Vol2 No 2

(IJCNS) International Journal of Computer and Network Security, 1
Vol. 2, No. 2, February 2010
Multiple region Multiple Quality ROI Encoding

using Wavelet Transform
P.Rajkumar1 and Madhu Shandilya2
12
, Department of Electronics, Maulana Azad National Institute of Technology,
Bhopal, India
madhu_shandilya@yahoo.in
Abstract: The Wavelet Transform, which was developed in the video is huge. A typical procedure of 5 minutes, taken at
last two decades, provides a better time- frequency 30 frames per second for 512x512 pixel images results in
representation of the signal than any other existing transforms. approximately 2.5GB of raw data. [3]
It also supports region of interest (ROI) encoding. This allows
different regions of interests to be encoded at different bit rats
with different quality constraints, rather than encoding the
2. Need for Image compression
entire image with single quality constraints. This feature is A digital compression technique can be used to solve both
highly desirable in medical image processing. By keeping this in the storage and the transmission problems. An efficient
mind we propose multiple regions multiple quality (MRMQ) lossy compression scheme to reduce digital data without
encoding technique which facilitates different ROIs according to significant degradation of medical image quality is needed.
the priority of enhancement. The paper utilizes the adaptive
Compression methods are important in many medical
biorthogonal wavelet application to keep quality in the highest
priority ROI, as they posses excellent reconstruction features.
applications to ensure fast interactivity during browsing
The paper also proposes bit saving by pruning the detail through large sets of images (e.g. volumetric data sets, time
coefficients in the wavelet decompositions and truncating the sequences of images, image databases). In medical
approximation coefficients outside ROI by using space imaging, it is not acceptable to lose any information when
frequency quantization method. Simulation results obtained storing or transmitting an image. There is a broad range of
shows that, by proposed method the image quality can be kept medical image sources, and for most of them discarding
high up to loss less in the ROI region while limiting the quality small image details might alter a diagnosis and cause severe
outside the ROIs. human and legal consequences. [1]
In addition to high compression efficiency, future
Keywords: Wavelet Transform, supports region of interest. moving image coding systems will require many other
features. They include fidelity and resolution scalability,
1. Introduction to medical imaging region of interest enhancement, random access decoding,
Medical images are now almost always gathered and stored and resilience to errors due to channel noise or packet loss,
in digital format for easy archiving, storage and fast encoding/decoding speed, low computational and
transmission, and to allow digital processing to improve hardware complexity.
diagnostic interpretation. Recent developments in pattern recognition (regions of
Recently, medical diagnostic data produced by hospitals interest segmentation) and image processing have advanced
increase exponentially. In an average-sized hospital, many the state-of-the-art of image processing. During the past few
tetra or 1015 bytes of digital data are generated each year, years, we have applied some of these latest image-
almost all of which have to be kept and archived. processing techniques on image enhancement classification
Furthermore, for telemedicine or tele browsing applications, and compression.
transmitting a large amount of digital data through a
bandwidth-limited channel becomes a heavy burden [1]. 3. Region of interest (ROI) encoding
Three-dimensional data sets, such as medical volumetric Support of region of interest (ROI) access is a very
data generated by computer tomography (CT) or magnetic interesting feature of image compression, in which an image
resonance (MR), typically contain many image slices that sequence can be encoded only once and then the decoder can
require huge amounts of storage. directly extract a subset of the bit stream to reconstruct a
A typical mammogram must be digitized at a resolution of chosen ROI of required quality [4]
about 4000 x 5000 pixels with 50 µm spot size and 12 bits, The arbitrarily shaped regions inside an image will be
resulting in approximately 40Mb of digital data. Such high encoded at different quality levels according to their
resolution is required in order to detect isolated clusters of importance or, as per diagnostic relevance. The whole
micro calcifications that herald an early stage cancer. The image is transformed and coefficients associated to the ROI
processing or transmission time of such digital images could are coded at higher precision (up to lossless) than the
be quite long. Also, archiving the amount of data generated background. Especially in medical imaging ROI coding help
in any screening mammography program becomes an compression method s to focus on those regions that are
expensive and difficult challenge. [2] important for diagnosis purpose.
The storage requirement for digital coronary angiogram
2 (IJCNS) International Journal of Computer and Network Security,
For some applications, only a subsection of the image

