face alignment Computer Vision and
Pattern Recognition, 2005. CVPR 2005. IEEE Computer Society
conference 2005

India.
ISBN: 978-81-920575-9-0 :: doi: 10. 72955/ISBN_0768
On A Graph Theory Based Approach For Improved Computer Vision-II
V.Yegnanarayanan
1
1
Senior Professor, Department of Mathematics,
Velammal Engineering College,Ambattur-Red Hills Road,
Chennai - 600 066, India.

Abstract - Given any information about the object
under observation, the various tasks a vision system is
expected to achieve like deduction of features, recognition
of objects, manipulation of objects etc can be
accompanished by making use of a graph model based
approach. In this paper we point out how certain
techniques of graph theory can be diligently used
especially in the main problem of sensor planning. The
various techniques discussed here includes graph cuts for
binary image restoration, bipartite matching algorithms,
minimum cut algorithms, shortest path algorithms etc. We
also touch upon the importance of discrete optimization
techniques for solving computer vision problems. More
over we also indicate other applications and related
survey on how sensor planning is being done for
computer vision mainly for the task of feature
deductibility.

Keywords
Computer vision, graph, graph Algorithms, graph
cuts

I. MOTIVATION
Sensor planning in computer vision is an emerging
research area that tries to understand and quantify the
relationship between objects to be viewed and the sensors
observing them in a model-based, task directed way. The
importance of the viewpoint in this object-viewer
relationship is evident. It is the viewpoint, for the most
part, that creates the quality of the resulting image [34],
determining the feasibility of the vision task and
facilitating its execution. Previous work in computer
vision, however, has generally placed an emphasis on the
object being observed [21] assuming that the viewpoint is
given, suitable for the task, and not controlled. The issues
related to determining viewpoints that will be most
suitable for the vision task at hand have received
considerably less attention. This latter area includes
questions such as What should the observer pose be? or
What values should other observer attributes have?
Even in currently employed vision systems that
observe known objects in known poses (e.g., visual
inspection, surveillance, or monitoring systems),
appropriate sensor parameter values are determined by
often laborious and time-consuming techniques.
Generally, a trial-and-error approach involving human
interaction is taken. Sensor locations and settings are
chosen and then tested in order to verify whether they
meet the requirements of the task at hand. The resulting
parameter values are valid for only a specific setup and
can potentially become unsatisfactory when errors (e.g.,
robot inaccuracy) alter the environment. Such procedures
constitute a major bottleneck in system installation and
result in a design cost which often exceeds that of the
equipment. As a result, these applications are costly and
have limited intelligence and flexibility.
Sensing strategies are also required for vision systems
that perform object recognition since a single sensor
configuration may not always result in a sufficiently
informative view. For instance, in the cases of occlusion
and near-symmetrical or similar objects, more than one
interpretation may be consistent with the sensory data. As
a result, additional sensor configurations are needed from
which observed details of the scene can help recognize
the unknown objects by disambiguating among multiple
interpretations.
In addition to the sensor parameters themselves,
another set of key parameters in reconfigurable vision
systems that can be controlled, and thus may need to be
planned, are those associated with the illumination of the
scene. If the interaction between lighting and the object
surface to be imaged is carefully considered, the image
quality can be substantially improved, leading to more
informative images and making later processing of the
image much easier. For instance, by planning the
illumination in a visual inspection application, features of
interest can be made to appear more prominent, noise due
to extraneous features can be reduced or eliminated, and
the accuracy with which an object is measured, can be
increased. While techniques that vary the
illumination(e.g., illuminator location, polarization of
illumination, etc.) have been employed in several vision
systems, illumination planning has been overlooked to a
great extent. Since image
acquisition is much less costly than the subsequent
image analysis, it seems advantageous to dedicate
computational effort to determine appropriate values for
both the viewpoint and the illumination parameters.
Sensor planning is pertinent to a number of areas of

robotics and computer vision that have been studied
extensively in the past.For instance, the general problem
of task planning in robotics and its component areas of
motion planning, `grasp planning, and assembly planning
can be viewed as different facets of the sensor planning
problem. In addition, recent research has underlined the
importance of an area very much related to sensor
planning, namely, that of active sensing. In active sensing
sensor parameters are controlled in response to the
requirements of the task [6], [53]. It has been shown in
[3]that active sensing can take ill-posed vision problems
and render them well posed through constraints
introduced by the dynamic nature of the sensor. Also, by
actively reconfiguring a sensor system its effective scope
is increased, and as a result, a range of sensing situations
can be accommodated. For example, the field of view of a
mobile observer is far more extensive than that of a
stationary one, while required object detail can be
resolved by either reducing the viewer-object distance or,
when possible, by modifying the associated lens settings.
In order to purposefully alter sensor configurations, active
sensing requires a sensor planning component that
determines appropriate sensor parameter values. The goal
of sensor planning is to automatically, rather than
manually, generate appropriate sensor configurations
based on any a priori known information that is often
available. For instance, the required geometric and
physical information of objects can be extracted from
CAD/CAM models which are often available in todays
manufacturing environment. Camera and illumination
models approximating their physical and geometric
properties can also provide the planning system with the
required sensor and illumination characteristics. The
planning algorithms can use this model information and
augment it with knowledge regarding the functions the
system is to perform. In this way, the sensor-based system
will be able to reason about its own configuration in order
to achieve the task at hand.

II. INTRODUCTION
In a letter to C.Huygens of 1679,G.W.Leibnitz
expressed his classification with the standard coordinate
geometry treatment of geometric figures and maintained
that we need yet another kind of analysis, geometric or
linear, which deals directly with position, as algebra deals
with magnitude[9]. In fact, Leibnitz initiated the so called
geometry of positions(geometria situs) which , as
L.Euler clearly put it in his1736 konigberg bridges paper
which had to mark the beginning of graph theory, is
concerned only with the determination of position, and its
properties; it does not involve measurements or
calculations made with them[23]. After about two
centuries, this study developed into two of the richest
branches of modern Mathematics: graph theory and
combinatorial topology.
Mutatis Mutandis, an analogous discontent is
nowadays being felt among many researchers working in
computer vision, a field that is currently dominated by
purely geometric methods, who are increasing making use
of sophisticated graph-theoretic concepts, results and
algorithms. Indeed, graphs have long been an important
tool in computer vision, especially because of their
representational power and flexibility. However, there is
now are renewed and growing interest toward explicitly
formulating computer vision problems as graph problems.
This is particularly advantageous because it allows vision
problems to be cast in a pure, abstract setting with solid
theoretical underpinnings and also permits access to the
full arsenal of graph algorithms developed in comport
science and operations research.
Graph theoretic problems which have proven to be
relevant to computer vision include maximum flow,
minimum spanning tree, maximum clique, shortest path,
maximal common subgraph /sub tree etc. In addition, a
number of fundamental techniques that were designed in
the graph algorithms community have recently been
applied to computer vision problems. Examples include
spectral methods and fractional rounding.
Scope of Graph Theory ideas to computer vision
The idea of graph partitioning, poses the problem of
making cuts in a weighted graph according to an
appropriate minimum weight criterion. Typical
applications in computer vision includes image
segmentation or perceptual grouping; The method of
graph indexing, addresses the problem of efficiently
selecting a small number of candidate graphs(from a large
database) that may account for a query graph; The idea of
graph matching, attempts to compute correspondence
between two graphs representing underlying image
structure. Graph matching is common in problems
ranging from object recognition to image registration;
Then the idea of graph generalization, involves computing
a prototype graph from a number of exemplar graphs, an
important problem in object recognition and object
modeling.
3. Binary Image Restoration Via Graph cuts
There are a number of algorithms for solving low
level vision problems that compute a minimum s-t cut on
an appropriately defined graph. This technique is usually
called graph cuts. The idea has been applied to a
number of problems on computer vision, medical imaging
computer graphics etc that can be formulated in terms of
energy minimization. Such energy minimization problems
can be reduced to instances of the maximum flow
problems in a graph. Consider a binary image where each
pixel has been independently computed (for example, the
output of an error prone document scanner). Our goal is to
clean up the image. This can be naturally formulated as an
optimization problem. The labels are binary, so the search
space is { }
n
1 , 0 where n is the number of pixels in the

image, and there are
n
2 possible solutions. The function
i
D provides a cost for labeling the i -th pixel using one of
two possible values: in practice, this cost would be zero
for it to have the same label that was observed in the
corrupted image, and a positive constant for it to have
the opposite label .The simplest choice for ( )
j i ij
y x V , is
a 0-1 valued function, which is 1 just in case pixels i and j
are adjacent and
j i
y x = . Putting these together we see
that ( )
n
x x x E ,..., ,
2 1
equals times the number of
pixels that get assigned a value different from the one
observed in the corrupted image, plus the number of
adjacent pixels with different labels. By minimizing this
energy we find a spatially coherent labeling that is similar
to the observed data.
This energy function is often referred to as Ising
model(its generalization to more than 2 labels is called the
potts model). The problem of minimizing this energy can
be reduced to the min-cut problem on an appropriately
constructed graph. With the appropriate choice of edges
weights, the cost of a graph cut is the energy of the
corresponding labeling.
It is important to realize that this construction depends
on the specific form of the Ising model energy function.
Since the terminals in the graph correspond to the labels,
the construction is restricted to 2 labels. Note that the
natural multiterminate variants of the min-cut problem are
NP-hard. In addition, we cannot expect to be able to
efficiently solve even simple generalizations of the of the
Ising energy function, since minimizing the potts energy
function with 3 labels is NP-hard;
4. Bipartite matching Algorithms
In the minimum weight bipartite matching problem
we have a bipartite graph with B A V = and weights
e
w associated with each edge. The goal is to find a
perfect matching M with minimum total weight. Perfect
matchings in bipartite graphs are particularly interesting
because they represent one-to-one correspondences
between elements in Aand elements in B .In computer
vision the problem of finding correspondences between
two sets of elements has many applications ranging from
three-dimensional reconstruction to object recognition.
The bipartite matching problem described here is
also known as the assignment problem. We also note that
there is a version of the problem that involves finding a
maximum weight matching in a bipartite graph, without
requiring that the matching be perfect.
We can solve the bipartite matching problem in
( )
3
n O time using the Hungarian algorithm where
B A n = = . Given a set U and a function f defined
on all subsets of U , consider the relationship between
( ) ) (B f A f + and ) ( ) ( B A f B A f + . If the two
quantities are equal (as such as when f is set cardinality),
the function f is said to be submodular. Submodular
functions can be minimized in high-order polynomial
time. There is a close relationship between sub modular
functions and graph cuts.
Given a graph, let u be the nodes of the graph, and
( ) A f be the sum of the weights of the edges from nodes
in A to nodes not in A.Such an f is called a cut
function.
A cut-function is sub-modular. For, Each term in
) ( ) ( B A f B A f + also appears in ( ) A f and
) (B f . Now consider an edge that starts at a node that is
in A, but not not in B, and ends at a node in B, but not
in A.Such an edge appears in ( ) ) (B f A f + but not in
) ( ) ( B A f B A f + .Also note that a cut function is
not modular. For, a similar argument shows that the cost
of a s-t cut is also a sub modular function: that is, u
consists of the non terminal nodes, and ( ) A f is the
weight of the outgoing edges from { } s A , which is the
cost of the cut. While cut functions are sub modular, the
key to computing min-cuts efficiently is the famous
reduction of Ford and Fulkerson. The reduction uses
maximum flow problem (max flow), which is defined on
the same directed graph, but the edge weights are now
capacity constraints. An s-t flow is an assignment of non-
negative values to the edges in the graph, where the value
of each edge is no more than its capacity. Further, at
every non-terminal vertex the sum of values assigned to
incoming edges equals sum of values assigned to
outgoing edges. From the stand point of computer vision,
max-flow algorithms are very fast, and can even be used
for real time systems. The most well known algorithm for
the single-source shortest paths problem is due to
Dijkstra. For a directed graph the Dijkstras algorithm
runs in | | V E O log time assuming that there is at least
one edge touching each node in the graph. It assumes that
all edge weights are non-negative and builds shortest
paths from a source in order of increasing length. In fact,
computing shortest paths on an arbitrary graph with
negative edge-weights is NP-hard.(Since this would
allows as to find Hamiltonian paths).However, there is an
algorithm that can handle graphs with some negative edge
weights, as long as there are no negative length cycles.
The Bellman-Ford algorithm can be used to solve the
single-source problem on such graphs in ( ) V E O time.
The method is based on dynamic programming. It
sequentially computes shortest paths that uses at most i
edges in order of increasingi .It can also be used to deduct

negative length cycles on arbitrary graphs. This is an
important subroutine for computing minimum ratio cycles
and has been applied to image segmentation and
deformable shape matching.
The all pairs shortest paths problem can be solved
using multiple calls to Dijkstras algorithm. We can
simply solve the single-source problem starting at every
possible node in the graph. This leads to an overall
method that runs in | | V V E O log time. This is a
good approach to solve the all-pairs shortest paths
problem in sparse graphs. If the graph is dense we are
better off using the Floyd-Warshall algorithm. That
algorithm is based on dynamic programming and runs on
| |
3
V O time. Moreover, like Bellman-Ford algorithm, it
can be used as long as there are no negative length cycles
in the graph. Another advantage is that it can be
implemented in just a few lines of code and does not rely
on an efficient priority queue data structure.
5. Minimum cut Algorithms
The minimum cut problem(min cut) is to find the
minimum cost s-t cut in a directed graph. On the surface,
this problem may look intractable. It is possible to show
that there is a polynomial-time algorithm to compute the
min-cut, by relying on results from sub modular
optimization. If the solution to the relaxation happens to
occur at an element of S , then by solving the relaxation
we have solved the original problem. However, if the
solution to the relaxation does not lie in S ,it provides no
information about the original problem beyond a lower
bound.(Since there can be no better solution in
'
S .In
computer vision, the most widely used relaxations involve
linear programming or spectral graph partitioning.
Spectral graph partitioning methods can efficiently
cluster the vertices of a graph by computing a specific
eigen vector of an associated matrix. These methods are
exemplified by the well known normalized cut algorithm
in computer vision. Typically a discrete objective
function is first written down, which is NP hard to
optimize. With careful use of spectral graph partitioning,
a relaxation of the problem can be solved efficiently.
Presently, there is no technique that converts a solution to
the relaxation into a solution to the original discrete
problem that is provably close to the global minimum.
Despite these lack of guaranteethese methods can
perform well in practice.
6. Problem Reductions
Problem reduction, a powerful technique show how an
algorithm that solves one problem can also be used to
solve a different problem, usually by transforming an
arbitrary instance of the second problem to an instance of
the first. Suppose the second problem is NP-hard, and
hence widely believed to require exponential time, the
method gives a proof that the first problem is at least as
difficult as the second one. However, Discrete
optimization methods, typically perform a reduction in the
opposite direction, by reducing a difficult appearing
problem to one that can be solved quickly. Many NP-hard
problems falls within the realm of discrete optimization.
On the other hand, problems with easy solutions can
also be phrased in terms of optimization: for example,
consider the problem of sorting n numbers. The search
space consists of the set of ! n permutations, and the
objective function might count the number of pairs that
are mis-ordered.
7. Graph Algorithms
Many discrete optimization methods have a natural
interpretation in terms of a graph ( ) E V G , = , given by a
set of vertices V and a set of edges E.We will often
associate a non-negative weight
e
w to each edge e in a
graph. In computer vision, the vertices v usually
correspond to pixels or features such as interest points
extracted from an image, while the edges e encode spatial
relationship. The length of a path in a weighted graph is
the sum of the weights associated with the edges
connecting consecutive vertices in the path. A tree is a
connected acyclic undirected graph.
Now consider a directed graph with two
distinguished vertices s and t called terminals. An t s
cut is a partition of the vertices into two components S
and T such that S s e and T t e .The cost of the cut is
the sum of the weights on edges going from vertices in S
to vertices in T .A matching M in an undirected graph is
a subset of the edges such that each vertex is in atmost
one edge in M. A perfect matching is one when every
vertex is in some edge of M .The weight of a matching
M in a weighted graph is a sum of the weights associated
with edges in M . A Perfect matching in a bipartite graph
defines a one-to-one correspondence between vertices of
one partite set to vertices of the other partite set.
8. Shortest path Algorithms
This finds minimum length paths between pairs of
vertices in a graph. There are two common versions of the
problem: 1)In the single-source case we want to find a
shortest path from a source vertex s to every other vertex
in a graph 2)In the all pairs case we look for a shortest
path between every pair of vertices in a graph. The former
is used often in computer vision. The main property that
we can use to efficiently compute shortest paths is that
they have an optimal substructure property: a subpath of a
shortest path is itself a shortest path. All of the shortest
path algorithms use this fact Vision problems are
inherently ambiguous and in general, we can only expect
to make educated guesses about the content of images.
A natural approach to address this issue is to devise an
objective function that measures the quality of a potential

solution and then use an optimization method to find the
best solution.
Nowadays, there is an increasing emphasis on
discrete optimization methods such as graph algorithms
for solving computer vision problems.
9. Discrete optimization concepts
An optimization problem involves a set of candidates
solution S and an objective function R S E : that
measures the quality of a solution. In general, S is defined
implicitly and consists of a very large number of
candidate solutions. E measures badness, the
optimization problem is referred to as an Energy
Minimization Problem, and Eis called an Energy
function. Since many papers in computer vision refer to
optimization as energy minimization, we assume that an
optimal solution S x e is one that minimizes an energy
function E.
The ideal solution to an optimization problem would
be a candidate solution that is a global minimum of the
energy function, ) ( min arg
*
x E x
S xe
= .This suggests
designing an energy function that fully captures the
constraints of a vision problem, and then applying a
general purpose energy minimization technique. But such
an approach fails to pay heed to computational issues,
which have enormous practical importance. No algorithm
can find the global minimum of an arbitrary energy
function without exhaustively enumerating the search
space, since any candidate that was not evaluated could
turnout to be the best one.
Note that in general global minimum of an energy
function might not give the best results in terms of
accuracy, because the energy function might not capture
the right constraints. But if we use an energy
minimization algorithm that provides no theoretical
guarantees it can be difficult to decide if poor
performance is due to the choice of energy function as
opposed to weakness in the minimization algorithm.
Most of the energy functions that are used in vision
have a general form that reflects the observed data and the
need for a bias towards a candidate solution called prior.
( ) 1 ) ( ) ( ) ( + = x E x E x E
prior data
The first
term penalizes the candidate solutions that are
inconsistent with the observed data, while the second term
imposes the error.
We will often consider an n -dimensional search
space of the form
n
L S = , where Lis an arbitrary finite
set. We will refer to Las a set of labels, and will use k to
denote the size of L. This search space has an
exponential number,
n
k of possible solutions. A
candidate solution
n
L x e will be written as
( )
n
x x x ,..., ,
2 1
where L x
i
e . One can also consider
search spaces
n
L L L S = ...
2 1
whICH allows the
use of different label set for each
i
x . A particularly
common class of energy functions in computer vision can
be written as
( ) ( ) ( ) ) 2 ( , ,...,
,
2 , 1
+ =
j i
j i
ij i
i
i n
x x V x D x x x E
.The terms
i
D are used to ensure that the label
i
x is
consistent with the image data, while the
ij
V terms ensure
that the labels
i
x and
j
x are compatible.
10. Continuous Vs Discrete Methods
The fundamental difference between continuous and
discrete optimization methods concerns the nature of the
state space. In the former, we are looking for a set of real
numbers and the number of candidate solutions is
uncountable. In a discrete optimization problem we are
looking for a set of real numbers and the number of
candidate solutions is uncountable. In a discrete
optimization problem we are looking for an element from
a discrete (often, finite)set . Continuous methods are
powerful, but it does not produce guarantees concerning
the absolute quality of the solutions they find(unless, the
energy function is convex). But instead, they guarantee
about the speed of convergence towards a local minimum.
Sometimes it is difficult to minimize E over original
discrete set S , but easy to minimize E over a continuous
set
'
S that contains S . Minimizing E over
'
S instead of
S is called relation of the original problem.
11. Other Applications
Several researchers have recognized the importance of
such a planning component for a new generation of
automated visual inspection systems [44], [59], [61]. In
[58] a system is defined to automatically generate
dimensional measurement applications from CAD models
of parts. In a concept based inspection system is outlined
in which a model-based inspection system would be
augmented with a generic set of principles in order to
determine appropriate and flexible behavior in new
situations. The need for machine vision development tools
that will assist machine vision designers and allow less
specialized labor to be employed is underlined
in [61]. The tool proposed would automatically
generate the configuration of a machine vision system by
selecting sensors, illumination, optics, and image
processing algorithms. Vision sensor planning is also
useful for robot-controlled vision systems in which
cameras and light sources are mounted on robot
manipulators (e.g., [8], [59]). In order for these systems to
perform their task reliably (e.g., vision-guided remote
assembly/disassembly or manipulation of objects in space
or in hazardous environments), selection of the proper
vision sensor parameter values is critical. The developed

sensor planning techniques could be used to automatically
position and orient the cameras and light sources as well
as to control the camera optics (e.g., controlling the zoom,
focus, and aperture settings of programmable zoom
lenses).Sensor planning techniques are also applicable to
areas such as the automatic synthesis of vision programs
from task specifications and model information [25]
[39].For example, a vision program to inspect an object
can be automatically generated, to a certain extent, based
on sensor planning techniques that determine appropriate
camera and illuminator poses, optical settings, and image
processing algorithms. It is important to note that
planning techniques developed for vision sensors can
prove useful in other areas of automation, such as the
automated machining and dimensional inspection of
mechanical parts [16], [51], [53]. For instance, planning
for accessibility of a surface in order to machine or probe
it by tactile means is equivalent to planning for visibility
of this surface assuming orthographic projection. In these
latter domains of automation, parameters such as work
piece and probe orientation, machine selection and cutter,
or probe type selection are determined by the planning
systems. Vision sensor planning techniques can also be
used in the area of computer graphics for the automatic
generation of viewing specifications that result in scene
renderings that achieve a certain intent .
By employing a sensor planning component in these
application domains:
1) The development cycle and thus the cost of a sensor
system is reduced since sensor configuration can be done
automatically;
2) Sensor parameter values can be found that are
robust (i.e., satisfying the task requirements at hand even
in the presence of uncertainty);
3) Sensor parameter values can be determined that
optimize the sensor output with respect to a given task
criterion, (e.g., camera and illuminator poses that enhance
the contrast
between task relevant features and the background);
4) A sensor-based system can adaptively reconfigure
itself to accommodate variations in the workplace.
12.Sensor Planning for vision
In an attempt to limit the scope of this survey, we have
focused on sensor planning for vision sensors (e.g.,
cameras, range finders, illuminators, etc.). Furthermore,
within the area of vision sensing, we have tried to limit
our study to systems using higher-level model-based
approaches as opposed to pixel-level active vision. We
are concerned with finding a generalized viewpoint that
includes sensor parameters other than camera placement
(e.g., focus, aperture, illumination
placement). Active vision systems that vary their
parameters by an assumed a priori schedule-randomly or
heuristically rather than in a knowledge-based manner
with an identifiable selection criterion-is outside the scope
of this survey. Following the approach, we can classify
the approaches to the sensor planning problem by the
vision task to be achieved or similarly by the amount of a
priori known information about the scene. This allows us
to identify three distinct areas of research: object feature
detection, model-based object recognition and
localization, and scene reconstruction. The first area,
object feature detection, seeks to automatically determine
vision sensor parameter values for which particular
features of a known object in a known pose satisfy
particular constraints when imaged [5], [7], [20],
[21],[38],[44]-[48], [55], [56], [57]. For example, the
features are required to appear in the image as being
visible, in-focus, and magnified to a given specification.
These planning techniques draw on the considerable
amount of a priori knowledge of the environment, the
sensors, and the task requirements. Because the identities
and poses of the viewed objects are known, the sensor
parameters are usually preplanned off-line and then used
on-line when the object is actually observed. Our own
interest is in this first area, and this survey reflects this
interest as well as our belief that it is an important
emerging research area. In the remainder of this section
references to, and a brief description of, the other two
areas is presented. The second area is concerned with
developing sensing strategies for the tasks of model-based
object recognition and localization [14], [15], [26], [32],
[36], [40]. Here, sensing operations are chosen that will
prove most useful when trying to identify an object or
determine its pose. In this work, the a priori known
information about the world in the form of models of
possible objects, sensor models, and information acquired
to this point are compiled into recognition/localization
strategies (see Fig. 1). Most approaches in this area follow
a common theme. Namely, a search is performed in the
space of object identities/poses employing the
hypothesize-and-verify paradigm:
1) Hypotheses are formed regarding the object
identities and poses;
2) These hypotheses are assessed according to certain
metrics;
3) New sensing configurations are proposed based on
a given criterion until a stopping condition is met.
Since after the first step in the above approach the
identities and poses of the objects in the scene have been
hypothesized, the sensor planning techniques of the first
area are also
applicable in this second area. In order to limit the
search of sensor parameter space in this
hypothesize-and-verify paradigm, a discrete
approximation of this space is commonly employed. For
example in [26], sensor configurations are only chosen
from a known finite set. In [13],[32] a discrete
approximation of viewing space based on the aspect graph
representation of the object is used. The individual
approaches following this general methodology basically

differ in the hypothesis evaluation metrics, the criterion
for the next sensing operation or the stopping condition.
1) The quality of the match between image and scene
features is evaluated by Cameron in [14] using a cross
correlation measure. In the work by Hutchinson and Kak
[32], confidence in matches between model and scene
features are based on the similarity of the area and the 3D
shape attributes of the surfaces.
2) In forming the hypotheses, Magee [40] takes a
theorem proving approach to establish a symbolic
correspondence between scene and image points, while in
Kims work, the
correspondence between object images from one view
to the next is determined by employing correspondence
techniques used in stereo. In the work by Grimson [26], a
set of possible
interpretations of the original sensory data is assumed
to have been determined by some other means.
3) The criterion for the next sensing operation is based
on optimally verifying the current hypotheses according
to a metric. A very similar metric is employed in most of
the approaches. For example, in [32] this metric involves
reducing the maximum uncertainty of the hypothesis set,
while in [14],[26], [36], [40] it entails viewing particular
disambiguating features. Similar to the aforementioned
sensor planning work for object recognition, there has
also been work [10], [11], [24], [28], [33] in automatically
determining strategies that guide the search for matching
image and scene features in Step 2 of the hypothesize-
and-verify paradigm. In this work, however, sensing
configurations are not planned, but rather, for whatever
the sensing configuration happens to be, a strategy is
given to match image and scene features based on this
initial set of sensory data without often acquiring
additional sensory data.
The third area addresses the problem of determining
sensing strategies for the purpose of scene reconstruction
[l], [12],[17], [18], [27], [42], [51], [59], [62]. In this case,
a model of the scene is incrementally built by
successively sensing the unknown world from effective
sensor configurations using the information acquired
about the world to this point. At each step of the
algorithm, new sensor configurations are chosen based on
a particular criterion (e.g., ability to explore the largest
area of unknown space). The sensory information
acquired at each step is then integrated into a partially
constructed model of the scene and new sensor
configurations are generated until the entire scene has
been explored. While there is no a priori known scene
information that can be used in this problem, the iterative
sensing is guided by the information acquired to each
stage. With respect to the planning component of this
work, the various approaches differ in the criterion with
which a new sensor configuration is chosen. The
approaches also differ in the way the multiple views are
integrated into a scene model. The work described in [2],
[29], [41] is related in that it addresses the problem of
reconstructing a scene from multiple views, however, the
viewpoints are predefined (i.e., assumed to be given) and
are not plannedin any manner.
13. Sensor Planningfor the Task of Feature
Detectability
This problem is fairly new in the area of computer
vision but has recently received considerable interest.
Several vision planning systems are being developed that
draw on the a priori known information regarding the
observed object and the employed sensors in order to
automatically determine vision sensor parameter values
that satisfy certain feature delectability constraints. The
developed techniques differ in their general approach to
determining sensor parameter values.
Several systems [46], [47], [48] take a generate-and-
test approach, in which sensor configurations are
generated and then evaluated with respect to the task
constraints. In order to limit the number of sensor
configurations that are considered, the domain of sensor
configurations is discretized. The viewing space
discretization in these systems is performed by
surrounding the object with a tessellated viewing sphere
and limiting the possible sensor locations to either groups
or individual tessels on the sphere. Determination of the
sensor parameter values is formulated as a search over
this discretized domain of sensor configurations guided
by task-related heuristics.The sensor planding methods
described in [4], [5],[19]-[21], [55], [56], [57] take a
synthesis approach. In this approach, the task
requirements are characterized analytically and the sensor
parameter values that satisfy the task constraints
are directly determined from these analytical
relationships. There has also been work related to sensor
planning in the area of sensor simulation systems [31],
[35], [45]. In such systems, a scene is visualized given the
description of the objects, sensors, and light sources.
These systems provide a framework for planning of
sensor configurations. For instance, by taking a generate-
and-test approach, satisfactory sensor configurations can
be found by creating a simulated view of the scene and
evaluating the task constraints in the simulated
image.Finally, there has been related work that follows
the expert systems paradigm [7], [38], [44]. In such
systems, expert
knowledge of viewing and illumination techniques is
incorporated into an expert system rule base. After
acquiring information regarding the particular object to be
observed, the expert system provides advice regarding the
appropriate sensor configuration.
Defining the Problem
In this section we describe the problem of sensor
planning for feature detectability in more detail. In
Section I the problem was summarized as automatically
determining camera and illumination parameters (e.g.,
position, orientation, settings) that satisfy certain feature

detectability task requirements (e.g. visibility, focus,
contrast) by drawing on the knowledge of the
environment, the sensors, and the task requirements
themselves. From this description, it can be seen that the
problem is model based and task driven. The models
include information of the objects in the environment and
the employed sensors.This a priori knowledge, together
with knowledge of the task requirements that need to be
satisfied, are incorporated into the planning algorithms.
In the following sections, we discuss each of the
above mentioned components of the problem. These are
1) The sensor parameters,
2) The sensor and object models, and
3) The feature detectability constraints.
1 ) Sensor Parameters: The parameters that need to be
determined for vision sensors are basically associated
with the point from which the scene is observed (i.e.,
viewpoint) and the point from which the scene is
illuminated. Customarily, a viewpoint is considered to be
the viewer location alone.However, it is useful to define
the concept of a viewpoint in a broader sense that includes
not only the viewer orientation, but also the optical
settings associated with the viewpoint. These settings are
also observer attributes affecting the resulting image of
the scene. This broader definition, then, identifies two
types of sensorparameters: geometric and optical. The
geometric parameters are independent of any of the lens
optics. Parameters of this
type the following:
- The three positional degrees of freedom of the
sensor (x,y,z)-this parameter is the position vector of a
point associated with the sensor. An example of a
convenient point for this purpose is a point on the C-
mount surface of the lens since the relationship between
the C-mount and the sensor does not vary. The
characteristic points of the lens itself (e.g., the principal
points, the nodal points, the pupils, etc.) can also be used.
However, since these points are not rigidly fixed to the
sensor, the relative movement of these points as the
optical settings of the lens varymust be accounted for.
- The three orientational degrees of freedom of
the sensor-this parameter can be specified by the pan, tilt,
and swing angles of a vector rigidly attached to the
sensor. An example of a convenient vector of this type is
a unit vector along the viewing direction (i.e., optical
axis). While the characteristic points of the lens change as
the optical settings of the lens vary, the viewing direction
remains constant.
The optical parameters, on the other hand, depend on
the lens settings. Parameters of this type include the
following:
- The back principal point
1
to image plane
distance, d-this parameter d is varied in the several focus
control mechanisms that are employed in lenses in order
to achieve a focused image. Also, d is the image distance
that is employed in the Gaussian lens formula governing
focusing of a lens.
- The entrance pupil diameter
2
depends on the size
of the aperture (controlled by the aperture setting of the
lens) and the lens optics that image the aperture. Among
other things, the aperture, and hence the entrance pupil,
affect the focus and the brightness of the image.
- The focal length f of the lens-the focal length is a
fundamental property of a lens and it provides a measure
of its refractive ability. In cases where it can be varied
(e.g., zoom lenses) the focal length is a sensor parameter;
otherwise, it is a sensor constant. Other optical camera
parameters that can also be controlled in some cases [51]
include the following:
1) The exposure time of the sensor, which determines
how long the sensor will be exposed to light from the
scene;
2) The gain of the video signal by the camera
amplifier;
3) Parameters adjusting the spectral responsivity of
the imaging system (e.g., the spectral transmittance of
filter) .The illumination parameters are also of two types:
geometric
and radiometric. The geometric parameters are
independent of any of the physical characteristics of the
illumination. Parameters of this type include the
following:

1
The principal points of an optical system are two
conjugate points on the optical axis such that planes
drawn perpendicular to the optical axis through these
points are planes of unit magnification. For the case of a
thin lens camera model, the principal points coincide at a
single point.
2
The entrance pupil of a lens is the image of its
aperture with respect to all elements preceding it in the
lens system.

- The three positional degrees of freedom of the
illuminator-this parameter is similarly the position vector
of a point associated with the illuminator. For example, in
the case of a point light source, this parameter can
obviously be the location of the point light source itself.
- The three orientational degrees of freedom of the
illuminator in the case of directional illuminational-this
parameter can be specified by the pan, tilt, and swing
angles of a vector rigidly fixed to the illuminator. An
example of a convenient vector of this type is the unit
vector along the axis of symmetry of the illumination
beam. The geometric characteristics of the illumination
beam for example, the illumination beam may have the
shape of a conical solid angle with a constant or variable
apex angle.
- The radiometric parameters depend on the
physical and optical characteristics of the illumination.
Parameters of this type include the following:

- The radiant intensity-this parameter describes the
power output of the light source.
- The spatial distribution of intensity-for example,
the above parameter is the most important characteristic
of the illumination in the case of structured lighting. The
spectral distribution of intensity-this parameter can be
varied by employing spectral filters with a certain spectral
transmittance.
- Parameters describing the polarization of the
illumination- the polarization state of the illumination can
be controlled by employing polarizers.
Thus, in the most general, planning of camera, lens,
and illumination parameters is to be done in a high-
dimensional imaging space [50]. However, only a subset
of the above set of sensor and illumination parameters is
typically considered in any of the existing sensor planning
systems. For example, in many systems the camera is
assumed to be located at a fixed distance away from the
object or the optical axis is assumed to pass through a
fixed point on the object.
2) Sensor and Object Models: The sensor models,
camera, lens, and illuminator, embody information that
characterizes the operation of the sensor. Such
information includes thefollowing: The sensor and
illumination parameters and constants- for example, the
field-of-view angle of the sensor, the sensor pixel size, the
lens focal length, the sensor noise, the spatial extent of
illumination, its intensity distribution within this extent,
etc. The governing relationships between the sensor
parameters and the sensor constants-for example, the
perspective projection imaging model, the lens formula
for focusing, etc. Object models that are commonly used
(e.g., CAD/CAM models) contain geometric and
topological information, which may be sufficient for some
aspects of the sensor planning problem. However, the
photometric properties of the object, such as color and
reflectivity of its surface, are needed for illumination
planning in order to accurately model the interaction
between light and the object surface and then approximate
the resulting image. In this case, currently used object
models need to be augmented to include such
information.
3) Feature Detectability Constraints: The feature
detectability constraints discussed in this section are fairly
generic to most vision tasks. An initial set of such
constraints for sensor
planning was introduced by Cowan et al. in [19], [21].
Similar to the way sensor parameters have been classified
in Section IIB. 1, the feature detectability constraints can
also be collected into two groups depending on whether
the illumination source plays a role in the constraint or
not.
The first group consists of the purely sensor
constraints for which illumination is not a factor. Such
constraints include the following:
- Visibility-for a feature to be detectable by the
sensor, it must first be visible to the sensor. This means
that all lines of sight from the camera to each point of the
feature are not obstructed (i.e., occluded) by anything in
the environment.
- Field of view-while the visibility constraint
requires that the rays of light from the feature reach the
sensor, the field-of-view constraint requires that these
rays must then image the features onto the active area of
the sensor. If this does not happen, the image of the
features will either be clipped by, or will lie outside, the
active sensor area and thus will not be observable.
- Focus-the goal of this constraint is to guarantee
that the features are in focus. While there is only one
object distance for which points are in perfect focus, there
is a tolerance in position for which a feature is still
considered acceptably focused based on the resolution of
the image sensor. This tolerance is referred to as depth of
field. Thus the focus constraint requires that the features
of interest lie within the depth of field of the lens.
- Magnification or pixel resolution-this constraint
governs the size of the feature images. Often, a linear
feature is required to appear in the image with a certain
length (e.g., 5 pixels). This constraint is referred to as the
magnification constraint. A related constraint is that of
resolution for which any two points of the feature must be
imaged by distinct pixels on the sensor plane.
- Perspective distortion-in many applications (e.g.,
graphics) the images of a scene are considered
undesirable when the scene is severely distorted under the
projection. In such applications a constraint that
minimizes the perspective distortion of features is
appropriate.
The second group of feature detectability constraints
depends on the illumination. These constraints are
referred to as illumination or radiometric constraints and
include the following:
- Illuminability-for a feature to be detectable by
the sensor, it is not sufficient that it be visible to the
sensor alone. It is also necessary that the feature be visible
to at least some point of the light source as well. If this is
not the case, the feature will not be illuminated and as a
result, it cannot be detected. If a feature point is visible to
only some but not all points of the light source, it will be
illuminated by these points and will either be back-facing
or in shadow for the remaining part of the light source. A
related constraint is shadow avoidance. For example, such
a constraint is important in photometric stereo and-in
cases where shadows cast over the feature-may be
mistakenly interpreted as the feature itself.
- Dynamic range of the sensor-an object point will
also not be detectable in the image when the image
irradiance from that object point is outside the dynamic
range of the

sensor. For instance, the image irradiance from
an object point may be too weak to
sensitize the corresponding photoreceptor cells
and as a result will appear black in
the image, On the other hand, the image
irradiance from an object point may be too high
and thus may saturate the photoreceptor area.
In both cases, the object point will be
considered undetectable.
- Contrast-detectability of an edge feature in a
scene is often determined by the edge contrast in the
image, that is, the disparity in image intensity values at
points in the neighborhood of the image of the edge. For
example, anedge feature may not be detected
in the image when the contrast between its
adjacent faces is not sufficient for the edge
operator at hand.
The above constraints can be further classified as
geometric and optical constraints. Geometric constraints
depend only on the geometric sensor and illumination
parameters. Examples
of such constraints are visibility and illuminability,
where these depend only on the geometric parameter of
position. Optical constraints depend on both the geometric
and the optical parameters. Examples of such sensor
constraints include resolution, focus, field of view,
dynamic range, and contrast.
It is important to note that when the sensor and
radiometric constraints are formulated in terms of the
sensor and illumination parameters, the sensor constraints
involve only sensor parameters, while in general the
radiometric constraints include both sensor and
illumination parameters. For example, the sensor
parameter of aperture is involved in both the sensor
constraint of focus and the radiometric constraint
regarding the dynamic range of the sensor. This coupling
of sensor and illumination parameters can also be seen in
the case in which the reflectance model of the object
surface includes a specular component or when lens
collection is taken into account. In both cases, the sensor
location is coupled into the image irradiance equation.
This coupling results from the bidirectional reflectance
distribution function in the former case, and in the latter
from the fourth power law for the cosine of the off-axis
angle [30]. There are cases in which such couplings can
be neglected. For example, in the case of a diffuse object,
the scene radiance is the same irrespective of the camera
position. Similarly, if the field of view is narrow, the
dependence of the image irradiance on the off-axis angle
is negligible. In general, however, the sensor and
illumination planning problems are not separable. See
[37] for more.

ACKNOWLEDGEMENT

I thank my Ph.D Student Ms G.K.Umamaheswari for
several useful discussions and for the preparation of this
expository article.