sequence is selected for analysis or diagnosis. Therefore, it
is very important to have region of interest retrievability that While JPEG 2000 uses Wavelet transform based
can greatly save decoding time and transmission bandwidth. compression. The DCT based compression; the algorithm
An important part of genetics analysis starts with breaks the image into 8x8 pixel blocks and performs a
researchers breaking down the nucleus of a human cell into discrete cosine transform on each block. The result is an 8x8
a jumbled cluster of chromosomes that are then stained with block of spectral coefficients with most of the information is
dye so that they can be studied under a microscope. This concentrated in relatively few coefficients. Quantization is
jumble of stained chromosomes, which carry the genetic performed, which approximates the larger coefficients;
code of the host individual, is then photographed, creating a smaller coefficients become zero. These quantized
chromosome spread image (see Fig. 1 (a)). This image is coefficients are then reordered in a zig zag manner to group
subsequently subject to a procedure called chromosome- the largest values first, with long strings of zeroes at the end
karyotyping analysis. The result of this procedure is a that can be efficiently represented.
karyotype image (see Fig. 1 (b)), the standard form used to While this algorithm is very good for general purposes,
display chromosomes. In this configuration, the it has some draw backs when applied to medical images .It
chromosomes are ordered by length from the largest degrades ungracefully at high compression ratios, producing
(chromosome 1) to the smallest (chromosome 22 in prominent artifacts at block boundaries, and it can not take
humans), followed by the sex chromosomes. Karyotype advantage of patterns larger than the8x8 blocks. Such
images are used in clinical tests, such as amniocentesis, to artifacts could potentially be mistaken as being
determine if all the chromosomes appear normal and are diagnostically significant. A wavelet approach was also
present in the correct number. [5] Unlike some other types suggested by Li who considered the problem of accessing
of medical imagery, chromosome images (see Fig. 1) have still, medical image data, remotely over low bandwidth
an important common characteristic: the regions of interest networks. This was accomplished using a region of interest
(ROIs) to cytogeneticists for evaluation and diagnosis are all (ROI) based approach that allocated additional bandwidth
well determined and segmented prior to image storage. The within a dynamically allocated ROI, whilst at the same time
remaining background images, which may contain cell providing an embedded bit stream, which is useful for
nuclei and stain debris, are kept as well in routine progressive image encoding.[3] Some of the key features of
cytogenetics lab procedures for specimen reference rather the wavelet transform which make it such a useful tool are:
than for diagnostic purposes. Since the chromosome ROIs • Spatial-frequency localization,
are much more important than the rest of the image for • Energy compaction,
karyotyping analysis, loss less compression for the former is • Decaying magnitude of wavelet coefficients across
required while lossy compression for the latter is acceptable. sub-bands.
This calls for lossy and loss less region-of-interest (ROI) Wavelet based compression schemes generally out
coding. In contrast, commercial chromosome karyotyping perform JPEG compression in terms of image quality at a
systems fail to utilize the ROI information by compressing given compression ratio, and the improvement can be
entire chromosome spread or karyotype images. dramatic at high compression ratios. [9]
The Wavelet Transform, which was developed in the
last two decades, provides a better time- frequency
representation of the signal than any other existing
transforms. It supports ROI (region of interest) encoding.
Multiple regions multiple qualities ROI encoding facilitates
Figure. 1(a) A metaphase cell spread image[5] different ROIs according to the priority to be enhanced with
different qualities (i.e. at different compression ratios, both
by lossless and lossy methods), while limiting the bit size
outside ROI by compressing heavily (lossy). This paper
proposes the application of adaptive biorthogonal wavelets
to keep quality in the highest priority ROI as they posses
excellent reconstruction features.
Figure. 1(b) A metaphase cell’s karyotype[5] 4.1 Classification of wavelets
In the karyotype, all chromosomes in the spread are rotated We can classify wavelets into two classes: [7]
and copied onto an image with constant background and (a) Orthogonal and (b) biorthogonal.Based on the
positioned according to their classes. The label annotation is application, either of them can be used.
drawn separately.
4.1.1 Features of orthogonal wavelet filter banks
4. Wavelet transform
The coefficients of orthogonal filters are real numbers. The
Discrete cosine transform and Wavelet transform are more filters are of the same length and are not symmetric.
commonly used for compression. The popular JPEG & The low pass filter, G and the high pass filter, H are related
MPEG uses discrete cosines transform based compression; to each other by
0 0
-N -1
H (z) = z G (-z ) (1) coefficients for all pictures for all images while
0 0
decomposing the image using the wavelet transform.
The two filters are alternated flip of each other. The Keeping this in the mind, Adaptive wavelet basis
alternating flip automatically gives double-shift function is additionally proposed. Amongst the all important
orthogonality between the low pass and high pass filters, biorthogonal wavelet basis functions, a particular wavelet
i.e., the scalar product of the filters, for a shift by two is zero basis function is chosen on the basis of comparison of the
i.e, ∑G [k] H [k-2l] = 0, where k,lεZ . Perfect highest coefficients produced by all the different wavelet
reconstruction is possible with alternating flip. basis functions at the first level of the decomposition
Also, for perfect reconstruction, the synthesis filters are process. The proposed MRMQ Encoding uses the Discrete
identical to the analysis filters except for a time reversal. Wavelet Transform, chooses the right Biorthogonal basis
Orthogonal filters offer a high number of vanishing function on the basis of the highest coefficient produced in
moments. This property is useful in many signal and image the highest priority ROI amongst of all the ROIs.
processing applications. They have regular structure, which
leads to easy implementation and scalable architecture. 5. Multiple Regions Multiple Quality ROI
Encoding
4.1.2 Features of biorthogonal wavelet filter banks
For certain applications, only specific regions in the
In the case of the biorthogonal wavelet filters, the low pass volumetric data or video are of interest. For example, in
and the high pass filters do not have the same length. The MR imaging of the skull, the physician is mostly concerned
low pass filter is always symmetric, while the high pass about the features inherent in the brain region. In video
filter could be either symmetric or anti-symmetric. The conferencing, the speaker’s head and shoulders are of main
coefficients of the filters are either real numbers or integers. interest and need to be coded at higher quality whereas the
For perfect reconstruction, biorthogonal filter bank has background can be either discarded or encoded at a much
all odd length or all even length filters. The two analysis lower rate. High compression ratios can be achieved by
filters can be symmetric with odd length or one symmetric allocating more bit rate for region(s) of interest (ROI) and
and the other antisymmetric with even length. Also, the less bit rate for the remaining regions, i.e. the background.
two sets of analysis and synthesis filters must be dual. The Region-based image coding schemes using
linear phase biorthogonal filters are the most popular filters heterogeneous (multiple) quality constraints are especially
for data compression applications attractive because they not only can well preserve the
The analysis and synthesis equations for the biorthogonal diagnostic features in region(s) of interest, but also meet the
DWT of any x (t) ε L2(R) are requirements of less storage and shorter transmission time
for medical imaging applications and video transmission.
Analysis A main advantage of the proposed technique is that it
supports multiple-region multiple-quality (MRMQ) coding.
ã j,κ = ∫ x(t)2j/2φ̃(2jt-k)dt (2) By this method, total bit budget can be allocated among
multiple ROIs and background depending on the quality
b̃j,κ = ∫ x(t)2j/2ψ̃(2jt-k)dt (3) constraints. Experimental results show that this technique
offers reasonably well performance for coding multiple
Synthesis
ROIs.
M-1
x(t)= 2 N/2Σ ãN,k φ(2 Nt-k) + Σ 2j/2 To support region of interest (ROI) coding, it is
k j=N necessary to identify the wavelet transform coefficients
associated with the ROI. We have to keep track of the
Σ b̃j,k ψ(2 jt-k) (4) coefficients that are involved in the reconstruction of ROI
k through each stage of decomposition.
5.1 ROI Coding

We use region mapping which trace the positions of
{Φi,j (t)Ψi,j (t)} and {Φi,j(t) Ψ˜i,j(t)} are the biorthogonal
pixels in an ROI in image domain back into transform
basis functions[8]. a˜j,k and b˜j,k are the scaling and
domain by inverse DWT. The coefficients of greater
wavelet coefficients respectively; together they form the
importance are called ROI coefficients, the rest are called
biorthogonal DWT coefficients of x(t). The DWT starts at
background coefficients.
some finest scale j = M and stops at some coarsest scale j =
N. M - N is the number of levels of decomposition in the
The coding algorithm, fig.2 needs to keep track of the
biorthogonal DWT of x (t).
locations of wavelet coefficients according to the shape of
4.1.3 Selection of wavelet from the biorthogonal family: the ROI. To obtain the information about ROI coefficients, a
mask image, which specifies the ROI, is decomposed by the
Different wavelets will be suited for different types of wavelet decomposition of the image. In each decomposition
images. One wavelet basis may produce highest coefficient stage, each subband of the decomposed mask contains
for a picture, may not necessarily produce highest information for specifying the ROI in that subband. By
successively decomposing the approximation coefficients
(LL subband) for a number of decomposition levels,

information about ROI coefficients is obtained.
Figure 3. Original medical Image of

abdominal operation (2.32bpp)
Figure 2. ROI Encoding using Wavelet Transform Figure 4. ROI Enhanced medical image of fig. 3
Here no. of ROI =2(0.92bpp, PSNR =25.2db)
5.2 Outside ROI
Space –Frequency Quantization (SFQ) technique exploits