REFERENCES
[1]. A. L. Abbott and N. Ahuja, Active surface reconstruction by
integrating focus, vergence, stereo, and camera calibration, in Znt.
Proc. Conf.Comput Vision, 1990, pp. 489-492.
[2]. N. Ahuja and J. Veenstra, Generating xtrees from object silhouettes
in orthographic views, IEEE Trans. Putt. Anal. Mach. Intell., vol.
11,pp. 137-149, Feb. 1989.
[3] J. Aloimonos, I. Weiss, and A. Bandyopadhyay, Active vision, in
Proc. Conf Comput. Vision, 1987, pp. 35-54.
[4]. D. P. Anderson, An orientation method for central projection
programs, Comput.& Graphics, vol. 6, no. I , pp. 35-37, 1982.
[5] D. P. Anderson, Efficient algorithms for automatic viewer
orientation, Comput. & Graphics, vol. 9, no. 4, pp. 407-413, 1985.
[6]. R. Bajcsy, Active perception, Proc. IEEE, vol. 76, no. 8,
pp.9961005, Aug. 1988.
[7]. B. G. Batchelor, A Prolog lighting advisor, in Proc. SPZE Intell.
Robotsand Vlll: Syst. and Applicat., vol. 1193, pp. 295-302, 1989.
[ 8]. A. K. Bejczy, W. S. Kim, and S. C. Venema, The Phantom robot:
Predictive displays for tele-operation with time delay, in Proc. 1990
IEEE Int. Conf Robotics and Automat., 1990, pp. 546-551.
[9 ]. N.L.Biggs, E.K.Lloyd, and R.J.Wilson, Graph theory:1736-1936,
Oxford, U.K.: Oxford Univ.Press.1976.
[10]. R. C. Bolles and R. A. Cain, Recognizing and locating partially
visible objects: The local-feature-focus method, Znt. J. Robot. Res.,
vol. I , no.3, pp. 57-82, 1982.
[11]. R. C. Bolles and P. Horaud, 3DPO: A three-dimensional part
orientation system, Int. J. Robot. Res., vol. 5, no. 3, pp. 3-26, 1986.
[12]. A. Califano, R. Kjeldsen, and R. M. Bolle, Data and model driven
foveation, in Proc. 10th Int. Conf Putt. Recognition, 1990, pp 1-7.
[13]. A. Cameron and H. Durrant-Whyte, A Bayesian approach to
optimal sensor placement, Rept. OUEL 1759/89, Department of
Engineering Science. Oxford Universitv. Oxford. A subtle, but
important problem in this area is that of UK. 1989.
[14]. A. Cameron and H. L. Wu, Identifying and localizing electrical
components: A case study of adaptive goal-directed sensing, Tech.
Note TN-90-085, Philips Laboratories, Briarcliff, NY, 1990.
[15]. 0. I. Camps, L. G. Shapiro, and R. M. Haralick, PREMIO: An
overview, in Proc. IEEE Workshop on Directions in Automat.
CADBased Vision, 1991, pp. 11-21,
[16]. L. L. Chin and T. C. Woo, Computational geometry on the sphere
with application to automated machining, Tech. Rept. TR 89-30,
Industrial and Operations Engineering, Univ. of Michigan, Ann Arbor,
MI, Aug. 1989.
[17]. J. J. Clark and N. J. Fenier, Modal control of an attentive vision
system, in Int. Con$ Comput. Vision, 1988, pp. 514-523.
[18]. C. I. Connolly, The determination of next best views, in Proc.
1985 IEEE Int. Conf Robotics and Automat., 1985, pp. 432-435.
[19]. C. K. Cowan, Model based synthesis of sensor location, in Proc.
1988 IEEE Int. Con$ Robotics and Automat., 1988, pp. 900-905.
[20]. C. K. Cowan and A. Bergman, Determining the camera and light
source location for a visual task, in Proc. IEEE Int. Conf Robotics and
Automat., pp. 509-14, 1989.
[21]. C. K. Cowan and P. D. Kovesi, Automatic sensor placement from
vision task requirements, IEEE Trans. Putt. Anal. Mach. Intell., vol. 10,
pp. 407-16, May 1988.
[22]. R. I. D. Cowie, The viewers place in the theories of vision, in
Int. Joint Cant Artificial Intell., 1983, pp. 952-958.
[23]. L.Euler, Solution Problematis ad Geometrian Situs Pertinentis.
Commentric Academiae Scientiarum Imperialis Petrolitanae, Vol-8,
128-140,1736.

[24] C. Goad, Special purpose, automatic programming for 3D model-
based vision, in Proc. DARPA Image Understanding Workshop, 1983,
pp.94-104.
[25].J. L. Gordillo and A. LUX, Synthesizing vision programs from
robot task specifications, in 3rd In?. Symp. Robotics Res., 1985, pp.
149-154.
[26]. W. E. L. Grimson, Sensing strategies for disambiguating among
multiple objects in known poses, ZEEE J. Robot. Automat., vol. RA-2,
no. 4, Dec. 1986.
[27] G. D. Hager, Task Directed Sensor Fusion and Planning. Norwell,
MA: Kluwer Academic Publishers, 1990.
[28]. C. Hansen and T. Henderson, CAGD-based computer vision, in
Proc.IEEE Workshop on Computer Vision, 1987, pp. 100-105.
[29]. T. H. Hong and M. 0. Schneier, Describing a robots workspace
using a sequence of views from a moving camera, IEEE Trans. Pan.
Anal. Mach. Intell., vol. 7, pp. 721-726, Nov. 1985.
[30] B. K. P. Hom, Robot vision.New York: McGraw-Hill, 1986.
[29]. M. Huck, J. Raczkowsky, and K. Weller, Sensor simulation in
robot applications, in Proc. Advanced Robotics Program, Workshop on
Manipulators, Sensors and Steps Towards Mobiliry 1987, pp. 197-209.
[31]. S. A. Hutchinson and A. C. Kak, Planning sensing strategies in
robot work cell with multi-sensor capabilities, IEEE Trans. Robot.
Automat., vol. 5, no. 6, pp. 765-783, Dec. 1989.
[32]. K. Ikeuchi, Generating an interpretation tree from a CAD model
for 3D-object recognition in bin picking tasks, Int. J. Comput. Vision,
vol.I , no. 2, pp. 145-165, 1987.
[34]. K. Ikeuchi and J. C. Robert, Modeling sensors detectability with
the VANTAGE geometrickensor modeler, IEEE Trans. Robot.
Automat., vol. 7, pp. 771-784, Dec. 1991.
[35]. J. R. Kender and D. G. Freudenstein, What is a degenerate
view? in Proc. 1987 DARPA Image Understanding Workshop, 1987,
pp. 589-598.
[36]. H. S. Kim, R. C. Jain, and R. A. Volz, Object recognition using
multiple views, in Proc. 1985 IEEE Int. Con$ Robotics and Automat.,
1985, pp.28-33.
[37]. Konstantinos A. Tarabanis, Peter K. Allen, Member, IEEE, Roger
Y. Tsai A Survey Of Sensor Planning In Computer Vision IEEE
Transactions On Robotics And Automation, Vol 11, No 1. February
1995.
[38]. Y. Kitamura, H. Sato, and H. Tamura, An expert system for
industrial machine vision, in Proc. 10th Int. Con$ Patt. Recognition,
1990, pp.771-773.
[39]. L. Lieberman, Model-driven vision for industrial automation, in
Advances in Digital Image Processing, Stucki, Ed. New York: Plenum,
1979, pp. 235-248.
[40]. M. Magee and M. Nathan, Spatial reasoning, sensor repositioning
and disambiguation in 3d model based recognition, in Proc. Workshop
on Spatial Reasoning and Multi-Sensor Fusion, 1987.
[41]. W. N. Martin and J. K. Aggarwal, Volumetric description of
objects from multiple views, IEEE Trans. Patt. Anal. Mach. Intell., vol.
5, pp. 15&158, Mar. 1983.
[42]. J. Maver and R. Bajcsy, How to decide from the first view where
to look next, in Proc. 1990 DARPA Image Understanding
Workshop,1990, pp. 482-496.
[43]. J. L. Mundy, Industrial machine vision-Is it practical? in
Machine Vision, H. Freeman, Ed. San Diego, CA: Academic, 1988.
[44]. A. Novini, Lighting and optics expert system for machine vision,
in Proc. Optics, Illumination, Image Sensing, pp. 1005-1019, 1988.
[45]. J. Raczkowsky and K. H. Mittenbuehler, Simulation of cameras in
robot applications, Comput. Graphics Applicat., pp. 16-25, Jan. 1989.
[46].S. Sakane, M. Ishii, and M. Kakikura, Occlusion avoidance of
visual sensors based on a hand eye action simulator system: HEAVEN,
Adv. Robot., vol. 2, no. 2, pp. 149-165, 1987.
[47]. S. Sakane, R. Niepold, T. Sato, and Y. Shirai, Illumination setup
planning for a hand-eye system based on an environmental model, Adv.
Robot., vol. 6, no. 4, pp. 461482, 1992.
[48]. S. Sakane, T. Sato, and M. Kakikura, Model-based planning of
visual sensors using a hand-eye action simulator system, in 3rd Int.
Con$ Advanced Robotics (ICAR), 1987.
[49].D. D. Seligmann and S. Feiner, Automated generation of intent-
based 3D illustrations, Comput. Graphics, vol. 25, no. 3, July 1991.
[50]. S. A. Shafer, Automation and calibration for robot vision
systems,Tech. Rep. CMU-CS-88-147, Camegie Mellon University,
Pittsburgh, PA, May 1988.
[51].A. Shmuel and M. Werman, Active Vision: 3D from an image
sequence, in Proc. 10th Int Conf Patt. Recognition, 1990, pp. 48-54.
[52]. A. J. Spyridi and A. G. Requicha, Accessibility analysis for the
automatic inspection of mechanical parts by coordinate measuring
machines, in Pmc. 1990 IEEE Int. Ciwf Robotics and Automat., 1990
[53]. M. J. Swain and M. Stricker, Promising directions in active
vision,Tech. RePt. Cs 91-27, University of Chicago, Chicago, IL, Nov.
1991.
[54]. K. Tang, T. C. Woo, and J. Gan, Maximum intersection of
spherical polygons and workpiece orientation for 4- and 5-axis
machining, Industrial and Operations Engineering, Univ. of Michigan,
Ann Arbor, MI, 1990.
[55] K. Tarabanis and R. Y. Tasi, Computing viewpoints that satisfy
optical constraints, in Proc. CVPR 91: The Comput. Soc. Con$ on
Comput.Vision and Patt. Recognition, 1991.
[56]. K. Tarabanis, R. Y. Tsai, and S. Abrams, Planning viewpoints
thatsimultaneously satisfy several feature detectability constraints for
robotic vision, in 5th Int. Con$ Advanced Robotics (ICAR), 1991.
[57]. K. Tarabanis, R. Y. Tsai, and P. K. Allen, Automated sensor
planning for robotic vision tasks, in Proc. 1991 IEEE Int. Conf
Robotics and Automat., 1991.
[58]. R. Y. Tsai and M. A. Lavin, Three-dimensional mechanical part
measurement using a visiodrobot, system, Tech. Rept. RC 10506, IBM
T.J. Watson Research Center, Yorktown Heights, NY, May 1984.
[59]. P. Whaite and F. P. Feme, From uncertainty to visual
exploration, in Proc. Int. Con$ Comput. Vision, 1990, pp. 69 M97.
[60]. W. J. Wolfe, D. W. Mathis, M. Magee, and W. Hoff, Task panel
sensing with a movable camera, in Proc. SPIE Intell. Robots and
Comput. VisionVIII: Syst. and Applicat., vol. 1193, 1989.
[61]. K. H. Womack, Front-end foresight, MVNSMEs Vision
Technol. Quarter., p. 4, Jan. 1990.
[62] S. Xie, T. W. Calvert, and B. K. Bhattacharaya, Planning views for
the incremental construction of body models, in Proc. 8th Int. Con$
Patt. Recognition, 1986, pp. 154-157.

India.
ISBN: 978-81-920575-9-0 :: doi: 10. 72962/ISBN_0768
Simulation Of Spread Spectrum Modulation Techniques

Trima Piedade Fernandes e Fizardo
Assistant Professor
Electronics and Telecommunication Engineering
Don Bosco College Of Engineering
Fatorda-Goa

Abstract- In any communication system, transmitted power
and channel bandwidth are of considerable importance. The
task is to utilize these considerations effectively. In this paper,
simulation as well as comparison of two spread spectrum
modulation techniques, Direct Sequence Spread
Spectrum(DSSS) as well as Frequency Hop Spread
Spectrum(FHSS) are put forth. The simulation process is
done using Matlab simulator.
Keywords-DSSS, FHSS,Spectrum
I. INTRODUCTION
Over the last eight or nine years a new commercial
marketplace has been emerging. Called spread spectrum,
this field covers the art of secure digital communications
that is now being exploited for commercial and industrial
purposes. In the next several years hardly anyone will
escape being involved, in some way, with spread spectrum
communications.
How will you define a system to be a spread spectrum
system[4]? Spread spectrum is a transmission technique in
which pseudo noise code independent of information data
is employed to spread the signal energy over the bandwidth
much greater than the signal information bandwidth. At the
receiver the signal is despread using the synchronized
replica of the pseudo noise code. A spread spectrum
system should fulfill the following requirements[3]:
The modulated signal occupies a bandwidth much in
excess of the minimum bandwidth necessary to send
the data.
The spectrum spreading is accomplished by means of
wideband spread signal, often called a code signal,
which is independent of data.
At the receiver, despreading is accomplished using the
same code sequence operating in synchronism with the
transmitter.
Standard modulation techniques like frequency
modulation and pulse code modulation also spread the
spectrum of an information bearing signal, but they do not
qualify as spread spectrum systems because they do not
satisfy all conditions stipulated above.
Spread spectrum modulation was originally developed
to meet military needs or to have a reliable communication
in a hostile jamming environment. It provides multipath
rejection in a mobile radio system and has asynchronous
multiple access capability and hence very useful in cellular
telephony,radio-paging,wireless local area networks etc.

Spread Spectrum uses wide band, noise-like signals.
Because Spread Spectrum signals are noise-like, they are
hard to detect. Spread Spectrum signals are also hard to
Intercept or demodulate. Further, Spread Spectrum signals
are harder to jam (interfere with) than narrowband signals.
These Low Probability of Intercept (LPI) and anti-jam (AJ)
features are why the military has used Spread Spectrum for
so many years.
Spread signals are intentionally made to be much wider
band than the information they are carrying to make them
more noise-like. Spread Spectrum signals use fast codes
that run many times the information bandwidth or data
rate. These special "Spreading" codes are called "Pseudo
Random" or "Pseudo Noise" codes. They are called
"Pseudo" because they are not real Gaussian noise[2].
Spread Spectrum transmitters uses similar transmit
power levels to narrow band transmitters. Because Spread
Spectrum signals are so wide, they transmit at a much
lower spectral power density, measured in Watts per Hertz,
than narrowband transmitters. This lower transmitted
power density characteristic gives spread signals a big
plus. Spread and narrow band signals can occupy the same
band, with little or no interference. This capability is the
main reason for all the interest in Spread Spectrum today.

There are two types of spread spectrum techniques:
Direct Sequence spread spectrum
Frequency hop spread spectrum

II. DIRECT SEQUENCE SPREAD SPECTRUM
Direct sequence spread spectrum, also known as direct
sequence code division multiple access (DS-CDMA), is
one of two approaches to spread spectrum modulation for
digital signal transmission over the airwaves. In direct
sequence spread spectrum, the stream of information to be
transmitted is divided into small pieces, each of which is
allocated across to a frequency channel across the
spectrum. A data signal at the point of transmission is
combined with a higher data-rate bit sequence (also known
as a chipping code) that divides the data according to a
spreading ratio. The redundant chipping code helps the
signal resist interference and also enables the original data
to be recovered if data bits are damaged during
transmission[3].
The DSSS process is performed by effectively
multiplying an RF carrier and a pseudo-noise (PN) digital
signal. First the PN code is modulated onto the
information signal using one of several modulation
techniques (eg. BPSK,). Then, a doubly balanced mixer is
used to multiply the RF carrier and PN modulated
information signal. This process causes the RF signal to be
replaced with a very wide bandwidth signal with the
spectral equivalent of a noise signal.

Figure 1. DSSS Spreading

Input:
Binary data d(t) with symbol rate Rs=1/Ts
(=bitrate Rb for BPSK
Pseudo-noise code Pn(t) with chiprate
Rc=1/Tc(an integer of Rs)

Spreading :
In the transmitter[1], the binary data d(t) (for BPSK) is
directly multiplied with the PN sequence Pn(t),which is
independent of binary data, to produce the transmitted
baseband signal Txb
Txb= d(t) * Pn(t)

The effect of multiplication of d(t) with a PN sequence
is to spread the baseband Rs of d(t) to a baseband
bandwidth of Rc.
Spread spectrum systems are spreading the information
signal d(t) which has a BWi,over a much larger bandwidth
BWss:
BWi Rs BWss Rc

The Spread Spectrum signal is white noise like. The
amplitude and thus the power in the Spread spectrum
signal Txb is the same as in original information signal
d(t).The power spectral density of spread spectrum signal
is lower due to the increased bandwidth.
The bandwidth expansion factor is the ratio of chip rate
Rc and data symbol rate Rs.
SF=(BWss/BWi)=Rc/Rs=Tb/Tc=Nc

III. SIMULATION OF DIRECT SEQUENCE SPREAD
SPECTRUM
By using Matlab version 7 simulator, simulation of
Direct Sequence Spread Spectrum is done and is shown as
follows:

Figure 2. Original bit sequence

Figure 3. Pseudorandom bit sequence

Figure 4. DSSS Signal

Input
data
d(t)
PN
Code
Pn(t)
Tx b
Modulator
RF

Figure 5.Fast Fourier transform of BPSK Signal

IV. FREQUENCY HOP SPREAD SPECTRUM
In frequency hop spread spectrum communication,the
binary data sequence modulates the frequency of the
carrier signal resulting in an FSK signal,or groups of K
binary bits and and a set of 2^K subcarrier frequencies are
chosen to generate a multilevel FSK or M-ary
FSK(MFSK)signal.The MFSK signal finally modulates a
carrier hops randomly from one frequency to another is
called frequency hope spread spectrum(FHSS)[3].
By choosing a large number of randomly hopping
discrete carrier frequencies, it is possible to have a
modulated signal of wide bandwidth, which in turn results
in large processing gain. The spectrum spreading in case of
frequency hop spread spectrum is sequencial.Frequency
Hop Spread Spectrum is divided into 2 types, depending on
the rate at which the carrier frequency hops from one value
to another.
The carrier frequency changes every Th seconds.
Therefore
Rh=1/th
Rh is the hopping rate. If Ts and Tb represent symbol
and bit durations of MFSK and FSK signals respectively,
then Rs=1/Ts,is called the symbol rate of MFSK system
and Rb=1/Tb is called bit rate of FSK system.

Figure 6. Block Diagram of FHSS

V. SIMULATION OF FREQUENCY HOP SPREAD SPECTRUM

Figure 7.Original bit sequence and BPSK modulated signal

Figure 7.FFT of BPSK modulated signal

Figure 8.Frequency Hope to form a spread signal

Carrier
+ Data
Frequency Synthesiser
PN Code Generator

Figure 9: Transformation of BPSK signal into wider range

VI. ADVANTAGES OF SPREAD SPECTRUM
Secure communication privacy
Mitigation of multipath, hold only direct path
Protection to intentional interference(Jamming)
Rejection of unintentional interference(Narrow
band
Low probability of detection and interception
Availability of license free
ISM(Industrial,Scientific and Medical)frequency bands

VII. DISADVANTAGES OF SPREAD SPECTRUM
Increased bandwidth(frequency Usage)
Increased computational load

VIII. COMPARISON
Direct Sequence Spread Spectrum:
Short latency time
No re-synch with other radio necessary
Constant Processing gain
Long outdoor range(40 m)
Greater overall data throughput

Frequency Hop Spread Spectrum:
Long latency time
Must re-synch with other radio necessary
No processing gain
Short outdoor range(10 m)
Lower overall data throughput

IX. CONCLUSION
In todays world, spread spectrum plays an important
role in various applications such as security applications
military applications etc.In this paper, the main concept of
spread spectrum has been put forth.The various spread
spectrum modulation techniques like direct sequence
spread spectrum and frequency hop spread spectrum has
been explained as well as simulation has been performed
using Matlab software simulator. The comparison of both
the modulation spread spectrum techniques is done.

REFERENCES

[1]J.G. Prokais, Communication Systems Engineering,Prentice
Hall,1994,chap 11
[2] M.K. Simon Spread Spectrum Communications Handbook,Mc-
Graw-Hill,1994
[3]Dr.K.N.Hari Bhat,Dr.D.Ganesh Rao, Digital
Commuincations,Sanguine Technical Publications,2006
[4]Randy Roberts, Introduction to Spread Spectrum
[5]Kevin M. Coumo,Alan V.Oppenheim,Steven H.Isabelle, Spread
Spectrum Modulation and Signal Masking using Chaotic Signals,1992

India.
ISBN: 978-81-920575-9-0 :: doi: 10. 72969/ISBN_0768
Brain Computer Interface an innovative and emerging engineering technique of
speechless thought recognition
Mr. Gaurav Vishnu Londhe
Department of Electronics Engineering,
Datta Meghe College of Engineering
Airoli, Navi Mumbai. Maharashtra,India.

Abstract The BrainComputer Interface (BCI) research
community has acknowledged that researchers are
experiencing difficulties when they try to compare the BCI
techniques described in the literature. In response to this
situation, the community has stressed the need for objective
methods to compare BCI technologies. Suggested
improvements have included the development and use of
benchmark applications and standard data sets. However,
lacks a common vocabulary. As a result, this deficiency leads
to poor intergroup communication ,which hinders the
development of the desired methods of comparison. One of
the principle reasons for the lack of common vocabulary is
the absence a common functional model of a BCI System. BCI
transforms brain activity to control commands, constructing
a direct communication path between human and external
devices. People with severe motor disabilities can benefit a lot
from the technology. BCI research is an intelligent
information processing technique incorporating multiple
disciplines knowledge like neural science, signal
processing, machine learning and intelligent control, etc. Non-
invasive scalp electroencephalogram (EEG) is widely used in
BCI researches because of the advantages of safety and easy
operation. In EEG-based BCI systems, event-related
potentials, event-related synchronization and de
synchronization, slow cortical potentials, and visual evoked
potentials (VEPs) are commonly used systems. In the past few
years, the traditional power spectrum estimation method has
played important roles in SSVEP-based BCI researches.
However, the Fourier transformation(FT)-based methods use
sinusoids and cosines to analyze all waveforms, in spite of the
peculiarities of the shapes of those signals. Moreover, FT
method is originally designed for stationary signals. As
frequency field analysis technology, the FT breaks signals
down into a series of component frequencies, eliminating all
time information. So wavelet transform, on the other hand, is
capable of adjusting the window size of its wavelet functions
and providing a flexible way. The prominent property of
wavelet transform is that it provides an effective way to
analyze non stationary signals through decomposing signals
into a set of time-scaled and time-shifted versions of wavelet
coefficients. EEG signals are non stationary whose frequency
components vary as a function of time. The analysis of such
signals can be facilitated by wavelet transform which provide
flexible timefrequency resolution. Wavelet transform has
widely been employed for EEG signals analysis in recent
years. Another advantage of wavelet analysis over Fourier
techniques is that it is easy to choose different mother
wavelet functions to analyze different types of signals. Proper
selection of mother wavelet is thus important to obtain good
performance in analyzing EEG signal using CWT method.

Keywords- Brain computer interface a speechless thought
recognition system
I. INTRODUCTION (BRAIN COMPUTER INTERFACE)
A brain-computer interface (BCI) is intended to help
disabled subjects gain control over their environment with
the use of their brain activity. A computer maps this
electrical activity to functions the subject is in need of. It
collects the data from an EEG amplifier and by using
signal processing techniques, analyzes and makes a
decision of what to do with the data.
A BCI is most useful at helping disabled subjects
make choices about things such as their needs (medication,
nurse, pain etc), channels in a TV remote, answers to basic
YES /NO questions, or maybe letters of the alphabet. In
this paper, the performance of a BCI application, a P300
speller is tackled. P300 is an event related potential (ERP)
that occurs in brain signals when the subject is exposed to
visual or auditory stimulation. The P300 speller paradigm
we use was first introduced by Farwell and Donchin in [1].
They reported their results on 4 healthy subjects, with a
rate of 2.3 letters per minute with 95% accuracy.
Since then, various aspects of this paradigm are tackled
to increase performance; electrode selection, stimulus
shape, timings, and presentation, data sampling, feature
extraction, filtering, classifier algorithm and other
processing procedures. Donchin et al. increased this rate up
to 4.3 letters/min with 95% accuracy [2]. Meinicke
increased this rate up to 5.5 letters/min with above 90%
accuracy [3]. Kaper reported a rate of 47.26 bits/min [4]
and Serby [5] reported 5.45 letters/min and 23.77 bits/min.
These results are summarized in Table II, along with a re-
interpretation with respect to this work. We have
developed a new end-to-end P300-based realtime BCI
system to explore possibilities to increase this speed. Our
system offers flexible stimulus mechanisms and evaluation
possibilities assessable via user defined preferences. We
have explored various choices in the classifier, in search of
better classification results. The speed letter/min. we
achieved in this study improves the
o
a
r
t
s
p
d
c

s
w
A
w
2
r
u
s

A
m
T

c
c
A
c
p
t
c
b
o
i
c
r
c
m
t
l
c
l
m
m

e
t
d
published
offline experim
and 48.4073
result is 11.1
types what he
simple guidel
paper look ex
do this is simp
content with y
A. HAR
All the exp
same hardwar
with a 64-chan
Faraday C
Active electro
with conducti
2048 Hz
and sent to
records the in
used for stimu
signals to the
B. SOFT
For offline
ActiView softw
modified versio
The classificatio
done in MA
C. STIM
Our flexib
cell content
coloring and s
Also, each fla
can be specif
presets to be u
the most we
characters, so
based on this
originally pro
incorporates le
columns of
randomized f
column is fla
ms, and after
the cells are h
stimulus l
letter, there sh
column, wher
letter. Offline
matrix. Onlin
matrix where

D. TERM
In the con
experimenter
training and t
data are recor
work so far. T
ments for 100
bits/min, whe
4 letters/min,
e/she wants. W
lines. In essen
xactly like this
ply to downlo
your own mate
RDWARE SET
periments in th
re setup. The
nnel Bio semi
Cage, void o
odes are utilize
ive gel. The
o a laptop wit
coming data t
ulation and is
amplifier duri
TWARE SETU
e analysis, the
ware, and for o
on of a MEX
on and other an
ATLAB and the
MULUS
ble stimulus s
customization
stimulation sc
ash duration a
fied. Overall,
used again lat
ell known sti
that they can
s stimulus. T
oposed by Fa
etters and num
the matrix
fashion; i.e.
ashed exactly
each one is a p
highlighted. Th
asts 300 ms.
hould be at lea
re the cell at t
analyses are
ne analyses a
each highlight
INOLOGY
ntext of this pa
has prior kno
test sets, and
rded. Online a
The average sp
0% accuracy is
ere in online
, making sure
We ask that au
nce, we ask y
s document. T
oad the templa
erial.
TUP
his study are c
data are recor
Active two E
f electromagn
ed, attached to
recorded dat
th a dual-core
o a hard disk.
responsible f
ing the experim
UP
e data are rec
online analysis
interface deve
nalyses are also
visual stimulus
system allows
n (letters or
chemes as disp
and ISI (inter-
these options
ter on. We pre
mulus type,
be compared
This stimulus
arwell and Do
mbers in each
are highligh
in 12 flashe
once. Each f
period of 175
herefore, each
Note that in
ast two flashes
the intersectio
done in the st
are done in
t is in a differe
aper, offline a
owledge on th
the analysis
analysis means
peed we achiev
s 9.363 letters
experiments,
e that the su
uthors follow s
you to make
The easiest wa
ate, and replac
conducted with
rded and digi
EEG amplifier
netic interfere
o the electrode
ta are digitize
e processor, w
The laptop is
for sending tri
ment.
corded in Bio
in MATLAB,
eloped byHoffm
s is developed in
s any matrix
shapes), diffe
played in Figu
-stimulus inte
s can be save
esent our resul
a 6x6 matri
to existing re
is a 6x6 m
onchin in [1]
cell. The rows
hted in a bl
es, each row
flash lasts for
ms where non
order to defi
s, one row and
on holds the ta
tandard grey/w
a random-col
ent color.
analysis mean
he letters for
is done after
s the experime
ve in
s/min
this
ubject
some
your
ay to
e the
h the
tized
in a
ence.
e cap
ed at
which
s also
igger
oSemi
via a
mann.
n C.
size,
ferent
ure1.
erval)
ed as
lts in
x of
esults
matrix
that
s and
lock-
and
r 125
ne of
ine a
d one
arget
white
lored
ns the
both
r raw
enter
dic
kn
pr
rea
W
all
Ac
tri
gr
of
gr
an
ru
is
tar
nu
8
tha
se
rec
re
ind
to
alo
pe
Fz
to
ge
inc
cla
in
oth
the
D
fea
by
ctates only the
nowledge of
roduces estima
al time. Each
With block rand
l the rows and
ccording to ti
ial group lasts
roups make up
ffline experime
Online expe
roups since th
nd stimulation
un, and the targ
a brief period
rget letter, an
umber of runs
runs (8 letters
at includes m
ssion and on
cordings of se
st and prepare
dication of th
the acquisi
ongside with r
eriod of record

E DATA AC
The electrod
z, Cz, Pz and
each masto
enerally ignor
cluded them
assification.
F. PRILIMIN
For offline a
a session gro
her, the test s
e training sess
D E D E D E
atured 8 runs
y the subject o
e letters for th
letters in th
ated letter and
flash of a ro
domization in
d columns fla
imings report
s for 3.6 s. A
p a run. In thi
ents.
eriments hav
hey depend on
n goes on with
get is the same
d where the us
nd then the ne
constitute a s
s) in a session
more than on
e test session
essions in a se
e for the next
he highlighted
ition device.
regular EEG d
ded data that in
Figure 1. D
CQUISITION
des used are F
Oz. Two refe
id channel. A
red due to e
in our analys
NARIES
analyses in thi
oup, one being
ession. Other
sion of each su
E D E as ta
and included
or the experim
he training set
he test set\ a
d displays it t
w or column
mind, 12 flas
ashing constitu
ed in the pre
A determined
is study, this n
ve variable nu
n classifier ou
hout any inter
e letter. After
ser is informe
ext run begin
ession. In this
n. A session gr
ne session (e
n). There are
ession group, t
session. A tri
row/column
Trigger dat
data. An epoch
ncludes a trial
Different stimuli
N
Fp1, Fp2, P3,
erence electro
Although Fp
eye-blink arti
sis to explore

s study, there
g the training
than a few m
ubject featured
argets. The te
random letter
menter beforeh
and has no pr
and the syst
to the subject
is called a tri
shes that inclu
ute a trial grou
evious section
number of tr
number is 10
umbers of tr
utput. Recordi
rruptions in ea
a run ends, th
ed about the n
s. A determin
s study, there
roup is a data
.g. one traini
breaks betwe
to let the subj
gger signal is
and is sent ov
ta are record
h is a determin
.
, P4, PO7, PO
odes are attach
1 and Fp2
ifacts, we ha
e their effect
are two sessio
g session and t
minor exceptio
d 8 runs that h
est sessions a
rs, chosen eith
and. Each epo
rior
em
t in
ial.
ude
up.
n, a
rial
for
rial
ing
ach
ere
ext
ned
are
aset
ing
een
ect
an
ver
ded
ned

O8,
hed
are
ave
on
ons
the
ons,
had
also
her
och
lasts for 1 second. The classifier is trained on the first
session and tested on the second.

G. DATA PREPROCESSING
Proper pre-processing is an important factor in
classification performance. We have conducted several
different pre-processing schemes and observed that no
scheme is best for all subjects. The definitive scheme used
in all offline analyses is similar to [6] and is as follows: To
get rid of irrelevant frequency components, the data are
filtered with a 6th order Butterworth band-pass filter with a
pass-band of 1 12 Hz. ActiView saves the data with
respect to the common-mode sense (CMS) electrode. To
obtain a greater SNR, the data are re-referenced to the
average of two mastoid channels. For better performance
of the classifier, the data should be normalized. But data
with peaks lose resolution when normalized; therefore the
data are first winsorized in a 10% frame, and zero-mean
normalization follows next. Lastly, the data are decimated
by 64. After decimation, each epoch is represented with 32
samples. The feature vector for each epoch is then the
concatenation of filtered data from each electrode, i.e. a
vector of 320 samples for 10 electrodes. We found out that
in general, subjects blink rarely during each run and Fp1
and Fp2 contribute positively to the classification
performance, especially when eye-blink artifacts are
removed by winsorization.We have observed that half of
the subjects performed better with normalization and
winsorization, and the other half performed better without
them. In offline analysis, the results are generated
according to the scheme the subject was best at. In online
analysis, normalization and winsorization are applied to all
subjects.
H. CLASSIFICATION
For the classification algorithm, we used Bayesian
Linear
Discriminant Analysis (BLDA), mentioned in [7], A
derivative of Fishers LDA, BLDA gives probabilistic
output of test data, incorporates feature selection based on
discriminative power and learns regularization
parameters
automatically from the training set. Averaging of
multiple trials is frequently used to increase the SNR of
P300 waves. Rather than using averaging, in our work, we
incorporate information from multiple trials by
probabilistic updates as new trial data are received. In
particular, BLDA calculates a score for each epoch of test
data, reflecting its similarity to the underlying classes.
Scores are added up in consecutive trial groups until a firm
separation between scores is present.
For offline analysis, the sum of scores are checked at
the
end of each trial group and the row and column with
the
maximum scores are selected as answers of
classification,
and are compared with actual targets to generate the
accuracy plots in Figure 2. Since actual targets are
unknown to the experimenter in online analyses, the
classifier has to decide by itself when to end each run. This
is done by using margins in scores. A safe margin is
determined and when the column and the row with the
highest scores have that margin between themselves and
the next best ones, the character at the intersection of these
two is presented as the decisive answer of the classifier.

II. NETWORK ARCHITECTURE
In this study, we use a network comprised of two layers
of LIF neurons, similar to the work of Liaw and Berger
The neurons across the two layers are fully connected to
each other, as shown in Fig. 3. The input layer consists of
sixteen neurons, each one receiving one of the wavelet
packet decomposed frequency sub-bands. One of the
output layer neurons is the inhibitory interneuron, which
sends negative feedback to all synapses. Each one of the
other output neurons indicates one class of the EEG signal
to be recognized. While the network is exposed to the EEG
signal, the output neurons fire action potentials. Each
neuron is supposed to response exclusively to its
determinate class of EEG signal (right hand movement, left
hand movement) however, in practice, the most spikes
each neuron does fire, the EEG class associated with that
neuron is recognized. The genetic algorithm is exploited to
regulate
the network parameters in order to maintain this
behavior. In the next section, we will give more
explanation on the 1 Network parameters being trained
under genetic algorithm.
Network Training By Genetic Algorithm
The genetic algorithm utilizes a fixed length vector,
called
chromosome, in order to represent a possible solution
for a given problem domain. In this study, chromosomes
are made up of real numbers, called genes, each number
corresponding to one of the model parameters. The GA
consists of a population of chromosomes - from which
possible solutions are picked up and two operations: (1)
parent selection, and (2) mutation.
The initial population consists of several randomly
chosen
vectors. The fitness (a measure of appropriateness of
the solution) of each chromosome is evaluated by a fitness
function. The parent selection operator selects
chromosomes from the population based on their fitness. A
chromosome with higher level of fitness has more chance
to be chosen. After parent selection, each selected
chromosome (parent) is altered through some deterministic
and stochastic a result, a population of parents and children
is created, on which the parent selection operation is
performed again. The process continues cyclically until a
stop criterion is reached.

o
I
U
d
d
c
b
o
c
c
b
q
o
a
(
s
c
c
c
p
a
(
C
b
w
s
t
u
t
p
d
w
5
p
i
A
w
n
p
Experimen
In the fo
operated on D
Informatics,
University of
dataset was re
during a feed
chair with arm
by means of
order of left
consists of 7
conducted on
between. Give
quite, at t=2s
of the trial, th
and a cross +
(left or right)
subject was a
cue, The fee
channel #1 (C
combined wit
parameter. Th
and a Ag/Ag
(anterior +, p
C4. The EEG
between 0.5 a
were random
systematic eff
tree first secon
usable inform
these second
performance
dimension of
with omitting
512 samples.
preprocessor.
interneuron, w
After decomp
we feed these
network. And
parameters of

nts On Eeg Da
ollowing expe
Data set provi
Institute fo
f Technology
ecorded from
dback session.
mrests. The tas
imagery left
and right cue
7 runs with 4
the same day
en are 280 tria
an acoustic s
he trigger chan
+ was display
was displaye
asked to move
edback was b
C3) and #3 (C
th a discrimi
he recording w
gCl electrodes
posterior -) w
G was sample
and 30Hz. Th
mly selected.
fect due to th
nds and two la
mation about
ds of data b
of classific
data the spee
unusable data
Fig. 2. Arch
The black
which affects
position of the
e sub bands in
d then train t
f the network.
ata
eriments, thes
ided by Depar
or Biomedic
Graz. (Gert P
a normal sub
. The subject
sk was to cont
or right hand
es was random
40 trials each
y with several
als of 9s length
stimulus indica
nnel (#4) went
yed for 1s; then
ed as cue. At
e a bar into th
based on AA
C4), the AAR
inant analysis
wasmade using
s. Three bipol
were measure
ed with 128H
he trials for tr
. This shou
e feedback [1
ast seconds of
control inten
because they
cation and w
ed of network
a the samples
hitecture of D
node indicat
all of the syn
e EEG signal
nto the 16 inp
the network
se methods
rtment of Me
cal Enginee
Pfurtscheller).
bject (female,
t sat in a rela
trol a feedback
d movements.
m The experim
h. All runs
l minutes brea
h. The first 2s
ates the begin
t from low to h
n at t=3s, an a
the same time
he direction o
AR parameter
R parameters
s into one ou
g a G.tec amp
lar EEG chan
ed over C3, Cz
Hz, it was filt
raining and te
uld prevent
10].As you see
f datadoes not
d and so we
dont affect
with reduce
k will improve
of data reduc
SNN and wa
tes the inhib
naptic connect
into 16 sub b
put neurons o
for tuning of
were
dical
ering,
This
25y)
axing
k bar
The
ment
were
ak in
s was
nning
high,
arrow
e the
f the
rs of
were
utput
lifier
nnels
z and
tered
sting
any
e the
have
eomit
t the
the
e. So
ces to
avelet
bitory
tions.
bands
f the
f the
no
con
a
of
im
nu
ind
ax
su
19
co
is
ele
an
as
let
bit
62
pr
an
P3
re
ep
us
If
de
an
of
pe
cla
va
Fig. 2. Archite
de indicates the in
nnections.
A. Offline
The offline a
format compa
ffline classific
mproves upon
umber of tria
dicate the per
xis and dashe
ubjects took pa
9 and 26. No
omputation of
the accuracy
ements in the
nd a pause of
sumed, there
tters can be w
t rate of our
2.0391 bits/mi

B. Online
We have dev
that relies
robabilistic sco
nd column rec
300 wave, the
flecting the re
pochs get a ne
sually no need
the score of
ecision can alr
nalyses with 6
ffline analyse
erformance i
assification re
alues. The firs
ecture of DSNN a
nhibitory interneu
III. R
e analysis.
analysis result
atible with [6
cation perform
the results i
al groups, the
rcentage of co
ed lines indi
art in this study
subject had an
the bit rate is
y of classifica
matrix. Since
1.4 seconds fo
can be 12 tria
written at maxi
r system for
n.
e analysis.
veloped a gree
on the fact
ores. In the be
eives a score
ey get a posit
esemblance to
egative score.
d to evaluate al
an epoch alre
ready be mad
6 subjects, 5
es. Table I
in detail, L
esults and tw
t one allows e
and wavelet prep
uron, which affec
RESULTS
ts in Figure 3
6]. Figure 3(c)
mance of our
in [6]. The x
e left y-axis
orrect results
cate the bit
y, whose ages
ny prior BCI
performed as
tion and N is
e one trial grou
or displaying
al groups in a
imum. The ma
offline classi
edier version o
that the cla
eginning of eac
of 0. If their e
tive score, wit
the training se
. With this ap
ll the 12 epoch
eady satisfies
de. We have c
of whom also
I presents
Listing Righ
wo kinds of ac
errors in result
processor. The bl
cts all of the synap
are presented
) shows avera
r system, whi
x-axis shows t
and solid lin
and the right
rate. 7 healt
s ranged betwe
experience. T
in [8]: Swhere
s the number
up lasts for 3.
the next letter
a minute, i.e.,
aximum possib
ification is th
of our algorith
ssifier produc
ch run, each r
epoch include
th its magnitu
et, and irrelev
pproach, there
hs for a decisio
the margin, t
conducted onli
o participated
average onli
ht and Wro
ccuracy vs. r
ts, so typing r

lack
aptic
d in
age
ich
the
nes
y-
thy
een
The
e P
of
6 s
r is
12
ble
hen
hm
ces
ow
es a
ude
ant
e is
on.
the
ine
d in
ine
ong
rate
rate
i
i
l
r
t
a
s
s
o
a
c
c
c
p
c
t
t
l
e
W
o
9
g
o
l
s
a
t
p
e
r
i
o
o
h
o
f
s
a
o
is calculated o
is faster. The
letter he/she w
results are tak
this informatio
a given set of
shows that eac