both spatial and frequency compaction property of the
wavelet transform through the use of two simple PSNR
quantization modes.
To exploit the spatial compaction property, a symbol is
defined, that indicates that a spatial region of high
frequency coefficients has zero value. Application of this
symbol is referred to as zero-tree quantization. This is done Bits per pixel
in the first phase called Tree Pruning Algorithm. In the
next phase called Predicting the tree, the relation between a
spatial region in image and the tree- structured set of Figure 5. PSNR vs. bits per pixel of the ROI enhanced
coefficients is exploited. Zero tree quantization can be Image “abdomen”
viewed as a mechanism for pointing to the location where Note: ROI area is 17230pixels and Image is 448x336 pixels.
high frequency coefficients are clustered. Thus, this
quantization mode directly exploits the spatial clustering of
high frequency coefficients predicted.
For coefficients that are not set to zero by zero tree
quantization, a common uniform scalar quantization,
independent of coefficient frequency band is applied.
Uniform quantization followed by entropy coding provides
nearly optimal coding efficiency.
Figure 6. Teratoma affected baby
6. Results
In Fig. 3 an Image is taken showing the abdominal

operation. In the above Image MRMQ ROI encoding is
applied. The result is shown in fig. 4.Here the no. of ROIs
are 2, as shown the area outside the ROI is heavily degraded Figure 7. ROI enhanced Teratoma affected baby
.The PSNR of this image is 25dB.while the quality is (Background degraded)
lossless in the ROI. The PSNR vs. bits per pixel of the ROI
enhanced Image at various compression points is plotted in
fig.5. This shows that although the PSNR difference is less
whiles the bits per pixel reduces from 1bpp to 0.5 bpp using
Matlab® simulation.
Fig. 6 an image of a newborn baby affected by

sacrococcyged teratoma (before surgery) is taken. In fig. 7
the area outside the ROI is heavily degraded. In fig. 8,
Outside the ROI is excluded for further reducing the bit size
requirement. The PSNR vs. bits per pixel at various Figure 8. ROI enhanced Trachoma affected baby
compression points are plotted in fig. 9. (Background excluded)
Electronics, Vol. 46, No. 4, pp. 1103-1127, November

2000.
[7] K.P.Soman and K.I.Ramachandran -Insight Into

Wavelets –From Theory to Practice,PHI Publications.
PSNR [8] Sonja Grgic, Mislav Grgic, and Branka Zovko-Cihlar,
(db) Performance Analysis of Image Compression Using
Wavelets, IEEE Transactions on Industrial Electronics,
vol.48, no. 3, June 2001.
[9] Michael W. Marcellina, Margaret A. Lepleyb, Ali
Bits per pixel Bilgina, Thomas J. Flohrc, Troy T. Chinend, James H.
Kasner, An overview of quantization in JPEG 2000
Figure 9. PSNR vs. bpp of the ROI enhanced Teratoma Signal Processing: Image Communication 17 (2002) 73–
affected baby 84.
7. Conclusions Authors Profile

The simulation results give bits per pixel with
corresponding PSNR values for the proposed MRMQ ROI P.Rajkuma received the M.Tech. Degree in Electronics
encoding. It has been observed that for still images, the bit Engineering from Maulana Azad National Institute of
Tech.Bhopal, India in 2006.He is doing his research in the area of
size requirement was made less by degrading the image
medical image processing.
quality out side the ROI, while keeping the quality of image
uncompromised in the ROI region. Madhu Shandilya received the M.Tech. and Ph.D. degrees in
In the present paper, still image results are simulated Electronics Engineering from Maulana Azad National Institute of
with the biorthogonal wavelet transform family for the Tech.Bhopal, India in 1994 and 2005, respectively. From last 22
multiple region multiple quantization (MRMQ) ROI coding, years she is associated with Electronics Engineering from Maulana
therefore further analysis can be extended in the field of Azad National Institute of Tech.Bhopal. She has
video processing. Also the present analysis is carried on the published/presented about 20 papers in various national and
medical still images, which can be extended for the video international journals /conferences. Her area of specialization is
digital image processing.
processing with graphical and natural images.
References
[1] Shaou-Gang Miaou, Shih-Tse Chen, Shu-Nien Chao ,

Wavelet-Based Lossy-to-Lossless Medical Image
Compression Using Dynamic VQ And SPIHT Coding,
Biomedical Engineering applications, Basis
&Communications. Pg 235-242 Vol. 15 No. 6 December
2003.
[2] Mónica Penedo, William A. PearmanPablo G.
Tahoces, Miguel Souto, Juan J. Vidal,
EmbeddedWavelet Region-Based Coding Methods
Applied To Digital Mammography IEEE International
Conference on Image Processing, vol.3 Barcelona,
Spain 2003.
[3] David Gibson, Michael Spann, Sandra I. Woolley, A
Wavelet-Based Region of Interest Encoder for the
Compression of Angiogram Video Sequences, IEEE
Transactions on Information Technology in Biomedicine
8(2): 103-113 (2004)
[4] Kaibin Wang, B.Eng., M.Eng. Thesis: Region-Based
Three-Dimensional Wavelet Transform Coding,
Carleton University, Ottawa, Canada, May, 2005.
[5] Zixiang Xiong, Qiang Wu and Kenneth R.,Castlemen,
Enhancement, Classification And Compression Of
Chromosome Images. Workshop on Genomic Signal
Processing, 2002 , Cite seer.
[6] Charilaos Christopoulos, Athanassios Skodras ,Touradj
Ebrahimi - The JPEG2000 Still Image Coding
System:An Overview , IEEE Transactions on Consumer
2- Dimension Automatic Microscope Moving Stage

Ovasit P.1 Adsavakulchai S.1,* Srinonghang W1., and Ueatrongchit, P. 1
1
School of Engineering, University of the Thai Chamber of Commerce
126/1 Vibhavadi Rangsit Rd., Bangkok 10400
*Corresponding author, e-mail: suwannee_ads@utcc.ac.th
Abstract: Currently, microscope stage has controllable by An automated slide system is disclosed which organizes
manual. In medical laboratory, the specimen slide examination microscope slides in cassettes, automatically and positioned
system processes one at a time for microscopic examination. each slide under the microscope as provided by the protocol,
The main objective of this study is to develop a two-dimension and after examination returns the slide to its proper cassette
automatic microscope moving stage. This equipment is designed [3]. A slot configured for holding slides in spaced parallel
by microcontroller PIC 16F874 using stepping motor as configuration using the mechanism for removing and
horizontal feed mechanism. There are three function modes, the
replacing a slide housed. A feed arm containing a
first one is manual, the second is automatic scan specimen
which transfers the specimen slide onto a microscope stage
longitudinal to draw-out spring wire surrounding an
which has controllable X and Y axis positioning to move the imaginary longitudinal axis having at the first end and a
specimen slide into the optical viewing field of the microscope second end, the first and second end being bent orthogonal
and examination over the desired area of the specimen and the to one another and to the imaginary longitudinal axis of said
last one is to examine specimen slides may be automatically draw-out spring wire, said longitudinal draw-out spring wire
returned to the microscope stage for reexamination. The result being positioned in said longitudinal channel in said feed
of this study can be concluded that the accuracy of this arm such that bent ends protrude from the channel and
equipment for reexamination the specimen slide is 86.03 % wherein said longitudinal draw-out spring wire is
accuracy. operatively positioned in said longitudinal channel such that
the draw-out spring wire is rotatable therein, allowing for
each bent end to change orientation in respect to the feed
Keywords: Microscope Automatic Moving Stage, microcontroller
arm. [1].
PIC16F874, stepping motor
The main objective of this study is to develop an
automatically returned to the microscope stage for
1. Introduction reexamination. These technologies dramatically increased
The microscope is a conventional laboratory microscope the accuracy of measurement results and contributed greatly
with attached to actuate the stage and control is affected to the modernization of testing and medical care (medical
through manual [1] as shown in Figure 1. testing).
2. Materials and Methods