Figure 3.
subject (S5), (
of 7 subjects E
at intervals of
classified sep
correct classi
classifications
Discussion
Figure 2 t
predicts the
classifier has
time and so o
then on the a
letters/min. In
each run to di
When we tak
offline rate b
93% accuracy
groups with w
of 11.111 le
letters/min w
speed is 12.4
accuracy of 8
taken into con

In this pap
performance
experiments d
rate achieved
is 12.1 letters/
offline analys
online analys
highest rate f
165.44 ts/min
online analys
for 90.91% a
system can ac
accuracy) tha
offline and for
omitting the er
latter makes
wants, so time
ken into calcu
on, on average
f 100 runs wil
ch run is on av
Offline perfo
(b)Best perfor
Evaluation of
f 3.6 s. For ca
arately and a
fications in a
s.
n
tells us that
right answer
correct answe
on. If we assu
average our sy
n practice, w
splay the next
ke that extra
becomes 7.684
y, 100 runs
wrong results
etters/min wi
with pauses. In
48 letters/min
88%, and 11.
nsideration.
IV. CO
per, we have d
of our en
done with 8 ab
by a subject u
/min and 62.55
is, and 15.5 le
sis. On the o
for offline an
n for 100% ac
is was 23.19
accuracy. We
chieve higher
an current sta
r online exper
rror in classifi
sure the subje
e spent on wr
ulation as lost
e, 100% corre
ll last for 178
verage classifi
ormance. (a)
rming subject
performance i
alculating accu
accuracy is th
a session over
48% of the t
in the first
ers in 2 trial g
ume no delay b
ystem achieve
e also spend
t letter to be ty
time into acc
44 letters/min
will be class
in 7 runs, wh
ithout interru
n online anal
for error ign
14 letters/min
ONCLUSIONS
demonstrated
nd-to-end BC
ble-bodied sub
using our syste
5 bits/min for
etters/min and
ther hand, fo
nalysis was 3
ccuracy and th
letters/min a
e have demo
rates (for a g
ate-of-the-art
riments.
cation; therefo
ect types the e
rong classifica
time. Conside
ect classificatio
trial groups.
ied in trial gro
Worst perform
(S6), (c) Ave
in Figure 2 is
uracy, every ru
he total numbe
r total numbe
time the class
trial group.
groups 81% o
between each
es a rate of 9
1.4 sec. betw
yped on the scr
count, the ave
n. If one assu
sified in 150
hich yields a r
uption and 8
lysis, the ave
norant results
n when errors
S
the flexibility
CI system
bjects. The hig
em with 300m
100% accura
d 63.71 bits/m
or 125ms ISI
32 letters/min
he highest rat
and 98.88 bits
onstrated that
given classifica
systems both
ore it
exact
ation
ering
on of
This
oups.

ming
erage
done
un is
er of
er of
sifier
The
of the
h run,
9.363
ween
reen.
erage
umes
trial
result
8.823
erage
with
s are
y and
with
ghest
ms ISI
cy in
min in
, the
and
te for
s/min
t our
ation
h for
Te
M

[1]
tow
Ele
[2]
P3
pp
[3]
tra
11
[4]
bra
pp
[5]
IEE
[6]
int
Jan
[7]
bra
44
[8]
and
[9
Ev

[1]
[2]
[3]
[4]
[5]
11
9.3
20
20
Th

This work
echnically, by
Meghe College
] L. A. Farwell
ward a mental
ectroencephalogr
] E. Donchin, et
00-based brain-c
. 174-9, Jun 2000
] P. Meinicke, M.
ansfer rates in brai
14, 2002.
] M. Kaper an
aincomputer inter
. 4363-6, 2004.
] H. Serby, et al.,
EE Trans Neural
] U. Hoffmann,
terface for disable
n 15 2008.
] U. Hoffmann,
aincomputer inter
6-9, 2004.
] J. R. Wolpaw,
d control," Clin N
9] Armagan Am
valuation of a Rea
] 2.3 10.67 500ms
] 4.3 19.83 125ms
] 5.5 24 300ms 90
] - 47.26 140ms 4
] 5.45 23.77 125m
.111 49.39 300ms
363 48.41 300ms
.48 105.88 125ms
.44 91.58 125ms
his work 24.42 12
ACKNOW
was supporte
y professors
of Engineerin
REFER
and E. Donchin,
prosthesis utiliz
Clin Neurophysi
al., "The mental
computer interfac
0.
. Kaper, F. Hoppe
in computer inter
nd H. Ritter
rfacing, Conf Pr
An improved P3
Syst Rehabil Eng
et al., "An e
ed subjects," J Ne
et al., "Applica
rfaces," Conf Proc
et al., "Brain-com
Neurophysiol, vol
mcalar, Mujdat
al-time P300-base
s 95% unk. unk.
s 95% 9.367 100%
0% 11.037 95%
44% 19.7 92%
ms 92% 15.209 10
s 93% Offline tes
100% Offline tes
s w/o err Online t
94% Online tests
6.25 125ms 100%
WLEDGMENT
ed by the sc
and staff of
ng) Airoli. Nav
RENCES
"Talking off th
zing event-related
ol, vol. 70, pp. 51
l prosthesis: asses
e," IEEE Trans R
e, M. Huemann, H
rface: A case stud
Generalizing to
roc IEEE Eng M
300-based brain-
g, vol. 13, pp. 89-
efficient P300-ba
eurosci Methods, v
ation of the evid
c IEEE Eng Med
mputer interfaces
. 113, pp. 767-91
Cetin Design, I
ed Brain-Compute
%
00%
sts
sts
tests
s
% Offline tests120
T
cientifically a
f DMCE (Da
vi Mumbai.
e top of your he
d brain potentia
10-23, Dec 1988.
ssing the speed o
Rehabil Eng, vol
H. Ritter, Improv
dy. NIPS, pp. 11
o new subjects
Med Biol Soc, vol
computer interfac
98, Mar 2005.
ased brain-compu
vol. 167, pp. 115-
dence framework
Biol Soc, vol. 1,
s for communicat
, Jun 2002.
Implementation
er Interface System
0.
and
atta
ead:
als,"
of a
l. 8,
ving
07
in
l. 6,
ce,
uter
-25,
k to
pp.
tion
and
m.
Proc. of the Intl. Conf. on Computer Applications Vol-
ume 1. Copyright 2012 Techno Forum Group, India.
ISBN: 978-81-920575-9-0 :: doi: 10. 72976/ISBN_0768
Information hiding method based on Steganography and Image morphing

Bhushan Zope
Pune Institute of Computer Technology
Pune University
Pune, Maharashtra, India
bhushan.zope@hotmail.com

AbstractSteganography is used to hide the information inside
the cover image. In Steganography cover image can only hide
the image having less size than itself. Data hiding can also be
achieved using Image Metamorphosis (also known as image
Morphing) technique. Amount of data that can be hidden us-
ing morphing technique is more than the amount of data hid-
den using Steganography. We can further increase the cove-
rate by combining the both techniques. So here new data hid-
ing technique is proposed based on Image Morphing and Ste-
ganography. The intermediate images generated in the morph-
ing process are very much similar to both source and destina-
tion image. So if we use one of such intermediate image as the
secret image and destination image as cover image, then by
using steganography we can hide more data using same cover
image.
Keywords - Image Representation : Morphological ; Image Pro-
cessing and Computer Vision.
I. INTRODUCTION
In recent time the use of digital media is increased in
many folds. With increase in popularity of digital media,
security related issues of digital media have gain im-
portance. The confidentiality of digital data can be achieved
by cryptography. But main problem with cryptography is
that, it reveals the existence of important data. If encrypted
Message is caught by intruder then even if he cant under-
stand the confidential data, he can gain the knowledge of
certain secret communication between two parties. Also
because of this knowledge, cipher text attracts the interest of
cryptanalysts. In many system, revelation of existence of
secret message leads to failure of that system. Because of
this drawback, another approach, data hiding also called as
steganography, has gained importance in recent time.
The word Steganography is derived from the Greek
words Steganos meaning "covered" and graphy meaning
"writing or drawing". Steganography is the art and science
of communicating in such a way that the presence of a mes-
sage cannot be detected[7].
In steganography important information (secret) is hid-
den in the innocent looking data (cover). The study of ste-
ganography can be traced to [8], in which the Prisoners
Problem was proposed. In this scenario, Alice and Bob were
in jail, and attempted an escape plan. However, all their
communications must go through the warden, Willie. If de-
tecting any encrypted messages, Willie would frustrate their
plan by putting them into solitary confinement. Therefore,
Alice and Bob must find a way to conceal their secret in an
innocuous looking cover text. Steganography is typically
achieved by altering some nonessential information in the
host message. Host message can be anything like text, im-
age, audio-video file. There are number of techniques dis-
covered to hide the data. Various Steganography techniques
are reviewed in [2].
The use of image morphing for information hiding is
proposed in [4]. In [4], authors proposed that the intermedi-
ate images generated in morphing process can be used as
stego data to hide the source and destination images. They
considered secret image as source image and cover image as
destination image, in morphing process. Also both the
source image and the target image can be hidden in the in-
termediate image. To ensure the security, the cover image is
used as one of the stego keys, and is kept secretly.
In this paper we proposed the new method for image se-
curity based on morphing and steganography.
Section 2 describes design and model of proposed sys-
tem. Design part also explains the block diagram of the pro-
posed system. Section 3 is for conclusion of the proposed
model. It also gives some idea about the future work related
to proposed model.

II. PROPOSED SYSTEM
A. Design concept:
Our technique is based on image morphing and ste-
ganography. Image morphing is a process which changes
the one image into other through seamless transition. It is
the fluid visual transformation of one image into another
image using the sequence of intermediate images. These
intermediate images are similar to source image and destina-
tion image. First image in the sequence is very similar to
source image while last image is similar to destination im-
age. As we move from first image to last image the similari-
ty with source image decreases while destination image in-
creases. We can select one of such intermediate image as
secret image and destination image as cover image. Both
these images have strong correlation between each other. So
information contained in both images is similar. So we have
to hide very less information which is different from cover
image. Thus using steganography techniques on these both
images we can achieve more hiding capacity.

Figure 1. Block diagram for proposed model

Figure1 describes the block diagram of proposed model.
In the proposed system, both secret image and cover images
are morphed using available morphing techniques and num-
bers of intermediate images are generated. Then one of the
images out of them is selected using correlation as a selec-
tion criterion. The selected image is then becomes the new
secret image for the steganography process. Then using
cover image and new secret image and using any existing
steganography process, stego data is produced, which can be
sent across the public network. Once receiving the stego
data at other end, the secret image is extracted using extrac-
tion techniques. This extracted secret image is the interme-
diate image generated in morphing technique. The actual
secret image can be obtained by de-morphing the extracted
image with the help of morphing keys.

B. Model for proposed system
1) Morphing :
Consider two images, I
s
and I
c
, where Is and Ic are secret
image and cover image respectively. Then morphing func-
tion can be defined as:

Where:

,

for any other
,

Where, I[i] is the set of intermediate frames transformed
from I
s
to I
c
.

2) Selecting one image out of r intermediate images:
So after generating intermediate images I[i], select one
image out of them as secret image for steganography. The
function can be described as:

F : I I
m

Function F selects the one image I
m
out of i images
generated in first stage. The image selection should be such
that it should have maximum correlation between cover
image but as the correlation increases, secret gets more
exposed. So there is tradeoff between correlation and secret
exposure.

3) Steganography:
Once the image is selected from set of intermediate
images, it is used in the steganography process along with
the cover image. The process of steganography is carried
out using any of the existing steganography technique.
Steganography function can be defined as:

S : (I
m
, I
c
, S
k
) W

Where, S
k
is set of stegokeys and W is the stegodata
generated.

4) Image extraction:
This generated stegodata, W is then sent across the
private network. Image I
m
is extracted from stegodata W,
using extraction method associated with steganography
technique used earlier to generate W. the extraction function
can be written as:

E : (W, S
k
) I
m

5) De-morphing:
To reconstruct the secret image I
s
from I
m
, de-morphing
technique is used which is inverse of morphing. The de-
morphing function can be defined as:

D : (I
m
, M
k
) I
s

Where, M
k
is set of morphkeys.

III. CONCLUSION
This paper presents a new information hiding technique,
on the base of image morphing and steganography. It uses
one of generated intermediate image as the secret image and
destination image as cover image. So, as to increase amount
of hidden data in the intermediate image. In future, we want
to explore exact results of this new technique and capacity
of data hiding is need to be evaluated.

REFERENCES

[1] Tseng, Y.-chee, Chen, Y.-yuan, & Pan, H.-kuang, A Secure
Data Hiding Scheme for Binary Images, IEEE transactions on
communications, vol. 50(8), pp. 1227-1231, 2002.

[2] Abbas Cheddad, JoanCondell, KevinCurran, PaulMcKevitt, Digital
image steganography: Survey and analysis of current methods,
Signal Processing, Vol. 90(3), pp.727-752, 2010.
[3] H.B.Kekre,A.A. Athawale, & U.A. Athawale, Increased Capacity for
Information Hiding using Walsh Transform, International Confer-
ence and Workshop on Emerging Trends in Technology (Icwet), pp.
96-101, 2010.
[4] H. Nakamura, & Q. Zhao, Information Hiding Based on Image
Morphing, 22nd International Conference on Advanced Infor-
mation Networking and Applications, pp.1585-1590, 2008.
[5] Hong, W., Chen, T.-shou, & Shiu, C.-wei, Reversible Data Hiding
Based on Histogram Shifting of Prediction Errors, International
Symposium on Intelligent Information Technology Application
Workshops, pp. 292-295, 2008.
[6] XIE Qing, XIE Jianquan, XIAO Yunhua, A High Capacity Infor-
mation Hiding Algorithm In Color Image, E-business and infor-
mation system security, pp. 1-4, 2010.
[7] Christian Cachin, An information-theoretic model for steganogra-
phy, Information and Computation, vol. 192(1), pp.41-56, 2004.
[8] G. J. Simmons, The prisoners problem and the subliminal channel,
in Proc. CRYPTO83, pp. 5167, 1983.

India.
ISBN: 978-81-920575-9-0 :: doi: 10. 72983/ISBN_0768
Dimensionality Reduction Techniques for Face
Image Retrieval: A Review
Fousiya K.K
M.Tech Scholar, Department of CSE
M.E.S. College of Engineering, Kuttippuram,
Kerala, India

AbstractNowadays, face recognition and face image
retrieval systems have attained much importance in
security systems. Image retrieval becomes a challenging
issue in real world applications because the image
collections are often high dimensional. Curse of
dimensionality is a major issue in Content-Based image
Retrieval (CBIR). By providing better dimensionality
reduction and higher class discrimination prior to the
classification process, a classifier must provide higher
classification accuracy. So, dimensionality reduction has a
significant role in the classification process. This
preprocessing step must provide a better dimensionality
reduction of high dimensional data as well as high
discrimination between classes. The aim of this paper is to
provide an overview of the various dimensionality
reduction methods and its relevance to improve the
classification accuracy of the existing face retrieval
systems.

Keywords-CBIR;Dimensionality Reduction;PCA; LDA;
SVM.
I. INTRODUCTION
Content-Based image Retrieval (CBIR) has got much
importance in the recent years because of the dramatic
increase in the volume of the digital images that are stored
and exchanged over the web. CBIR is a process of
retrieving the similar images corresponding to a given
query images, from a large image database. Because of
the image collections are non homogeneous and large in
volume, the image retrieval has become a challenging
problem.
In CBIR, image contents are represented by low-level
visual features including color, texture and shape. These
image contents cannot be interpreted by human beings
because the low-level visual features have limited
capacity for representing the high-level concepts which
are conveyed by the image. This is the major problem of
CBIR, referred to as the large semantic gap between the
low-level visual features and the high-level semantic
content. Normally, the images including digital
photographs, scientific images, medical images, hyper
spectral images etc. are high dimensional and there exist
hundreds or even thousands of features to represent the
image. When use with machine learning and data mining
algorithms, these larger number of features lead to major
problem in CBIR, known as the curse of dimensionality.
Curse of dimensionality causes degradation in the
efficiency and performance of the system. For the
efficient processing of high dimensional images, its
dimensionality has to be reduced without a loss in the
original properties of the high dimensional space.
Automatic face recognition and face retrieval is an
active research topic in the field of pattern recognition.
Face recognition has to address the dimensionality
reduction problem because, a face image having m*n
pixels is represented by a vector in R
m*n
space for
computational purposes. This m*n-dimensional space is
too large for recognition process. When the number of
images in the data set increases, the complexity of
representing the data set also increases. Analyzing the
data set with a large number of variables increases the
computational complexity and consumes a large amount
of memory. Thus, for solving the curse of dimensionality,
various dimensionality reduction methods have been
introduced.
The remainder of this paper is organized as follows. In
section II, we describe about dimensionality reduction
techniques. Section III describes about face image
retrieval system, section IV presents related works and
conclusion is described in section V.
II. DIMENSIONALITY REDUCTION TECHNIQUES
Dimensionality reduction is the transformation of high
dimensional patterns into a lower dimensional
representation. The goal of the dimensionality reduction is
to preserve the local structure of the original high
dimensional space. Dimensionality reduction removes the
irrelevant and redundant features, and extracts a small
number of relevant features. The minimum number of
parameters which are required for representing the
properties of the data is known as the intrinsic
dimensionality. So, the dimensionality of the reduced
representation should correspond to the intrinsic
dimensionality. The lower dimensional representation
provides the efficient visualization of the data. The
analysis of
efficiently by
representation
hyper spectr
bands. For r
bands are a m
a reduction i
Figure 1 [1].

Figure 1. Prepr

Dimension
linear and
transforms th
dimensional
dimensionalit
dimensional
dimensional d
reduced repr
original variab

S
i
=w
i,1
x

Where W
k*p
i

Among th
dimensionalit
Component A
Analysis (LD
easy to implem
The major
face recogniti
various factor
effects etc,. T
linear in natu
structure of
overcome thi
Locally Line
been introdu
complex non
the observed
of the origina
variations tha
the high dim
y projecting t
n. For exampl
al satellites
remote sensi
major problem
n the number
rocessed dimensi
derivation from h
nality reductio
non-linear te
he high dim
subspace usi
ty reduction
representatio
data X, where
resentation is
ables as given
x
1
+..+w
i,p
x
p
S=W
X
s the linear tra
he previously
ty reduction, t
Analysis (PCA
DA) [4]. The li
ment.
r problem tha
ion system are
rs like pose, i
These variation
ure. Linear m
these kind
s limitation, v
ar Embeddin
uced. The no
-linear structu
data lies on an
al high dimen
at occur in the
mensional da
them into a l
le, the image
contain hund
ing applicatio
m and the prep
r of bands, w

ionality reduction
hyper spectral im
on techniques
echniques. A
mensional da
ing a linear
techniques
on S for
e k < p and ea
a linear co
in (1) or in (2
p
, for i=1,
ansformation
investigated m
the most popu
A) [3] and Li
inear techniqu
at has to be f
e the variation
illumination, e
ns make the f
methods fail to
of non-lin
various non-li
ng (LLE) [5],
on-linear met
ures efficientl
n abstract spa
nsional space.
face images c
ata can be
ower dimensi
s produced by
dreds of spe
ons, these hi
processing req
which is show
n prior to informa
agery.
are classified
A linear me
ata into a l
mapping. L
finds a low
the origina
ach variable o
ombination of
2).
,2,,k

(1)
(2)
weight matrix
methods for l
ular are: Prin
inear Discrimi
ues are simple
faced in real
ns that occur in
expressions, a
face database
o find the intr
ear datasets.
inear methods
Isomap [6]
thods handle
ly. In some c
ce called man
For example
can be represe
done
ional
y the
ectral
igher
quires
wn in

ation
d into
ethod
ower
inear
wer k
al p
of the
f the
x [2].
inear
ncipal
minant
e and
time
in the
aging
non-
rinsic
To
s like
have
e the
cases,
nifold
e, the
ented
usin
dim
dim
learn
non-
emb
dim
in t
dim
struc
do s
W
retri
imag
perf
face
The
imag
and
appr
R
on
syst
Face
base
appr
the
dim
sepa
wide
Eig
dim
Ana
Fis
dim
Ana
P
repr
of th
PCA
min
for p
the
min
syst
high
[3][4
syst
less
P
reco
dim
com
ng a low dime
ensional ima
ensionality re
ning techniqu
-linear struc
bedded in the
ensional embe
the original
ensionality re
cture efficient
so.
III. F
With regard t
ieval may be
ges into dif
formance metr
e recognition s
Feature E
Classific
feature extrac
ges. Then the
finally, the c
ropriate classe
Recently a sign
face detectio
ems have wid
e recognition
ed methods
roaches extrac
eyes, nose, m
ension of th
aration betwee
Eigenface [
ely used subs
genface app
ensionality re
alysis (PCA)
herface ap
ensionality r
alysis (LDA).
PCA finds a
resentation, wh
he data varia
A provides the
imum reconst
projection [4]
discriminatio
imizing the
ems, LDA is
her class discr
4] and so, L
ems. Class di
compared to
PCA and LD
ognition [3][8]
ensional repr
mponent most
ensional face
age space. T
duction metho
ues [5][6]. Ma
tures or lo
high dimensi
edding preserv
space. In su
duction techn
tly, whereas t
FACE IMAGE R
o the machin
viewed as t
fferent classe
ric is classifica
ystem consist
Extraction
cation
ction step extr
e classifier is
lassifier class
es.
IV. RELAT
nificant numbe
on and reco
de spread appl
methods ca
and subspac
ct the local fea
mouth etc. Sub
e data, whil
en distinct clas
13] and Fis
space method
proach uses
eduction met
for subspace
pproach u
reduction me
a projection
here along the
tion occurs in
e accurate rep
truction error
. The main a
on between
within class
s superior to
rimination by
DA is widely
iscrimination
LDA [3].
DA have bee
][9]. PCA fin
resentation, w
t of the dat
manifold, ly
This form
hods are know
anifold learnin
ow-dimension
ional data sets
rves the struct
uch cases, t
niques can find
the linear tech
RETRIEVAL SYS
ne learning ap
the problem
es. In this
cation accuracy
ts of two majo
racts the featu
trained for tr
sifies the face
TED WORK
er of works h
ognition. Fac
lication in sec
an be classifi
ce methods.
atures such as
ubspace metho
le retaining t
sses.
sherface [4]
ds for face rec
the linear
thod Principa
e generation,
uses linear
ethod Linear
on a lower
e principal co
n an unsuper
presentation of
and also find
aim of LDA i
different c
distance. In
PCA becaus
using the cla
y used in fac
capability of
en widely u
nds a projectio
where along
ta variation
ying on a high
of non-linea
wn as manifold
ng extracts the
nal manifold
s and this low
tures that exis
the non-linea
d the manifold
hniques fail to
STEM
pproach, image
of classifying
context, the
y. Normally, a
or steps.
ures of the fac
raining image
images to the
have been don
ce recognition
curity systems
ied as feature
Feature-based
the position o
od reduces the
the maximum
are the two
cognition. The
unsupervised
al Componen
whereas the
supervised
Discriminan
r dimensiona
omponent mos
rvised manner
f the data with
ds the best axi
s to maximize
classes, while
classification
se, it provide
ss information
ce recognition
f PCA is very
used for face
on on a lowe
the principa
occurs in an
h
ar
d
e
ds
w
st
ar
d
o
e
g
e
a
e
s
e
e
n
s.
e
d
of
e
m
o
e
d
nt
e
d
nt
al
st
r.
h
s
e
e
n
s
n
n
y
e
er
al
n
unsupervised manner. PCA provides the accurate
representation of the data with minimum reconstruction
error and also finds the best axis for projection [4]. The
main aim of LDA is to maximize the discrimination
between different classes, while minimizing the within
class distance. In classification systems, LDA is superior to
PCA because, it provides higher class discrimination by
using the class information and so, LDA is widely used in
face recognition systems. Class discrimination capability
of PCA is very less compared to LDA. LLE and Isomap
are the non linear dimensionality reduction techniques that
has been used for face recognition [5][11]. LLE is suitable
for processing the high dimensional data that contains non
linear structures and the key merit of LLE is that it
preserves the neighborhood structure of the high
dimensional data. LLE has been applied for face
recognition, but the discrimination capability is less
compared to LDA [11]. LLE has wide spread application
in many areas of science including data analysis and data
visualization. In order to recover the true dimensionality
and geometric structure of the high dimensional data,
Isomap is preferred [6]. Isomap retains the global structure
but, computationally expensive for large datasets.
Experiments show that, when Isomap is applied for face
recognition, it has the worst performance [11].
Furthermore, many work has been done in the literature by
combining different dimensionality techniques for face
recognition [8][12]. However, the classification accuracy
of such combinations is not so high.
Mahesh Pal et al. [7] discussed about the need of
dimensionality reduction as a preprocessing step to
classification. Support Vector Machine (SVM) is a widely
used method for classification. SVM constructs a hyper
plane in which the margin between different classes
should be high and so, it provides better classification
accuracy. Experiments are conducted for classification of
hyper spectral data using SVM. Experimental results
show that the classification accuracy of SVM can be
increased by reducing the dimensionality of the data [7].
Thus, the authors proved that, dimensionality reduction is
an essential pre-processing stage for classification by
SVM. By the addition of features, the classification
accuracy of SVM decline significantly and this is best
apparent for small training sets. However, even for large
training data set, dimensionality reduction have an
important role. Experimental results show that there is a
strong dependence between dimensionality reduction and
classification accuracy of SVM. When the feature
dimensions considered were 55, 60 and 65, the
classification accuracy was 92.24%, 92.11% and 91.76%.
Besides providing classification accuracy, dimensionality
reduction offers the following advantages to SVM:

Improved speed of the classification by reducing
feature set size
Reduction in data storage requirements
. Jianke Li et al. [8] introduced a method for face
recognition system based on the combination of Principal
Component Analysis (PCA) and Linear Discriminant
Analysis (LDA) for feature extraction. In order to perform
feature extraction, a commonly used method is PCA and it
provides better dimensionality reduction. After performing
the PCA transformation, the resulting components are
uncorrelated and have less reconstruction error. LDA
increases the class separation and extracts the LDA
features, which is advantageous for classification. So, for
providing higher discrimination of the samples, LDA is
used after PCA transformation. Thus PCA forms a PCA
subspace and the combination of PCA and LDA form the
LDA subspace having high discrimination. For
classification purpose, Nearest Neighbor Classifier (NNC)
is used. Experiments are conducted on the ORL face
database containing 400 images of 40 individuals.
Classification accuracy for different training sample size
and for various feature dimensions was calculated. An
increase in training sample size causes an increase in the
classification accuracy of both PCA method and combined
PCA and LDA method. Also, it can be concluded that the
PCA and LDA combination method provides better
classification accuracy than the PCA alone method for face
recognition [8].
In [9], Md. Omar Faruqe et al. proposed a face
recognition system using Principal Component Analysis
(PCA) and support vector machine (SVM). Since the
major components of a face recognition system are feature
extraction and classification, its performance strongly
depends on the factors including the way in which the
features are extracted, how accurately the features are
classified into a group and the selection of a classifier.
Therefore, the selection of the feature extractor and the
classifier is very important in a face recognition system. In
this work, the authors have used PCA for feature
extraction and SVM for classification.
In SVM based classification systems, the classification
is done by the construction of a new hyper plane that
provides maximum separation between the data item that
is, the margin of the hyper plane should be maximum. The
vectors which are lying near to the hyper plane are called
support vectors. Every point in the input space is non-
linearly mapped into a high dimensional feature space.
This mapping is done using kernel functions. In [9], the
kernel functions used for experiments include linear,
polynomial and radial basis function kernels. Experiments
are conducted on the ORL face database containing 400
images of 40 individuals. Randomly selected 200 samples
serves as the training set. This sample set is used for
constructing Eigen faces and also for training the SVM. A
change in feature vector size causes a change in input
vector size of the classifier because; the output after PCA
transformation is given directly to the classifier. From the
experiments conducted it can be concluded that, PCA and
SVM combination provides better classification accuracy
for face recognition on the ORL face database.
Sangwoo Moon et al. in [10] proposed a dimensionality
reduction technique using support vector machine. The
decision vectors which are used for classification purpose
in SVM are used as the mapping vectors for
dimensionality reduction. By using these mapping vectors,
it produces a feature set having high efficiency and better
classification ability. Normally, the mapping vectors which
are derived from the SVM is highly redundant. So, further
reduction of redundancy is done based on a factor called
asymmetric decorrelation measure based on angular
distance between two mappings and the quality of
mapping, which is obtained from its classification
performance. Redundancy removal process eliminates the
less meaningful mapping vectors. The mapping vectors
which are survived after the removal process are far away
from each other and also provide higher classification
accuracy. Experiments are conducted on the handwritten
numeric characters dataset. The feature dimensionality of
SVM was 30 and the classification accuracy was 98.2%. It
can be concluded that, SVM provides better
dimensionality reduction as well as higher classification
accuracy.
Table I. summarizes the different dimensionality reduction
methods used for face recognition in the literature survey.
Experiments are conducted on the ORL database
containing 400 images of 40 individuals.

TABLE I. CLASSIFICATION ACCURACY VS FEATURE DIMENSION

Method Classifier Feature
Dimension
Classification
Accuracy

PCA NNC 40 85%

PCA +
LDA
NNC 40 93%

PCA SVM 40 97%

A classification system provides higher classification
accuracy only if it is preceded by a better dimensionality
reduction stage and better class discrimination stage. From
the literature survey, it is observed that whatever be the
classifier used for classification, dimensionality reduction
is an essential step prior to the classification stage [7].
When the Nearest Neighbor Classifier (NNC) is used for
classification of face images in ORL database, PCA alone
method provided a classification accuracy rate of 85%,
which is satisfactory. But, this rate is significantly
improved into 93% by using PCA and LDA combination.
It is clear that, combination of PCA and LDA is very much
better than PCA alone for improving classification
accuracy of SVM [8]. From the experiments conducted on
the ORL face database using SVM as a classifier, it is clear
that PCA alone method provides better classification
accuracy of 97% for the same feature dimension [9].
V. CONCLUSION AND FUTURE WORK
Curse of dimensionality can be overcome by using an
effective dimensionality reduction technique as a
preprocessing step. Studies show that, the feature
dimensionality has an adverse impact on the classification
accuracy of SVM. Therefore, it is beneficial to perform a
dimensionality reduction method prior to the SVM. PCA is
found to be suitable dimensionality reduction method for
SVM. By providing better dimensionality reduction and
higher class discrimination prior to the classification
process leads to higher classification accuracy. PCA is an
extensively used dimensionality reduction method. But,
the class discrimination capability is limited. It demands a
dimensionality reduction method with high class
discrimination. LDA gives higher class discrimination
among the different classes. By the induction of LDA
between PCA and SVM improves the classification
accuracy. It proposes the combination of PCA and LDA
prior to SVM improves classification accuracy. Future
work also needs to explore ways of eliminating the
redundant mapping vectors of SVM in order to bring down
the overall computational cost of the proposed system.
REFERENCES
[1] G. Chen and S.E. Qian, Dimensionality reduction of hyper
spectral imagery using improved locally linear embedding ,
Journal of Applied Remote Sensing 1, 2007.
[2] Imola K. Fodor, A Survey of Dimension reduction techniques,
June 2002.
[3] Aleix M. MartoAnez and Avinash C. Kak, PCA versus LDA,
IEEE Transactions on Pattern Analysis and Machine Intelligence,
vol. 23, no. 2, pp. 228-233, February 2001.
[4] Peter N. Belhumeur, Joao P. Hespanha, and David J. Kriegman,
Eigenfaces vs. Fisherfaces: Recognition Using Class Specific
Linear Projection", IEEE Transactions on Pattern Analysis and
Machine Intelligence, Vol. 19, NO. 7, July 1997
[5] S.T. Roweis and L. Saul, Nonlinear dimensionality reduction by
locally linear embedding", Science, vol 290 22, No. 5500,
December 2000, pp. 2323-2326.
[6] Joshua B. Tenenbaum, Vin de Silva and John C. Langford, A
Global Geometric Framework for Nonlinear Dimensionality
Reduction", science, VOL 290 22, December 2000.
[7] Mahesh Pal and Giles M. Foody, Feature Selection for
Classification of Hyperspectral Data by SVM", IEEE Transactions
on Geoscience and Remote Sensing, Vol. 48, No. 5, May 2010
[8] Jianke Li, Baojun Zhao and Hui Zhang, Face Recognition Based
on PCA and LDA Combination Feature Extraction", The 1st
International Conference on Information Science and Engineering,
2009
[9] Md. Omar Faruqe and Md. Al Mehedi Hasan, Face Recognition
Using PCA and SVM", The 3rd International Conference on
Antiounterfeiting, Security, and Identi_cation in Communication,
2009
[10] Sangwoo Moon and Hairong Qi, Effective Dimensionality
Reduction based on Support Vector Machine", International
Conference on Pattern Recognition, 2010
[11] A.Zagouras, A.Macedonas, G.Economou and S.Fotopoulos, An
application study of manifold learning-ranking techniques in face
recognition", Multimedia Signal Processing, 2007, Pages: 445
448
[12] Junping Zhang, Huanxing Shen and Zhi-Hua Zhou, Unified
Locally Linear Embedding and Linear Discriminant Analysis
Algorithm for Face Recognition", Sinobiometrics, LNCS 3338, pp.
269-304, Springer 2004
[13] M.Truk and A. Pentland, Eigen faces for recognition, Journal of
CognitiveNeuroscience,3:72-86,1991
India.
ISBN: 978-81-920575-9-0 :: doi: 10. 72990/ISBN_0768
Human Age Classification using 2-D Gabor Wavelet and Neural Network
Jayant Jagtap
Electronics and Telecommunication Department
SGGSIE&T
Nanded, India
jagtapjayant@yahoo.co.in

AbstractComputer-based age classification, age synthesis and
estimation via faces become interesting topics recently because of
their lot of real-world applications, such as forensic art,
electronic customer relationship management, security control
and surveillance monitoring, biometrics, and entertainment.
Coarse age classification such as classifying human face image
into baby, child, young or adult people is much easier for human
than age estimation. One attempt is done here to provide this
intelligence of human, to computer by an algorithm. In this
paper 2-D Gabor wavelet is used for extracting the aging face
features and feed forward back propagation neural network is
used for classification of face images into three age groups:
child, young and adult. Face images are taken from FG-NET
aging database for training and testing of age classification
algorithm.
Keywords-Age classification, Gabor Wavelet, Neural
Network.
I. INTRODUCTION
Processing human face image has lot of real world
applications such as multimedia communication, Human
Computer Interaction, security etc. as it conveys a significant
amount of non verbal information such as identity, gender,
expression, age, ethnic origin etc. Most of the information
has been used by researcher for different applications but
there is relatively less work concerning the age information
for applications such automatic age estimation, age
classification etc. Estimating age from face image is not only
challenging for computer but also difficult for human being,
because the aging progress depends upon the factors such as
gene, health, living style, weather condition, living location
etc. Males and females may also have different age
progression.
There are only few publications on age estimation, age
classification and age synthesis in the last several decades,
because the aging progress is uncontrollable and
personalized, it is harder to collect large size aging database.
The current age estimation performance is still not good
enough for practical use and more effort has to be put into
this research direction. However, coarse age classification
such as classifying human face image into baby, child, young
or adult people is much easier for human. Age classification
has lot of real world applications such as, an age specific
human computer interaction (ASHCI) system for secure
network/ system access control, e.g. a vending machine,
secured by the ASHCI system can refuse to sell alcohol or
cigarettes to the underage people. ASHCI denies access
to internet pages with adult materials, creating statistics in
the rating studies of entertainment or amusement park
visitors, television viewer etc. age-based retrieval of face
images from databases, Electronic Customer Relationship
Management (ECRM) etc.
In this paper a novel approach to classify the human face
image by using 2-D Gabor wavelet for facial feature
extraction and feed forward back propagation neural network
for classification is proposed. The proposed method
classifies the human face images into three groups: child,
young and adult.
The rest of the paper is organized as follows: Section 2
gives an overview of related work. 2-D Gabor wavelet and
Neural Network is described in Section 3. Proposed method
is explained in Section 4. Experimental results are reported
in Section 5 and Section 6 draws the conclusion.
II. RELATED WORK
There exists some earlier work on age synthesis, age
rendering in last several decades but less work on age
estimation. Burt and Perrett [1] simulated aging variations by
superimposing typical aging changes in shape and color on
face images. Tiddeman et al. [2] further extended their work
by using wavelet methods to prototype facial textures across
age. Ramanathan and Chellappa [3] proposed a craniofacial
growth model that characterizes growth related shape
variations observed in human faces during young ages.
Ramanathan and Chellappa [4] proposed a method for face
verification across age based on a Bayesian classifier. Lanitis
et al. [5] constructed an aging function based on a parametric
model for human faces and performed automatic age
progression, age estimation, face recognition across age etc.
Further, Lanitis et al. [6] compared the above age estimation
process with that of neural networks based approaches. First
time age classification problem was studied by Kwon and
Lobo [7]. They proposed a method to classify input face
images into one of the following three age groups: babies,
young adults and senior adults. Their study was based on
geometric ratios and skin wrinkle analysis. Geometric ratios
were first computed from facial features to distinguish babies
from adults. The ratio of the distance between the eyes to the
distance between the eyes and nose worked best for
separating babies from adults. After that, seniors were
distinguished from young adults and babies by detecting and
measuring skin wrinkles using snakelets. Finally, a fusion
rule was used to combine the ratios and wrinkle information
to judge age category. Their method was tested on a database
of only 47 high resolution face images containing babies,
young and middle aged adults. Kwon and Lobo reported
100% classification accuracy on these data. Several issues
are pertinent to their research. First, the method was tested
on only 47 images, making it difficult to evaluate. Second,
the method required a relatively high resolution image (at
least 256 x 256) to perform wrinkle analysis. Realizing the
shortcomings of Kwon and Lobo, Horng et al. [8] proposed a
faster and more robust system for age group classification.
Their method was composed of the same fundamentals but
different approaches to solving geometric ratios and wrinkle
analysis. Neural Networks were employed to perform
classification, achieving 81.6% accuracy for 230
experimental images.
In this paper 2-D Gabor wavelet is used for extracting the
face features and feed forward back propagation neural
network is used for classification of face images into three
age groups: child, young and adult. Face images are taken
from FG-NET aging database [9] for training and testing of
age classification algorithm.
III. GABOR WAVELET & NEURAL NETWORK
A. Gabor Wavelet
It is originally introduced by Dennis Gabor [10] for 1-D
signals and Daugman [11], [12] extended the Gabor filter to
2-D, have same physiology as that of receptive field receptor
profiles of mammalian visual systems. Gabor filter responses
are widely and successfully used as features in many
computer vision tasks, such as in texture segmentation, face
detection and recognition, and iris recognition. In a typical
feature construction the Gabor filters are utilized via multi-
resolution structure, consisting of filters tuned to several
different frequencies and orientations. Multiresolution
structure of Gabor filters is similar to wavelets, but it lacks
the important orthogonality property. Gabor filter do not
form a basis. They form instead a frame, which is a
generalization of the basis without orthogonality and unique
dual transform properties. However, a major disadvantage of
Gabor filter is computational heaviness. Fortunately, many
arithmetic tricks exist which can be employed to
significantly improve the computational complexity with
negligible loss in accuracy.
Gabor filters are principally band pass filters used for
feature extraction. Gabor filters are similar to a STFT or
windowed Fourier transform, satisfies the lower-most bound
of the time-spectrum resolution Gabor filters have both
frequency-selective and orientation selective and have
optimal joint resolution in spatial and frequency domains.
2-D Gabor filter is a sinusoidal plane of particular
frequency and orientation modulated by a Gaussian
envelope. Typical Gabor feature, such as Simple Gabor
feature space, consists of responses calculated with Gabor
filters at several different orientations and scales
(frequencies): a filter bank. Bank of filters is used with
different orientations so as to extract frequency information
and hence the features at different orientations since all facial
features are not present at the same orientation. Scaling is
done at each orientation so as to get maximum frequency
information at each orientation i.e. orientation and scaling
helps in extracting maximum frequency information.
A 2-D Gabor kernel is given by,
( )
2 2 2
2 ,
,
, 2
/ 2
,
2
( )
k z
ik z
k
z e e e
v
v
v o
o
v
o

( =

(1)
where and define the orientation and scale of the
Gabor kernels , ( , ) z x y = , . denotes the norm operator,
and the wave vector , k v is defined as follows:

,

i
e k k
|
v v = (2)
where

max k k f
v
v = and 8 | t = . max k is the
maximum frequency, and f is the spacing factor between
kernels in the frequency domain.
A filter bank consisting of several filters needs to be used
because relationships between responses provide the basis
for distinguishing objects. The selection of discrete rotation
angles is such that the orientations must be spaced uniformly.
{ }
2
, 0,1,....., 1 n
n
t
| = = (3)
where | is the
th
orientation and is the total number
of orientations to be used. The computation can be reduced
to half since responses on angles [ 2 ] t t are complex
conjugates of responses on [0 ] t in a case of a real valued
input.
Frequency selection is given by,
{ }
-
max , 0,1,....., 1 m k k f
v
v v = = (4)
Useful values for f include 2 f = for octave spacing
and 2 f = for half-octave spacing.
The Gabor kernels in (1) are all self-similar since they
can be generated from one kernel, the mother wavelet, by
scaling and rotation via the wave vector , k v . Each kernel
is a product of a Gaussian envelope and a complex plane
wave, while the first term in the square brackets in (1)
determines the sinusoidal part of the kernel and the second
term is used to get a zero DC response i.e. to make Gabor
filter insensitive to background luminance level by
subtracting the output of a low-pass Gaussian filter. o
determines the ratio of the Gaussian window width to
wavelength.
Gabor wavelets of five different scales, { } 0,1, 2, 3, 4 v = ,
and eight orientations, { } 0,1, 2, 3, 4, 5, 6, 7 = are used for
face feature extraction. Therefore a filter bank consists of 40
filters. Fig.1 shows the real part of the Gabor kernels at five
scales and eight orientations with half octave frequency
spacing in radial direction and one octave spatial bandwidth ,
for o t = , max / 2 k t = and 2 f = .