Sample preparation:
2.1 To prepare the blood smear sample and set up the

microscope working area at 1000x with 0.2 mm.
dimension [2] as shown in Figure 2. In order to the area
of the specimen slide is viewed during examination of
the specimen without sliding it.
Figure 1. Microscope moving stage
The basic principal for diagnostic in red blood cell is using

microscope manually. In such cases, electronic systems may
be used to automatically examine and analyze the optical
images of the microscope [2]. Where electronics systems are
used for rapid analysis of microscope specimen images it
becomes desirable to automatically regularly and rapidly Figure 2. Microscope working area
feed the specimens to the microscope optics. After analysis a
specimen would be removed to make room for the next 2.2 To set up the scope of microscope moving stage
specimen and would be collected for either further scanning area with 40 x 26 mm. as shown in Figure 3.
examination, reference, record keeping or disposal. About the specimen stage this opening in the specimen
stage is made as large as possible and exposes the full

width of the specimen slide.
Figure 3. Scanning area

Figure 5. Digital Electronics design
2.3 The automatic sequential examination of a group of
microscope specimen slides comprising:
Figure 5 is to illustrate the overall digital electronics
using two microcontrollers with serial communication and
synchronous Serial Peripheral Interface (SPI).
2.3.1 Moving stage design:
2.3.3 Microcontroller in Slave Board:
Apparatus comprised a substage directing serves to move
stage in a horizontal plane and there is provided further
2.3.3.1 Encoder: using sequential logic to control
positioning means supporting said translation means and
moving stage. The characteristic of encoder 2 signal using
operable for moving said stage with a specimen slide
microcontroller PIC 16F873 via RA0-RA3 port as shown in
supported therein vertically as shown in Figure 4.
Figure 6. The signals moving stage is to set up into 3
statuses 1. No movement 2. Increase the distance and 3.
Reduce distance as shown in Figure 7.
Figure 6. The characteristic of encoder 2 signal
Figure 4. Moving stage design
2.3.2 Digital Electronics Design:
The digital electronics architecture has two main

functional blocks, Master Board and Slave Board. ICP
(Instrument Control Processor): used a PIC 16F873
processor to perform all instrument control and event
processing functions as shown in Figure 5. The ICP will be
responsible for the following tasks: processing commands;
monitoring source and adjusting the LCD readout mode as
required; calculating centroids and transmitting centroid
positions [4],[5],[6]. Figure 7. Logical control
2.3.3.2 Stepping motor: control moving stage

using microcontroller PIC 16F873 via RB0-RB7 port and
working together with IC ULN2803 to control stepping
motor as shown in Figure 8.
Microscope Scan.
Version 1.00
Status 1a 1b 2a 2b
Wait
Home 1:Home 2:Manual

1 4
Please Wait. 3:Scan 4:Config
2 3
#
Start. 1:Posit. #:Exit

Please Wait. 2:Option Config
1
Wait
X 0000 #:Exit
Y 0000 *:Scan
#
Figure 8. Characteristics of stepping motor control * 2
Start 0000x0000y
Final 0000x0000y
X 0000 #:Exit
Y 0000 *:Stop
2.3.3.3 Serial Peripheral Interface: using Master X 0000 #:Exit 1: 0sec #:Exit
Y 0000 *:Jump 2: 0unit Option
Synchronous Serial Port in microcontroller PIC 16F873 * : X :Y
# : 1 or 2 1:T 2: S
using Microcontroller in Slave Board as shown in Figure 9.
1:Start #:Exit X _000 #:Exit T:>_sec #:Exit T: 0sec #:Exit
2:Final < Save Y 0000 < Jump S: 0unit Time. S:>_unit Space
Figure 10. Main menu of control program

Register Register SSPBUF
SSPBUF
3. RESULTS AND DISCUSSION

Register SSPSR Register SSPSR
To test the points and the results is shown in table 2.
The microscopic examination of the specimen slide can
Master Slave take place either visually or automatically. Motorized
microscope components and accessories enable the
Figure 9. Serial Peripheral Interface (SPI) investigator to automate live-cell image acquisition and
are particularly useful for time-lapse experiments about
20 milliseconds [7],[8]. For this purpose the X and Y
2.3.4 Microcontroller in Mater Board positioning systems can be controlled manually or
2.3.4.1 Input from keyboard using RB1-RB7 in automatically. Thus, the specimen slide carried by the
term of matrix 4 x 3 stage may be moved to any desired location relative to the
2.3.4.2 Display result using RA0-RA5 for Liquid optical axis by actuation of the Y-axis drive 43 and the X-
Crystal Display axis drive 44. For automatic examination the drives 43,
2.3.4.3 Serial Peripheral Interface SCK port 44 would be energized under scan or other program
2.3.4.5 Data communication using SPI as shown control. Finally, the specimen slide reexamination is
in table 1 automatically returned to the microscope stage for
reexamination with very high accuracy.
Table 1 : Data communication using SPI
Table 2: Sample testing results
To design main menu as shown in Figure 10 to control the

microscopic stage.
4. Conclusion microscopes, accepted for publication on Nucl. Instr.

Meth. A(2005).
After the examination of a particular series of specimen [5] Powell, Power and Perkins, The Study of Elementary
slides has been completed any individual specimen slide Particles by the Photo-graphic Method, Pergamon
that requires re-examination can by either operator Press (1959).
signals or by predetermined control signals be fed [6] W.H. Barkas, Nuclear Research Emulsion, Academic
automatically back into the microscope viewing optics for Press Inc. (London) Ltd. (1963).
further examination [9]. [7] John F. Reid et al., A Vision-based System for
Upon completion of the examination of a slide the Computer Control and Data Acquisition in
horizontal positioning Y-axis drive returns the specimen fermentation Processes, 38th Food Technology
slide on the stage to the position. It can be concluded that Conference, 1992.
the accuracy of this equipment for reexamination the [8] J.M. Shine, Jr., et al., Digital Image Analysis System
specimen slide is 86.03 % accuracy. for Determining Tissue-Blot Immunoassay Results for
Ratoon Stunting Disease of Sugarcane, Plant Disease,
Acknowledgements vol. 77 No. 5, pp. 511-513, 1993.
[9] Jinlian Ren et al., Knowledge-based Supervision and
This project is supported by University of the Thai Chamber Control of Bioprocess with a Machine Vision-based
of Commerce grant. Sensing System, Journal of Biotechnology 36 (1994)
25-34.
References
Authors Profile
[1] C.R.David, Microscopy and related methods from
http://www.ruf.rice.edu/~bioslabs/methods/microscop
Suwannee Adsavakulchai received the M.S. degrees in Computer
y/microscopy.html Information Systems from Assumption University in 1994 and
[2] Robert H., et.al, Handbook of Hematologic Doctoral of Technical Science from Asian Institute of Technology
Pathology, Marcel Dekker, Inc. (2000) in 2000, respectively. She now works as lecturer in the department
[3] Qin Zhang et al., A Prototype Neural Network of Computer Engineering, School of Engineering, University of the
Supervised Control System for Bacillus thuringiensis Thai Chamber of Commerce.
Fermentations, Biotechnology and Bioengineering,
vol. 43, pp. 483-489 (1994).
[4] N. Armenise et al., High-speed particle tracking in
nuclear emulsion by last generation automatic
A Large Block Cipher Involving a Key