B. Neural Network
Feed-forward back propagation neural network is used
for classification .It consists of three layers of neurons i.e.
input layer,hidden layer and output layer.These three layers
are connected to each other.Input layer has neurons equal to
number of features for each face image, ouput layer has
three neurons equal to number of output classes as shown
in Fig. 2.

Figure1. Real Part of Gabor filters correspond to 5
spatial frequencies and 8 Orientation
(Time Domain)

Figure 2. Neural Network

IV. PROPOSED METHOD

Fig. 3 Block Diagram of Age Classification System

Age classification system shown in Fig. 3 consists of face
detection, feature extraction and age classifier. Face
detection is done using skin color feature [19]. Age feature
extraction is done using Gabor filters. Lot of redundant
features are generated in feature extraction therefore
dimension reduction is done by simply deleting some rows
and columns of feature vector. This is not a standard method
for dimension reduction but results are good and this process
requires less time. Classification is done using neural
network.
For feature extraction Gabor kernels are convolved with
each input face image.
, ( ) ( ) ( ) H z z I z v = - (5)
where ( ) H z is convolution result, , ( ) z v is Gabor
kernel, ( ) I z is input face image, - denotes the convolution
operator and ( , ) z x y = . Fig. 4 shows the result of convolving
input face image with real part of Gabor filter.

Figure 4. Output of convolving face image with real part of Gabor filter

Figure 5. Steps involved in feature extraction

Fig. 5 shows the steps involved in feature extraction.
Input color face image is cropped and scaled to 27 18 and
converted into gray scale image. This image is convolved
with Gabor filter bank of 40 filters with each filter of size 27
18. The result of convolution is stored in cell with 40
matrices each of size 2718.These 40 matrices are
concatenated to form a matrix of size 135144 .This matrix
is resized to 4548 by deleting some rows and columns
randomly. This matrix is normalized and converted into a
vector of size 2160 1.This vector is the feature vector of
input face image.
V. EXPERIMENTAL RESULTS
The FG-NET (Face and Gesture Recognition Research
Network) Aging Database [9] is a public available age
database used for the experiments. The database contains
1002 high-resolution color or grey scale face images with
large variation of lighting, pose, and expression. There are
82 subjects (multiple races) in total with the age ranges from
0 to 69 years. Only color images are used from database.
A face detector is used to find the face area in each
image; the face images are cropped and resized to 27 18
and converted to gray scale. These images are applied to the
feature extraction algorithm as explained earlier. For feature
extraction Gabor wavelet and DWT are used and there
results are compared as shown in Table I & II. For DWT,
input face images from database are cropped and resized to
64 64 and converted to gray scale and decomposition is
done up to level 1. Extracted features are applied to neural
network classifier for classifying the input face image to any
one of three age groups: child, young and adult. Neural
network classifier is a feed forward back propagation neural
network. Input layer with 2160 neurons for Gabor wavelet
and 4096 neuron for DWT, hidden layer with 60 neurons
and output layer with 3 neurons. 20 images of each age class
are used for training and 10 images of each class are used
for testing. In training step of neural network parameters
like no. of epochs, no. of neurons in hidden layer, transfer
function for hidden and output layer etc. are selected
experimentally. In testing step features of face images that
are not used for training is given as input to the trained
neural network for classification. Experiments on male and
female database are done separately. Table I and II shows
experimental results of Age classification on both male and
female database respectively. This shows that database
using Gabor wavelet and neural network for both male and
female efficiency is 86.66% and 83.33% respectively.
TABLE I. EXPERIMENTAL RESULTS FOR MALE DATABASE
Age
Class
DWT Gabor
DWT
+
NN
Gabor
+
NN
Child 80% 70% 80% 90%
Young 70% 80% 70% 80%
Adult 60% 90% 80% 90%
Total 70.00% 80.00% 76.66% 86.66%

ADULT CHILD YOUNG CHILD ADULT
TABLE II. EXPERIMENTAL RESULTS FOR FEMALE DATABASE
Age
Class
DWT Gabor
DWT
+
NN
Gabor
+
NN
Child 70% 80% 80% 90%
Young 60% 70% 70% 80%
Adult 60% 80% 60% 80%
Total 66.66% 76.66% 70.00% 83.33%

CHILD ADULT CHILD YOUNG ADULT
VI. CONCLUSION
A novel approach is proposed to classify the human face
images into three groups: child, young and adult by using
2-D Gabor wavelet for facial feature extraction and feed
forward back propagation neural network for classification.
The proposed method gives better efficiency than the other
methods such as DWT, Gabor, DWT and Neural Network.
ACKNOWLEDGMENT
The author would like to thank Dr. M. B. Kokare for his
valuable guidance. Also the author would like to thank the
FG-NET consortium for providing the FG-NET Aging
Database.
REFERENCES
[1] D.M. Burt and D.I. Perett, Perception of Age In Adult Caucasian
Male Faces: Computer Graphic Manipulation of Shape and Color
Information, In Proc. Royal Society London, Series B, Volume:
259, Page(s): 137-143, 1995.
[2] B. Tiddeman, M. Burt, and D. Perret, Prototyping and
Transforming Facial Texture For Perception Research, IEEE
Computer Graphics and Applications, vol. 21, no. 5, pp. 42-50,
Sept. /Oct. 2001.
[3] N.Ramanathan and R. Chellappa, Modeling Age Progression in
Young Faces, Proc. IEEE Conf. Computer Vision and Pattern
Recognition, pp. 387-394, 2006.
[4] N. Ramanathan and R. Chellappa, Face Verification Across Age
Progression, Proc. EEE Conf. Computer Vision and Pattern
Recognition, pp. 462-469, 2005.
[5] A. Lanitis, C.J. Taylor and T.F. Cootes, Toward Automatic
Simulation of Aging Effects on Face Images, IEEE Transactions on
Pattern Analysis and Machine Intelligence, Volume: 24 Issue:
4,Page(s): 442 - 455, April 2002.
[6] A. Lanitis, C. Draganova, and C. Christodoulou, Comparing
Different Classifiers for Automatic Age Estimation, IEEE Trans.
Systems, Man, and Cybernetics B, vol. 34, no. 1, pp. 621-628, 2004.
[7] Y.H. Kwon and N. da Vitoria Lobo, Age Classification from Facial
Images, Computer Vision and Image Understanding, Volume: 74,
Page(s): 1-21, April 1999.
[8] Wen-Bing Horng, Cheng-Ping Lee and Chun- Wen Chen,
Classification of Age Groups Based on Facial Features, Journal of
Science and Engineering, vol. 4, no. 3, pp. 183-192, 2001.
[9] The FG-NET aging database available on:
http://www.fgnet.rsunit.com.
[10] Gabor, D. (1946), Theory of Communication, Journal of the
Institute of Electrical Engineers, 93, 429457.
[11] Daugman J. (1985), Uncertainty Relation for Resolution in Space,
Spatial Frequency, and Orientation Optimized by Two-dimensional
Visual Cortical Filters, Journal of the Optical Society of
America-A, 2(7), 11601169.
[12] Daugman J.G. (1988),Complete Discrete 2-D Gabor Transforms by
Neural Networks for Image Analysis and Compression, IEEE
Transactions on Acoustics, Speech, and Signal Processing 36 (7),
11691179.
[13] G.J. Edwards, A. Lanitis, and C.J. Cootes, Statistical Face Models:
Improving Specificity, Image Vision Computing, vol. 16, no. 3,
pp. 203-211, 1998.
[14] X. Geng, Z.H. Zhou, Y. Zhang, G. Li, and H. Dai, Learning from
Facial Aging Patterns for Automatic Age Estimation, in Proc.ACM
Conf. Multimedia, 2006, pp. 307316.
[15] V. Vapnik, Statistical Learning Theory. John Wiley and Sons, 1998.
[16] Chengjun Liu, Harry Wechsler, Gabor Feature Based Classification
Using the Enhanced Fisher Linear Discriminant Model for Face
Recognition, IEEE Transaction on Image Processing, vol.11, no.4,
April 2002.
[17] Guodong Guo, Yun Fu, Charles R.Dyer, Thomas S.Huang: Image-
Based Human Age Estimation by Manifold Learning and Locally
Adjusted Robust Regression. IEEE Transaction on Image
Processing, vol.17, no.7, July 2008.
[18] Xin Geng, Zhi-Hua Zhou, Kate Smith-Miles: Automatic Age
Estimation Based on Facial Aging Patterns. IEEE Transaction on
Pattern Analysis and Machine Intelligence, vol.29, no.12, December
2007.
[19] Y. T. Pai, S. J. Ruan, M. C. Shie and Y.C. Liu, A Simple and
Accurate Color Face Detection Algorithm in Complex
Background",IEEE 2006 International Conference on Multimedia
and Expo (ICME 2006), July 2006, pp.1545-1548.
[20] Feng Gao,Haizhou Ai :Face Age Classification on Consumer
Images with Gabor Feature and Fuzzy LDA Method, work
supported by National Science Foundation of China under grant
no. 60673107, and also supported by a grant from HP Corporation.
[21] J.M. Zurada, Introduction to Artificial Neural Systems, West
Publishing Company, St. Paul, 1992.
[22] Minsky M., Papert S.: Perceptons An Introduction to
Computational Geometry, MIT Press, Cambridge, Massachusetts,
1969.

India.
ISBN: 978-81-920575-9-0 :: doi: 10. 72997/ISBN_0768
LOW POWER COMPRESSED CGRA ARCHITECTURE BASED
PROCESSOR DESIGN
Kishore S., M.E.(VLSI DESIGN),
Veltech Multitech Dr.Rangarajan Dr.Sakunthala Engineering College, Chennai

Abstract - This paper proposes a novel technique of
dynamic context management scheme for coarse-
grained reconfigurable compressed architecture, which
effectively reduces the power consumption and speedup
reconfigurable process. frequent memory-read
operations for dynamic reconfiguration cause much
power consumption. Thus, reducing power in
configuration cache has become critical for CGRA to be
more competitive and reliable for its use in embedded
systems. However, the existing reconfigurable
architectures require too much area and power. In this
paper, we propose a new design space exploration flow,
optimizing CGRA to reduce area and power with
enhancing performance for digital signal processing
(DSP) application domain Such a design flow is based
on pipelining and sharing of area/delay-critical
resources in the processing element array.

Key words-Coarse-grained reconfigurable architecture
(CGRA),configuration cache, reconfigurable
computing, context register, low power, embedded
system.

I. INTRODUCTION

An application specific architecture solution is
too rigid, and a general purpose processor solution is
too inefficient. Neither general purpose processors
nor application specific architectures are capable of
satisfying the power and flexibility requirements of
future mobile devices. Instead, we want to make the
machine fit the algorithm, as opposed to making the
algorithm fit the machine. This is the area of
reconfigurable computing systems.

This paper proposes a scheme of dynamic context
management aiming to minimize the reconfiguration
overhead by reusing and switching contexts. The
technique permits background loading of
configuration data without interrupting the regular
execution. It prevents read/write operation for
redundant part of context words dynamically and
overlaps computation with reconfiguration. And
stored configurations can be switched dramatically
reducing reconfiguration overhead if the next
configuration is present in one of the alternate
contexts. In this paper, we address the power
reduction issues in CGRA and provide a framework
to achieve this. A new design flow and a new
configuration cache structure are presented to reduce
power consumption in configuration cache. The
power saving is achieved by dynamic context
compression in the configuration cacheonly
required bits of the context words are set to enable
and the redundant bits are set to disable. Therefore,
the new design flow for CGRA has been proposed to
generate architecture specifications that are required
for supporting dynamically compressible context
architecture without performance degradation. The
validation of the proposed approaches is
demonstrated through the use of real application
benchmarks and gate level simulations. The proposed
CGRA architecture has shown to reduce power
consumption by up to 39.72% in configuration cache
compared to conventional context architecture with
negligible area overhead of only 2.16%.

II. FINE GRAIN VS. COARSE GRAIN

Reconfigurable processors have been widely
associated with Field Programmable Gate Array
(FPGA)-based system designs. An FPGA consists of
a matrix of programmable logic cells with a grid of
interconnecting lines running between them. In
addition, there are I/O pins on the perimeter that
provide an interface between the FPGA, the
interconnecting lines and the chip's external pins.
However, FPGAs tend to be somewhat fine-grained
in order to achieve a high degree of flexibility. This
flexibility has its place for situations where the
computational requirements are either not known in
advance or vary considerably among the needed
applications. However, in many cases this extreme
level of flexibility is unnecessary and would result in
significant overheads of area, delay and power
consumption.
Contrasted with FPGAs, the data-path width of
coarse grained reconfigurable architectures is more
than one bit. Over the last 15 years, many projects
have investigated and successfully built systems
where the reconfiguration is coarse-grained and is
performed within a processor or amongst processors.
In such systems the reconfigurable unit is a
Specialized hardware architecture that supports logic
reconfiguration. The reconfiguration procedure is
much faster than that found in FPGAs. Because the
application domain is known, full custom data paths
could be designed, which are drastically more area-
efficient. Typically, a CGRA consists of a main
processor, reconfigurable array architecture (RAA),
and their interface, as shown in Fig. 1. The RAA has
identical PEs containing functional units and a few
storage units such as ALU, multiplier, shifter, and
register file. The data buffer provides operand data to
PE array through a high-bandwidth data bus. The
configuration cache (or context memory) stores the
context words used for configuring the PE array
elements. The context register between a PE and a
cache element (CE) in configuration cache is used to
keep the cache access path from being the critical
path of the CGRA.

Fig 1 General block diagram of CGRA

III. CONTEXT ARCHITECTURE

The configuration cache provides context words
to the context register of each PE on a cycle-by-cycle
basis. From the context register, these context words
configure the PEs. Fig. 2 shows an example of PE
structure and context architecture for MorphoSys
[12]. Thirty-two bit context word specifies the
function for the ALU-multiplier, the inputs to be
selected from MUX_A and MUX_B, the amount and
direction of shift of the ALU output, and the register
for storing the result as Fig. 2(a). Context architecture
means organization of context word with several
fields to control resources in a PE, as shown in Fig.
2(b). The context architectures of other CGRAs such
as are similar to the case of MorphoSys although
there is a wide variance in context width and kind of
fields used by different functionality. For particular
coarse-grained reconfigurable array, the performance
improvement depends on the inherent parallelism of
target applications. In many cases, the parallelism of an
application is smaller than the number of processing
element (PE) arrays. Accordingly, PE array shows lots
of redundant PEs not used at runtime.
If the configuration cache can provide only
required bits (valid bits) of the context words to PE
array at runtime, it is possible to reduce power
onsumption in configuration cache. The redundant
bits of the context words can be set to disable and
make these invalid at runtime. That way, one can
achieve low-power implementation of CGRA without
performance degradation while context architecture
dynamically supports both the cases at runtime: one
case is uncompressed context word with full bit
width and another case is compressed context word
with setting unused part of configuration cache
disabled. In order to support such a dynamic context
compression, we propose a new context architecture
and configuration cache structure in this paper.

Fig. 2. PE structure and context architecture of MorphoSys. (a) PE
structure. (b) Context architecture.

IV. COMPRESSED ALU ARCHITECTURE

In order to design and evaluate compressible
context architecture, we propose a new context
architecture design flow. This design starts from
context architecture initialization, which is similar to
the architecture specification stage of general CGRA
design flow. If the configuration cache can provide
only required bits (valid bits) of the context words to
PE array at runtime, it is possible to reduce power
consumption in configuration cache. The redundant
bits of the context words can be set to disable and
make these invalid at runtime. That way, one can
achieve low-power implementation of CGRA without
performance degradation. One more method to
reduce power is by dividing the ALU into two parts
thats in compressed and non-compressed method.
The normally repeating instructions are executed in
compressed mode and instructions that are frequently
not used such as branching instructions are executed
in non-compressed mode.The normal instructions are
add, sub etc are executed in compressed mode.In
compressed mode we use a carry look ahead adder
for execution the normal instructions. The use of
carry look ahead adder is helpful in increasing the
speed and saving power than that of ripple carry
adder. Also it reduces area than the ripple carry
adder.

Fig 3. Compressed ALU Architecture

Fig 4. Shows the output obtained using the logic
based on the compressed alu idea explained in this
paper.

Fig 4.Output

V. CONCLUSION
Power consumption is very crucial for the
CGRA for embedded systems and all reconfigurable
architectures have a configuration cache for dynamic
reconfiguration, which consumes significant amount
of power. In this paper, we introduced new context
architecture with its design flow. We are compressing
the architecture in each instruction cycle from the
decoding section of the processing of a instrtuction.
The exploration flow efficiently rearranges PEs with
reducing array size, and changes interconnection
scheme to save area and power. In addition, we
suggest the design scheme which splits the
computational resources into two groups (primitive
resources and critical resources). Primitive resources
are replicated for each processing element of the
reconfigurable array, whereas area-critical resources
are shared among multiple basic PEs.

REFERENCES

[1] H. Reiner, A decade of reconfigurable computing: A visionary
retrospective, in Proc. Des. Autom. Test Eur. Conf., Mar. 2001,
pp. 642649.

[2] H. Reiner, M. Herz, T. Hoffmann, and U. Nageldinger,
KressArray Xplorer: A new CAD environment to optimize
reconfigurable datapath array architectures, in Proc. Asia South
Pacific Des. Autom. Conf., Jan. 2000, pp. 163168.

[3] B. Mei, S. Vernalde, D. Verkest, and R. Lauwereins, Design
methodology for a tightly coupled VLIW/reconfigurable matrix
architecture: A case study, in Proc. Des. Autom. Test Eur. Conf.,
Mar. 2004, pp. 12241229.

[4] N. Bansal, S. Gupta, N. Dutt, and A. Nicolau, Analysis of the
performance of coarse-grain reconfigurable architectures with
different processing element configurations, presented at
theWorkshop Appl. Specific Processors, San Diego, CA, Dec.
2003.

[5] A. Lambrechts, P. Raghavan, and M. Jayapala, Energy-aware
interconnect- exploration of coarse-grained reconfigurable
processors, presented at the Workshop Appl. Specific Processors,
New York, Sep.2005.

[6] R.Hartenstein. A decade of reconfigurable computing: A
visionary retrospective. In Proc. DATE, pages 642-649, 2001.

[7] Hideharu Amano, Yohei Hasegawa, Satoshi Tsutsumi
"MuCCRA chips: Configurable Dynamically-Reconfigurable
Processors" IEEE Asian Solid-State Circuits Conference
November 12-14,2007 / Jeju, Korea [3] Sudang Yu, Leiba
Liu,"Automatic Contexts Switch in Loop Pipeline for Embedded
Coarse-grained Reconfigurable Processor" Communications,
Circuits and Systems, 2008, ICCCAS 2008, International
Conference.

[8] F.Veredas,M.Scheppler,W.Moffat, and B.Mei, "Custom
Implementation of the Coarse-Grained Reconfigurable ADRES
Architecture for Multimedia Purposes," in Proc. ofFPL, Aug. 2005,
pp. 106-111.
India.
ISBN: 978-81-920575-9-0 :: doi: 10. 73004/ISBN_0768
Mathematical Heuristics for Feed Forward Artificial Neural Network Design
Aakash Goel
CSIS, Birla Institute of Technology and Science
Pilani, Rajasthan, India - 333031
Vandit Bhatia
CSIS, Birla Institute of Technology and Science
Pilani, Rajasthan, India - 333031

AbstractTime series data prediction is complex because of
the large number of variables associated, and limited
computational power available. Traditional prediction
methods are slow and thereby infeasible for any real-time
predictions. We study the problem from the ground up and
present a solution to the major impediments in deciding the
neural network configuration. A mathematical heuristic based
method to design a neural network for predicting time series is
presented. The method described has been applied and tested
on stock and commodity prices.

Keywords-neural networks; artificial intelligence; forecast;
time series; commodity price prediction; stock price prediction;
feedforward network; backpropagation
I. INTRODUCTION
Artificial Neural Networks mimic the structural and
functional aspects of human brain's efficient neural network
model. They can find subtle mathematical relationships
among data and adapt themselves to surface trends. A
computing system modeled on a similar approach can adapt
itself to the past results by changing its structure and
enhancing its own accuracy.
Neural Network learning is a function optimization
problem at its core. The aim is to determine the network
configuration parameters that minimize the errors. Once the
network is trained, it can predict data points similar to what
it was trained on. The accuracy of prediction will depend on
factors including amount of training, number of adaptive
layers in the network, and input configuration. The problem
with these systems is the sheer number of variables
associated [1], which make them highly impractical to use
for any real time purpose, like weather forecasts,
commodity prices and stock markets etc.

With numerous variables, it comes down to brute force
while trying out the configurations to find out which one
works the best. However, the complexity of this
optimization problem can be reduced mathematically. We
here present the findings of our attempt to tackle this very
problem. A step by step method of choosing a network
configuration is discussed.
II. CONTENT
A. Assumptions
Without any loss of information or generality, for
simplicity, we have assumed in this study that
1. The value of a time series at a future point
in time is dependent solely on its past values.
2. Each neuron in the first layer will have the
same number of inputs. Using different number of
inputs is an added complexity with no known
benefits.

B. Structure of our Neural Network

Figure 1 neural network structure

There are four elements that make up a neural network:
1. number of layers in the system
2. number of nodes in each layer
3. mapping of inputs to the nodes
4. transfer and training functions used

Each of these is discussed in following subsections.

1) Number of layers in the system
The first layer of a network is the input layer and the last
layer is the result or output layer. These layers are special
because they not only interface with neural network nodes,
like other layers, but also with the actual data. The hidden
layers, however, only interface with other nodes. The total
number of layers in a system has a huge impact on the
behavior and accuracy of the network. Number of layers can
be thought of as the 'length' of the network. Specifically,
more the number of hidden layers,
1. more time is taken in training the network
as more variables need to be trained
2. lesser the speed of prediction, even after
the network is trained
3. more the chances of the network being
overtrained
The network, when overtrained, adapts the weights to
account even for the errors specific to the training set, in
addition to learning only the basic mapping, associated with
input and output data. The resulting network performs well
only on the data that it was trained on, but shows poor
results on any other dataset.
Two layer networks are widely used primarily because
of the ease of training, and speeds of prediction. In
discussion that follows, in order to focus on a solution that
can be used for real time prediction using minimal
computing resources, we have limited ourselves to a two
layer network.
2) Number of nodes in each layer
Number of nodes in each layer decides the 'breadth' of
the network. The output layer has only one node because the
value of the time series at a future point in time is the only
monitoring point. Since we are focusing on a two layered
network, we only need to decide how many nodes to use in
the first layer. The number of nodes in the network has a big
impact on the accuracy of the prediction, but at the same
time a large number of nodes also increases the chances of
over-training the network. Generally, more the number of
nodes in the first layer,
1. more learned the network becomes
2. more accurate the network becomes
3. greater are the chances that any factors
which would randomly affect the data, cancel out
3) Mapping of inputs to the nodes
For all the hidden layers, the number of inputs and
number of outputs are predetermined by the layers around
them. For the first layer, we can adjust this number to decide
how inputs are mapped to the nodes. Generally, more the
number of inputs to each node in first layer
1. better trained the network becomes
2. more time it takes to train the network
3. more fault tolerant the network becomes;
meaning that there is a less chance that any outliers
will disturb the stability of our system
4) Transfer and Training Functions used
We have used tansig, hyperbolic tangent sigmoid
transfer function, as our training function for the network.
This is the most widely used transfer function for
backpropagation neural networks [4,16]. The sigmoid
function is represented by the mathematical relationship
1/(1+e
-x
). The hyperbolic counterpart to the sigmoid
function is the hyperbolic tangent function, which is similar
but can exhibit different learning dynamics during training,
including accelerated learning for some models, and also
affecting predictive accuracy. Experimenting with the
transfer functions for each individual model is the only
conclusive method to determine whether a transfer function
will offer both good learning and accuracy characteristics. A
general rule of thumb is that sigmoid produces the most
accurate mode, albeit it takes longer [3].
C. Method
The method described here is based on mathematical
foundations of statistical relationship between data points.
The current values depend on the past values in a
mathematically definable manner. Our goal is to find this
relationship and use it as a trend. The four-step process to
train a neural network, with gold prices data as an example,
is described below.
1) Fetching Time Series
The first step is to obtain the past data points for the time
series we want to predict. This can be done in two ways -
using absolute prices as data points or using relative change
in prices as data points. Our results have shown that relative
data points do not show good results [III.B]. Instead
absolute prices are normalized [III.A] and adjusted to take
into account the time range, and these work well with
mathematical models. The data set used is:
Absolute gold prices for 40 years,
beginning January 1, 1970, were obtained [5], and
normalized to 8192 data points.

Figure 2 line plot of gold prices over time
2) Autocorrelation Analysis
Autocorrelation is a mathematical method to identify
hidden patterns in a time series by identifying the degree to
which the values at one point in time V
t
are linearly
associated to the values at different point of time V
T
, as a
function of time t or the time difference (T-t). Therefore,
autocorrelation can be expressed as a function of time-lag.
Its value lies between -1 and 1, where values close to 1
denote a strong positive correlation.
Since we are training the network on the same series of
data which we want to predict, the value of lag needs to be
decided. 'lag' is defined as the relative time difference
between the input and output data points. If we have n data
points, our range of lag values is [1:n].
Using autocorrelation is the first step in refining the
range of lag values to train our network. Autocorrelation on
the data values collected in previous step is calculated for a
full range of lag values, and plotted as a graph to bring out
the properties. The lag range is restricted at the point where
the first good minimum is seen on the autocorrelation graph.
Conversely, if no minimum is found on the autocorrelation
graph, we restrict the lag to the minimum value where the
curve reaches zero.

Figure 3 autocorrelation is minimum at 1090 points

In this case, such a point occurs at 1090, and hence the
range of lag values becomes [1:1090]. Note that we do not
necessarily need to look for the absolute minimum on the
graph, but only for a local minimum with neighboring
points limiting to the same values. It is costly to train a
network with a high lag value [7,9].
3) Autoregression Analysis
In the previous step, autocorrelation gives us a
condensed range of lag values to work with. In this step, we
aim to further shorten this range. Using linear regression to
determine the relationship between the data points obtained
in the previous section, we see that a clear minimum occurs
at 77. Autoregression should be used carefully, since it is
possible to have no clear minimum for a data set, and in that
case we should consider autocorrelation as the defining
metric. For this example, the huge range of data points has
now been reduced to a small [1:77] range. This gives us the
freedom to use the computational power available for
targeted efforts towards training in a small range.

Figure 4 arithmetic absolute mean and root mean
square error plots for autoregression analysis shows a
minimum at a lag value of 77
4) Training
Having trimmed the training range by 93% for this
dataset, the next step is to train the network. This involves
preparing a training environment and choosing the
appropriate number of training points. To train the network,
we have used MATLABs Neural Network Toolbox [13],
although various other options are available. It does not
make much sense to train on all the 77 points available.
Rather, we focus on choosing a good distribution of these
points. We select {3, 5, 7, 11, 15, 20, 25, 30, 50, 80, 105} to
demonstrate the concept. Each of these points are trained
over the same set of {3, 5, 7, 11, 15, 20, 25, 30, 50, 80, 105}
N:1 network configurations, which means 121
configurations are trained. Our aim here is to choose only 1
out of these 121 training configurations, to be used as our
final prediction network.
The training is done 100 times each to make sure that the
trained values are strongly optimized.
5) Results
Results are summarized in the following table:
A. Number of layers in the first layer, i.e. X in X:1
network
B. Number of inputs to each neuron in the first layer
C. Arithmetic Mean error after training 100 times
D. Root Mean Square error after training 100 times
E. Average time taken to train in seconds

Table 1
A B C D E
3 3 64.058373 72.937403 52.761383
3 5 52.801828 61.566203 60.672377
3 7 63.137499 73.061638 55.282218
3 11 64.406633 75.409743 49.696921
3 15 62.971018 72.215519 52.996465
3 20 63.136811 73.897121 54.188111
3 25 91.123199 121.21388 63.182732
3 30 59.173213 70.127381 65.123839
3 50 52.761383 68.891729 98.716211
3 80 63.122224 76.988997 91.876221
3 105 50.187388 60.912112 88.714113
5 3 65.476576 75.578657 35.992023
5 5 53.645759 61.839732 20.9756
5 7 69.190414 79.96803 25.711641
5 11 62.842623 71.615146 19.253819
5 15 50.840191 59.691703 22.653489
5 20 60.278351 70.958331 19.91673
5 25 60.80465 71.454713 21.616475
5 30 56.030011 65.289571 26.29414
5 50 46.647964 54.573395 34.768715
5 80 46.830645 58.180139 35.405876
5 105 56.618929 68.073364 37.231074
7 3 99.590535 112.405465 17.406199
7 5 83.516799 93.929783 14.580129
7 7 64.333939 74.990874 13.904058
7 11 79.741741 90.648792 14.685388
7 15 49.844658 58.211306 14.062603
7 20 76.421303 88.203641 14.364573
7 25 75.782031 86.138902 18.035175
7 30 71.069109 82.753432 21.681789
7 50 71.171946 82.417079 22.66282
7 80 70.661231 83.016256 31.812667
7 105 72.123551 85.881191 28.827132
11 3 68.240894 78.007037 12.395941
11 5 97.749137 109.673272 8.885981
11 7 90.953448 103.800556 9.665743
11 11 67.615102 76.772765 11.707351
11 15 82.702772 94.672651 11.065689
11 20 102.150101 115.96265 12.138648
11 25 51.895601 60.312444 17.574037
11 30 48.309852 57.622438 20.353891
11 50 71.501027 82.749908 30.157576
11 80 78.970984 95.9293 60.868444
11 105 71.007809 83.2744 94.634986
15 3 128.996396 146.000142 7.244708
15 5 105.966477 121.586333 8.481663
15 7 88.148095 99.888281 8.197472
15 11 58.763395 67.512271 12.043291
15 15 62.572929 71.5883 13.371354
15 20 72.167468 85.143136 16.029325
15 25 77.247509 88.819669 18.891057
15 30 76.98312 90.063183 22.803532
15 50 81.647629 94.910865 34.384787
15 80 69.062664 80.589544 81.744294
15 105 67.65931 80.317575 114.661788
20 3 131.40366 149.967446 3.989495
20 5 111.747183 128.919533 4.467168
20 7 75.682607 87.851128 5.27315
20 11 62.493225 74.62496 7.586665
20 15 92.792637 106.762297 7.36797
20 20 58.805897 69.279191 10.973172
20 25 73.472862 84.313976 12.548845
20 30 83.734198 97.968049 18.878972
20 50 76.414005 90.170984 37.799556
20 80 73.728414 89.070481 57.444669
20 105 87.364901 104.656269 56.488825
25 3 118.246119 137.98397 4.537291
25 5 115.783342 131.617816 4.976524
25 7 95.603223 109.24571 5.568128
25 11 71.684276 84.275781 8.281219
25 15 67.519255 80.170343 11.409522
25 20 71.295617 82.78931 14.31234
25 25 104.471715 119.136509 17.483916
25 30 76.144007 88.076986 23.551046
25 50 75.758601 89.450342 42.767205
25 80 113.361166 133.263816 55.989393
25 105 123.322179 145.059979 62.968225
30 3 120.128785 138.449159 3.95824
30 5 115.746547 133.053803 5.224467
30 7 98.085685 110.989434 5.175894
30 11 89.641507 105.321227 7.695538
30 15 118.540043 132.042454 9.987406
30 20 88.440777 100.786733 14.057005
30 25 95.108972 108.771116 19.685758
30 30 99.561092 113.395305 26.064313
30 50 81.224136 96.399093 51.647079
30 80 80.782137 97.126322 54.716882
30 105 105.877882 129.187223 71.834723
50 3 139.847492 158.53748 5.783928
50 5 88.82151 105.572644 9.659026
50 7 81.731867 96.960944 8.776727
50 11 127.135102 145.180957 12.407364
50 15 94.093821 104.377353 15.984795
50 20 81.435468 93.757256 24.614266
50 25 105.433283 119.130803 34.131767
50 30 130.621803 150.332241 56.752297
50 50 150.132894 171.812632 63.817323
50 80 108.123871 132.237892 82.283179
50 105 98.761662 128.213788 79.983224
80 3 102.612613 119.377828 11.218344
80 5 106.75996 120.527003 10.12673
80 7 96.748146 109.653981 17.30835
80 11 98.105181 108.460675 17.455198
80 15 135.137897 150.872628 30.1918
80 20 120.823145 135.960956 41.018839
80 25 122.429877 139.35457 49.223802
80 30 132.92682 152.068892 58.927672
80 50 146.372973 166.368025 121.489073
80 80 146.372973 166.368025 121.489073
80 105 144.956322 160.178631 132.234552
105 3 122.940403 140.459144 14.044496
105 5 117.399806 133.050817 17.960893
105 7 106.691071 120.358136 18.30904
105 11 134.375532 150.085168 27.586359
105 15 116.368113 130.116786 39.593592
105 20 149.774159 166.619761 54.600875
105 25 158.237268 178.681186 59.92438
105 30 186.73764 212.455809 67.665644
105 50 199.447561 228.721311 97.764047
105 80 190.748288 212.127787 107.728221
105 105 188.128312 219.872211 165.896532

6) Configuration Selection
The training data with errors averaged over 100 runs
provides good statistical base to choose the optimum
configuration. Our aim is to minimize the errors while
keeping the training time reasonable.

From the results above, 5:1, with 50 inputs, is one of the
most accurate network configurations, with an acceptable
average training time of 34.8 seconds.
III. TRAINING RESULTS FOR FOR YHOO
Here, we have demonstrated our method with the dataset
- daily closing prices for the stock of Yahoo! Inc., listed as
YHOO on NASDAQ. Prices from the IPO date April 12,
1996 and up to August 20, 2009 are taken.
A. Normalization
Stock prices can fluctuate abruptly due to stock splits,
stock reverse splits, dividend payments, or bonus stocks
offered by the company. In a 5-for-1 stock split, 10 million
outstanding shares at $50 each would now become 50
million shares outstanding at $10 per share. In both cases,
the company is still worth $500 million [17]. These
situations can cause the stock price to change significantly,
though the stock value remains unchanged. The daily
closing prices for the stock (YHOO) are adjusted by
multiplying the stock prices with the splitting factor from
the date when the stock price was affected due to such
corporate decisions. Following were the stock dividends
given on YHOO [18].
1. 2 for 1 - May 11, 2004
2. 2 for 1 - February 14, 2000
3. 2 for 1 - February 8, 1999
4. 2 for 1 - August 3, 1998
5. 3 for 2 - September 2, 1997
Yahoo! Inc. does not give cash dividends [19]. After
adjusting for stock splits, a total of 3363 data points were
generated.
B. Absolute and Relative data sets
Another way of generating the data points is to take the
relative daily changes in the stock price as the data. Only the
data pertaining to the above mentioned dates need to be
adjusted.
C. Autocorrelation Analysis
Both the methods of generating data, as mentioned
above, are used. The relative changes in daily prices do not
show any positive autocorrelation (Figure 5). Hence, we
resort to autocorrelation on adjusted daily closing prices for
YHOO (Figure 6).

Figure 5 - autocorrelation for relative daily changes for YHOO

Figure 6 - A minimum is achieved at a lag value of 600, but for simplicity,
we consider only positive correlation values. The range of lag values here
is taken as [1:350].
D. Autoregression Analysis
Using linear regression to determine the relationship
between the data points in the range determined in the
previous section, we see that a local minimum occurs at a
lag value of 80.

Figure 7- arithmetic absolute mean and root mean square error plots for
autoregression analysis shows a minimum at a lag value of 80

The range of data points has now been reduced to [1:80].
E. Training
N:1 neural network was chosen to be trained with {4, 7, 15,
25, 35, 50, 65} data points, with {4, 7, 15, 25, 35, 50, 65,
80} inputs to each configuration. Each configuration was
trained 100 times to optimize the results.
F. Results
Results are summarized using the following table:
A. First later, i.e. X in X:1 network
B. Number of inputs to each neuron in the first layer
C. Mean error for 100 runs
D. Average Time taken to train in seconds

Table 2
A B C D
4 4 178.291109 6.845449
4 7 153.080146 4.284876
4 15 139.483301 4.072542
4 25 135.054335 4.626625
4 35 183.754112 4.842254
4 50 153.314875 4.834237
4 65 185.293036 5.622460
4 80 194.413232 4.898141
7 4 143.554341 4.834318
7 7 200.791427 3.071382
7 15 175.790194 3.152921
7 25 209.464811 3.633478
7 35 132.328143 4.781792
7 50 163.343392 5.311094
7 65 174.801690 7.493499
7 80 183.635575 10.790012
15 4 212.246016 2.754282
15 7 240.983437 3.844774
15 15 212.052935 3.707138
15 25 267.669869 5.506480
15 35 233.214922 8.471368
15 50 267.554857 13.813672
15 65 261.999518 21.148093
15 80 238.191344 34.338015
25 4 283.993137 2.837547
25 7 281.374600 2.904754
25 15 239.178522 5.119674
25 25 259.889920 11.142585
25 35 298.157645 17.092290
25 50 231.844002 40.661015
25 65 279.474281 75.592995
25 80 285.510894 102.564736
35 4 293.293833 5.676183
35 7 284.663287 5.339431
35 15 272.789741 11.644197
35 25 258.558890 22.003486
35 35 268.609620 36.880377
35 50 305.517884 86.721427
35 65 274.309392 149.152628
35 80 225.588507 300.623322
50 4 289.457048 7.668525
50 7 291.362147 9.164058
50 15 242.452930 19.932377
50 25 281.666233 32.429125
50 35 353.521047 59.425432
50 50 281.433629 139.484074
50 65 231.515438 347.425292
50 80 229.497454 666.139941
65 4 255.582739 8.085872
65 7 310.734540 10.971058
65 15 326.853050 25.845619
65 25 248.681492 67.081598
65 35 261.526714 144.085232
65 50 250.352986 284.411717
65 65 237.813826 598.447416
65 80 237.095144 1112.514843

Just as for the gold data, a minimum in the errors is
observed; here for 7:1 network with 35 inputs, with an
acceptable average training time of 4.78 seconds.
IV. REMARKS
Choosing the suitable network configuration for learning is
a crucial step for the system's prediction accuracy and the
set of heuristics described above optimize our resources by
filtering out the poor configurations. These methods have
been tested to provide good results for two different time
series - gold prices and stock prices.

The accuracy of the trained neural network is vastly affected
by the network configuration. The set of heuristics described
filter out a large number of poor configurations by
employing simple mathematical concepts of autocorrelation
and autoregression effectively, thus favoring the optimal use
of the available resources. The results for the tested datasets
of gold and stock prices are consistent and promising.
V. FURTHER SCOPE
Simple heuristic methods are powerful tools in the hands of
researchers. As we have shown, they can be used to reduce
the complexity of the problem at hand effectively, and help
focus computing resources in neural network training.
Following are a few ways that the prediction methods can
be tested for better or worse results:
1. Using different training algorithms, e.g. Gauss
Newton method, Gradient descent method, etc.
2. Using higher degree regression functions instead of
linear regression.
3. Training a network more number of times.
4. Using fuzzy logic to make the network use
continuous variables rather than discrete.
ACKNOWLEDGMENT
We would like to acknowledge and extend our heartfelt
gratitude to Dr. Himanshu Agrawal, Birla Institute of
Technology and Science, Pilani, who has made the
completion of this paper possible by guiding and motivating
us in the difficult initial stages of the project.
REFERENCES
[1] HS Hippert, CE Pedreira, Neural networks for short-term load
forecasting: A review and evaluation Proc. IEEE Transactions on
Power Systems, Feb 2001.
[2] Odom, M.D., Sharda, R., A neural network model for bankruptcy
prediction, Proc 1990 IJCNN International Joint Conference on
Neural Networks, Jun 1990.
[3] K.C. Luk, J.E. Ball*, A. Sharma, A study of optimal model lag and
spatial inputs to articial neural network for rainfall forecasting,
2000.
[4] Xindi Cai, Nian Zhang, Ganesh K. Venayagamoorthy, Donald C.
Wunsch II, Time series prediction with recurrent neural networks
trained by a hybrid PSOEA algorithm, Neurocomputing, Volume
70, Issues 13-15, August 2007, Pages 2342-2353.
[5] Daily Gold Prices - London Gold Fix Proce History,
http://www.usagold.com/reference/prices/history.html, October 2011.
[6] J.S. Armstrong, Principles of forecasting: A handbook for researchers
and practitioners, Kluwer Academic.
[7] David Mendes, Jos A. Marengo, Temporal downscaling: a
comparison between artificial neural network and autocorrelation
techniques over the Amazon Basin in present and future climate
change scenarios, Theoretical and Applied Climatology Journal,
2010.
[8] In Santa Fe Institute Studies in the Sciences of Complexity,
Proceedings of the NATO Advanced Research Workshop on
Comparative Time Series Analysis, held in Santa Fe, New Mexico,
May 14-17, 1992, Reading, MA: Addison-Wesley, |c1994, edited by
Weigend, Andreas S.; Gershenfeld, Neil A. (1994).
[9] James Devillers, EVA/PLS versus autocorrelation/neural network
estimation of partition coefficients, Perspectives in Drug Discovery
and Design Journal, 2000.
[10] Fan-Yong Liu, Hangzhou Dianzi, Forecasting Renminbi Exchange
Rates Based on Autocorrelation Shell Representation and Neural
Networks, Proc. Spring Conference, 2009. IACSITSC '09.
International Association of Computer Science and Information
Technology, April 2009.
[11] Jian Chen, Lizy K. John, Autocorrelation Analysis: A New and
Improved Method for Measuring Branch Predictability, Proc. IEEE
International Symposium on Workload Characterization . November
2011.
[12] Mehmet Kurban, Ummuhan Basaran Filik, Next Day Load
Forecasting Using Artificial Neural Network Models With
Autoregression And Weighted Frequency Bin Blocks, Proc.
International Journal of Innovative Computing, Information and
Control, April 2009.
[13] MathWorks India - Neural Network Toolbox - MATLAB,
http://www.mathworks.in/products/neural-network/index.html,
October 2011.
[14] AI FAQ/neural Nets Index, http://www.faqs.org/faqs/ai-faq/neural-
nets/, October 2011.
[15] Evolving Neural Networks Through Augmenting Topologies,
http://nn.cs.utexas.edu/?stanley:ec02, November 2011.
[16] Neural Networks - A Systematic Introduction, Chapter 7: The
backpropagation algorithm, http://page.mi.fu-
berlin.de/rojas/neural/chapter/K7.pdf, November 2011.
[17] Common stock split-upsMotives and effects JC Dolley, Harvard
Business Review, 1933.
[18] YHOO Historical Prices | Yahoo! Inc. Stock Yahoo! Finance -
http://finance.yahoo.com/q/hp?s=YHOO+Historical+Prices, 2011.
[19] Yahoo! Inc. Investor FAQs -
http://yhoo.client.shareholder.com/faq.cfm, September 2011.

India.
ISBN: 978-81-920575-9-0 :: doi: 10. 74369/ISBN_0768
IMAGE RESOLUTION ENHANCEMENT BASED ON EDGE DIRECTIONAL
INTERPOLATION IN DUEL TREE-COMPLEX WAVELET TRANSFORM
L.suresh
Computer Science and Engineering
Pondicherry Engineering College
Pondicherry, India

Proff. D. Loganathan
Computer Science and Engineering
Pondicherry Engineering College
Pondicherry, India

Abstract - Image resolution enhancement is a usable process
for many image processing applications such as geosciences
studies, astronomy and geographical information systems.
Image interpolation is popularly used to increase the image
resolution, but the potential problem associated with it is to
magnify the image many times without loss in image clarity.
Most of the classical linear interpolation techniques like
bilinear & bi-cubic interpolation methods generate blurred
images. By employing dual-tree complex wavelet transform
(DT-CWT) on a edge directional interpolation, it is possible to
recover the high frequency components which provides an
image with good visual clarity and thus super resolved high
resolution images are obtained.

Keywords: Dual-tree complex wavelet transform (DT-CWT),
resolution enhancement, Edge directed interpolation, Super
resolved images.
1 INTRODUCTION
With the recent advances in low-cost imaging solutions
and increasing storage capacities, there is an increased
demand for better image quality in a wide variety of
applications involving both image and video processing.
While it is preferable to acquire image data at a higher
resolution, practically one can imagine a wide range of
scenarios where it is technically not feasible. In some cases,
it is the limitation of the sensor due to low-power
requirements as in satellite imaging, remote sensing, and
surveillance imaging. The low-resolution data can exist in
the form of still images. Furthermore, the observations can
be corrupted by motion-induced artifacts either in the case
of still images or videos. An improvement in the spatial
resolution for still images directly improves the ability to
discern important features in images with a better precision.
Interpolation is the technique that which estimate the
new pixel from the surrounding pixels available in the low
resolution image. But the computational problem is
increased as the order of interpolation factor increases. This
problem can be resolved by interpolating the image in
wavelet domain. But, super-resolution image reconstruction
is an ill-conditioned inverse problem, since the observation
process of the original high
Resolution object consists of a noisy blurred low
resolution observation affected by sampling artifacts, blur
and noise. This is mathematically modeled as a nonlinear
process consisting of a convolution operator acting on the
image, followed by a down sampling operation and the
mixing of additive noise in the frequency domain approach
using (discrete) Fourier transform and wavelet-transform
based methods(using real wavelets).
This difficulty can be recovered by using complex
wavelets because it has the properties of shift invariant and
good directionality and by constructing the filters in the
form of dual tree structure which provide complex-valued
wavelets it eliminates the inverse problem by perfectly
reconstructing the both forward and reverse filter banks and
such a transform is referred to as called DT-CWT (dual-tree
complex wavelet transform) [5-7].
In earlier works the linear interpolations such as
nearest neighbor, bilinear and bi-cubic interpolations are
used to enhance the resolution of the image. However these
techniques produce many artifacts like blurring, blocking
etc., To avoid these problems to a greater extent, a good old
nonlinear interpolation like edge directional interpolation
[3] can be applied to images. This is the prime motivation of
using edge directed interpolation technique. However, the
major drawback of applying interpolation to achieve super
resolved image is the smoothed edges leading to blurring
effect. In one of the recent work [7], the authors have
proposed a technique which is aimed at generating sharper
edges and detailed super resolved satellite images than the
dual tree complex wavelet transform (DT-CWT) with bi-
cubic interpolation.
To the best of our knowledge, the features of edge
directed interpolation is very less exploited and hence it is
used in combination with DT-CWT in this paper and
appreciable enhancement has been observed in terms of
PSNR. This is the major highlight of this paper. The obtained
simulation results are compared with the results shown in
the very recent work [7].
This paper is organized as follows: section 2 gives a brief
review of the technique edge directional interpolation.
Section 3 describes the proposed technique DT-CWT based
satellite image resolution enhancement algorithm.. Section 5
concludes this paper.

2 .PRINCIPLE OF EDI TECHNIQUE

The researchers have so far employed linear interpolation
techniques such as nearest neighbour, bilinear and bi-cubic
interpolations to improve image resolution. But other
techniques suffer from blurring edges or from introducing
artifacts around edge area. In [7], the authors have applied
bi-cubic interpolation, which again, suffers from blocking
and blurring. This paper aims and attempts to rectify all the
limitations of the work suggested in [7], [2] and [6] by
employing a novel edge directional interpolation technique
[EDI], to obtain a high resolution image. EDI estimates the
pixel along the edges by estimating the covariance of
surrounding neighbours while it uses the image data
themselves to direct the interpolation. The edge directional
interpolation works as follows:

Fig 1 Flow chart of Edge directional interpolation

From the flow chart, the sliding windowing technique like
kernel function is applied and sobel is applied for the edge
detection. If a pixel is related to edge, bilinear interpolation
is applied otherwise it will perform geometric
transformations based on the estimated elevation angle to
find the edges. After that, the interpolation is applied along
these edges by estimating the covariance of surrounding
neighbours. The logic employed for computing covariance
of surrounding neighbours is described under four steps.
Step1-Computing Covariance: Fig.2 illustrates the
approach used for computing the covariance in order to
estimate the high resolution pixels. By applying eq(1) at
pixels locations (1 to 4), denoted in the figure 2, the
covariance is estimated .

Fig.2 Finding covariance between pixels to estimate the
new pixel

(1)

Where, , | | ( ) 3 , 0 , s < = l k R R
kl

| | ( ) 3 0 , < < = k r r
k
, denotes local covariance
characteristics at high resolution.
Step2-Replacing High Resolution
kl kl
r R , with Low
Resolution: After finding the covariance of the high
resolution pixels
kl kl
r R , is replaced by their low resolution
pixels
,
kl kl
R r which couples the pair of pixels along the
same direction but at different resolution. As the difference
between the covariance of high resolution pixels and the low
resolution pixels become negligible the resolution gets
higher and higher.
Step3-Interpolate through 45Rotation: Interpolate the
other half by rotating the image by 45as indicated in Fig.3
and repeat the steps 1and 2 until the new interpolated pixels
achieves with increased resolution.

Fig.3 Finding covariance between pixels to estimate the new pixel by
rotating 45
Step4-Magnified High Resolution Image: Finally, the
image is interpolated by the factor of two which results in a
high resolution image. The above steps are applied to the
pixels related to edge areas and for non edge pixels, a
simple linear interpolation technique such as bilinear
interpolation is applied. Finally, by combining both the
pixels related to edge areas and non-edge areas, it forms a
new high resolution image with good visual clarity and with
fewer artifacts such as blurring, smoothening is obtained.

3. DUAL TREE COMPLEX WAVELET TRANSFORM

( )
1 1
2 1,2 1 2 1 2 .2( )
0 0
i j k i k j l
k l
y Y
+ + + + +
= =
= o
1
XX X
R r
o =

Kingsbury in his work [6] introduced the dual-tree
complex wavelet transform, stating that it is orthogonal and
also can be computed faster. The frequency responses of the
CWT have 6 orientations at each of the 4 scales (the number
of scales is arbitrary, but the number of orientations is fixed)
which provide more features in the image. The main
advantages of the DT CWT over the real DWT are that the
complex wavelets are approximately shift invariant and
have separate sub bands for positive and negative
orientations. The conventional separable real wavelet suffers
from the lack of shift invariance and thus provides just three
orientations leading to poor directional selectivity.
The CWT attains these properties by replacing the
tree structure of the conventional wavelet transform with a
dual tree. At each scale, one of the trees produces the real
part of the dual-tree complex wavelet coefficients, while the
other tree produces the imaginary part. The extra
redundancy allows a significant reduction of aliasing terms
and causes the complex wavelets to become approximately
shift invariant. Translation causes large changes to the phase
of the wavelet coefficients, but their magnitude is much
more stable. By using even and odd filters alternately in
these trees, it is possible to achieve overall complex impulse
responses with symmetric real parts and anti-symmetric
imaginary parts. In Fig. 4, the structure of the dual tree
CWT is illustrated.

Fig 4: Implementation of DT-CWT

The two sets of filters are jointly designed so that the overall
transform is approximately analytic. Let h
0
(n), h
1
(n) denote
the low-pass/high-pass filter pair for the upper FB, and g0
(n), g
1
(n) denote the low-pass/high-pass filter pair for the
lower FB. Let, the two real wavelets associated with each of
the two real wavelet transforms be denoted as
h
(t) and
g
(t). In addition to satisfying the Perfect Reconstruction
conditions, the filters used in DT-CWT are designed such
that the complex wavelet (t ) :=
h
(t ) + j
g
(t ) is
approximately analytic. Equivalently, they are designed
such that
g
(t) is approximately the Hilbert transform of
h
(t
) [denoted
g
(t ) H{
h
(t )}].
Furthermore the DT CWT also gives much better
directional selectivity when filtering multi-dimensional
signals. The above mentioned fact has lead to the following
properties [6].
Approximate shift invariance.
Good directional selectivity in 2-dimensions (2-D)
with Gabor-like filters.
Perfect reconstruction (PR) using short linear-phase
Filters.
Limited redundancy, independent of the number of
Scales, 2: 1 for 1-D.
Efficient order-N computationonly twice the
simple DWT for 1-D (2m times for m-D).

4. PROPOSED EDGE DIRECTIONAL INTERPOLATION
USING DT-CWT
The major loss in an image after being super resolved
through interpolation technique, is on its high-frequency
components (i.e., edges), which is mainly due to the
smoothing effects, caused by interpolation. Hence, in order
to increase the quality of the super resolved image,
preserving the edges is essential. DT-CWT has been
employed in order to preserve the high-frequency
components of the image. The DT-CWT has good
directional selectivity and has the advantage over discrete
wavelet transform (DWT). It also has limited redundancy.
The DT-CWT is approximately shift invariant, unlike the
critically sampled DWT. The redundancy and shift
invariance of the DT-CWT implies that DT-CWT
coefficients are inherently interpolable.
In the proposed technique, DT-CWT is used to
decompose an input image into different sub-band images as
shown in Fig.5. Complex-valued high-frequency sub-band
images contain the high-frequency components of the input
image. The interpolation is applied only to the high-
frequency sub-band images. In the wavelet domain, the low-
resolution image is obtained by low-pass filtering of the
high-resolution image.
The two up-scaled images are generated by interpolating
the low-resolution original input image and the shifted
version of the input image in horizontal and vertical
directions. These two real-valued images are used as the real
and imaginary components of the interpolated complex low
resolution image, respectively, for the IDT-CWT operation.
By interpolating the input image by /2 and the high-
frequency sub-band images by and then by applying IDT-
CWT, the obtained output image will contain more sharper
edges than the interpolated image obtained without DT-
CWT. This is due to the fact that the interpolation of the
isolated high-frequency components in the high-frequency
sub-band images will preserve more high-frequency
components after the interpolation of the respective sub
bands separately than interpolating the input image directly.

In
im
Low frequency
sub-band
Input image

Fig 5: Block diagram of the proposed resolution enhancement algorithm.

In summary, the proposed technique interpolates the
input image as well as the high frequency sub-band images
obtained through the DT-CWT process .The final high-
resolution output image is generated by applying IDT-CWT
of the interpolated sub-band images and the input image.
The Performance measure like visual clarity and the
peak signal to noise ratio (PSNR) are used to evaluate the
proposed technique.

5. CONCLUSION

This paper proposes image resolution
enhancement based on edge directional interpolation in duel
tree-complex wavelet transform. Image interpolation is
popularly used to increase the image resolution, but the
potential problem associated with it is to magnify the image
many times without loss in image clarity. Most of the
classical linear interpolation techniques like bilinear & bi-
cubic interpolation methods generate blurred images. By
employing dual-tree complex wavelet transform (DT-CWT)
on a edge directional interpolation, it is possible to recover
the high frequency components which provides an image
with good visual clarity and thus super resolved high
resolution images are obtained.

References

[1] J.Allebach and P.W. Wong, Edge-directed interpolation,
proceeding of ICIP1996, pp.707- 710.
[2] M.Unser, A.Aldroubi and M.Eden, image interpolation and
resampling, IEEE Trans. on image processing, vol.6, pp.1322-1326,
September, 1997.
[3] X. Li and M.Orchrard, New edge-directed interpolation, IEEE
Trans. image process, vol.10,no.10,pp.1521-1527,Oct.2001.
[4] Y. Piao, I. Shin, and H. W. Park, Image resolution enhancement
using inter-sub band correlation in wavelet domain, in Proc.
International conference on image processing, 2007, vol. 1, pp.
445- 448.
[5] N. Kingsbury, Complex wavelets for shift invariant analysis and
filtering of signals, Appl. Comput. Harmonic Anal., vol. 10, no. 3,
pp. 234253, May 2001.
[6] T. H. Reeves and N. G. Kingsbury, Prediction of coefficients
from coarse to fine scales in the complex wavelet transform, in
Proc. IEEE International conference on acoustics, speech and
signal processing, Jun. 59, 2000, vol. 1,pp. 508511.
[7] Hasan Demirel and Gholamreza Anbarjafari,Satellite Image
Resolution Enhancement Using Complex Wavelet
Transform,IEEE Trans. geoscience and remote sensing
letters,vol.7,no.1,January 2010,pp 123- 126.

DT-CWT
IDT-CWT

A
p
G
a
t
p
i
T
c
n
c

K
s
s
c
c
s
m
b
c
c
t
i
s
o
t
o
o
c
r
b
m
n
f
a
b
Proc. of th
Volume 1.
ISBN: 978-8
(Imp
M
AbstractGra
problems that
Graph drawin
and so graph
the design and
paper explores
including grap
The emphasis
computer scie
network, we h
coloring in GS

Keywords-list
structure,Greed
Graph theo
science appli
computer scie
clustering, im
science, grap
mathematical
between objec
context refers
collection of e
theory has gai
it in the 18t
scientific field
of Fermat's li
theory. In com
of the famou
operates in fo
color theorem
regions share
be used in res
mining, image
networking et
for mail deliv
and planning
be done using
Informally
he Intl. Conf. o
Copyright
In
1-920575-9-0
ACM #: dber
Ap
plementing
Mahadevi Som
Information
SGGSIE&T
Nande
aphs are app
arise in compu
ngs are a usefu
visualization h
d analysis of
s different elem
ph representa
s of this pape
nce. In this p
have to apply
SM mobile Netw

structure;
dy algorith,Map
I. INT
oretical ideas a
ications. Espe
ence such data
mage capturing
ph theory i
structure use
cts from a cer
s to a collecti
edges that con
ined wide imp
th century. It
ds. For exampl
ittle theorem
mputer science
us standards f
ur different fr
m. This ensu
the same colo
earch areas of
e segmentation
tc. Problems
very, fault di
a LAN using
g graphs.
y, a graph is a
on Computer A
2012 Techno
ndia.
0 :: doi: 10. 730
.imera.10. 730
pplication
g Greedy A

mnath Namo
n Technology
T, Vishnupuri
ed,India

propriate mo
uter science an
l way of depic
has found man
communicatio
ments involved
ations using c
er is on grap
aper we focus
y a Greedy al
work.
Matrix s
p coloring;
TRODUCTIO
are highly utili
ecially in re
a mining, ima
g, networking
s the study
d to model pa
rtain collection
on of vertices
nnect pairs of
portance since
t can be app
le, in pure ma
can be given
e, in the field
for cell phone
requencies is b
ures that no
or on the map
f computer sci
n, clustering, i
of efficiently
iagnostic in co
efficient netw
a set of objec
Applications
Forum Group
011/ISBN_07
011
n Of Grap
Algorithm
ose
odels for m
nd its applicati
cting these mod
ny application
on networks. T
d in graph the
omputer system
h applications
s on GSM mo
gorithm for M
structure; D
ON
ized by compu
esearch areas
age segmentati
etc. In compu
y of graphs,
air wise relati
n. A graph in
s or nodes an
f vertices . Gr
Euler introdu
plied to differ
athematics a pr
n based on gr
of networks,
e (GSM1) wh
based on the f
two neighbor
Graph theory
ience such as d
image capturin
y planning rou
omputer netw
work topology
cts called verti

p,
768
ph Theory
m For map C
many
ons.
dels,
ns in
This
eory
ms .
s in
obile
Map
Data
uter
of
ion,
uter
, a
ions
this
nd a
raph
uced
rent
roof
raph
one
hich
four
ring
can
data
ng,
utes
ork
can
ices
(
D
h
m
a
d
th
d
T
s
u
im
p
c
in
r
I
e
e
e
g
s
a
y In Comp
Coloring i
Rac
I
S
or nodes) co
Depending on
have a directio
may not be al
assigned weigh
direction assoc
he graphical r
digraph.
F
The rest of this
ection II prov
using compute
mportant grap
paper is conclu
II. G
In general, t
computer syst
ncidence m
epresentation,
Incidence Lis
element in the
edge contains
endpoints of t
graphs by ens
ource or desti
as the opposit
puter Scie
in GSM m

chana Sandip
Information Te
SGGSIE&T, V
Nanded,I

onnected by l
n the applicati
on; edges join
llowed, and v
hts, that are n
ciated with th
representation)
Figure1. Directed
s paper is orga
vides an overv
er systems. In
ph theory app
uded in section
GRAPH REPR
COMPUTE
there are four
em: The inci
matrix repre
, and the adjac
t - This repre
array corresp
a list of size
the edge. Thi
suring the fir
ination while
te. Incidence
ence
mobile Netw
p Potpelwar
echnology
Vishnupuri
India
links called e
ions, edges m
ning a vertex
vertices and/o
numbers. If th
hem (indicated
) then it is a d
and Weighted Gr
anized as follo
view of graph
n section III
plications are
n IV.
ESENTATIO
ER SYSTEMS
ways to repre
dence list rep
esentation,
cency matrix
esentation uses
ponds with a si
two which co
is can also ap
rst vertex be
the second sh
matrix - Thi
work )
edges (or arc
may or may n
to itself may
or edges may
he edges have
d by an arrow
directed graph,
raph
ows:
h representatio
some of m
explained. T
NS USING

esent a graph in
presentation, t
adjacency
representation
s an array. Ea
ingle edge. Ea
orresponds to t
pply to direct
defined as t
hould be defin
is representati
cs).
not
or
be
e a
w in
, or
ons
most
The
n a
the
list
n.
ach
ach
the
ted
the
ned
ion
uses a matrix of M edges by N vertices. If the vertex is an
endpoint to the edge, a value of 1 is assigned to their
crossing, otherwise, a value of 0 is assigned. This is a
terrible waste of space as every column arrow represented
by the edge can only have two values of 1 while the rest
are labeled 0.These two representations are most useful
when information about edges is more desirable than
information about vertices.
Adjacency list - Much like the incidence list, each
node has a list of which nodes it is adjacent to. This can
sometimes result in "overkill" in an undirected graph as
vertex 3 may be in the list for node 2, then node 2 must be
in the list for node 3. Either the programmer may choose
to use the unneeded space anyway, or he/she may choose
to list the adjacency once. This representation is easier to
find all the nodes which are connected to a single node,
since these are explicitly listed. Adjacency matrix - there
is an N by N matrix, where N is the total number of
vertices in the graph. If there is an edge from some vertex
x to some vertex y, then the element Mx,y would be 1,
otherwise it would be 0. This makes it easier to find sub
graphs, and to reverse graphs if needed. These two
representations are most useful when information about
the vertices is more desirable than information about the
edges. Also these two representations are also most
popular since information about the vertices is often more
desirable than edges in most applications.
III. GRAPH APPLICATIONS
Graph theoretical concepts are widely used to study
and model various applications, in different areas. They
include, study of molecules, construction of bonds in
chemistry and the study of atoms. Similarly, graph theory
is used in sociology for example to measure actors
prestige or to explore diffusion mechanisms. Graph theory
is used in biology and conservation efforts where a vertex
represents regions where certain species exist and the
edges represent migration path or movement between the
regions. This information is important when looking at
breeding patterns or tracking the spread of disease,
parasites and to study the impact of migration that affect
other species. Graph theoretical concepts are widely used
in Operations Research. For example, the traveling
salesman problem, the shortest spanning tree in a
weighted graph, obtaining an optimal match of jobs and
men and locating the shortest path between two vertices
in a graph. It is also used in modeling transport networks,
activity networks and theory of games. The network
activity is used to solve large number of combinatorial
problems. The most popular and successful applications
of networks in OR is the planning and scheduling of large
complicated projects. The best well known problems are
PERT(Project Evaluation Review Technique) and CPM
(Critical Path Method). Next, Game theory is applied to
the problems in engineering, economics and war science
to find optimal way to perform certain tasks in
competitive environments. To represent the method of
finite game a digraph is used. Here, the vertices represent
the positions and the edges represent the moves.

A. Algorithms and graph theory
The major role of graph theory in computer
applications is the development of graph algorithms.
Numerous algorithms are used to solve problems that are
modeled in the form of graphs. These algorithms are used
to solve the graph theoretical concepts which intern used
to solve the corresponding computer science application
problems.
Some algorithms are as follows:
1. Shortest path algorithm in a network
2. Finding a minimum spanning tree
3. Finding graph planarity
4. Algorithms to find adjacency matrices.
5. Algorithms to find the connectedness
6. Algorithms to find the cycles in a graph
7. Algorithms for searching an element in a data structure
(DFS, BFS) and so on. Various computer languages are
used to support the graph theory concepts. The main goal
of such languages is to enable the user to formulate
operations on graphs in a compact and natural manner
Some graph theoretic languages are
1. SPANTREE To find a spanning tree in the given
graph.
2. GTPL Graph Theoretic Language
3. GASP Graph Algorithm Software Package
4. HINT Extension of LISP
5. GRASPE Extension of LISP
6. IGTS Extension of FORTRAN
7. GEA Graphic Extended ALGOL (Extension of
ALGOL)
8. AMBIT To manipulate digraphs
9. GIRL Graph Information Retrieval Language
10. FGRAAL FORTRAN Extended Graph Algorithmic
Language
B. Map Colouring and GSM Mobile Phone Networks
Map Coloring and GSM Mobile Phone Networks:The
Groupe Spcial Mobile (GSM) was created in 1982 to
provide a standard for a mobile telephone system. The
first GSM network was launched in 1991 by Radiolinja in
Finland with joint technical infrastructure maintenance
from Ericsson. Today, GSM is the most popular standard
for mobile phones in the world, used by over 2 billion
people across more than 212 countries. GSM is a cellular
network with its entire geographical range divided into
hexagonal cells. Each cell has a communication tower
which connects with mobile phones within the cell. All
mobile phones connect to the GSM network by searching
for cells in the immediate vicinity. GSM networks operate
in only four different frequency ranges. The reason why
only four different frequencies suffice is clear: the map of
t
f
m
f
f

s
c
G
c
c
g
e
a
m
w
o
s
s
p
g
s
r
b
m
s
c
t
s
v
n
I
I
2
t
A
C
4
s
the cellular re
four different
may be use
frequencies f
figure given b
Figure2
C. Map color
The goal o
so that region
colors, so th
Generally spe
color a map f
color a map f
greedy algorit
exhaustion is
algorithm rep
maximize the
with the hop
outcome for
strategy is gu
some other c
produces acc
greedy algori
simple to im
resources [10]
begin with a
measuring by
sure to obtain
considers one
the condition
solution value
values have
neglect.
In general, gre
I. A candidate
2. A selection
to be added to
A feasibility f
Candidate ca
4. An object
solution, or a p
egions can be p
t colors! So,
d for assign
for any GSM
below.
2. ************
ring practical
of a map color
ns sharing a c
hat we could
eaking, it is
for some regio
for many regio
thm is the best
s used. A.
eatedly execu
return based
e that the ou
the global p
uaranteed to o
cases it may
ceptable app
ithms employ
mplement and
]. The basic i
certain initia
y a certain op
n local optim
e data by loca
satisfies. If th
e are connect
no feasible
eedy algorithm
e set, from w
n function, wh
o the solution
function, that i
an be used to
tive function,
partial solutio
properly color
the vertex co
ning at mos
M mobile pho

**************
l algorithm
ring problem
common bord
d partition d
easy to use s
ons; however
ons. Under su
t optimal solu
Greedy algor
utes a procedu
on examining
utcome will l
problem. In s
offer optimal
y provide a c
proximations.
y simple str
d require min
deas of greed
al value and
ptimization ,
mal solution v
al optimal sol
he next data a
ted and their
solution, the
ms have five p
which a solu
hich chooses th
is used to dete
o contribute t
, which assig
on, and
red by using o
oloring algorit
t four differ
one network,
**********
is to color a m
der have differ
different regio
several colors
, it is difficul
uch circumstan
ution if method
rithm A gre
ure which trie
g local conditio
ead to a desi
some cases s
solutions, and
compromise
Typically,
rategies that
nimal amount
dy algorithm i
go step by s
each step ma
value. Each s
lution value u
and local opti
optimal solut
ese data sho
illars:
ution is creat
he best candid
ermine if a
to a soluti
gns a value
only
thm
rent
see
map
rent
ons.
s to
lt to
nce,
d of
eedy
s to
ons,
ired
such
d in
that
the
are
t of
s to
step,
akes
step
until
mal
tion
ould
ted
date
ion
toa
5
d
p
b
w
B
a
fe
in
c
c
c
a
n
d
m
a
r
im
a
1
r
n
in
2
r
le
3
c
4
a
c
5
e
6
c
p
{
fo
{
C
//
fo
{
if
r
5. A solution fu
discovered a
produce good
but not on oth
well have two
B. Map color
analysis of gre
feasible for ma
n the course
considered as
chosen decisio
coloring order
algorithm. So,
number of sh
dealing with th
map coloring
according to d
egions by gre
mproved map
are as follows:
1. Based on s
egion and its
number and its
nto hash table
2. Resort the v
egions accord
ess in order to
3. According to
color value 1.
4. Judge that
adjacent to th
conflict, if not
5. If the presen
exit circle, try
6. Check out th
core part of ma
paper: for( int
{
for( j = 1; j <=
{
Color( i, j )// co
/all the colored
for( k = 1; k <=
{
f( IsTouch(i,k
egion k is adja
function, which
complete s
solutions on s
hers. Most pr
properties:
ring algorithm
eedy algorithm
ap coloring. H
e of greedy
an optimal c
on, does not d
r has an imp
, the paper co
haring comm
he regions whi
region numb
descending, a
eedy algorithm
p coloring app

surface I surfa
adjacent reg
s adjacent reg
.
values in hash
ding to adjace
o satisfy adjace
o the resorting
whether the
he colored reg
color it.
nt region is con
to the next col
he next region
ap coloring alg
i = l;i <= regio
the max color
olor j colors re
d regions
= i; k++)
k) && r_color[
acent to the pr
h will indicate
solution Gre
some mathem
roblems for w
m by reorder
m, we could g
However there
algorithm, th
choice, which
epend on new
portant influe
olor regions h
mon borders
ich is easy to
ber by the a
and then color
m. The implem
proach propos
ace topology,
gions number,
gions number
table, namely
ent regions fr
ent regions to
g result to colo
e present col
gions, whethe
nflicting to co
lor.
n, repeat step 4
gorithm propo
on number; i+
ring number; j
egion i
[k] = r_color[i
resent region I
e when we hav
eedy algorith
matical problem
which they wo
region: By t
get that it is ve
e is no feedba
he later step
depends on t
w choice, so m
ence on gree
having the m
firstly, name
conflict, sort t
adjacent regio
r the descendin
ment steps of t
sed by the pap
calculate eve
and put regi
and the numb
resort the who
rom the more
color firstly.
or from the
loring region
er there is co
olored regions,
4 and step 5. T
ose by the
++ )
j++)
i] )//the colore
I, the colorin
ve
hms
ms,
ork
the
ery
ack
is
the
map
edy
most
ely,
the
ons
ng
the
per
ery
ion
ber
ole
to
is
olor
,
The
ed
ng
approach is the same.
break; // exit circle and try to the next color
}
if( k>=i )
break; //check out the next region
}
}
By implementing this Greedy Algorithm on GSM Mobile
Network we properly color the cells of mobile Network.
By the analysis of greedy algorithm, we could get that it is
very feasible for map coloring.
IV. CONCULSION:
The main aim of this paper is to present the importance
of graph theoretical ideas in various areas of compute
applications. By the analysis of greedy algorithm, we
could get that it is very feasible for map coloring. An
overview of most common applications of graph theory is
illustrated. These applications are presented especially to
project the idea of graph theory and to demonstrate its
importance in computer science.

REFERENCES
[1] [http://en.wikipedia.orglwiki/Greedy_algorithm]
[2] Cormen, Leiserson, Rivest, and Stein, Introduction to Algorithms
2001, Chapter 16 "Greedy Algorithms".
[3] Cormen, Leiserson, and Rivest ,Introduction to Algorithms
1990, Chapter 16 "Greedy Algorithms" p. 329
[4] http://en.wikipedia.orglwikilFour_color_theorem
[5] Lingli Zhao, Shuai Liu, Junsheng Li,Haichen Xu, Hao Chen,
Lingli Zhao,Shuai LiuA Practical Approach for Map Coloring
Considering of Greedy Algorithm 2010 2nd Conference on
Environmental Science and Information Application Technology.
[6] Ferozuddin Riaz, Khidir M. Ali Applications of Graph Theory in
Computer Science 2011 Third International Conference on
Computational Intelligence, Communication Systems and Networks.
[7] K. Rosen, Discrete Mathematics and its Application, 6th Ed.,
McGraw-Hill.

India.
ISBN: 978-81-920575-9-0 :: doi: 10. 73018/ISBN_0768

AN APPROACH FOR DISEASE TREATMENT IN MEDLINE USING SEMANTIC
RELATIONS
P.Menaka,
Department Of Computer science and Engineering,
Adhiyamman College Of Engineering
Hosur

Abstract -- The Machine Learning (ML) is almost used in any
domain of research and now it has become a reliable tool in
the medical domain.ML is a tool by which medical field is
integrated with the computer based systems to provide more
efficient medical care. The main objective of this work is to
show what Natural Language Processing (NLP) and Machine
Learning (ML) techniques used for representation of
information and what classification algorithms are suitable
for identifying and classifying relevant medical information
in short texts. It acknowledges the fact that tools capable of
identifying reliable information in the medical domain stand
as building blocks for a healthcare system that is up-to-date
with the latest discoveries. In this research, it focuses on
diseases and treatment information, and the relation that
exists between these two entities.