Applied on Both the Sides of the Plain Text
V. U. K. Sastry1, D. S. R. Murthy2 , S. Durga Bhavani3
1
Dept. of Computer Science & Engg., SNIST,
Hyderabad, India,
vuksastry@rediffmail.com
2
Dept. of Information Technology, SNIST,
Hyderabad, India,
dsrmurthy.1406@gmail.com
3
School of Information Technology, JNTUH,
Hyderabad, India,
sdurga.bhavani@gmail.com
Abstract: In this paper, we have developed a block cipher by considering an example. In section 4, we have dealt with the
modifying the Hill cipher. In this, the plain text matrix P is cryptanalysis of the cipher. Finally, in section 5, we have
multiplied on both the sides by the key matrix. Here, the size of presented the computations and arrived at the conclusions.
the key is 512 bits and the size of the plain text is 2048 bits. As
the procedure adopted here is an iterative one, and as no direct 2. Development of the cipher
linear relation between the cipher text C and the plain text P can
be obtained, the cipher cannot be broken by any cryptanalytic
Consider a plain text P which can be represented in the form
attack. of a square matrix given by
P = [Pij], i = 1 to n, j = 1 to n, (2.1)
Keywords: Block Cipher, Modular arithmetic inverse, Plain text, where each Pij is a decimal number which lies between 0
Cipher text, Key. and 255.
Let us choose a key k consisting of a set of integers,
1. Introduction which lie between 0 and 255. Let us generate a key matrix,
The study of the block ciphers, which was initiated several denoted as K, given by
centuries back, gained considerable impetus in the last K = [Kij], i = 1 to n, j = 1 to n, (2.2)
quarter of the last century. Noting that diffusion and where each Kij is also an integer in the interval [0 – 255].
confusion play a vital role in a block cipher, Feistel etal, [1] Let
–[2] developed a block cipher, called Feistel cipher. In his C = [Cij], i = 1 to n, j = 1 to n (2.3)
analysis, he pointed out that, the strength of the cipher be the corresponding cipher text matrix.
increases when the block size is more, the key size is more, The process of encryption and the process of decryption
and the number of rounds in the iteration is more. applied in this analysis are given in Fig. 1.
The popular cipher DES [3], developed in 1977, has a
56 bit key and a 64 bit plain text. The variants of the DES
are double DES, and triple DES. In double DES, the size of
the plain text block is 64 bits and the size of the key is 112
bits. In the triple DES, the key is of the length 168 bits and
the plain text block is of the size is 64 bits. At the
beginning of the century, noting that 64 bit block size is a
drawback in DES, Joan Daemen and Vincent Rijmen, have
developed a new block cipher called AES [4], wherein the
block size of the plain text is 128 bits and key is of length
128, 192, or 256 bits. In the subsequent development, on
modifying Hill cipher, several researchers [5]–[9], have
developed various cryptographical algorithms wherein the
length of the key and the size of the plain text block are
quite significant.
In the present paper, our objective is to develop a block
cipher wherein the key size and the block size are Figure 1. Schematic Diagram of the cipher
significantly large. Here, we use Gauss reduction method for
obtaining the modular arithmetic inverse of a matrix. In Here r denotes the number of rounds.
what follows, we present the plan of the paper. In the process of encryption, we have used an iterative
In section 2, we have discussed the development of the procedure which includes the relations
cipher. In section 3, we have illustrated the cipher by P = (K P K) mod 256, (2.4)
P = Mix (P), (2.5) Algorithm for Decryption

and P = P ⊕ K (2.6) 1. Read n, C, K, r
The relation (2.4) causes diffusion, while (2.5) and (2.6) 2. K–1 = Inverse (K)
lead to confusion. Thus, these three relations enhance the 3. for i = 1 to r
strength of the cipher. {
Let us consider Mix (P). In this the decimal numbers in C=C⊕K
P are converted into their binary form. Then we have a C = IMix (C)
matrix of size n x 8n, and this is given by C = (K–1 C K–1) mod 256
}
4. P = C
5. Write (P)
Algorithm for Inverse (K)
// The arithmetic inverse (A–1), and the determinant of the
matrix (∆) are obtained by Gauss reduction method.
Here, P111 , P112, …, P118 are binary bits corresponding 1. A = K, N = 256
to P11. Similarly, Pij1, Pij2, …, Pij8 are the binary bits 2. A–1 = [Aji] / ∆, i = 1 to n, j = 1 to n
representing Pij. //Aji are the cofactors of aij, where aij are elements
The above matrix can be considered as a single string in of A, and ∆ is the determinant of A
a row wise manner. As the length of the string is 8n2, it is 3. for i = 1 to n {
divided into n2 substrings, wherein the length of each if ((i ∆) mod N = 1)
substring is 8 bits. If n 2 is divisible by 8, we focus our d = i;
attention on the first 8 substrings. We place the first bits of break;
these 8 binary substrings, in order, at one place and form a }
new binary substring. Similarly, we assemble the second 8 4. B = [d Aji] mod N
bits and form the second binary substring. Following the // B is the modular arithmetic inverse of A
same procedure, we can get six more binary substrings in 3. Illustration of the cipher
the same manner. Continuing in the same way, we exhaust
all the binary substrings obtained from the plain text. Let us consider the following plain text.
However, if n2 is not divisible by 8, then we consider No country wants to bring in calamities to its own
the remnant of the string, and divide it into two halves. people. If the people do not have any respect for the
Then we mix these two halves by placing the first bit of the country, then the Government has to take appropriate
second half, just after the first bit of the first half, the second measures and take necessary action to keep the people in
bit of the second half, next to the second bit of the first half, order. No country can excuse the erratic behaviour of the
etc. Thus we get a new binary substring corresponding to people, even though something undue happened to them in
the remaining string. This completes the process of mixing. the past. Take the appropriate action in the light of this
In order to perform the exclusive or operation in P = P fact. Invite all the people to come into the fold of the
⊕ K, we write the matrices, both P and K, in their binary Government. Try to persuade them as far as possible. Let
form, and carryout the XOR operation between the us see!! (3.1)
corresponding binary bits. Let us focus our attention on the first 256 characters of
In the process of decryption, the function IMix the above plain text which is given by
represents the reverse process of Mix. No country wants to bring in calamities to its own
In what follows, we present the algorithms for people. If the people do not have any respect for the
encryption, and decryption. We also provide an algorithm country, then the Government has to take appropriate
for finding the modular arithmetic inverse of a square measures and take necessary action to keep the people in
matrix. order. No country can excuse the erratic (3.2)
Algorithm for Encryption On using EBCDIC code, we get 26 numbers,
1. Read n, P, K, r corresponding to 256 characters. Now on placing 16
2. for i = 1 to r numbers in each row, we get the plain text matrix P in the
{ decimal form
P = (K P K) mod 256
P = Mix (P)
P=P⊕K
}
3. C = P
4. Write (C)
Obviously, here the length of the plain text block is 16 x