I. INTRODUCTION
Medline is the richest and most used source of
information. It is a database that has enormous articles
regarding life sciences. This paper focuses on two tasks:
Automatically identifying sentences from Medline and
also provides the relation between the diseases and
treatments. The next task is based in three relations: Cure,
Prevent and Side effects. These tasks helped to develop the
Information Technology Framework that provides all the
healthcare information. Healthcare related information
(e.g., published articles, clinical trials, news, etc.) is a
source for both healthcare providers and people. These
Studies shows that people are searching the web and gain
the knowledge about the healthcare related information.
Natural Language Processing (NLP) and Machine
Learning (ML) are the techniques that are used here. The
ultimate aim is to show what representation of information
and algorithms used to identify and provide relevant
healthcare information in short texts. The tools that are
used to identify all medical related information are the
building blocks in medical domain that has all latest
discoveries up-to-date. The main entities involved in this
paper are disease and treatment. The main aim is to
focus on all information related to disease. It is also
possible with tendency of having personalized
medicine so that each patient has healthcare to provide
treatment to their particular disease. However it is not
enough to focus on certain study that provides
treatment to particular disease. Also the Healthcare
providers should focus on all recent discoveries in
order to identify any side effects for patients.
This paper provides the foundation for
development of technology framework that makes easy
to find all the relevant information regarding treatment
and diseases. The tool that is built with the techniques
such as Natural Language Processing (NLP) and
Machine Learning (ML) has capability to find all
relevant short text information regarding diseases and
treatments. This work presents various Machine
Learning (ML) and information for classifying short
texts and relation between diseases and treatments.
According to ML technique the information are shown
in short texts when identifying relations between two
entities such as diseases and treatment. Thus there is
improvement in solutions when using a pipeline of two
tasks (Hierarchical way of approaching). It is better to
identify and remove the sentence that does not contain
information relevant to disease or treatments. The
remaining sentences can be classified according to the
interest. It will be very complex to identify the exact
solution if everything is done in one step by classifying
sentences based on interest and also including the
sentences that do not provide relevant information.
II. RELATED WORK
This work presents various Machine Learning
(ML) and information for classifying short texts and
relation between diseases and treatments. According to
ML technique the information are shown in short texts
when identifying relations between two entities such as
diseases and treatment. Thus there is improvement in
solutions when using a pipeline of two tasks
(Hierarchical way of approaching). It is better to
identify and remove the sentence that does not contain
information relevant to disease or treatments. The
remaining sentences can be classified according to the
interest. It will be very complex to identify the exact
solution if everything is done in one step by classifying
sentences based on interest and also including the
sentences that do not provide relevant information.
The related work is done by Hearst and Rosario. The
data set used in this work are created and distributed by the
authors of this paper. The data set contains information
from medline with all relevant information including
diseases, treatments and eight relations between diseases
and treatment. Entity recognition is focused mainly for
diseases and treatments. The authors use two models for
entity recognition and relation details. The models are
Hidden markov model and maximum entropy model. The
representation of information depends on medical lexical
ontology, phrases and words in context. In this paper
different techniques and models are used which provides
better solution.
The main task in this work is to extract healthcare
information and the relation details. Extraction of relation
is done using three approaches. The approaches are: co-
occurrences analysis, rule based approaches, and statistical
methods. Rule based approach is used in solving relation
between entities. Co-occurrences analysis is based on
words in context and lexical knowledge. Rule based
approach use syntactic: part-of-speech and syntactic
structure or information in the form of fixed patterns that
has word that describes the relation between the entities.
The disadvantage of rule based approach is that it is mostly
driven with human efforts rather than using data driven.
One of the best rule based system is that rules constructed
by humans, auto-extraction and refined by human. The
main advantage of having this system is it provides
precision results while recall tends to be low.
Syntactic rule based systems are very complex since it
depends on some more tools used to assign POS tags. But
it is clearly mentioned that such tools are not yet reached
the state of art level since they are for designed only for
common English texts and therefore it does not provide
better solutions. Thomas et al., Yakushiji et al are the two
who presented the works on relation extraction in medline
abstracts. Although it does not provide accurate results
these systems are successful in this biomedical field.
The semantic rule based system has a disadvantage that
lexicon changes from domain to domain. So new rules
have to be followed each time. To obtain good results
semantic and syntactic based systems are combined so that
they provide flexibility of syntactic information and good
precision of semantic rule. Statistical approaches are used
to solve various tasks. So that rules can be automatically
extracted. This approach is used to solve various NLP
tasks. So this approach works well even with fewer
amounts of data. Concerning relation extraction the rule
checks whether the text information contains any relation
or not. For this relation extraction the statistical approach
uses bag-of-words technique. Some researchers
combined this technique with POS which provides two
sources of information such as relation between
entities and their specific contexts. And it is proved
that simple technique can produce accurate results.
III. THE PROPOSED APPROACH
A. Tasks and Data Sets:
The two tasks used in this paper are the basis
for the development of information technology
framework. This framework helps to identify the
medical related information from abstracts. The first
task deals with extraction all information regarding
diseases and treatments while the task deals with
extraction of related information existing between
disease and treatments. The framework developed with
these tasks are used by healthcare providers, people
who needs to take care of their health related problems
and companies that build systematic views. The future
product can be provided with browser plug-in and
desktop application so that it helps the user to get all
information related to diseases and treatments and also
the relation between those entities. It is also be useful
to know more about latest discoveries related to
medicine. The product can be developed and sold by
companies that do research in medical care domain,
Natural Language Processing (NLP), and Machine
Learning (ML), and companies that develop tools like
Microsoft Health Vault and Google Health.
This product is valuable in e-commerce fields by
showing the statistics that the information provided
here are accurate and also provide all the recent
discoveries related to health care. To make a product
more popular it should be trust worthy so that people
can buy it. It is the key factor foe any company to
make product successful. When coming to health care
products it should be more trust worthy since it is
dealing with health related problems. Companies that
wish to sell health care framework need to develop
tools that automatically extract the wealth of research.
For example the information provided for diseases
or treatments needs to be based on recent discoveries
on health care field so that people can trust. The
product quality also should be taken care so that it
provides dynamic content for users. The first task deals
with the identification of sentences from the Medline
abstracts that provide the information about the
diseases and treatments. In other words it also seems
like scanning the sentences from medline abstracts that
contain relevant information which the user wants.
Natural Language Processing (NLP), and Machine
Learning (ML) are used to extract accurate information
or it can also say that it perfectly removes the
unwanted information which are not related to disease
or treatment. Natural Language Processing (NLP) and
Machine Learning (ML) itself involve in extracting
informative sentences. It is difficult task to identify the
informative sentences in fields such as summarization and
information extraction. The work and contribution value
with this task is helpful in results and in settings for this
task in healthcare field.
In the first task data set are provided with information
such as label containing that the extracted sentence is
informative or label containing that the extracted sentence
is not accurate.
For the second task it is provided with information that
shows if the relation between disease and treatment is cure,
prevent or side effects. Further it is good to focus on
relations based on interest to provide accurate results. The
three relations (cure, prevent and side effects) are
mentioned in original data set and they are needed for
future reference. Two ways of identifying semantic
relations are:
Task 1: In this three models have been built. Each
model is focused on relation and provides the difference
between the sentences that contain information that do not
contain.
Task 2: Unlike in setting 1 it has only one model. This
model focus on three relation (cure, prevent and side
effects). It distinguishes relations in three-class tasks and
contains label in each sentence that has any one semantic
relation.
The most important thing is that they can be combined
in pipeline so that it provides solution to framework which
identifies all related information. This proposed pipeline
approach first deals task1 and then carries out with task2
so that it can give only informative sentences based on
three relations. The logic behind to go with pipeline
approach is to identify the best model to identify and
extract the reliable healthcare information. With this
pipeline task some problems can be removed due to the
fact that unwanted information can be the potential factor
to classify the sentences into three semantic relations.
B. Classification Algorithms and Data
Representations:
In Machine Learning the expertise and previous
research provides the guidance to solve new tasks. The
models described should be able to identify and provide
informative sentences and relation between entities. The
research should be made in a way to achieve high
performance.
While working with ML technique two challenges
should be considered. One is to find the exact model. ML
provides a better model that can be used. To find the
model one should rely on empirical studies and should
gain knowledge in healthcare domain. The next one is to
provide better data representation and to do feature
engineering, because feature increase the performance of
the model. Finding the exact feature for the predictive
model is a crucial task that should be considered. These
two challenges can be achieved by using various
algorithms and textual representation that suits for the
tasks.
As classification algorithms, set of six
representative models can be used. They are: adaptive
learning (Ada-Boost), decision-based models
(Decision trees), and probabilistic models (Naive
Bayes (NB) and Complement Naive Bayes (CNB),
which is adapted for text with imbalanced class
distribution), a linear classifier (support vector
machine (SVM) with polynomial kernel), and a
classifier that always predicts the majority class in the
training data (used as a baseline). These classifiers are
used to learn more algorithms and to work on long text
and short texts. Probabilistic models based on Naive
Bayes used in text classification and automatic text
classification tasks. Decision trees based on decision
models are used in short texts. Adaptive learning
algorithm is used to focus on hard concepts such as
unbalanced data sets, underrepresented in data. SVM
based algorithm involve in text classification
technique. All the classifiers are part of tool called
Weka. All the above mentioned steps to followed since
it is used to obtain database table that contains as many
columns as number of feature selected to represent
data and as many rows as the data points.
C. Bag-of-Words Representation:
The bag-of-words is the name commonly used to
classification of tasks. It is a representation in which
features are selected among the words that are present
in the training data. And then selection techniques are
used in order to identify the most suitable words as
features. Once the feature is identified, each training
and test instance is mapped to this feature
representation by providing values to each feature for a
specific instance. Two most common feature value
representations for Bag-of-Words (BOW)
representation are: binary feature values or frequency
feature values. The binary feature value is the value of
a feature can be either 0 or 1, where 1 represents the
fact that the feature is present in the instance and 0
otherwise. The frequency feature value is the value of
the feature is the number of times it appears in an
instance or 0 if it did not appear. Here we use
frequency feature value. There is no that much
difference between binary feature values and
frequency feature values because there is only twenty
words in each sentence of short texts. The advantage of
using frequency feature values is that the features
value will be greater than other features since it
captures the number of feature appeared once in a
sentence. It keeps words that appear at least three times
in training data and have at least one alphanumeric.
D. NLP and Biomedical Concepts epresentation
Unlike above representation this representation is
based on syntactic information such as nouns, phrases,
healthcare related points in the sentence. A tool called
Genia tagger is used to extract information. The tagger is
specifically designed for biomedical text such as Medline
abstracts. It analyzes the sentences to generate base forms,
chunk tags, part of speech tags. The phrases and nouns
obtained by tagger is used in second representation
method. While running the Genia tagger it extracts only
nouns, phrases and healthcare related concepts from each
sentence of data set. The steps to be followed to obtain the
final feature for classification are: removing features that
contains punctuations and considering lemma-based forms.
Lemma is used because there is lot of words that has plural
forms. Lemmatized form gives the base form of the word.
Another reason to use lemma is to reduce the sparseness
problem. Lemma forms remove the problem of
representing only a few features in short text forms.
E. Medical Concepts (UMLS) Representation
For more general features we use Unified Medical
Language system (UMLS). UMLS is a developed at the
US National Library of Medicine (NLM) .It contains a
metathesaurus, a semantic network, and the specialist
lexicon for biomedical domain. The metathesaurus deals
with concepts and meanings. It provides relation between
various different concepts.In addition to the UMLS
knowledge base, NLM created a set of tools that allow
easier access to the useful information. UMLS contains
over 1 million medical concepts, and over 5 million
concept names. Each concept is assigned with one
semantic type from the semantic between concepts. NLM
created new tool called MetaMap that maps text to
healthcare concepts in UMLS. With this text is processed
through entire data set and finally it provides ranked list of
all possible concepts for particular noun-phrase. For each
noun phrase in text various noun-phrases are generated.
For each variant noun-phrases UMLS met thesaurus are
obtained and evaluated. The obtained phrase is compared
with actual phrase using fit function. Fit function measures
the text overlap between obtained concepts and actual
phrase. The best of the candidates are then organized
according to the decreasing value of the fit function.

IV. CONCLUSION
The interests are inline with the tendency of having a
personalized medicine, one in which each patient has its
medical care tailored to its needs. It is not enough to read
and know only about one study that states that a treatment
is beneficial for a certain disease. The proposed system
used the top concept candidate for each identified phrase in
an abstract as a feature. It extracts diseases and treatments
given in and identifies the semantic relations between
them. The outcome of this proposed work could be
integrated with an application and used in health care field.

REFERENCES
[1] R. Bunescu and R. Mooney, A Shortest Path Dependency
Kernel for Relation Extraction, Proc. Conf. Human Language
Technology and Empirical Methods in Natural Language Processing
(HLT/ EMNLP), pp. 724-731, 2005.
[2] R. Bunescu, R. Mooney, Y. Weiss, B. Scho lkopf, and J. Platt,
Subsequence Kernels for Relation Extraction, Advances in Neural
Information Processing Systems, vol. 18, pp. 171-178, 2006.
[3] A.M. Cohen and W.R. Hersh, and R.T. Bhupatiraju, Feature
Generation, Feature Selection, Classifiers, and Conceptual Drift for
Biomedical Document Triage, Proc. 13th Text Retrieval Conf.
(TREC), 2004.
[4] M. Craven, Learning to Extract Relations from Medline,
Proc.Assoc. for the Advancement of Artificial Intelligence, 1999.
[5] I. Donaldson et al., PreBIND and Textomy: Mining the
Biomedical Literature for Protein-Protein Interactions Using a
Support Vector Machine, BMC Bioinformatics, vol. 4, 2003.
[6] C. Friedman, P. Kra, H. Yu, M. Krauthammer, and A. Rzhetsky,
GENIES: A Natural Language Processing System for the
Extraction of Molecular Pathways from Journal Articles,
Bioinformatics, vol. 17, pp. S74-S82, 2001.
[7] O. Frunza and D. Inkpen, Textual Information in Predicting
Functional Properties of the Genes, Proc. Workshop Current Trends
in Biomedical Natural Language Processing (BioNLP) in
conjunction with Assoc. for Computational Linguistics (ACL 08),
2008.
[8] R. Gaizauskas, G. Demetriou, P.J. Artymiuk, and P. Willett,
Protein Structures and Information Extraction from Biological
Texts: The PASTA System, Bioinformatics, vol. 19, no. 1, pp. 135-
143, 2003.
[9] C. Giuliano, L. Alberto, and R. Lorenza, Exploiting Shallow
Linguistic Information for Relation Extraction from Biomedical
Literature, Proc. 11th Conf. European Chapter of the Assoc. for
Computational Linguistics, 2006.
[10] J. Ginsberg, H. Mohebbi Matthew, S.P. Rajan, B. Lynnette, S.S.
Mark, and L. Brilliant, Detecting Influenza Epidemics Using
Search Engine Query Data, Nature, vol. 457, pp. 1012-1014, Feb.
2009.
[11] M. Goadrich, L. Oliphant, and J. Shavlik, Learning Ensembles
of First-Order Clauses for Recall-Precision Curves: A Case Study in
Biomedical Information Extraction, Proc. 14th Intl Conf. Inductive
Logic Programming, 2004.
[12] L. Hunter and K.B. Cohen, Biomedical Language Processing:
Whats beyond PubMed? Molecular Cell, vol. 21-5, pp. 589-594,
2006.
[13] L. Hunter, Z. Lu, J. Firby, W.A. Baumgartner Jr., H.L. Johnson,
P.V. Ogren, and K.B. Cohen, OpenDMAP: An Open Source,
Ontology-Driven Concept Analysis Engine, with Applications to
Capturing Knowledge Regarding Protein Transport, Protein
Interactions and Cell-Type-Specific Gene Expression, BMC
Bioinformatics, vol. 9, article no. 78, Jan. 2008.
[14] T.K. Jenssen, A. Laegreid, J. Komorowski, and E. Hovig, A
Literature Network of Human Genes for High-Throughput Analysis
of Gene Expression, Nature Genetics, vol. 28, no. 1, pp. 21-28,
2001.
[15] R. Kohavi and F. Provost, Glossary of Terms, Mac ine
Learning, Editorial for the Special Issue on Applications of Machine
Learning and the Knowledge Discovery Process, vol. 30, pp. 271-
274, 1998.
[16] G. Leroy, H.C. Chen, and J.D. Martinez, A Shallow Parser
Based on Closed-Class Words to Capture Relations in Biomedical
Text, J. Biomedical Informatics, vol. 36, no. 3, pp. 145-158, 2003.
[17] J. Li, Z. Zhang, X. Li, and H. Chen, Kernel-Based Learning for
Biomedical Relation Extraction, J. Am. Soc. Information Science
and Technology, vol. 59, no. 5, pp. 756-769, 2008.

India.
ISBN: 978-81-920575-9-0 :: doi: 10. 73032/ISBN_0768
THRESHOLDING TECHNIQUES FOR SKULL STRIPPING THE MRI
BRAIN IMAGES

Mr.Vijendra P.Meshram
1
Electronics Engineering Department
DBACER, Nagpur,India

Prof.Mahesh S.Pawar
2
Electronics Engineering Department
YCCE,Nagpur,India

Abstract Segmentation is very important to analyze the
processed image data. In MRI brain, for skull stripping
removal of its noncerebral tissues is needed.Because of
homogeneity intensities, it is difficult to segment the non-
cerebral and intracranial tissues. The morphological
methods to be implemented on MRI brain image requires
binary image.The binarization of MRI brain image is
performed by using Otsus Thresholding and gray level
thresholding.To remove non cerebral tissues from MRI
brain image, robust threshold values are required.This can
be done by using gray level thresholding or Otsus
Thresholding method. This experiment uses MRI brain
images collected from hospitals.The result shows promising
use of Otsus Thresholding as a robust threshold value as
compared to gray level thresholding.

Keywords : Morphology , Segmentation, Otsus thresholding,
etc

I. INTRODUCTION
To detect the brain abnormalities such as brain tumour
, stroke , paralysis and breathing difficulties Brain
imaging is widely used in many medical applicatins. For
further analysis of Magnetic Resonance Imaging (MRI)
brain images, skull stripping is required an it is the major
pre-processing phases in brain applications. Segmention
of images requires thresholding values. This paper helps
to find a robust threshold value, it will helpful in
diagnosing some brain irregularities. Non-cerebral tissues
such as skull, scalp, vein or meninges can be removed
form MRI brain image, and it requires the robust
threshold value. Skull stripping include various
techniques such as region-based segmentation techniques
such as watershed , region growing techniques and
mathematical morphology. the watershed algorithm based
on the concept of pre-flooding which is to avoid the over
segmentation and to overcome the problem of noise. The
advantages of the watershed algorithm are it is a simple
and fast technique as well as frequently produces
complete boundaries. That watershed technique can cause
over segmentation, sensitivity to noise, poor detection of
significant areas with low contrast boundaries, and poor
detection of thin structures. Region growing works by
appending neighbouring pixels of starting seed pixel to
form a region based on predefined criteria[8]. The
disadvantage of this algorithm is that user has toselect the
seed regions [14] and threshold values. As morphological
operations requires binary form images, it provides a
simple and efficient way for integrating distance,
neighbourhood information in segmentation [4] as well as
offers an unified and powerful approach to numerous
image processing problems. However, morphology
requires a prior binarization of the image into object and
background regions [4]. Thresholding creates binary
images from gray-level onesby turning all pixels below of
the threshold value to zero and all pixels above the
threshold value to one [9]. The selection of an adequate
threshold of gray-level for extracting objects from their
background is important and thresholding algorithm is an
intuitive properties and simplicity of implementation of
the application of image segmentation [8].However, it is
difficult to find the robust threshold values which produce
a good output for some of the images. However, it is
difficult to find the robust threshold values which produce
a good output for some of the images. Therefore, this
paper proposed skull stripping using gray level
thresholding and mathematical morphology
experimenting on using Otsu threshold values in order to
address the drawback of choosing the incorrect threshold
values.

II. LITERATURE SURVEY
The goal of this proposed algorithm is to find the
robust threshold value to obtained the binary image whose
non-cerebral tissues has to be removed by using
mathematical morphology. Thus, this paper experimented
on two thresholding methods: gray level thresholding and
Otsus thresholding.Figure1 shows the process flow of
the proposed algorithms.

Figure 1: Process flow of the proposed mathematical morphology
operations algorithm

A. Data Collection
Two-dimensional MRI data sets are collected from the
Hospital . Details of image sequences used in the
experiments are:
T1-Weighted of axial orientation (30 images)
T2-Weighted of axial orientation (30 images)
Fluid Attenuated Inversion Recovery (FLAIR) of
axialorientation (30 images)
B. Thresholding
Two thresholding methods are experimented in this
paper that is gray level thresholding and Otsus method.
The purpose of this experiment is to identify the robust
threshold values to remove the non-cerebral tissue from
MRI brain images.Figure 2 illustrates the non-cerebral
tissues (skull,cerebrospinal fluid, meninges) to be
extracted.

Figure 2: Anatomical of cerebral and non-cerebral tissues

III. GRAY LEVEL THRESHOLDING

This is the simplest and the oldest of all thresholding
processes. Hence we used GRAY colour space for the
study. Gray level thresholding method is very simple and
easy to implement in real time application. In this,
preprocessing the image included
1) Transforming the colored image to gray level
image, cropping the area of interest,
using the median filter to smooth the image, widen the
pixel gray level distribution.
2) Threshold input image.
3) Extract the features such as the region area,
boundary perimeter, number of holes etc. from the image.
4) Recognize three insect specimens using these
features.
Gray level thresholding suffers from serious
drawback, Intensity Variation, because of which the
intensity varies from time to time. Threshold values in
gray level thresholding are fixed using only one parameter
i.e. intensity. Therefore, successful segmentation using
gray level thresholding requires change of threshold
values periodically.

A complete segmentation of an image R is a finite set
of regions R
1
, . . .R
S
,
For binary images, there is a single threshold:

Thresholding algorithm
Search all the pixels f(i,j) of the image f. An image
element g(i,j) of the segmented image is an object pixel if
f(i,j) >=T, and is a background pixel otherwise.By trial
and error, determine the threshold segmenting the rabbit
from the background. To do this, adjust the upper and
lower thresholds (upperT, lowerT) of this function.
Determine the interval in which the threshold has to be
defined to perform the desired segmentation.

IV.OTSU THRESHOLD
Otsu threshold is a nonparametric and unsupervised
method used to select threshold value automatically.
An optimal threshold is selected by the discriminant
criterion, namely, so as to maximize the seperability of
the resultant classes in gray levels. The procedure is very
simple, utilizing only the zeroth- and the first-order
cumulative moments of the gray-level histogram. It is
straightforward to extend the method to multithreshold
problems. Several experimental results are also presented
to support the validity of the method.

It is important in picture processing to select an

adequate threshold of gray level for extracting objects
from their background. A variety of techniques have been
proposed in this regard. In an ideal case, the histogram
has a deep and sharp valley between two peaks
representing objects and background, respectively, so that
the threshold can be chosen at the bottom of this valley
[1].
However, for most real pictures, it is often difficult to
detect the valley bottom precisely, especially in such
cases as when the valley is flat and broad, imbued with
noise, or when the two peaks are extremely unequal in
height, often producing no traceable valley. There have
been some techniques proposed in order to overcome
these difficulties. They are, for example, the valley
sharpening technique [2], which restricts the histogram to
the pixels with large absolute values of derivative
(Laplacian or gradient), and the difference histogram
method [3], which selects the threshold at the gray level
with the maximal amount of difference. These utilize
information concerning neighboring pixels (or edges) in
the original picture to modify the histogram so as to make
it useful for thresholding. Another class of methods deals
directly with the gray-level histogram by parametric
techniques. For example, the histogram is approximated
in the least square sense by a sum of Gaussian
distributions, and statistical decision procedures are
applied [4]. However, such a method requires
considerably tedious and sometimes unstable calculations.
Moreover, in many mation of the real modes. In any
event, no "goodness" of threshold has been evaluated in
most of the methods so far proposed. This would imply
that it
could be the right way of deriving an optimal
thresholding method to establish an appropriate criterion
for evaluating the "goodness" of threshold from a more
general standpoint. In this correspondence, our discussion
will be confined to the elementary case of threshold
selection where only the gray-level histogram suffices
without other a priori knowledge. It is not only important
as a standard technique in picture processing, but also
ssential for unsupervised decision problems in pattern
recognition.A new method is proposed from the
viewpoint of discriminant analysis; it directly approaches
the feasibility of evaluating the goodness" of threshold
and automatically selecting an optimal threshold.

V. FORMULATION
Let the pixels of a given picture be represented in L
gray levels [1, 2, ,L]. The number of pixels at level i is
denoted by ni and the total number of pixels by N = n1 +
n2 + + nL* In order to simplify the discussion, the gray-
level histogram is normalized and regarded as a
probability distribution:

/ N , (1)

Now suppose that we dichotomize the pixels into two
classes C
O
and C
1
(background and objects, or vice versa)
by a threshold at level k; C
O
denotes pixels with levels [1,
, k], and C1 denotes pixels with levels [k + 1, , L]. Then
the probabilities of class occurrence and the class mean
levels, respectively, are given by
)=

)= 1 -
And
) =

Where

And

Are the zero
th
and first order cumulative moments of
the histogram up to the kth level, respectively, and
=
Is the total mean level of the original picture. We ca
easily verify the following relatin for an choice of k;
,
The class variances are given by,
I - )
2
=
2
,
I - )
2
=
2

These require second-order cumulative moments
(statistics).In order to evaluate the "goodness" of the
threshold (at level k), we shall introduce the following
discriminant criterion measures
(or measures of class separability) used in the
discriminant analysis [5]:

/ , K = /
Where

)
2
+ )
2

= (
2

and
)
2

are the within-class variance, the between-class
variance, and the total variance of levels, espectively.
Then our problem is reduced to an optimization problem

to search for a threshold k that maximizes one of the
object functions (the criterion measures) in (12). This
standpoint is motivated by a conjecture that well
thresholded classes would be separated in gray levels, and
conversely, a threshold giving the best separation of
classes in gray levels would be the best threshold. The
discriminant criteria maximizing A, K, and q,
respectively, for k are, however, equivalent to one
another, because the following basic relation always
holds:

It is noticed that are functions of
threshold level k, but is independent of k. It is also
noted that is based on the second-order statistics (class
variances), while ( is based on the first-order statistics
(class means). Therefore, is the simplest measure with
respect to k. Thus we adopt q as the criterion measure to
evaluate the "goodness" (or separability) of the threshold
at level k.
The optimal threshold k* that maximizes , or
equivalently maximizes is selected in the following
sequential search by using the simple cumulative
quantities (6) and (7), or explicitly using (2)-(5):
(k)= (k) /

and the optimal threshold k* is
B
2
(k
*
) = max
B
2
(k)
i k L
From the problem, the range of k over which the
maximum is sought can be restricted to
S
*
={k ;
0
1
= (k) [1- (k) ] > 0, or 0 <(k) <1 }
We shall call it the effective range of the gray-level
histogram. From the definition in (14), the criterion
measure
B
2
(or ) takes a minimum value of zero for
such k as k S - S* = {k; = (k) = 0 or 1} (i.e., making
all pixels either C
l
or C
O
, which is, of course, not our
concern) and takes a positive and bounded value for k
S*. It is, therefore, obvious that the maximum always
exists.

VI. EXPERIMENTAL RESULTS

Fig 3 : Original Image

a) Gray level thresholding
Gray Level Enhanced Image

Fig: 4 . lower threshold value:30
upper threshold value:60


Fig: 5. lower threshold value:80


Fig: 6. lower threshold value:150


Fig: 7. lower threshold value: 180

b) Otsus thresholding

OTSU Segmented Image, BW Threshold Level:0.244

Fig: 8. Binary image from Otsus thresholding
VII. CONCLUSIONS
The thresholding using manual gray level thresholding
and otsus thresholding produces threshold value. The
Otsus thresholding produces robust and accurate
threshold value which will help for accurate skull
stripping .Furthermore, morphology segmentation using
Otsus thresholding took shorter time to complete
compared to other method such as region growing For
future development, further research may focus on
overcoming problem of segmenting similar intensitiy
noncerebral and intracranial tissue (i.e. meninges) of the
brain. Next, mathematical morphology operations (i.e.
erosion,dilation and region filling) are applied to the
binary image to remove the non-cerebral tissue. The
concept is to convolve the binary image with a structuring
element to produce the skull stripped image.

REFERENCES

[1] Rosniza Roslan,Nursuriati Jamil , Rozi Mahumud , Skull Stripping
of MRI Brain Images using Mathematical Morphology , IEEE EMBS
conference on Biomedical Engineering and Science(IECBES 2010),30
th

November-2
nd
December 2010.

[2] Rosniza Roslan , Nursuriati Jamil , Rozi Mahumud , Skull
Stripping Magnetic Resonance Images Brain Images : Region Growing
versus Mathematical Morphology , International Journal of Computer
Information Systems and Industrial Management Applications , ISSN
2150-7988 Volume 3 (2011) pp.150-158.

[3] N. Otsu, A Threshold Selection Method from Gray-Level
Histograms, IEEE Transactions on Systems, Man, and Cybernetics,
Vol. 9, No. 1, pp. 62-66, Jan 1979.

[4]R.C. Gonzales and R.E. Woods, Digital Image Processing, Second
Edition. Prentice Hall, 2002.

1

India.
ISBN: 978-81-920575-9-0 :: doi: 10. 73046/ISBN_0768
IMPLEMENTATION OF DIFFERENTIATOR USING PFG
Ms. Sonalee P. Suryawanshi
Electronics
YCCE
Nagpur, India

Abstract - In this paper we represent pseudo floating gate
and its bidirectional property. We are using this
bidirectional property to implement the differentiator. We
are first implementing inverter using pseudo floating gate.
Inverter can be implemented using bidirectional property.
The inverter can be made bidirectional simply by
interchanging vdd and gnd and no need to add any circuitry
or any amplifier. The bidirectionality of the gate is further
evolved to be able to control signal flow conditions. And
finally using this inverter we are implementing
differentiator. Typical applications are in filter design and
IO ports in ICs. Linearity and AC simulations are presented
to show the good properties and versatility suited for Bi-
directional analog circuit design.
Keywords - Pseudo Floating Gate(PFG), Semi
Floating Gate(SFG), Nonvolatile Floating Gate(NFG)
I. INTRODUCTION
Capacitively connecting the input signals to the
transistor we achieve a floating-gate (FG). As the name
implies, the gate is actually floating gate technology has
beginning in late sixties , and now becoming mature
technology. Kahng and Sze first praposed the concept of
floating gate device in 1967. The first commercially
available floating gate based memory, the floating gate
avalanche MOS device was (FAMOS) was developed in
1970. The initial charge on the floating-gates may vary
significantly and therefore substantial inaccuracy may
occur unless some form of initialisation is applied. In
order to control or assert a desired voltage level different
techniques are available. Floating gate can be categarised
as follows.
Non-volatile floating-gate (NFG). Programmed
by use of Fowler-Nordheim Tunneling, hot carrier
injection or UV activated conductances. Used as non-
volatile analog memories in neural systems, and used for
threshold shift for low-voltage/low-power digital (binary
and multiple valued) and analog circuits. No leakage and
hence no requirement for keeping the non-volatile-
floating-gates at a specific dc level.
Semi-floating-gate (SFG). Initialized, or
recharged, frequently through active devices
(transistors).The SFG circuits resembles analog switch-
cap circuits or precharge logic. SFG circuits can be used
to implement digital (binary and multiple-valued) and
analog circuits. The frequency of the recharge is
determined by the global clock signals. There is a
significant leakage at the floating-gates due to the
switches connected to the gates.
Pseudo-floating-gate (PFG). Initialized through
passive components or transistors operating in weak
inversion. A local feedback will force the pseudo-
floating-gate and the output of a gate towards an
equilibrium state when dc inputs are applied. The weak
initialization is a continuous
process which will have no effect when high
frequency input signals are applied. Continuous leakage
forcing the floating-gate and the output of a gate towards
the equilibrium state, preferably Vpfg = Vout = Vdd/2.
The dc level for the output of the PFG gate will be equal
to the equilibrium voltage.
In section II we present a analog PFG inverter and
describe the inverters operation principle. In section III
the bi-directional inverter is introduced. In section IV a
CSPFG differentiator circuit is presented along with its
transient and AC simulation. In section V is conclusion of
the work. The simulation is done in this paper is done in
035 nm environment.
II. PFG INVERTER
The analog PFG inverter is shown in figure 1.
This circuit is an current-starved inverter with a weak
positive feedback. The positive feedback circuitry is
basically a current starved inverter where the PMOS
transistors are connected between the output and GND
and the NMOS transistors between output and VDD, the
opposite of the inverter. This positive feedback has 2
important functions in the circuit, it sets the operational
point (DC level) of the circuit and it decides the lower
cut-off frequency. Bias voltages on Vhf and Vlf limit the
current flowing through the inverter and the feedback
circuitry respectively. Input capacitances, C1 and C2,
block DC signals from the other circuitry connected to the
inputs and make the circuit floating C1 and C2can also be
used for weighting the input signals compared to each
other simply by choosing the right values.
o
n
b
b

III
Figure 3 s
obtained simp
no extra circ
bidirectional.
bidirectional P
Figure 1: Sym
Figure
I. BIDIRECTIO
Figure 3: B
shows bidirec
ply by swappi
cuit or ampl
The respon
PFG inverter i
mbol for PFG inv

2: PFG inverter
ONAL PFG INV
Bidirectional Inver

ctional PFG i
ng the vdd an
ifier is requi
nse of PFG
is shown in fig

verter

VERTER
rter
inverter which
nd gnd and he
ired to make
G inverter
gure 4.
h is
ence
e it
and
d
fo
c
p
c
im
d

Fi
Figure 5
differentiator i
forwarding in
consisting ana
produces an ou
change of the
mplement. Fig
differentiator. t

F
igure4: Input Out
IV. PFG D
shows a
is implemente
nverter A1
alog inverters
utput voltage t
e input volta
gure 6,7,8 sh
these all are ex
Figure 5: Differen
tput Waveform of
DIFFERENTIATO
PFG Diffe
ed using three
and a feed
s A2 and A
that is proport
age. This we
ows the vario
xpected respo
.
ntiator using PFG
f Inverter
OR
erentiator. T
inverter. i.e t
dback circui
A3. This circ
tional to the r
e are going
ous responses
nses.
inverter
2

The
the
itry
cuit
rate
to
of
m
h
h
T
u
t
c
c
Figur

Figure 7

Figu
The
multifunctiona
have the adv
high gain and
The bi-directi
useful in filt
theory these
computer wh
continuous tim
re 6: PFG differen
7: PFG differentia
ure8: PFG differen
V. C
PFG inverte
al, multi-direc
vantages offer
d ability to ad
ional different
ter and high
circuits can
hich can sol
me.
ntiator response f
ator response for t
ntiator response fo
CONCLUSION
er enables de
ctional circuit
red by PFG i
djust the freq
tiator presente
frequency ci
be used to
lve differenti
for pulse wave
triangular wave

for Sine Wave
esign of ana
ts. These circ
inverters such
quency behaiv
ed in this pape
ircuit design.
construct ana
ial equations
alog
cuits
h as
vior.
er is
. In
alog
in
[
[
[
[
[
[
[1]. M. Az
Pseudo
IEEE I
System
[2]. O. Mir
FLOAT
Interna
System
[3]. O. Mir
using p
on Elec
[4]. Shweta
CMOS
Interna
Techno
[5]. Shweta
Multipl
Interna
Techno
[6]. Y. Ber
pass ps
Confer
2007.

REF
zadmehr, Y. Ber
o-Floating-Gate
International Con
ms, ICESC 2008 M
rmotahari, Y. Be
TING-GATE
ational Conferen
ms,2010
rmotahari, Y. Ber
pseudo floating g
ctronic Design, Te
a P. Hajare and M
AD/DA Convert
ational Confere
ologies in Commu
a P. Hajare and
le-Valued Pseud
ational Confere
ologies in Commu
rg, M. Azadmehr
seudo floating
ence on electroni
FERENCES
rg. Bi-direction
differentiator /
nference on elect
Malta.
erg RECONFIGU
ANALOGCIR
nce on Electron
rg. Proposal for
gate, IEEE Inter
est and Applicatio
Monica V. Mankar
ter using Pseudo F
nce on Adva
unication and Com
Monica V. Mank
do Floating G
nce on Adva
unication and Com
r, O. Mirmotahar
g-gate amplifier
ics, Circuits a
nal Current star
integrator, Pr
tronics, Circuits
URABLE PSEU
RCUITS IE
nics, Circuits
a bidirectional g
rnational Symposi
ons, 2008.
r Implementation
Floating Gate IE
ances in Rec
mputing 2010.
kar Multiple-In
Gate DAC IE
ances in Rec
mputing 2010.
ri, S. Aunet. B
, IEEE Internatio
and Systems, ICE
3
rved
roc,
and
UDO
EEE
and
gate
ium
n of
EEE
cent
nput
EEE
cent
and
onal
ESC
India.
ISBN: 978-81-920575-9-0 :: doi: 10. 73060/ISBN_0768
Extraction of Signal features from Deaf-mute Voice Signal

J.NirmalaDevi Dr.K.BommannaRaja
Department of Information Technology Department of Electronics & Communication Engineering
Excel College of Engineering for Women Excel College of Engineering for Women
Namakkal, India Namakkal, India

Abstract Speech plays a vital role for a human to
communicate. Speech is produced by the acoustic excitation
of the vocal tract by an air stream derived from the lungs and
pulsed at a rate that is determined by the vibration of the
speakers vocal folds. Deaf-mute persons feel hard to produce
speech and they can be of partially-mute, where there is
possibility of partial communication and completely-mute,
where there is no way for communication. Various methods
like sign language and Alternative and Augmentative
communication (AAC) are used to help the deaf-mute people
to communicate. Existing research is using
communication/voice aid to help the deaf-mute community. In
this article, we present a method to produce synthetic speech
from deaf-mute voice which is equivalent to normal speech
quality using speech signal processing techniques.
Keywords- Speech Production; Deaf-mute; Speech
reconstruction; Speech quality; Feature Extraction; Text-to-
Speech Converter; Synthetic Speech.
I. INTRODUCTION
Lungs, vocal tract, and larynx are the main organs
related with generation of sound. The lungs are the source
of airflow. The vocal tract is an acoustic enclosure and acts
as acoustic filter shaping the spectrum of the generated
sound. The source of most speech occurs in the larynx.
There are two folds of the muscular bundle known as vocal
chords inside the larynx. These vocal folds obstruct the
airflow from the lungs and produce audible vibrations that
make the speech. The mechanism of sound production is
known as phonation. These vibrations consist of three
phases namely contact phase, separation phase and open
phase [1].
The frequency of the vocal fold vibration determines the
pitch of the voice which is directly correlated to intonation
[2].The manner in which the vocal folds vibrate contributes
to the sound quality produced by the speaker. Also, it has
been hypothesized that pathological larynx manifests itself
in an abnormal vibratory pattern [3].
During normal breathing, air passes the larynx and vocal
tract gets unobstructed, creating little or no sound. When
the vocal tract is opened or totally closed, the airflow is
interrupted to create turbulent pulses of air. When the vocal
folds are adducted and air is expired, sub glottal air-pressure
pushes the vocal folds apart, and air flows rapidly (Fig 1).
This immediately creates a partial vacuum between the
vocal folds by Bernoullis principle, which pulls them once
again towards each other. It stops the airflow, building
pressure again so that the folds again open and thus, a
vibratory motion is set in.

Figure 1. The vocal fold Vibration cycle
All vowels are voiced. Some consonants are
voiced(b,d,g,v,z,zh,j) while others are unvoiced
(p,t,k,f,s,sh,ch). In whispered speech this difference cannot
be detected and hence humans can generally understand
whispered speech. This shows the presence of substantial
redundancy in speech because the whole signal is not
required in order to determine the meaning.
The vital information can be obtained from the vocal
tract messages using the spectrogram analysis. This
provides an idea, with partial availability of speech
information, it is possible to synthesize speech signal for
Deaf-mute.
II. LITERATURE REVIEW
A large set of methods can be used to assess speech
disorders like discussion with the patient, endoscopic
examination of the larynx, postural behavior of the patient
[4], psychological and behavioral profile [5], and auto-
evaluation as Voice Handicap Index questionnaire [6],
perceptual judgment [7], or instrumental assessment [8].
With the available pathological and normal speech
samples it was possible to distinguish between normal and
pathological signal. Three level wavelet transform is used
and the normalized energy ratios between the levels and
normalized peak-to-peak values are used as parameters [9].
The visualized speech in a test result shows that the
method has very good robustness; the correct answer rate
reaches 94.56% of vowel and consonant [10]. In a novel
speech plot display approach that extracts combined time-
frequency feature as a new feature vector to present signal.
The combined feature is displayed on CRT by plot patterns
and the deaf-mute can make use of their own brain to
identify speech for training their oral ability effectively
[11].
The methods alaryngeal speech, esophageal speech, and
tracheoesophageal voice prosthesis cannot be applied to
the patients whose some parts of vocal folds or larynx has
been removed [12]. Several Systems that analyze and
enhance the characteristics of the esophageal speech and
speech using electrolarynx have been designed so far [13]
[17]. A MELP based real time system that produces
synthetic speech for patients with chronic Dysphonia has
also been presented [12].
The acoustic differences are observed between normal
and dysphonic speech is determined. Also the acoustic
parameters of the dysphonic voiced phonemes are
modified in order to obtain a synthetic speech that is closer
to normal speech. Enhanced speech is synthesized by
MELP [18].
From the literature there are many possible methods
available like sign language for the respective language for
the deaf-mute communication. Many of the previous
researchers have been focusing on visual
communication/visual aids to help the deaf-mute
community. All the possible features cannot be achieved
by the existing methods.

III. IMPLEMENTATION
The present implementation aims at providing a
methodology for identifying purposeful meaning of the
speech produced by a Deaf-mute patient. The steps
involved in this process includes,
A. Data Acquisition & Feature Extraction
B. Speech Registration
C. Comparison of Deaf-mute speech signal with the
database
D. Synthetic Speech generation
In this paper two different approaches are suggested to
synthesize the Deaf mute speech namely,
1. Synthetic speech generation by extracting normal
speech.
2. Synthetic speech generation with Text-to-Speech
Converter.
A. Data Acquisition
(a) Acquiring Speech: For training, speech is acquired
from a normal and Deaf mute Speech signals using
microphone and used for offline analysis. Then the speech
is continuously streamed into the environment for online
processing.
During training stage, it is necessary to record repeated
utterances. With a PC sound card the speech is captured for
10 Seconds from a microphone at 8000 samples Per
Second. However the speech is continuously acquired and
buffered. At the same time, the incoming speech is
processed frame by frame or in continuous groups of
samples.
(b) Analyzing the acquired speech & Feature extraction:
Deaf mute speech differs from normally phonated speech
in terms of voicing, pitch, and formant structure. Fig 2
clearly shows that, contrary to the voiced phonemes of
normal speech, there is no perceivable pitch period or
voicing observed in the voiced phonemes of deaf mute
speech. In addition to this, voiced phonemes of deaf mute
speech differ from the voiced phonemes of normal speech
in terms of formant distortion. Bandwidths of deaf mute
phonemes are larger, and their formant frequencies are
greater. However, in unvoiced phonemes of deaf mute
speech, there is no significant formant distortion observed
[5].
The Deaf mute speech detection algorithm is developed
by processing the prerecorded speech. Here Signal energy
works well for detecting voiced signals and Zero-Crossing
counts well for unvoiced signals. The acoustic features of
Deaf mute Speech signal is extracted using the parameters
like Mean Square (MSR), Excess Co-efficient (EX),
Spectral Flatness of Residue (SFR), JITTER Perturbation
of Fundamental Frequency, PEAK, Peak to
Autocorrelation (P/A) [19]. The extracted feature is used to
evaluate the Deaf mute Speech.
B. Speech Registration
Every Deaf mute Speech Signal is measured and
registered using extracted feature value before storing in
the database. The feature value is a numerical measure and
that correlates approximately to speech samples. Hence the
database has collection of Deaf mute speech samples along
with the respective feature values.
C. Comparison of Deaf mute and Normal Speech Signal
Whenever an input Deaf mute Speech signal is received
by the system, automatically it is measured and the feature
value is extracted along with the numerical measure. If the
feature value exactly matches the database either through
direct normal speech extraction or Text-to-Speech
converter, then it is used to reproduce the synthetic speech.
Or else approximate or nearest feature value is taken into
account.
If a new Deaf mute speech signal is received as an input
it will be automatically updated in the database along with
its extracted feature value. Finally the normal speech signal
is synthesized after comparison of the Deaf mute Speech
Signal with the extracted feature value in the database.
Deaf mute Speech Signal is measured and registered using
extracted feature value before storing in the database. The
feature value is a numerical measure and that correlates
approximately to speech samples. Hence the database has
collection of Deaf mute speech samples along with the
respective feature values.
D. Synthetic Speech Generation
Deaf mute speech differs from normally phonated
speech in terms of voicing, pitch and formant structure.
There is no perceived pitch period in deaf mute speech and
the voice is definitely noisy.

Figure 2. Implementation of Synthetic speech generation using direct feature extraction

Figure 3. Implementation of Synthetic speech generation using Text-to-Speech Converter

Update the database with
New Speech Signal
Input Deaf-mute
Speech
Signal
Feature
Extraction
Synthetic Speech
Generator
Normal Speech Signal
Database
Does Input Deaf-
mute Speech
Signal matches
with the database
Database containing the
sampled Deaf-mute Speech
Signals
YES
NO
Extract Equivalent Normal
Speech Signal
Update the database with
New Speech Signal
Input Deaf mute
Speech
Signal
Feature
Extraction
Synthetic Speech
Generator
Normal Speech Signal
Database
Does Input
Deaf mute
Speech Signal
matches with
Database containing the
sampled Deaf mute Speech
Signals
Reproduced Synthetic
YES
NO
Text-to-Speech Converter
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
-1
-0.5
0
0.5
1
x 10
-3
Time (s)
A
m
p
l
i
t
u
d
e

Waveform
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
-100
-50
0
Frequency (Hz)
M
a
g
n
i
t
u
d
e

(
d
B
)

Spectrum
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
0
50
100
Quefrency (s)
A
m
p
l
i
t
u
d
e

Cepstrum
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18
-2
-1
0
1
2
x 10
-3
Time (s)
A
m
p
l
i
t
u
d
e

Waveform
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
-80
-60
-40
-20
0
Frequency (Hz)
M
a
g
n
i
t
u
d
e

(
d
B
)

Spectrum
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
0
100
200
300
Quefrency (s)
A
m
p
l
i
t
u
d
e

Cepstrum
0 1000 2000 3000 4000 5000 6000
0
5
10
15
20
25
30
35
d
B
0 0.5 1 1.5 2 2.5 3
x 10
5
-600
-400
-200
0
200
400
600
A
m
p
l.
Samples
0 1000 2000 3000 4000 5000 6000
0
10
20
30
40
d
B
0 0.5 1 1.5 2 2.5 3
x 10
5
-4000
-3000
-2000
-1000
0
1000
2000
3000
A
m
p
l.
Samples
(a) Method 1: After comparison of the input Deaf mute
Speech Signal, direct feature extraction of a relevant normal
speech signal may be retrieved from the database. The Fig 2
illustrate the implementation of synthetic speech generation
using direct feature extraction is shown. The feature is
extracted using the basic acoustic features pitch,
fundamental frequency and formant analysis which can
differentiate the normal and deaf mute speech. The unvoiced
phoneme is unmodified and only the voiced phonemes are
enhanced using MELP [18] for synthetic speech generation.
Then the normal speech is transferred to the Synthetic
Speech generator which checks the quality and reproduces
the complete normal speech signal.
(b) Method 2: In this method a Text-to-Speech converter is
used to generate the synthetic speech. The implementation
diagram of this method is shown in the Fig 3. Speech
synthesis programs convert written input to spoken output
by generating synthetic speech. These are often referred to
as Text-to-Speech conversions (TTS). Synthesized TTS
creates speech by generating sounds through a digitized
speech format. This output sounds more like a computer
than a human. The speech signals are measured by their
intelligibility, naturalness and test preprocessing capabilities
(ability to convert acronyms into normal speech).
In this method a specified database is created which
contains number of natural speech signals of normal persons
from which the equivalent pattern of normal speech is
identified for the respective received deaf mute signal based
on the acoustic characteristics of the person. This may
provide synthetic speech production based on the persons
individuality. Then the normal speech is transferred to the
Synthetic Speech generator which checks the quality and
reproduces it as a complete normal speech signal.

Figure 4. Waveform, Spectrum and Cepstrum of Normal Signal

Figure 5. Waveform, Spectrum and Cepstrum of Deaf mute Signal

Figure 6. Energy Diagram of Normal Signal

Figure 7. Energy Diagram of Deaf mute Signal
IV. RESULT AND DISCUSSION
Spectrum analysis is used to calculate pitch,
fundamental frequency and formant analysis. Spectral
analysis also contains very useful information about the
speech signal.
It is possible to calculate the spectrum of the entire
sound or of part of it, and so to obtain a close realization of
the static and dynamic spectral analysis of the sound signal.
In a static analysis we assume a case where the spectrum is
calculated from the entire sound fragment from the
beginning to the end. In the case of a dynamic spectral
analysis we deal with a spectral calculation for a window of
defined size, thus using only part of the file. This window is
moving along the file. Hence, we can trace a frequency that
changes in amplitude in the time flow.
The turbulence in glottal flow resulting from the
malfunction of the vocal folds can be quantified by the noise
in spectral components of speech. Speech disorders reduce
the energy of the harmonic structure of the spectrum and
increase the non harmonic structures. This distortion relates
to the disease extension and several spectral measurements
in order to assess the speech disorder.
The relationship between harmonic components and the
remaining ones [20] [24], the difference among the low-
frequency harmonics [25], the amount of spectral energy
starting from a specific frequency[26], and the analysis of
the least spectral magnitude at specific frequency intervals
[27]-[28] are some of the spectral measurements proposed
in the literature.
A spectrogram is an image that shows how the spectral
density of a signal varies with time. Spectrograms are used
to identify phonetic sounds, to analyze the speech
processing. From the spectrogram the pitch frequency of the
normal and disordered speech is clearly identified and the
formant locations and bandwidths of deaf mute speech
differ from normally phonated speech is also seen.
A formant structure distortion can be observed in voiced
phonemes, while there will be no significant distortion
observed in unvoiced ones. It can be shown that voiced
frequency bands of the unvoiced phonemes, which are
pronounced by a deaf mute speaker, and normal words are
not different contrary to the voiced frequency bands of
voiced phonemes. Also unvoiced phonemes have no
perceived pitch when they are pronounced by a normal
speaker. Using the basic formant analysis we can generate
the synthetic speech.
The Fig 8 (a) and 8 (b) is shown above which gives the
speech spectrum of both normal and disordered voice where
the enhancement of synthetic speech has to be done on the
red color region of the spectrogram where the energy is
more. The amplitude of unvoiced segments is noticeably
lower than that of the voiced segments in Fig 6. Also in the
Deaf mute speech the person has to strain much to create
sound so that the energy level of the Deaf mute signal is
higher than the normal signal as in Fig 7.

Figure 8. Spectrogram for tamil word amma from (a) Normal speech
signal repeated three times (b) Deaf mute speech signal repeated three
timesergy Diagram of Deaf mute Signal
V. CONCLUSION
This article provides two different methods for
generation of synthetic speech using Deaf mute. The
suggested methods may be a solution to develop a complete
automatic assessment for Deaf mute speech. The voiced
segments are identified and replaced by equal segments of
normal speech. It follows from the fact that the
spectrograms of enhanced and normal speech are quite
similar. Feature extraction method proposed provides better
detection of deaf mute speech features than any other
methods like LPC and MELP. The generated speech signal
can be used to communicate with the normal persons and
can be applied in several practical situations and any
electronic gadgets. The suggested method will improve
voice and life quality of deaf mute person.
REFERENCES
[1] Anil Lurtthra,An Impedence Glottograph, M.Tech Dissertation,
Supervisor:Prof.P.C.Pandey, Dept.of Electrical Engineering, IIT Bombay,
July 2004.
[2] A.J.Fourcin and E.Abberton,First applications of a new larynograph,
Medical and Biological Illustration, Vol 21, pp 172-182, July 1971.
[3] D.G.Childers and J.N.Larar, Electroglottography for laryngeal
function assessment and speech analysis,IEEE Trans, Biomedical
Engineering, Vol BME-31, No.12, pp 802-817, Dec 1984.
[4] J. D. Hoit, Influence of body position on breathing and its implications
for the evaluation and treatment of speech and voice disorders, Journal of
Voice, vol. 9, no. 4, pp. 341347, 1995.
[5] N. Roy, D. M. Bless, and D. Heisey, Personality and voice disorders: a
multitrait-multidisorder analysis, Journal of Voice, vol. 14, no. 4, pp. 521
548, 2000.
[6] B. H. Jacobson, A. Johnson, C. Grywalski, et al., The Voice Handicap
Index (VHI): development and Validation, American Journal of Speech-
Language Pathology, vol. 6, no. 3, pp. 6669, 1997.
[7] P. H. Dejonckere, C. Obbens, G. M. de Moor, and G. H. Wieneke,
Perceptual evaluation of dysphonia: reliability and relevance, Clinical
Linguistics and Phonetics, vol. 45, no. 2, pp. 7683, 1993.
[8] P. H. Dejonckere, P. Bradley, P. Clemente, et al., A basic protocol for
functional assessment of voice pathology, especially for investigating the
efficacy of (phonosurgical) treatments and evaluating new assessment
techniques: guideline elaborated by the Committee on Phoniatrics of the
European Laryngological Society (ELS), European Archives of Oto-
Rhino-Laryngology, vol. 258, no. 2, pp. 7782, 2001.
[9] Cheol-Woo Jo, Dae-Hyun Kim, Analysis of Disordered speech signal
using wavelet transform.
[10] HAN Zhi-qiang et.al, Speech Visualization Research for Deaf-
mute, International Conference on Computer Application and System
Modeling (ICCASM 2010).
[11] Xu Wand et.al, Speech plot display for the Deaf-mute based on
Combined characters encoding of Speech signal, International Conference
on Complex Medical Engineering, IEEE/ICME 2007.
[12] H.Irem Turkmen and M.Elif Karsligil, Dysphonic Speech
Reconstruction, IEEE Engineering in Medicine and Biology Magazine,
pp. 135-138, Mar / Apr 2010.
[13] G. Aguilar and M. Nakano-Miyatake, Alaryngeal speech
enhancement using pattern recognition techniques, IEICE Trans. Inform.
Syst., vol. 88, no. 7, pp. 16181622, 2005.
[14] N. Bi and Y. Qi, Speech conversion and its application to alaryngeal
speech enhancement, in Proc. 3rd Int. Conf. Signal Processing (ICSP96),
1997, pp. 15861589.
[15] A. Pozo and S. Young, Continuous tracheoesophageal speech
repair, in Proc.European Signal Processing Conf. (ICSP96), Florence,
Italy, Sept. 48, 2006.
[16] Y. Qi, B. Weinberg, and N. Bi, Enhancement of female esophageal
and tracheoesophageal speech, J. Acoust. Soc. Amer., vol. 98, no. 5, pp.
24612465, 1995.
[17] H. Sawada, N. Takeuchi, and A. Hisada, A real-time clarification
filter of a dysphonic speech and its evaluation by listening experiments,
in Proc. Int. Conf. Disability, Virtual Reality and Associated Technologies
(ICDVRAT04), 2004, pp. 239246.
[18] H.Irem Turkmen and M.Elif Karsligil, Reconstruction of Dysphonic
Speech by MELP.
[19] Acoustic characteristics of normal and pathological voices, in
Speech and Language: Advances in Basic Research and Pratice, N. J. Lass,
Ed. New York: Academic, 1979, vol. 1, pp. 271335.
[20] S. N. Awan and M. L. Frenkel, Improvements in estimating the
harmonics- to-noise ratio of the voice, J. Voice, vol. 8, pp. 255262, 1992.
[21] T. Fukazawa, A. El-Assuooty, and I. Honjo, A new index for
evaluation of the turbulent noise in pathological voice, J. Acoust. Soc.
Amer., vol. 83, pp. 11891193, 1988.
[22] N. Hiraoka, Y. Kitazoe, H. Ueta, S. Tanaka, and M. Tanabe,
Harmonic intensity analysis of normal and hoarse voices, J. Acoust. Soc.
Amer.,vol. 76, pp. 16481651, 1984.
[23] G. de Krom, A cepstrum-based technique for determining a
harmonics- to-noise ratio in speech signals, J. Speech Hear. Res., vol. 36,
pp. 254266, 1993.
[24] E. Yumoto, W. J. Gould, and T. Baer, Harmonics-to-noise ratio as an
index of the degree of hoarseness, J. Acoust. Soc. Amer., vol. 71, pp.
15441549, 1982.
[25] G. de Krom, Some spectral correlates of pathological breathy and
rough voice quality for different types of vowel fragments, J. Speech
Hear.Res., vol. 38, pp. 794811, 1995.
[26] K. Shoji, E. Regenbogen, J. D. Yu, and S. M. Blaugrund, H-index: A
new measure of glottal efficiency for the pathologic voice, Laryngoscope,
vol. 102, pp. 11131117, 1992.
[27] F.W. Emanuel and R. L. Whitehead, Harmonic levels and vowel
roughness, J. Speech Hear. Res., vol. 22, pp. 829840, 1979.
[28] M. A. Lively and F. W. Emanuel, Spectral noise levels and roughness
severity ratings for normal and simulated rough vowels produced by adult
females, J. Speech Hear. Res., vol. 13, pp. 498502, 1970.
India.
ISBN: 978-81-920575-9-0 :: doi: 10. 72948/ISBN_0768
IMPROVING THE PERFORMANCE OF SPEECH RECOGNITION USING
GENETIC ALGORITHM BASED ON ARTIFICIAL NEURAL NETWORK

S. Narendran *
Final-B.E. ECE, Ranganathan Engineering College
Coimbatore, India

Abstract: This paper deals widely with the speech
recognition technology by making use of genetic
algorithm. This paper elaborates the role of genetic
algorithms in recognizing human speech and thus
prevails a further step forward in the speech
recognition.

I. INTRODUCTION
The development for speech recognition system
has been for a while. The recognition platform can
be divided into three types. Dynamic Time
Warping (DTW) (Sakoe, 1978), the earliest
platform, uses the variation in frame's time for
adjustment and further recognition. Later, Artificial
Neural Network (ANN) replaced DTW. Finally,
Hidden Markov Model was developed to adopt
statistics for improved recognition performance.
Besides the recognition platform, the process of
speech recognition also includes: recording of
voice signal, point detect, pre-emphasis, speech
feature capture, etc. The final step is to transfer the
input sampling feature to recognition platform for
matching. In recent years, study on Genetic
Algorithm can be found in many research papers
(Chu, 2003a; Chen, 2003; Chu, 2003b). They
demonstrated different characteristics in Genetic
Algorithm than others. For example, parallel search
based on random multi-points, instead of a single
point, was adopted to avoid being limited to local
optimum.
In the operation of Genetic Algorithm, it only
needs to establish the objective function without
auxiliary operations, such as differential operation.
Therefore, it can be used for the objective functions
for all types of problems. Because artificial neural
network has better speech recognition speed and
less calculation load than others, it is suitable for
chips with lower computing capability.
Therefore, artificial neural network was adopted
in this study as speech recognition platform. Most
artificial neural networks for speech recognition are
back-propagation neural networks. The local
optimum problem (Yeh, 1993) with Steepest
Descent Method makes it fail to reach the highest
recognition rate. In this study, Genetic Algorithm
was used to improve the drawback.
Consequently, the mission of this chapter is the
experiment of speech recognition under the
recognition structure of Artificial Neural Network
(ANN) which is trained by the Genetic Algorithm
(GA). This chapter adopted Artificial Neural
Network (ANN) to recognize Mandarin digit
speech. Genetic algorithm (GA) was used to
complement Steepest Descent Method (SDM) and
make a global search of optimal weight in neural
network. Thus, the performance of speech
recognition was improved. The nonspecific speaker
speech recognition was the target of this chapter.
The experiment in this chapter would show that
the GA can achieve near the global optimum search
and a higher recognition rate would be obtained.
Moreover, two method of the computation of the
characteristic value were compared for the speech
recognition. However, the drawback of GA used to
train the ANN is that it will waste many training
time. This is because that the numbers of input
layer and output layer is very large when the ANN
is used in recognizing speech. Hence, the
parameters in the ANN is enormously increasing.
Consequently, the training rate of the ANN
becomes very slow. It is then necessary that other
improved methods must be investigated in the
future research.
The rest of this chapter is organized as follows.
In section II, the speech pre-processing is
introduced. Then, in section III we investigate the
speech recognition by ANN which is trained by
genetic algorithm to attain global optimal weights.
Section IV presents the experiment result of the
speech recognition. Finally, in section V, we make
some conclusions about this chapter.

II. SPEECH PRE-PROCESSING
The speech signal needs be pre-processed prior
to entering the recognition platform. The speech
pre-processing includes point detection, hamming
windows, speech feature, etc. Each process is
illustrated as follows.
A. Fixed-size frame and Dynamic-size frame
With fixed-size frame, the number of frame
varies with speech speed due to different length of
voice signal. This problem does not exist in DTW
recognition system. However, this study with
artificial neural network ANN has to use dynamic-
size frame to obtain a fixed number of frames.
There are two methods to get a fixed number of
frames:
(1) Dynamic numbers of sample points
(2) dynamic overlap rates (Chen, 2002).
Either of the two methods can lead to a fixed
numbers of frames, meeting the requirement by
ANN recognition system.
B. Point Detection
A voice signal can be divided into three parts:
speech segment, silence segment and background
noise. How to differentiate between speech
segment and silence segment is called point
detection. After removal of unnecessary segments,
the time frame in comparison is narrowed and the
response time is shortened.

There are a number of algorithms for speech
end point detection. In general, there are three types
based on parameters: (1) time domain point
detection method, (2) frequency domain point
detection method, (3) Hybrid parameters point
detection method.
Among the three, the time domain point
detection is the simplest and the most frequently
used, but has the shortcoming of lower noise
resistance. On the other hand, frequency domain
point detection and hybrid parameters point
detection have higher noise resistance and better
accuracy, but is more complicated in calculation. In
this chapter, we adopted the time domain point
detection method for shortening the computation
time.
C. Hamming Window
The purpose to fetch hamming window is to
prevent discontinuity in every frame and both ends
of every frame. When the voice signal is multiplied
by hamming window, we can reduce the effect of
discontinuity (Wang, 2004).
D. Feature capture
In general, there are two types of features
capturing of voice signal for speech recognition,
time domain analysis and frequency domain
analysis. Of the two, time domain analysis is more
straightforward and involves less calculation, so it
saves time. Frequency domain analysis needs to go
through Fourier transform first, so it involves more
calculation and complexity and takes more time.
Common speech features capturing include Linear
Predict Coding (LPC) (Chen, 1994), linear predict
cepstrum coefficient (LPCC), Mel-frequency
Cepstrum coefficient (MFCC) (Chu, 2003b) etc.
With consideration of recognition accuracy, this
study selected MFCC as the speech feature
capturing method. Using MFCC to obtain solutions
involves three steps: 1. Using Fast Fourier
Transform (FTT) to obtain power spectrum of the
speech signal; 2. Applying a Mel-space filter-bank
to the power spectrum to get logarithmic energy
value; 3. We conduct the discrete cosine transform
(DCT) of log filter-bank energies to obtain MFCC.

III. SPEECH RECOGNITION PLATFORM
BPNN is the most commonly used structure in
ANN. Although ANN has fast recognition rate and
fault tolerance, it is not perfect because its SDM
has a problem with Local Optimum. To prevent
this from happening, GA is adopted to assist in
SDM for obtaining optimal weight and improved
recognition performance.
A. Back-propagation neural network
In principle, back-propagation neural network
uses Multiple-Layer Perception as system
framework and SDM as training rule. Such a
system is called back-propagation neural network.
Multiple-layer in Multiple-Layer Perception model
indicates it is composed of many layers of neurons.
Besides, the signal transmittance mode between
neurons in two layers is the same as that for a
single layer. The study adopted three- layer
structure (input layer, hidden layer, output layer) as
the basic framework for speech recognition, which
is depicted in Fig. 1.

B. Genetic algorithm
Genetic algorithm (Goldberg, 1989;
Michalewicz, 1999) involves Selection,
Reproduction and Mutation. The purpose of
selection is to determine the genes to retain or
delete for each generation based on their degree of
adaptation. There are two types of determination:
(1)Roulette-Wheel Selection
(2) Tournament Selection.
The study adopted tournament selection. It is to
follow their fitness function sequence for each gene
set to determine whether they are retained. The
fittest survives. Reproduction is a process to
exchange chromosomes to create the next
generation according to distribution rule. In
general, there are one-point crossover, two-point
crossover and uniform crossover, etc. The
evolutionary process of GA is depicted in Fig. 2.

Fig.1. Three layered structured ANN

The way of mutation is not very different from
that for crossover. There are one-point mutation
and two-point mutation. Mutation process depends
on conditions; for example, mutation can start as
adaptation function and stop changing after several
generations.
Mutation rate cannot be too high. Otherwise,
convergence will not occur.

Fig.2. Evolutionary process of Genetic Algorithm

IV. EXPERIMENTAL RESULTS
ANN with SDM training (i.e., BPNN) was first
adopted to be speech recognition platform. The
result was compared to GA assisted ANN platform.
Although genetic algorithm has advantages that
SDM cannot provide. It has one drawback, i.e. the
crossover, reproduction and mutation process needs
more time to seek optimal solutions.
In the initial speech, voice files for ten people
were collected. This file contains the voice of
numeral 1 ~ 9. Each people recorded four sets,
three of which were used for training and one of
which was for test. Recording format had sampling
frequency 8 kHz, mono channel 16 bit sampling
point. After recording, point detection was to
remove silence segment, followed by pre-emphasis.
Then, speech segment was divided to 20 fixed
frames. Feature parameter was extracted from each
frame. One frame had 10 features. Thus, each
number would have 200 features.
In the aspect of speech frame, because ANN
recognition platform was adopted, speech segment
needs the same number of frames regardless of its
length of time. The adopted dynamic-size frame
was different from the DTW fixed frame. The
study also adopted dynamic sampling point (fixed
overlap rate). The frame sampling point can be
expressed by the following equation (1) (Chen,
2002):

(1)
L represents the number of frame sampling
points, ls is the total signal length, N is the number
of frames, R is overlap rate (%), while Fix(x) the
maximum integer smaller than x. Through such
function, the same number of frames can be
obtained for different length of speech. Besides, the
point detection in the study still adopted time-
domain point detection due to the reason of less
calculation load. It took the average of energies for
the first few silence segments (background noise)
and added it to 5%~10% of the maximum frame
energy to set the threshold value for point
detection, as shown in the following equation (2)
(Chen, 2002):

(
2)
Threshold is the point detection threshold value.
N is the number of frames prior to point detection.
E(n) is the sum of energies for sampling points in
the nth frame prior to point detection. K represents
captured number of silence segment frames. E(i) is
the sum of energies for sampling points in ith
silence segment. 7.5% of the maximum frame
energy was added with the average energy for five
silence segment frames (K = 5) to establish point
detection threshold calculation.
After completion of point detection, pre-
emphasis and Hamming window were carried out
to capture features. With consideration of
recognition accuracy, MFCC parameter was
adopted for recognition. MFCC level in this
experiment was 10. After obtaining features, they
were input to recognition platform to start speech
recognition.
A. BPNN Experiment Results
In the design structure of artificial neural
network, there are two output modes. One used
binary coding to express output, for example, the
system has 16 corresponding outputs to four output
neurons. Thus, the number of output neurons was
reduced. The other was one-to-one output. For
example, 9 frames needed 9 outputs neurons.
Although binary coding can minimize the number
of neurons, it not only had low recognition rate, but
difficulty in convergence after experiment
comparison with one-to-one mode. Therefore, one-
to-one output was adopted here. The entire ANN
structure had 200 inputs, 30 neurons in hidden
layer and 9 neurons in output layer.
In the experiment, each speech had 20 frames.
Each frame had ten levels of features, indicating
200 input parameters as input layer, while 30
neurons were in the hidden layer.
The number of neurons cannot be too many;
otherwise, it cannot obtain convergence. If the
number is too small, recognition error will be large.
The number of neurons in the hidden layer (N_no)
is expressed by the following equation (3) (Yeh,
1993) :

(3)

In_number represents the number of input layer
units, while Out_number represents the number of
output layer units. Because the number of ANN
inputs is as high as 200, there will be a problem,
i.e. when the input approaches a relatively large
number, the output will be at the extreme condition.
To solve the problem, the speech features were all
downsized prior to input. It was to multiply the
speech feature by 0.01 to prevent transfer function
from over- saturation. From Fig. 3, it is known that
once the number of training generation is over
1,500, it then fails to breakthrough and the
recognition rate is 91% for the existing database.
Even further training is continued, root mean
square error does not progress, and recognition rate
does not improve either. Under this situation,
genetic algorithm is used to assist in seeking the
weight.
Figure.3. MSE for Output in ANN with SDM Training
Process
B. Experiment Results with Genetic Algorithm
to assist in Training
SDM in the above-mentioned artificial neural
network was used to seek the weight. Back
propagation neural network tended to have
problems with local optimum, so genetic algorithm
was used to seek the weight for the entire domain
(Chu, 2003a).
At first, the study used SDM to train ANN to
obtain the weight. Then converged weight and bias
were entered into genetic algorithm. It improved
the initial speed and also helped SDM out of the
local optimum. Fig. 4 shows the entire training
structure. The experiment has shown that SDM
training followed by GA training would help error
break the SDM limit and greatly improve
recognition rate. Fig. 5 shows the error range for
SDM training followed by GA.

Through 3000 generations of GA, MSE value
drops to 0.001 with recognition rate up to 95%, as
shown in Table 1.
Figure 4. The Proposed ANN Training Structure

Training after SDM Training

Figure 5. MSE for Output in Genetic Algorithm

V. CONCLUSION
In this chapter, we have seen that the
recognition rate through the SDM in BPNN is up to
91% under the MFCC feature. This recognition rate
is not the optimum because that the SDM can
always get local optimum. To solve this problem,
GA was adopted and following SDM to improve
MSE. By this two stage (SDM then GA) training
scheme, the recognition rate can be increasing up to
95%. However, under the condition of adopting
only MFCC parameters, speech recognition rate
still has room for improvement. For the future,
other modes of features capturing method can be
adopted, such as Cepstrum Coefficient or LPC
parameter together with pitch parameter, to
improve recognition rate.

Table.1. Speech Recognition Results after 10 sets of testing

VII. ACKNOWLEDGEMENT
I hereby feel responsible in thanking my sole
father and mother, who are marked in my every
walk of life. I also feel thankful to my staffs, who
are not only in current designations, but also had
imparted knowledge, since my primary classes. Its
my divine quality of being grateful to my college
management, who had sponsored me to participate
in this great event. I feel grateful to my beloved
principal, Prof. J. Amos Robert Jayachandran;
Head of the Department, Mrs. S. Kalyani Madam;
my project guide, Ms. M. Tamilselvi for guiding
me to this position. I also acknowledge my student
friends, who were of great support for my
knowledge gain. Last but not least, I thank Lord
Almighty for showering his blessings over me.

REFERENCES
[1] Chen, S. C. (2003). Use of GA in CSD Coded Finite
Impulse Digital Filter (FIR), Shu-Te University, MS Thesis,
Taiwan.
[2] Chen, M. Y. (1994). PC Computer Voice Operation, Chi
Biao Publication, 957-717-071-4, Taiwan.
[3] Combination of GA and SDM to Improve ANN
Training Efficiency, Shu-Te University, MS Thesis, Taiwan.
Chu, W. C. (2003b).
[4] Speech Coding Algorithms, John Wiley & Sons, 978-0-
471-37312-4, USA.
[5] Demuth, H. B., Beale, M. H., Hagan, M. T. (1996).
Neural Network Design, Thomson Learning, 0534943322, USA.
[6] Goldberg, D. E. (1989). Genetic Algorithms in Search,
Optimization, and Machine Learning, Addison-Wesley
Professional, 0201157675, Reading, Massachusetts.
Michalewicz, Z. (1999).
[7] Genetic Algorithms + Data Structures = Evolution
Programs, Springer-Verlag, 3540606769, Berlin. Sakoe, H. and
Chiba, S. (1978).
[8] Dynamic Programming Optimization for Spoken Word
Recognition, IEEE Transactions on Signal Processing, Vol. 26,
pp 43- 49. Wang, S. C. (2004).
[9] Voice Signal Processing, Chun Hwa Publication,
9572143840, Taiwan. Yeh, Y. C. (1993).
[10] Implementation and Application of Artificial Neural
Network, Ru Lin Publication, 957499628X, Taiwan.

India.
ISBN: 978-81-920575-9-0 :: doi: 10. 73053/ISBN_0768

A Hybrid Approach for Color Image Retrieval based on Color Autocorrelogram
and Gray-Level Spatial Dependency Matrix in wavelet domain

M.Kamarasan
Department of Computer Science and Engineering
Annamalai University,Annamalai Nagar-608002,
Tamilnadu, India

Dr. K.Seetharaman
Department of Computer Science and Engineering
Annamalai University,Annamalai Nagar-608002,
Tamilnadu, India

Abstract The advent of large scale of digital image
database has led to great challenges in content-based image
retrieval (CBIR). The CBIR is considered an active area of
research; however it comprises a strong backdrop for new
methodologies and systems implementations. Hence, many
research contributions are focusing on techniques enabling
higher image retrieval accuracy while preserving low level of
computational complexity. In this paper, we propose a novel
CBIR method, which is based on an efficient combination of
color and texture features in multiresolution domain. As its
color features, color autocorrelogram of the hue(H) and
saturation(S) components of HSV color space are used. As
its texture features, value (V) component of HSV color space
is used. These two image features are extracts from Gray
Level Co-occurrence Matrix (GLCM) of optimum level of
image, which is constructed based on wavelet coefficients. A
feature space is formed based on the features extracted. We
employed Minkowski-form distance method to computes
similarity between query and target feature vectors.
Experimental results show that the proposed method obtains
higher retrieval accuracy at optimum level resolution than
other conventional methods which are extracts features at all
resolution levels. In addition, the proposed method almost
always shows performance gain in precision and recall
methods.
Keywordsautocorrelogram, gray-level co-occurrence
matrix, Content-based image retrieval, multiresolution,
wavelet transform.

I. INTRODUCTION
Content-Based Image Retrieval (CBIR)[1] has
become a prominent research area and started from
publish in the early 1990s due to the increases of image,
video and sound data in digital form. In addition, the
increased bandwidth availability to access the internet
in the near future will allow the users to search for
and browse through image, video and sound databases
located at remote sites. Therefore fast retrieval of
image from huge databases is an important issue that
needs to be addressed. Image retrieval systems attempt to
search through a database to find images that are
perceptually close to the query image. The CBIR is a
key alternative and complement to traditional text-
based image retrieval and significantly enhance the
accuracy of retrieval. Most proposed CBIR [2,3,4]
techniques automatically extracts low-level features
such as color, texture, shape, motion and combination
of them and form the feature vector. Then distance
methods uses these features to find the similarity between
query image and target image.
Color is one of the most widely used low-level
visual features and is invariant to image size and
orientation [1]. As conventional color features used in
CBIR, there are color histogram, color correlogram,
and dominant color descriptor. Color histogram, which
represents the color distribution in an image, is one of the
most widely used color features and invariant to image
rotation, translation and viewing axis. But, a problem with
the color histogram is that it ignores the spatial
organization of colors. Color correlogram describes the
probability of finding color pairs at a fixed pixel
distance and provides an efficient spatial information.
Therefore color correlogram yields better retrieval
accuracy compare to that of color histogram. Huang et al
[5] have demonstrated that the autocorrelogram, the
statistical probability which captures the spatial
correlation between identical colors only. Since, color
autocorrelogram improves significant computational
benefits over color histogram and color correlogram [1].
Color space plays important roles in image retrieval and
choice of color space representation affects the efficiency
of image retrieval. Traditionally, RGB color space is
widely used in image retrieval system, but it is not
accurate in human visual perception and statistical
analysis [1]. In the recent literature of image retrieval
system, the HSV color space is adopted and yields good
retrieval accuracy [1][12]. The correlation of colors
used in [4], is adopted HSV color space, which
produces significant outperform over its original method
that uses the colors autocorrelation statistics of the
RGB color space[5]. The quantization process of HSV
color space were examined in [1], improving the
correlograms robustness with changes in color content
and varying illumination conditions. Therefore propoerly
quantized HSV color autocorrelogram produces
significant improvement over RGB color space
autocorrelogram and HSV color space histogram.
Texture is another important feature of a visible
surface and is more relevant for CBIR since texture
characteristics are powerful in discriminating between
images. It refers to the visual patterns that have property
of homogeneity or arrangement. In the early 1990s,
Haralick[6] explained a method for computing texture
features based on co-occurrence matrix of gray-scale
image, which extracts meaning statistics. Tamura et al[7]
took a approach of deriving texture features that
correspond to human visual perception. They defined six
textural features such coarseness, contrast,
directionality, line-likeness, regularity and roughness and
compared them with psychological measurements for
human subjects. The first three of them attained very
useful information about an image. Jue Jiang et al[8]
presented a multi-resolution approach for texture
classification based on the moment features that were
extracted in x and y directions, of a histogram in each
image resolution. Chuen-Horng Lin et al [9] proposed
three image features for image retrieval. The first and
second image features are based on color and texture
features, called color co-occurrence matrix (CCM) and
difference between pixels of scan pattern (DBPSP). The
third image feature is based on color distribution, called
color histogram for K-mean (CHKM). In addition, a
feature selection technique is also brought forward to
select optimal features that not only maximize the
detection rate but also reduces the computation
complexity. Aliaa et al[10] presents a new approach to
image retrieval based on color, texture, and shape by
using pyramid structure wavelet. The major advantage of
such an approach is that little human intervention is
required. Young, and Nam [11] proposed a CBIR
methods, which computes color autocorrelograms for
hue(H) and saturation(S) components and BDIP(Block
Difference of Inverse Probabilities)-BVLC(Block
Variation of Local Correlation coefficients) moments as
texture feature for value(V) component of HSV color
image in the wavelet domain. Pannirselvam et al[17]
proposed a new texture features for image retrieval
scheme using Full Range Auto Regression model.
Autocorrelation coefficients are computed with positional
difference to capture the texture present in the small
image region. In addition, autocorrelation functional and
directionality features are also extracted.
Even though considerable amount of works has done
in the combination of color and texture features of image
retrieval but still some issues needs to be addressed.
One of the issues is to extract efficient features that
improve the retrieval performance without increasing the
feature vector dimension. This is motivates us to develop
a new methodology, which uses color autocorrelogram
and Gray-level spatial dependence matrix for extracting
optimum feature in multi-resolution domain.
The next section describes proposed features which
are incorporated in the proposed method. Section III
explains the proposed retrieval method, Section IV
discusses experimental results, and Section V finally
provides the discussion with conclusion.

II. CONVENTIONAL FEATURES
In this section, we explain the conventional features
which is adopted in the proposed retrieval method; for
color features, color autocorrelogram of hue(H) and
saturation(S) components were used; for texture
features, Gray-level spatial dependencies of value(V)
component is used in HSV color space. The color
autocorrelogram and gray-level spatial dependence matrix
are useful for describing the relationship between colors
and textures in an image.
A. Color autocorrelogram
Color Correlogram describes the probability of finding
a pixel of another special color at a fixed pixel distance
for a given pixel of a color. The color autocorrelogram
captures the spatial correlation between identical colors
whereas color correlogram captures all possible
combination of color pairs.
Let be an image. The colors in I are quantized
into m colors
, ,
. Randomly pick a pixel
and pixel
form the image I. Let

random variables
represent the color values of
respectively in the following equation.
| = =

(1)
The numerator is the number of pairs which is k
distance away in the image and the colors are
, the
denominator is the total number of pairs of pixels which is
k distance away. The factor 8k is due to the eight
neighbors which is k distance (in
-norm) away from the

center pixel.
If we fix the color of pixel
to be
, definition of
conditional probability [14]
, |
| = |
8
=

=
,
|
| =

(2)
Therefore, the autocorrelogram describes the
conditional probability distribution of pairs of colors
changing with distance k. The HSV color space used in
the autocorrelogram for feature extraction provides better
human perception of color similarities compared to that of
RGB color space[12].
B. Gray-level spatial dependence matrix
Texture-based image retrieval relies strongly on
effective feature extraction. Grey-level co-occurrence
matrix (GLCM) is one of the most widely used statistical
texture measures. Haralick [6] et al proposed a statistical
method of examining texture that considers the spatial
relationship of pixels is the gray-level co-occurrence
matrix (GLCM), also known as the gray-level spatial
dependence matrix. The idea of the method is to consider
the relative frequencies for which two neighboring pixels
are separated by a distance on the image. Since the
GLCM collects information about pixel pairs instead of
single pixels and which is called by a name as second-
order statistics [13]. First order texture measures are
statistics calculated from the original image values, like
variance, and not consider pixel neighbor relationships.
Second order measures consider the relationship between
groups of two (usually neighboring) pixels in the original
image. Suppose an image I to be analyzed is rectangular
and has
resolution cells in the horizontal direction and
resolution cells in the vertical direction and the gray

tone appearing in each resolution cell is quantized to
levels. Let
= 1,2, ,
be the horizontal spatial

domain,
= 1,2, ,
be the vertical spatial domain

and = 1,2, ,
be the set of
quantized gray
tones. The set
is the set of resolution cells of the

image I ordered by their row-column designations can be
defined as

Image I
G
r
a
y

t
o
n
e

(GLCM) Gray tone
0 1 2 3
#
(0,0)
#
(0,1)
#
(0,2)
#
(0,3)
#
(1,0)
#
(1,1)
#
(1,2)
#
(1,3)
#
(2,0)
#
(2,1)
#
(2,2)
#
(2,3)
#
(3,0)
#
(3,1)
#
(3,2)
#
(3,3)
Fig.1. sample quantization image (a) 4x4 image with four gray-tone
values 0-3 (b) General form of any gray-tone spatial-dependence matrix
for image with gray-tone values0-3. #(m, n) stands for number of times
gray tones m and n have been neighbors.
In the above Fig.1 show, Gray-level co-occurrence
matrix (GLCM) is calculating how often a pixel with the
intensity (gray-level) value m occurs in a specific spatial
relationship to a pixel with the value n. By default, the
spatial relationship is defined as the pixel of interest and
the pixel to its immediate right (horizontally adjacent),
but specify other spatial relationships between the two
pixels. Each element (m, n) in the resultant GLCM is
simply the sum of the number of times that the pixel with
value m occurred in the specified spatial relationship to a
pixel with value n in the input image. The number of gray
levels in the image determines the size of the GLCM.

III. PROPOSED IMAGE RETRIEVAL METHODS
A. Overall process of proposed system
Spatial-frequency based image transform, also known
as multi-scale representations, decompose the query
image into a set of sub-image, exhibiting image
approximations and details at multiple scales and multiple
orientations. Each orientation corresponds to frequency
sub-bands. The decomposition process is continued until
to derive the optimum level and then the proposed system
considered the features at optimum level frequency sub-
bands. Fig.2 shows the block diagram of the proposed
retrieval method, when a query image submitted into the
retrieval system, is transformed to HSV color image,
whose components are H, S and V. Then each component
image
, = , , are decomposed into a wavelet

coefficients
,
, = , , is shown in Fig.3, where

k= , , , denote sub band orientation and
l= 1, , denotes wavelet decomposition level. In
HSV color space, the H and S components closely related
to chrominance information and V component represent
the wider bandwidth, which contains most texture
information. Hence, the color features extracts from
,


Fig.4. Show
the 2- level
decomposition of HSV image
Fig.2 Block diagram of proposed system

Fig.3. show the 3-level decomposition image.

and
,
and texture feature extract from

,
at
optimum level decomposition. The optimum level means
that not losing the dominant features/characteristics. The
subsequent section describes the color and texture feature
extraction, which is adopted by the proposed method.
B. Color autocorrelogram
An color autocorrelogram[18] of image I provides the
probability of finding identical colors at distance d and it
gives significant computation benefits over the color
correlogram[19], which is described as
= k

0,1, , 1 (3)
Where Pr[.] denotes the probability satisfying the
condition, the set of all pixels in an image, and the
set of pixels of color which is requantized with
levels,
represents the probability that a pixel of

distance k from the given pixel color . Fig.5 shows the
procedure for extracting color features from decomposed
frequency sub-bands such,
,
,
,
.
The
,
,
,
denotes color quantization of

frequency sub-bands. Each frequency bands (LL
components) are uniformly quantized by using Lloyd
algorithm [15].