16 x 8 (2048) bits.
Let us choose a key k consisting of 64 numbers. This only in one binary bit as the EBCDIC code of y is 168 and
can be written in the form of a matrix given that of z is 169.
On using the encryption algorithm, we get the cipher
text C corresponding to the modified plain text (wherein y is
replaced by z) in the form
by
The length of the secret key (which is to be transmitted)
is 512 bits. On using this key, we can generate a new key K
in the form
w On
T
here U = Q , in which T denotes the transpose of a matrix, comparing (3.9) and (3.10), we find that the two cipher texts
and R and S are obtained from Q and U as follows. On differ in 898 bits, out of 2048 bits, which is quite
interchanging the 1st row and the 8th row of Q, the 2nd row considerable. However, it may be mentioned here that, the
and the 7th row of Q, etc., we get R. Similarly, we obtain S impact of changing 1 bit is not that copious, as the size of
from U. Thus, we have the plain text is very large. Even then it is remarkable.
Now let us change the key K given in (3.6) by one
binary bit. To this end, we replace the 60th element 5 by 4.
Then on using the original plain text given by (3.3), we get
C in the form
whose size is 16 x 16.

On using the algorithm for modular arithmetic inverse
(See Section 2), we get
On comparing (3.9) and (3.11), we find that the cipher texts

differ in 915 bits, out of 2048 bits.
From the above analysis, we find that the avalanche
effect is quite pronounced and shows very clearly that the
cipher is a strong one.
4. Cryptanalysis
On In the literature of cryptography, it is well known that the
using (3.6) and (3.7), it can be readily shown that different types of attacks for breaking a cipher are:
K K–1 mod 256 = K–1K mod 256 = I. (3.8) (1) Cipher text only attack,
On applying the encryption algorithm, described in (2) Known plain text attack,
Section 2, we get the cipher text C in the form (3) Chosen plain text attack,
(4) Chosen cipher text attack.
In the first attack, the cipher text is known to us
together with the algorithm. In this case, we can determine
the plain text, only if the key can be found. As the key
contains 64 decimal numbers, the key space is of size
2512 ∼ (103)51.2 = 10153.6
which is very large. Hence, the cipher cannot be broken by
applying the brute force approach.
We know that, the Hill cipher [1] can be broken by the
known plain text attack, as there is a direct linear relation
On using (3.7) and (3.9), and applying the decryption between C and P. But in the present modification, as we
algorithm presented in section 2, we get the Plain text P. have all nonlinear relations in the iterative scheme, the C
This is the same as (3.3). can never be expressed in terms of P, thus P cannot be
Let us now find out the avalanche effect. To this end, determined by any means in terms of other quantities.
we focus our attention on the plain text (3.2), and modify Hence, this cipher cannot be broken by the known plain text
the 88th character ‘y’ to ‘z’. Then the plain text changes attack.
As there are three relations, which are typical in nature, [5]. V. U. K. Sastry, V. Janaki, “On the Modular Arithmetic
in the iterative process for finding C, no special choice of Inverse in the Cryptology of Hill Cipher”, Proceedings
either the plain text or the cipher text or both can be of North American Technology and Business
conceived to break the cipher. Conference, Sep. 2005, Canada.
[6]. V. U. K. Sastry, S. Udaya Kumar, A. Vinaya Babu, “A
5. Conclusions Large Block Cipher using Modular Arithmetic Inverse
In the present paper, we have developed a large block cipher of a Key Matrix and Mixing of the Key Matrix and the
by modifying the Hill cipher. In the case of the Hill cipher, Plaintext”, Journal of Computer Science 2 (9), 698 –
it is governed by the single, linear relation 703, 2006.
C = (K P) mod 26, (5.1) [7]. V. U. K. Sastry, V. Janaki, “A Block Cipher Using
while in the present case, the cipher is governed by an Linear Congruences”, Journal of Computer Science
iterative scheme, which includes the relations 3(7), 556 – 561, 2007.
P = (K P K) mod 256, (5.2) [8]. V. U. K. Sastry, V. Janaki, “A Modified Hill Cipher
P = Mix (P), (5.3) with Multiple Keys”, International Journal of
and P = P ⊕ K. (5.4) Computational Science, Vol. 2, No. 6, 815 – 826, Dec.
Further, it is followed by C = P. (5.5) 2008.
In the case of the Hill cipher, we are able to break the [9]. V. U. K. Sastry, D. S. R. Murthy, S. Durga Bhavani, “A
cipher as there is a direct linear relation between C and P. Block Cipher Involving a Key Applied on Both the
On the other hand, in the case of the present cipher, as we Sides of the Plain Text”, International Journal of
cannot obtain a direct relation between C and P, this cipher Computer and Network Security (IJCNS), Vol. 1, No.1,
cannot be broken by the known plain text attack. pp. 27 – 30, Oct. 2009.
By decomposing the entire plain text given by (3.1) into
blocks, wherein each block is of size 256 characters, the Authors Profile
corresponding cipher text can be obtained in the decimal
form. The first block is already presented in (3.9) and the Dr. V. U. K. Sastry is presently working as Professor in the Dept.
rest of the cipher text is given by of Computer Science and Engineering (CSE), Director (SCSI),
Dean (R & D), SreeNidhi Institute of Science and Technology
(SNIST), Hyderabad, India. He was Formerly Professor in IIT,
Kharagpur, India and worked in IIT, Kharagpur during 1963 –
1998. He guided 12 PhDs, and published more than 40 research
papers in various international journals. His research interests are
Network Security & Cryptography, Image Processing, Data Mining
and Genetic Algorithms.
Dr. S. Durga Bhavani is presently working as Professor in School

of Information Technology (SIT), JNTUH, Hyderabad, India. Her
research interest is Image Processing.
Mr. D. S. R. Murthy obtained B. E. (Electronics) from Bangalore

University in 1982, M. Tech. (CSE) from Osmania University in
1985 and presently pursuing Ph.D. from JNTUH, Hyderabad since
2007. He is presently working as Professor in the Dept. of
In this analysis, the length of the plain text block is 2048 Information Technology (IT), SNIST since Oct. 2004. He earlier
bits and the length of the key is 512 bits. As the worked as Lecturer in CSE, NIT (formerly REC), Warangal, India
cryptanalysis clearly indicates, this cipher is a strong one during Sep. 1985 – Feb. 1993, as Assistant Professor in CSE,
and it cannot be broken by any cryptanalytic attack. This JNTUCE, Anantapur, India during Feb. 1993 – May 1998, as
analysis can be extended to a block of any size by using the Academic Coordinator, ISM, Icfaian Foundation, Hyderabad, India
concept of interlacing [5]. during May 1998 – May 2001 and as Associate Professor in CSE,
SNIST during May 2001 - Sept. 2004. He worked as Head of the
Dept. of CSE, JNTUCE, Anantapur during Jan. 1996 – Jan 1998,
Dept. of IT, SNIST during Apr. 2005 – May 2006, and Oct. 2007 –
References Feb. 2009. He is a Fellow of IE(I), Fellow of IETE, Senior Life
Member of CSI, Life Member of ISTE, Life Member of SSI,
[1]. Feistel H, “Cryptography and Computer Privacy”, DOEACC Expert member, and Chartered Engineer (IE(I) &
Scientific American, May 1973. IETE). He published a text book on C Programming & Data
[2]. Feistel H, Notz W, Smith J, “Some Cryptographic Structures. His research interests are Image Processing and Image
Techniques for Machine-to-Machine Data Cryptography.
Communications”, Proceedings of the IEEE, Nov. 1975.
[3]. William Stallings, Cryptography and Network Security,
Principles and Practice, Third Edition, Pearson, 2003.
[4]. Daemen J, Rijmen V, “Rijdael: The Advanced
Encryption Standard”, Dr. Dobb’s Journal, March
2001.
Electrocardiogram Prediction Using

Error Convergence-type Neuron Network System
Shunsuke Kobayakawa1 and Hirokazu Yokoi1
1
Graduate School of Life Science and Systems Engineering, Kyushu Institute of Technology
2-4 Hibikino, Wakamatsu-ku, Kitakyushu-shi, Fukuoka 808-0196, Japan
{kobayakawa-shunsuke@edu., yokoi@}life.kyutech.ac.jp
Abstract: The output error of a neuron network cannot outputs, for ECNNS is a single output. The highest output
converge at zero, even if training for the neuron network is accuracy of NN system cannot be expected for even if
iterated many times. “ Error Convergence-type Neuron Network ECMNNS is simply applied to BP network (BPN) [26],
System” has been proposed to improve this problem. The output because the BPN has the mutual interference problem of
error of the proposed neuron network system is converged at zero learning between outputs [18-22]. Then “Error Convergence
by infinitely increasing the number of neuron networks in the Parallel-type Neuron Network System (ECPNNS)” which
system. A predictor was constructed using the system and applied ECNNS was applied to a parallel-type NN which does not
to electrocardiogram prediction. The purpose of this paper is to
have the above-mentioned problem and which outputs
prove the validity of the predictor for electrocardiogram. The
accuracies are high to deal with plural outputs has been
neuron networks in the system were trained 30,000 cycles. As a
designed. Furthermore, “Error Convergence-type Recurrent
result, averages of root mean square errors of the first neuron
network was 2.60×10-2, and that of the second neuron network
Neuron Network System (ECRNNS)” which ECNNS was
was 1.17×10-5. Prediction without error was attained by this
applied and “Error Convergence Parallel-type Recurrent
predictor, so that its validity was confirmed. Neuron Network System (ECPRNNS)” which ECPNNS was
applied also have been designed. It is theoretically shown
Keywords: Volterra neuron network, Predictor, Error free, that output accuracy of NN system is elevated by them.
Electrocardiogram. In general, BPN which learns a time series signal using a
teacher signal and input signals is that two teacher signal
1. Introduction values or more occasionally correspond to same values of
input signals. In such a state, the BPN cannot be used for
A neuron network (NN) cannot be learned if there is not a ECNNS because it cannot be learned. There are means
correlation between input signals to each layer and teacher using input-delay NN and Volterra NN (VNN) to eliminate
signals of it. Therefore, uncorrelated components between this problem. These means are used for usual researches on
them cannot be learned when correlated and uncorrelated compression for nonlinear signal using predictive coding
components intermingled between them. NN cannot learn [10,27]. Learning for a nonlinear predictor using NNs is
completely correlated components between them when easier by strengthening of causality between signals from
learning capability of NN is low. Elevating learning past to present and a prediction signal. Therefore, Learning
capability of NN as a means to learn completely these for NN at the first step in ECNNS is comparatively easy.
correlated components is thought. However, such NN has However, learning for NN at high step in ECNNS is
not existed. Various researches have been actively done to difficult because the causality weakens by rising of steps in
elevate output accuracy of NN up to now. Improvements on ECNNS. Then, ECNNS is redesigned for improvement on
the I/O characteristics of a neuron [1-11,17], the structure of learning capability. The redesigning is that NN at each step
NN [13-23] and NN system [24] are effective as means of in ECNNS can be used as a predictor to strengthen causality
the output accuracy elevation. However, output errors of between an input signal and a teacher signal of it. As a
their NNs cannot converge at zero, even if trainings for their result, predictor using ECNNS [28] has been designed. This
NNs are iterated many times. Therefore, usual NNs are used is called “Error Convergence-type Predictor (ECP)”
with output error in tolerance. However, it is insufficient The purpose of this paper is to prove the validity of ECP
when applying NN to predictive coding [10-12] and the with rounding which is constructed of 2nd-order VNNs
flight control for an aircraft [18,19] and a spacecraft etc. (2VNN) of two steps for a normal sinus rhythm
which smaller output error is advisable. electrocardiogram (ECG). This ECP is called “Error
Then, NN system with a possibility which this problem Convergence-type 2nd-order Volterra Predictor (EC2VP)”.
can be solved even with NN which learning capability is As a result, prediction without error was attained by this
low if it is used by plural has been proposed. As a result, predictor, so that its validity was confirmed.
“Error Convergence Method for Neuron Network System
(ECMNNS)” which output error of single output NN system 2. Single Output Error Convergence-type
using NNs of multi-step converges and “Error Convergence
Neuron Network System
Neuron Network System (ECNNS)” which is designed using
it have been proposed by S. Kobayakawa [25]. The output 2.1 Principle
error is theoretically converged at zero by infinitely Here, it thinks about a single output NN which error of the
increasing the number of NNs in ECNNS. However, it is output signal to a teacher signal does not converge at zero
necessary to devise ECNNS when it is used as plural though it becomes smaller than one before training for the
NN under the result of executing until the training Teacher signal to ECNNS
converges to all input signals with correlation between the
teacher signal. Gain tuning to the teacher signal
The training is executed by this NN at the beginning.
Next, other training is executed by another NN using an error
signal obtained from an output signal and a teacher signal of Training
-
+ Extracting the error
the NN as a teacher signal. Thus it is thought that an error of
an output signal to a teacher signal cannot be converged at Restoration to an output signal Spreading the error
zero, even if it is kept training an error signal obtained from
training of previous NN one after another by another NN. Sum of restored output signals
That is, the error obtained from a difference between the sum
of the output signals of all NNs and the teacher signal of the
first NN does not converge at zero. This is shown in
expressions from (1) to (4). Iteration until the error converges at zero
Figure 1. Concept of processing for error convergence-type