,
,
,
,
,
Query
image
HSV
Color
Wavelet
Decompo
Feature
extraction
Similarity
measure
Feature
extraction
Wavelet
Decompos
HSV
Color
Retrieved
Image
Imag
e DB
Featu
re DB
Color & Texture
features

Fig.5. Process of color feature extraction.
After constructing the numbers of quantization levels
for given image I, then color autocorrelogram in (3) is
employed to extract the color features at optimum
level
,
,
,
. Then color feature vector
is
formed as
=
,
, , , , 0,1, ,
,
1,
, , , , ,n 1, , .
(4)
C. Gray-level spatial dependence matrix
A statistical method of examining texture that
considers the spatial relationship of pixels is the gray-
level co-occurrence matrix (GLCM), also called the gray-
level spatial dependence matrix is discussed in section
2.2. We construct the spatial dependence matrix or
GLCM from optimum level decomposed image
,
for
considering texture features extraction. The process of the
texture features extraction is given in Fig. 6.

Fig. 6. process of texture features extraction
The following Statistical texture features are
computed on GLCM, they are given in equations as
= | |
,
, (5)

,
(6)
= ,
,
(7)
=

||
,
(8)

where
represents mean value;

denotes the standard deviation. Features
measure the
intensity contrast between a pixel and its neighbor over
the
,
measure correlated a pixel is to its neighbor

over
,
represent the sum of squared elements in

the
,
and also known as uniformity, non-uniformity

of energy and angular to moment and
measure the
closeness of distribution of elements in the
,
to the
diagonal. We combine all these features and

construct texture feature vector is given in the equation

(9)
The proposed system combine both visual feature
vector
in eq. (4) and texture feature vector
in eq. (9)
for retrieve the similar image from image database is
given in eq.(10)
=
(10)
The eq. (10) represents the combined features for both
visual and texture representation which is known as
feature vector space.
The goal of feature selection is to select the best
features from original ones. It can not only achieve
maximum recognition rate but can also simplify the
calculation of image retrieval.
D. Similarity measure
Texture-based image retrieval relies strongly on
efficient feature extraction and similarity measurement
steps. The proposed system uses the generalized
Minkowski-form distance [16] of metric order one for
compute the similarity between query and target features,
is given as
= |
(11)
where q and t represent the query and target image;
and
the i
th
components of
and
,
respectively and computed distance values are shown in
Table I.
Table I shows that the sample distance values
obtained from each decomposition levels (L
0
to L
5
)
between query and target image using the equation in
(11). It is seen that the distance values are significantly
almost same for all decomposition levels (L
0
to L
5
).
Therefore the proposed CBIR consider to extracts the
features at optimum level is enough for retrieving image
from image database. Since optimum level has few
dominant wavelet coefficients and not loses
features/characteristics. This is the main advantage of the
proposed system.

Decomposed
Images
,

Color
Quantization
of
,
and

Color
autocorrel
ogram
extraction
Color
feature
vector

Optimum
level
Decomposed
image
,

Texture
Feature
Extraction
Texture
feature
vector

Spatial
dependence
matrix
using
GLCM

Table I
THE DISTANCE VALUES BETWEEN QUERY IMAGE AND FEATURE DATABASE IMAGE USING MINKOWSKI-FORM DISTANCE
FOR PYRAMID IMAGE UP TO LEVEL 5.
Query vs.
Target images
L0
(1024x1024)
L1
(256x256)
L2
(128x128)
L3
(64x64)
L4
(32x32)
L5
(16x16)
Plant Vs. Plant2 1.7453 1.6402 1.6394 1.6155 1.5327 1.5053
GrassLand Vs.
GrassLand1
1.4242 1.4933 1.4125 1.4168 1.3956 1.3860
Paintings Vs.
Paintings1
0.3376 0.3524 0.3689 0.3601 0.3090 0.3414
Fabric Vs.
Fabric1
0.7106 0.7687 0.7844 0.7789 0.7755 0.7073
Food Vs. Food1 1.6608 1.6812 1.6005 1.7354 1.7559 1.6562
Leaves Vs.
Leaves1
1.4465 1.5023 1.4602 1.4030 1.5298 1.4253
Paintings Vs.
Paintings1
0.4376 0.4524 0.4689 0.4601 0.4090 0.4414
Sand Vs. Sand1 1.6576 1.6647 1.5996 1.6693 1.6035 1.6554
Terrain Vs.
Terrain1
0.8949 0.8024 0.7931 0.7208 0.7388 0.7525
Water Vs.
Water1
1.3789 1.2836 1.3038 1.4268 1.3560 1.3241
Bark Vs Bark 0 0 0 0 0 0

Fig.7. Perfomance curve of proposed progressive method using Vistex database.

In Fig.7, curves shows that distance values are
significantly almost same for all decomposition levels(L0

to

L5) between query and target image. Therefore the
proposed system extract features at optimum level only.
The entire process of the proposed system is
presented in the algorithmic form as follows:
Step 0 :Start
Step 1: Read image I and convert into HSV color
space.
Step 2: Decompose an HSV color image using
Daubechie-4 and construct optimum level
pyramid.
Step 3: Computes color autocorrelogram from W
,

and W
,
and a set of texture feature from

W
,
at optimum level.
Step 4: Find the similarity between computed feature
vectors of query image and feature DBs using
Minkowski-form distance.
Step 5: The recall-precision method is adopted to
compare the performance accuracy of
proposed system with existing system.
Step 6: stop.
Algorithm-2. The overall process of the proposed system.

VI. EXPERIMENTAL RESULTS
The proposed retrieval method is experimented with
texture and natural images from different databases such
Vistex and Brodatz album. The image considered in this
experiment with various pixel size 512 x 512, 256 x 256,
and 128x128. This paper adopts image features including
texture of color, spatial texture of color and color feature.
We consider 64 gray-tone bins for constructing GLCM on
HSV color space. First, the HSV color image is
decomposed up to the result oriented optimum level using
Daubechie-4 wavelet transform and feature vector is
constructed using eq. (10). Then, query feature vector is
compared with features vector database using generalized
Minkowski-form distance is discussed in section 3.3. The
distances values are arranged in ascending order and then
top 10 relevant images from image database are retrieved
and presented in Fig.8. From Fig.8, first column consider
query image and adjacent columns are relevant images
according to rank order and actual distance value of
relevant images is presented in the bottom of each case.
The proposed system is implemented on an Intel Core 2
Duo with a 2.66 GHz processor and 2 GB of memory,
Matlab 7 with Microsoft Windows XP Professional.
Performance of proposed CBIR
In the proposed CBIR, the performance is measured
in terms of Precision (P) and Recall (R), which are
defined
P(q)=

R(q)=

(12)
P(q) represents the ratio of the number of images relevant
to the query q among retrieved images to the number of
retrieved images; recall R(q) represents the ratio of the
number of images relevant to the query q among retrieved
images to the number of images relevant to the query in a
image database. Hence the precision versus recall jointly
represents that the percent of the images relevant to the
query are retrieved and that the percent of the retrieved
images are relevant.
The proposed system measure the retrieval
performance in terms of precision and recall by
considering each image in the database as a query image
and above said method is repeated for computing them. It
is observed from Table II, our system obtained higher
retrieval performance over the other conventional
methods.
V. CONCLUSION
The proposed CBIR method uses the two image
features, namely color autocorrelogram and gray-level
spatial dependency matrix to describe the relationship
between color and texture of an image. The color
autocorrelogram features extracted from H and S
components and Spatial dependency matrix constructed
from V component of HSV color space in the
multiresolution domain. Through the optimum feature
selection, the number of features required has been
significantly reduced and retrieval speed increased.
Experimental results of Vistex database showed that the
proposed method obtained higher retrieval accuracy
compared to that of other conventional methods. The
proposed system shows that it is conceptually simple,
memory efficient and significantly reduces computational
complexity. In addition, the proposed method almost
always shows performance gain in precision versus recall
method for Vistex database. As further studies, the
proposed retrieval method is to be evaluated for various
multimedia databases and to be applied to video
classification, indexing and retrieval.

REFERENCES
[1] Ritendra Datta, Dhiraj Joshi, Jia Li, James Z. Wang, Image
retrieval: ideas, influences, and trends of the new age, ACM
Computing Surveys 40 (2) (2008) 160.
[2] W.Niblack et al., The QBIC Project: Querying Images by
Content using Color, Texture, and Shape, in Proc. SPIE,
vol. 1908, San Jose, CA, pp. 173187, Feb. 1993.
[3] A.Pentland, R. Picard, and S. Sclaroff, Photobook:
Content-based Manipulation of Image Databases, in Proc.
SPIE Storage and Retrieval for Image and Video
Databases II, San Jose, CA, pp. 3447, Feb. 1994.
[4] J. Huang, S. Kumar, M. Mitra & W. Zhu: Image indexing using
color correlograms. In: Proc. IEEE Computer Society
conference on Computer Vision and Pattern Recognition (San
Juan, Puerto Rico),1997, pp. 762-768.
[5] J.Huang, S. Kumar, M. Mitra & W. Zhu: Spatial color indexing
and applications. In: Proc. Sixth International conference on
Computer Vision (Bombay, India), 1998, pp. 602-607.
[6] R.M.Haralick, K. Shanmugam, and I. Dinstein, Texture
features for image classication, IEEE Trans. Syst. Man
Cybern, vol. SMC-8, pp.610621, Nov. 1973.
[7] Tamura, H., Mori, S., Yamawaki, T.: Textural features
corresponding to visual perception. IEEE Trans on Systems,
Man and Cybernetics 8 (1978) 460472.
[8] Jue Jiang, Runsheng Wang and Peng Zhang, Texture description
based on multiresolution moments of image histogram, optical
Eng., 47(3):037005. DOI:10.1117/1.2894149, 2008.
[9] Chuen-Horng Lin , Rong-Tai Chen and Yung-Kuan Chan, A
smart content-based image retrieval system based on color and
texture feature, Image and Vision Computing 27 (2009) 658665.
[10] Aliaa.A.A.Youssif, A.A.Darwish and R.A.Mohamed, Content
based medical image retrieval based on pyramid structure wavelet,
IJCSNS International Journal of Computer Science and Network
Security, VOL.10 No.3, March 2010.
[11] Young Deok Chun, Nam Chul Kim, Content-Based Image
Retrieval using Multiresolution color and Texture Features, IEEE
Trans. on Multimedia, Vol.10.No.6, pp. 1073-1084 October 2008.
[12] T.Ojala, M. Rautiainen, E. Matinmikko, and M. Aittola, Semantic
image retrieval with HSV correlogram, Proc. 12
th
Scandinavian
Conf. On Image Analysis, Bergen, Norway, Jun. 2001, pp. 621-
627.
[13] A. Akono, E. Tony
,
A. N. Nyoungui ,J. -P. Rudant, Nouvelle
m@thodologie d' valuation des param tres de texture d'ordre
trios, International Journal of Remote Sensing ,Volume 24,Issue
9,2005,pages 1957-1967.
[14] W. Feller. An introduction to probability theory and its
applications. New York,Wiley, 1966.
[15] A. Gersho and R. M. Gray, Vector Quantization and Signal
Compression. Norwell, MA: Kluwer, 1992.
[16] D. Androutsos, K. Plataniotiss, and A. Venetsanopoulos,
Distancemeasures for color image retrieval, in Proc. IEEE Int.
Conf. ImageProcessing, Chicago, IL, Oct. 1998, vol. 2, pp. 770
774.
[17] S.Pannirselvam, R.Krishnamoorthi,A New Texture Image
Retrieval Scheme with Full range Auto Regressive(FRAR)
Model, International Journal of Soft Computing 4(5):229-
235,2009.
[18] Wichian Premchaiswadi, Anucha Tungkatsathan, On-line
Content-Based Image Retrieval System using Joint Querying and
Relevance Feedback Scheme, WEAS Transaction on Compters,
Issue 5, Vol. 9, May 2010.
[19] T. Ojala, M. Rautiainen, E. Matinmikko & M. Aittola: Semantic
image retrieval with HSV correlograms, In: Proc. 12
th

Scandinavian conference on image analysis (Bergen Norway)
2001, pp. 621-627.


Query image 0.0000 0.7915 0.958 0.9822 1.0073 1.1452 1.2884 1.4075 1.4135 1.4373

Query image
0.0000 0.7487 0.8704 0.9886 0.9886 0.9956 1.0582 1.2376 1.3278 1.4562

Query image
0.0000 0.7003 0.9002 0.9253 1.0506 1.0775 1.192 1.2811 1.3612 1.5759

Query image
0.0000 0.8001 0.8414 0.8871 1.0649 1.1378 1.2045 1.2123 1.3854 1.4199

Query image
0.0000 0.3710 0.7958 0.9355 1.7092 1.7436 1.7556 1.7866 1.7989 1.8842

Query image
0.0000 0.4525 0.8946 0.8998 1.072 1.1789 1.2964 1.3255 1.4462 1.5487

Query image
0.0000 1.2916 1.3897 1.3915 1.4935 1.5538 1.6906 1.7085 1.7452 1.9318

Query image
0.0000 1.1500 1.1895 1.2233 1.2305 1.3145 1.3281 1.386 1.4793 1.9572

Query image
0.0000 0.7266 0.7601 0.8100 0.8277 0.8434 0.9024 0.9039 1.2587 1.3215

Query image
0.0000 0.8028 1.0149 1.0733 1.1599 1.1772 1.1867 1.5002 1.5778 2.2729
Fig. 8. First column represent query image and neighboring columns are relevant images in rank orders.
TABLE II
COMPARISON OF PROPOSED METHOD WITH EXISTING METHODS
Proposed method FRAR method GLCM method Laws method
Recall Precision Recall Precision Recall Precision Recall Precision
0.94 0.79 0.92 0.78 0.72 0.54 0.81 0.68

1

India.
ISBN: 978-81-920575-9-0 :: doi: 10. 73039/ISBN_0768
Analysis of Computational Errors in Digital Signal Processing with Voltage Scaling
Mr. Vishwas V. Balpande
Electronics
DBACER
Nagpur,India

Abstract It has been studied that voltage overscaling
(VOS) is used to reduce energy consumption while
maintaining satisfactory signal processing performance. Due
to the computation-intensive nature of most signal processing
algorithms, the energy saving potential largely depends on the
behavior of computer arithmetic units in response to
overscaled supply voltage. In this paper we apply the voltage
scaling on CORDIC (COordinate Rotation DIgital Compter)
processor which is the Digital Signal Processing System. In
Digital Signal Processing, trigonometry and complex
multiplications are used in many signal equations, such as
synchronization and equalization. Therefore, a fast and an
efficient method to calculate trigonometry and complex
multiplications are required. Coordinate Rotation Digital
Computer (CORDIC) is trigonometric algorithm [1] that is
used to transforming data from rectangular to polar and vice
versa. CORDIC also can be used other to compute several
trigonometry functions, either directly or indirectly. The
design process is started by modelling CORDIC function,
design datapath and control unit, coding to hardware
description language using Verilog HDL, synthesized using
Leonordo Spectrum and implemented using Mentor
Graphics tool.
Keywords- Voltage Over Scaling(VOS), Coordinate Rotation
Digital Computer (CORDIC).
I. INTRODUCTION
Voltage Scaling is an effective technique to reduce the
energy consumption in CMOS integrated circuits [1],[2].
In conventional practice, voltage scaling is lower bounded
by , under which the critical path delay equals the target
clock period. Voltage overscaling (VOS), i.e., overscaling
the supply voltage below , can result in transient circuit
timing errors, which is generally not allowed in current
design practice. It is well known that there is a large degree
of discrepancy between the average-case and the worst-
case circuit delay in practice, particularly in many
computer arithmetic functions. This suggests that, even
under an overscaled supply voltage, computer arithmetic
functions may have a relatively low probability to produce
errors in each clock cycle. Intuitively, this feature may be
exploited to enable the use of VOS to reduce the energy
consumption of computational datapath. In this context, the
key issue is how to maintain the satisfactory functionality
in the presence of the transient errors incurred by VOS
while ensuring that the overall system energy consumption
is reduced.
In most signal processing systems, various computer
arithmetic functions, particularly addition and
multiplication, are major building blocks and typically
constitute the critical paths. Therefore, the signal
processing performance degradation incurred by VOS
heavily depends on the output transient error
characteristics of those computer arithmetic functions in
response to overscaled supply voltage. All the prior works
[3] on voltage-overscaled signal processing system design
assumed the use of carry-ripple adder architecture in the
realizations of adders and multipliers. Intuitively, different
computer arithmetic architectures (e.g., carry-ripple adder,
carry-select adder, and carry-lookahead adder) may
respond to overscaled supply voltage differently, leading to
different output transient error characteristics, even though
they have the same critical path delay. This will further
result in different signal processing performance
degradation and, hence, different energy saving potentials
in voltage-overscaled signal processing systems.
Therefore, computation error characteristic analysis of
computer arithmetic units in voltage-overscaled signal
processing systems is an important issue, which
nevertheless has not been addressed in the open literature
to the best of our knowledge.
In Digital Signal Processing, trigonometry and complex
multiplications are used in many signal equations, such as
synchronization and equalization. The big problem in
trigonometry is how to transform data in rectangular
domain to polar domain and vice versa. The big problem in
complex multiplications is the need of multiplier which too
big. Coordinate Rotation Digital Computer (CORDIC) is
one of a method to solve this problem. The CORDIC is a
simple algorithm to transform data in rectangular domain
to polar domain and vice versa. The CORDIC algorithm
provides an iterative method of performing vector rotations
by arbitrary angles using only shifts and adds.

II. CORDIC THEORY
CORDIC (COordinate Rotation DIgital Computer) is
an simply algorithm to compute a wide range of functions
including certain trigonometric, hyperbolic, linear, and
logarithmic functions [4]. The CORDIC algorithm
provides an iterative method of performing vector rotations
by arbitrary angles using only shifts and adds, and all this
2

component is available in hardware. The algorithm, credit
to Volder, is derived from the general (Givens) rotation
transform:

This rotates a vector in a Cartesian plane by the
angle . There can be rearranged so that:

= cos

If the rotation angles are restricted so that tan () = 2-
i, the multiplication by tangent terms is reduced to simple
shift operation. Arbitrary angles of rotation are obtainable
by performing a series of successively smaller elementary
rotations. If the decision at each iteration, i, is which
direction to rotate rather than whether or not to rotate, than
the cosh (i) term becomes a constant. The iterative
rotation can now be express as:

. 2
. 2

Where:
= costan
= 1

1 + 2
= 1
The angle of a composite rotation is uniquely defined
by the sequence of the directions of the elementary
rotations. To calculate the angles, an additional adder-
subtractor that accumulates the elementary rotation angles
at each iteration.

. tan

The CORDIC rotator is normally operated in one of
two modes. The first, called rotation by Volder, rotates the
input vector by a specified angle (given as argument). The
CORDIC equations are:

. 2
. 2

..(1)

. tan

where

<
+

which provide the following result:

= +

The second mode, calling vectoring, rotates the
input vector to the x axis while recording the angle
required to make that rotation. The CORDIC equations are:

. 2
. 2
. tan

where

=
+
<

which provide the following results:

= +

The CORDIC rotation and vectoring algorithms as
stated are limited to rotation angles between /2 and /2.
This limitation is due to use of 20 for the tangent in the
first iteration. This gives the correction iteration:

a. Rotation mode

= .

= .

= .
2

where
d
=
+1 > 2
1 < 2

b. Vectoring mode

= .

where

=
+1 > 0
1 < 0

III. CORDIC ALGORITHM
CORDIC is a popular algorithm for trigonometric
calculations. The processor has three registers, namely,
X(X coordinate), Y(Y coordinate) and Z(Angle). It has two
modes 1. Vector mode
2. Rotation mode
The rotation mode is used to find the sine or cosine of
an angle iteratively, using only simple math operations
such as add, subtract, compare, shift, and table look-up.
The angle whose sine or cosine is to be found is first stored
in the Z register. X and Y are made 1 and 0 respectively,
thus representing a horizontal vector. After sufficient
number of iterations, Z nearly becomes 0 and at this time,
X represents cosine and Y represents sine of the original
angle.
The vector mode is used to find arc tangent of a
number. In this mode, the Z register is initialized to 0 and
X, Y registers are initialized such that Y/X=tan(), where
is the desired angle. The vector is likewise rotated so that it
3

eventually becomes horizontal. At each step, the rotation
angle is accumulated in the Z register. After sufficient
number of steps, Y becomes almosy 0 and Z contains .

IV. CORDIC ARCHITECTURE
The CORDIC algorithm has traditionally been
implemented using bit serial architecture with all iterations
executed in the same hardware [5]. This slows down the
computational device and hence, is not suitable for high
speed implementation. The word serial architecture is an
iterative CORDIC architecture obtained by realizing the
iteration equations (1). In this architecture, the shifters are
modified in each iteration to cause the desired shift for the
iteration. The appropriate elementary angles, are
accessed from a lookup table. The most dominating speed
factors during the iterations of word serial architecture are
carry/borrow propagate addition/subtraction and variable
shifting operations, rendering the conventional CORDIC
implementation slow for high speed applications. These
drawbacks were overcome by unfolding the iteration
process, so that each of the processing elements always
perform the same iteration as shown in Figure 1. The main
advantage of the unfolded pipelined architecture compared
to folded architecture is high throughput due to the
hardwired shifts rather than time and area consuming
barrel shifters and elimination of ROM. It may be noted
that the pipelined architecture offers throughput
improvement by a factor of n for n-bit precision at the
expense of increasing the hardware by a factor less than n.

Figure 1. Unfolded Pipelined Architecture of CORDIC
V. CONCLUSION
In this paper we studied about the voltage scaling
and how it is helpful in lowering the power consumption.
Also we saw the CORDIC algorithm, which is useful in
solving trigonometric calculation in Digital Signal
Processing System. CORDIC algorithm also used to
convert the polar coordinate to rectangular and vice-versa.
Also we used the voltage overscaling funda in designing
the CORDIC processor. By using voltage scaling in
designing CORDIC processor the power consumption is
reduced without affecting the processor operation on large
scale.
There are some error because of transient time,
this error analysis is mine next task.
REFERENCES
[1]. A. P. Chandrakasan and R. W. Brodersen, Minimizing power
consumption in digital CMOS circuits, Proc. IEEE, vol. 83,
no. 4, pp. 498523, Apr. 1995.
[2]. R. Gonzalez, B. M. Gordon, and M. A. Horowitz, Supply and
threshold voltage scaling for low power CMOS, IEEE J.
Solid-State Circuits, vol. 32, no. 8, pp. 12101216, Aug. 1997.
[3]. R. Hegde and N. R. Shanbhag, A voltage overscaled low-
power digital filter IC, IEEE J. Solid-State Circuits, vol. 39,
no. , pp. 388391, Feb. 2004.
[4]. Ray Andraka, A Survey of CORDIC algorithms for FPGA
based computers, January 1998, ACM 089791-978-5/98/01.
[5]. J. E. Volder, The CORDIC trigonometric computing
technique, IRE Transactions on Electronic Computers, vol. 8,
no. 3, pp. 330334, 1959.
[6]. Yang Liu, Tong Zhang and Keshab K. Parhi Computation
Error Analysis in Digital Signal Processing Systems With
Overscaled Supply Voltage in IEEE transactions on very
large scale integration (VLSI) systems, vol. 18, no. 4, April
2010

India.
ISBN: 978-81-920575-9-0 :: doi: 10. 73025/ISBN_0768
PAPR Reduction Technique in MIMO-OFDM Based Wireless Communication

Debasis S. N Das
1
, Arun Kumar Ray
2
, Mihir N. Mohanty
3

KIIT University, Bhubaneswar, KIIT University, Bhubaneswar, ITER, Siksha O Anusandhan
Odisha, India , Odisha, India University, Bhubaneswar, Odisha,

Abstract - Orthogonal Frequency Division Multiplexing
(OFDM) system gives more attentions due to its high data
rate and its capability to combat multi-path interference. In
this paper, initially, the concept for wireless communication
using MIMO-OFDM is considered & tested stepwise. Then
channel estimation based on MIMO-OFDM has been
performed. Besides this the noise analysis has also evaluated
& comparison result of LS, MMSE shows the performance
in fading environment. The paper also describes
multicarrier(MC) BPSK-OFDM as real valued modulation
as well as MC QPSK-OFDM as complex modulation.
Another technique for evaluation of peak-to-average-power
ratio (PAPR) using complex modulation is presented. The
proposed scheme is over sampled PAPR that shows the
efficacy of the result.
Keywords Orthogonal Frequency Division Multiplexing
(OFDM), MIMO, peak-to-average-power ratio (PAPR),
multicarrier (MC)

I. INTRODUCTION
Wireless communication is an interesting area of
research for the modern society in many aspects like,
efficient technology in terms of error rate and occupation
bandwidth. A MIMO communication system, combined
with OFDM (MIMO-OFDM) modulation technique can
achieve a reliable high data rate transmission over
wireless channels. It achieves this by higher spectral
efficiency and link reliability or diversity (reduced
fading). The requirement to provide reliable high data
rate communication over the wireless channel has led to
the development of efficient modulation and coding
schemes. A wireless channel suffers from time-varying
impairments like multipath fading, interference and
noise. Diversity is an effective method to combat the
fading of the wireless channel [1]. Coherent detection
and accurate channel estimation is required at the
receiver to obtain reasonable performance. The
estimation of channel response can be obtained by LS
estimation scheme but in wireless application, the
MMSE, BER and SER for data transmission is most
important.
One of the major challenges of Orthogonal Frequency
Division Multiplexing (OFDM) is that the output signal
may have a potentially very large peak-to-average power
ratio (PAPR, also known as PAR). The resulting
technical challenges, as well as PAPR-reduction
techniques and related issues, have been widely studied
and reported in the research literature [1], [2].Since the
actual signal that enters the power amplifiers is a
continuous-time signal, we ultimately want to reduce the
PAPR of the continuous-time OFDM signal (we call this
the continuous-time PAPR for convenience).However,
the evaluation of the continuous-time PAPR is
analytically non- trivial and computationally expensive.
Therefore, most PAPR- reduction techniques focus on
discrete-time approximations of the continuous-time
PAPR. The discrete-time approximations result in what
we call the discrete-time PAPR. In this paper, a
computational method is introduced to find the peaks for
OFDM signals with arbitrary complex-valued
modulations.
The paper is summarized as follows. Section II
describes the design of the model. Section III is for the
channel estimation of the model. Section IV describes
PAPR reduction technique for OFDM. Section V follows
the result. Finally section VI concludes the paper with
the scope for future work.

II. DESIGN OF MIMO-OFDM MODEL
Figure 1 depicts a high level block diagram of the
MIMO-OFDM system. We consider MIMOOFDM
systems with two transmit antennas and two receive
antennas. The total number of sub carriers is N.
Basically, the MIMO-OFDM transmitter has Nt, Nr
parallel transmission & reception paths which are very
similar to the single antenna OFDM system, each branch
performing serial-to-parallel conversion, pilot insertion,
N-point IFFT and cyclic extension before the final TX
signals are up-converted to RF and transmitted.


Figure1. MIMO-OFDM BLOCK DIAGRAM

Subsequently at the receiver, the CP is removed
and N-point FFT is performed per receiver branch.

Next, the transmitted symbol per TX antenna is
combined and outputted for the subsequent
operations like digital demodulation and decoding.
Finally all the input binary data are recovered with
certain BER.
Here the channel impulse response h(t) is
treated as follows
( ) ( )
m s
h t t T
m
m
a =

(1)
Where
m
are the complex amplitudes.
The system is then modeled using a N-point
discrete time Fourier Transform (DFT
N
) as
( )
%
N N
y DFT (IDFT x )
h
n
n
= +
(2)
The system is described in above equation can
be written as set of N-independent Gaussian
Channels, y
k
=h
k
x
k
+n
k
, k=0..N-1.
Where h
k
= complex channel attenuation given
by h=(h
0
,h
1..
.h
n-1
)
T
and n=(n
0
,n
1..
.n
n-1
)
T
is an
i.i.d complex Zero mean Gaussian noise vector. so
in matrix notation the ofdm system is described as
y XFg n = +
(3)
where X is the input data matrix and elements
of X are on its diagonal and
00 0( 1)
( 1)0 ( 1)( 1)

N
N N
N N N
N N
W W
F
W W

| |
|
=
|
|
\
K
M O M
L

(4)
Is the DFT matrix with
2 /
1/
nk j nk N
N
W Ne

=

II. CHANNEL ESTIMATION OF MIMO-
OFDM MODEL
A wireless channel model can be considered
both in time domain and frequency where the
features are stationary and ergodic in nature.
Therefore, the estimation of channel is one of the
emerging areas of resources.
A. MMSE AND LS CHANNEL ESTIMATOR
Let the channel vector h is Gaussian and un-
correlated with channel noise n, the MMSE
estimate of h becomes

Figure 2. General Estimator Structure

1
MMSE hy yy
h R R y
=
(5)
Where

{ }
H H H
hy hh
R E hy R F X = =

{ }
H H 2
yy hh
R E yy XFR FHX I
n N
= = +

are the cross covariance matrix between h and y
and the auto covariance of y. Further R
hh
is the
auto covariance matrix of h and

denotes the
noise variance E{ l n
k
l

2
}. These two quantities are
assumed to be known. So the MMSE estimates
(h
MMSE
) is represented by

H H

F X y
MMSE MMSE MMSE
h Fh FQ = =
(6)

Where

( )
1
H H 2 1 H H 1
hh
[ F X XF R ] (F X XF)
MMSE hh n
Q R

= +
(7)

The LS estimator for channel impulse response
h is analyzed as follows:-

H H
LS
FQ F X y
LS
h =
(8)

Where

( )
1
H H
LS
Q F X XF
=
(9)
Considering the two equations we have

1
X y
LS
h

=
(10)

IV. PAPR reduction for OFDM/OFDMA

There are many factors that should be
considered before a specific PAPR reduction
technique is chosen such as reduction of PAPR
capability, increase of power & BER at the
transmitter & receiver, loss in data rate,
computational complexity increase etc. It is
absolutely an important factor while choosing a
PAPR reduction technique as it may result in other
harmful effects. There are some techniques that
require a power increase in the transmit signal after
using PAPR reduction method. This is because
some power gets used for the PRCS. Increase in
BER is closely related to the power increase in the
transmit signal. In some cases the data rate gets
reduced after applying PAPR technique as one
among the information symbols is to be dedicated
for controlling PAPR. Besides all the above,
computational complexity also plays a vital role in
choosing a PAPR reduction method.
The evolution of OFDM to OFDMA
completely preserves the advantages of OFDM.
The drawbacks associated with OFDM, however,
are also inherited by OFDMA. Hence, OFDMA
also suffers from high PAPR. Some existing PAPR
reduction techniques, which were originally
designed for OFDM, process the whole data block
as one unit, thus making downlink demodulation of
OFDMA systems more difficult since only part of
the subcarriers in one OFDMA data block are
demodulated by each users receiver [7]. If
downlink PAPR reduction is achieved by schemes
designed for OFDM, each user has to process the
whole data block and then demodulate the assigned
subcarriers to extract their own information. This
introduces additional processing for each users
receiver. In the following we describe some
modifications of PAPR reduction techniques for an
OFDMA downlink. The PAPR problem for an
OFDMA uplink is not as serious as that for
downlink transmission since each users transmitter
modulates its data to only some of the subcarriers
in each data block.
A. PTS for OFDM/OFDMA:
The PTS technique can easily be modified for
OFDM/OFDMA. Subcarriers from one user are
grouped into one or more subblocks, and then PTS
is applied to subblocks from all users. One
subcarrier per subblock is reserved, and the phase
factor for the subblock is embedded into this
subcarrier. When applying PTS, the reserved
subcarrier does not undergo phase rotation, and this
reserved subcarrier is used as a reference for each
subblock at the receiver. The phase factor for each
subblock is extracted from the reserved subcarrier.
Using these phase factors, each user recovers the
data in the subblocks for that user.
B. SLM for OFDM/OFDMA:
The SLM technique can also be modified for
OFDM/OFDMA. Some of the subcarriers are
dedicated to transmitting side information for SLM.
All users use the information on these dedicated
subcarriers to obtain information on which phase
sequence is used. Using this knowledge, the data
for each user can be restored from the subcarriers
of that user only.
C. TR for OFDM/OFDMA:
In the TR technique for OFDM, the symbols in
PRCs are optimized for the whole data block in
both amplitude and phase. On the other hand, a
number of PRCs are assigned to each user only in
the TR technique for OFDMA. In order to reduce
the computational complexity, the PRCs for each
user are optimized for the subcarriers of that user
only, making the optimization for the whole
OFDMA data block suboptimal.

V. PROPOSED METHOD FOR PAPER
EVALUATION

All trigonometric functions of an angle may
be expressed as rational expressions in terms of
t=tan(/2). Let x =tan(t).Substituting(1 x2)/(1+
x2 ) for cos(2t) and 2x/(1+ x2 ) for sin(2t), and
letting k =cos(/2k) and k =sin(/2k), we have

2 2
1 1
2 2 2 1 1
2 1 2 1 2
( ) 1
1 1 1
N N
a k k k k k k
K K
x x x
p x T T
N x N x x

= =
( (
= + + +
( (
+ + +

We need only to find the roots of Pa (x)/x,
since Pa (t)/t = Pa (x)/x( sec2 (t)). Because
Tk (x) is an order-k polynomial, the highest power
of 1/(1+ x2 ) in(9) is N 1. Hence we can remove
the denominator and thus obtain a polynomial Q(x)
by writing
(12)

2
( )
( ) (1 )
N a
P x
Q x x
x
= +

(13)

Q(x) is a polynomial of degree at most 2N in x
and all roots of Pa (x)/x are also roots of Q(x).
Thus, Pa (x)/x has at most 2N roots. Pa (x) can
be routinely computed from (9) by expanding the
Chebyshev polynomials, factoring out 1/(1+x2)N
,and collecting terms. We may then evaluate the
values of Pa (x) at the real roots, and the maximum
is c.
A. NUME RICAL PROCEDURE:
The proposed method for computing the
continuous-time PAPR for a given symbol set {sn }
and number of sub carriers N is summarized as
follows.1) Compute k and k for k = 1, 2, ,N
1according to (6) and (7);2) Compute Pa (x)
according to (9), expanding and collecting the
coefficients of the different powers of x;3) Find the
derivative of Pa (x);4) Find the roots of Q(x), and
ence of Pa (x)/x using standard polynomial root
finding algorithms;5) Keep only the real roots of
Q(x);6) Evaluate and compare the values of Pa (x)
at the real roots, and obtain c .Each step is
straightforwardly handled by common
mathematical software like Mathematica or Matlab.
In our experiments, we have found that step 2
(expanding and simplifying Pa (x)), while
conceptually easy, may dominate the computation
time, especially for large N. In particular,
expanding and simplifying Tk [k(1
x2)/(1+x2)+k(2x)/(1+x2 )] is a time consuming
operation for large k. For a given N pre-computing
these terms helps to significantly reduce the
computation time.

VI. RESULTS

Simulation results demonstrate that the
performance of LS and MMSE method are quite
good in wireless communication. Even though
MMSE estimator is more complex than LS, it
provides reduced SER and BER. The analysis of
noise performance has also done. Beside this the
result shows evaluation of the proposed scheme
using QPSK-OFDM system for N=512 with
different sampling rates along with performance of
PAPR of different real and complex modulation
schemes used in OFDM systems.

Figure 3. Comparison of LS & MMSE estimator

Figure 4. Noise analysis of MIMO OFDM system


Figure 5. Simulation of PAPR of QPSK-OFDM Figure 6. Comparison of PAPRs (with different
(for different sampling rate) modulation schemes)

VII. CONCLUSION

The MIMO-OFDM based channel model has
been estimated successfully. Using the proposed
scheme, we have shown for MIMO based OFDM
modulations that the discrete-time PAPR obtained
from two-times or more oversampled signals may
be considered a sufficiently accurate approximation
of the continuous-time PAPR.

REFERENCES

[1] Chao-Cheng Tu and Benot Champagne, Subspace-Based
Blind Channel Estimation for MIMO-OFDM Systems With
Reduced Time Averaging, IEEE Trans. on Vehicular
Technology,, VOL.59, NO. 3, MARCH 2010.

[2] Kala Praveen Bagadi, Prof. Susmita Das, MIMO-OFDM
Channel Estimation using Pilot Carriers International Journal
of Computer Applications (0975 8887), Volume 2 No.3,
May 2010.

[3] Jan Jap van de beek, ove edfors, Magnus Sandell, Sarah kate
Wilson, Per ola Berjesson., On channel estimation of OFDM
IEEE Conf, 1995.

[4] V. Tarokh, H. Jafarkhani, and A.R. Calderbank, Space-
Time Block Codes from Orthogonal Designs, IEEE Trans.
Info. Theory, Vol45, pp. 1456-67, July 1999.

[5] Proakis, J.G, and Salehi, M, Communication Systems
Engineering Prentice Hall, 1994.

[6] J. C. Guey, M. P. Fitz, M. R. Bell, and W.-Y. Kuo, Signal
design for transmitter diversity wireless communication systems
over Rayleigh fading channels, in Proc. IEEE VTC96, pp.
136140, 1996.

[7] Mihir Narayan Mohanty, Laxmi Prasad Mishra, Saumendra
Kumar Mohanty, Design of MIMO Space-Time Code for High
Data Rate Wireless Communication International Journal on
ComputerScience and Engineerin,. Vol. 3 No. 2 Feb 2011.

[8] V. Tarokh, A.Naguib, N. Seshadri, and A.R. Calderbank,
Space-time codes for wireless communication: Combined away
processing and space-time coding, IEEE Trans. Inform.
Theory, Mar. 1998.

[9] Jin-Goog Kim and Jong-Tae Lim , MAP-Based Channel
Estimation for MIMOOFDM Over Fast Rayleigh Fading
Channels, IEEE Trans. on Vehicular Technology, Vol. 57, No.
3, May 2008.

[10] K.Daniel Wong, Man-On Pun, andH. Vincent Poor, The
Continuous Time Peak-to-Average Power Ratio of OFDM
Signals Using Complex Modulation Schemes IEEE
transactions on communications, vol. 56, no. 9, pp.1390-1393,
september 2008.

[11] S. H. Han and J. H. Lee, An overview of peak-to-average
power ratio reduction techniques for multicarrier transmission,
IEEE Wireless Commun., vol. 12, no. 2, pp. 5665, Apr. 2005.

ICCA Volume 6

Uploaded by

Document Information

Copyright

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

ICCA Volume 6

Uploaded by

Copyright:

I nt er nat i onal Conf er ence on Comput er Appl i cat i ons 2012

Gives complete spanning

Function 4 F forwards the query message to one of its

Proc. of the Intl. Conf. on Computer Applications

It is important in picture processing to select an

Then our problem is reduced to an optimization problem

Gray Level Enhanced Image

Proc. of the Intl. Conf. on Computer Applications

. Randomly pick a pixel

form the image I. Let

represent the color values of

respectively in the following equation.

-norm) away from the

resolution cells in the horizontal direction and

resolution cells in the vertical direction and the gray

be the horizontal spatial

be the vertical spatial domain

is the set of resolution cells of the

, = , , are decomposed into a wavelet

, = , , is shown in Fig.3, where

and texture feature extract from

represents the probability that a pixel of

denotes color quantization of

. Then color feature vector

represents mean value;

measure correlated a pixel is to its neighbor

represent the sum of squared elements in

and also known as uniformity, non-uniformity

diagonal. We combine all these features and

in eq. (4) and texture feature vector

and a set of texture feature from

You might also like