yi = zi + ε i ( i = 1, 2,L , n ) (1)
neuron network system
ε i = yi +1 ( i = 1, 2,L , n − 1) (2) ε
ε n = lim =0 (8)
n ←∞ An
n
y1 = ∑ zi + ε n (3) n
i =1
y = lim ∑ zi (9)
n ←∞
i =1
lim ε n ≠ 0 (4)
n ←∞
2.2 Single Output Neuron Network System

where y is the teacher signal, z is the output signal, ε is the Figure 1 shows a concept of processing for ECNNS. Figure
output error, suffix of each sign is steps of NN. 2 shows that ECNNS is designed based on this concept.
Moreover, trainings for error signals used as a teacher Symbol NN in this figure also contains the state of a neuron.
signal for NN at the second step or more become difficult This ECNNS can be built in freely selected NN and the
because it is thought that error signals become small along learning rule of the processible type for the I/O relation
with the number of steps. Error signals which are amplified which internal each NN should be achieved. Here, the
are used for teacher signals to NN at the second step or more general I/O characteristics of ECNNS are discussed without
to improve this problem. As a result, these trainings become touching concerning the I/O characteristic of NN concretely
easy. An output error of NN after training is reduced when it applied to ECNNS.
is restored to signal level before the amplification, if the ECNNS equips with amplifiers to tune amplitudes of
amplification factor is larger than 1. Moreover, it input signals and a teacher signal to NN at each step, to
approaches zero when the amplification factor is very large. execute the appropriate training. Furthermore, ECNNS also
Here, an output error obtained from training under an equips with amplifiers for restoration which amplification
amplification factor of an error signal which is a teacher factor is a reciprocal of an amplification factor for
signal to NN at the n step as An is assumed to be ε. Output amplification of the teacher signal to each NN on its output
error εn to a teacher signal given to a whole of NN system is part, to restore its output signal level. The I/O
shown like expression (5) and (6). Moreover, it is shown characteristics of ECNNS are shown in expressions from
like expressions (7) and (8) when steps of NN are infinitely, (10) to (16). Moreover, the relation of their teacher signals
and the output error converges at zero. Therefore, an error is shown in expressions from (17) to (20). Here, conditions
obtained from a difference between a sum of output signals of the amplification factors used by expressions (16), (18)
after restoration of all NNs and the teacher signal given to and (19) are expressions (21) and (22).
the whole of NN system converges at zero from expressions
(3) and (8) when the steps of NN are infinitely, and they
become equal. This is shown in expression (9). Thus, this (
x = x1 , x2 ,L , xnin ) (10)
means to improve the above-mentioned problem is effective
and necessary to obtain highly accurate output. This means (
Aini = ai1 , a i 2 ,L , a inin ) ( i = 1, 2,L , n ) (11)
is called ECMNNS, and NN system which applies this is
called ECNNS.
ε
xAij = aij x j ( i = 1, 2,L , n; j = 1, 2,L , nin ) (12)
εn = (5)
An
(
x i = xAi1 , xAi 2 ,L , xAinin ) ( i = 1, 2,L , n ) (13)
εn < ε (6)
zAi = fi ( xi ) ( i = 1, 2,L , n ) (14)
lim An = ∞ (7)
n← ∞
y1 y1
A1 x ECNNS1
z1
z1
-
1/A1 + y2
- ECNNS2 z2
yA1 + +
x1 y2 + -
x Ain1 zA1 + y3
NN1 - + ECNNS3 z3
A2 +
ε1 z2 z3
1/A2 y3
x2 yA2
Ain2 zA2
NN2 -+ A3
ym
ε2 ECNNSm
1/A3 zm
yA3
x3 zA3
Ain3
-+
NN3
Figure 3. Error convergence parallel-type neuron network
ε3
system
n −1 - y
∑ zi +
i =1 yn
+ z
An 3. Plural Outputs Error Convergence-type
zn
1/An
yAn
Neuron Network system
xn zAn
Ainn
-+
NNn
3.1 Principle
εn
ECNNSs are applied to the parallel-type NN when applying
ECNNS to a plural outputs NN system, for ECNNS is a
Figure 2. Error convergence-type neuron network system single output. NN system of this type is called ECPNNS.
n Moreover, it can be said a general type of ECNNS. Figure 3
z = ∑ zi (15) shows ECPNNS. This is constructed of parallel units which
i =1
apply ECNNS of the same number as the outputs. Therefore,
it is thought that high output accuracy is obtained, because
zAi
zi = ( i = 1, 2,L , n ) (16) the mutual interference problem of learning between outputs
Ai of BPN is not caused either, for training for each output is
executed independently by ECNNS.
y1 = y (17) 3.2 Plural Outputs Neuron Network System
I/O characteristics of ECPNNS are shown in expressions
yA1 = A1 y1 (18) from (23) to (28). Moreover, relation of teacher signals is
shown in expressions from (29) to (32). Here, conditions of
 i −1  amplification factors used by expressions (28), (30) and (31)
yAi = Ai  y − ∑ z j  ( i = 2,3,L, n ) (19) are expressions (33) and (34).
 j =1 
ε i = yAi − zAi ( i = 1, 2,L , n ) (20)

(
Ainij = a ij1 , a ij 2 ,L , a ijnin )
( i = 1, 2,L , m; j = 1, 2,L , ni ) (23)
Ai ≠ 0 ( i = 1, 2,L , n ) (21)
 i = 1, 2,L , m; j = 1, 2,L , ni 
xAijk = a ijk xk   (24)
lim An = ∞ (22)  k = 1, 2, L , nin 
n ←∞
where x is the input signal vector, x is the input signal, xAi is (

x ij = xAij1 , xAij 2 , L , xAijnin )
the input signal after amplification at the ith step, xi is the ( i = 1, 2,L , m; j = 1, 2,L , ni ) (25)
input signal vector after amplification at the ith step, Aini is
the input signal amplification factor vector at the ith step, a i
is the input signal amplification factor at the ith step, zAij = fij ( xij ) ( i = 1, 2,L, m; j = 1, 2,L, ni ) (26)
suffixes of x, xAi and a i are input signal numbers, zAi is the
output signal of NN at the ith step, fi is a nin variables ni
function which shows the I/O relation of NN at the ith step, y zi = ∑ zij ( i = 1, 2,L , m) (27)
j =1
is the teacher signal to ECNNS, z is the output signal of
ECNNS, yAi is the teacher signal to NN at the ith step after
amplification, zi is the output signal to NN at the ith step zAij
after restoration, Ai is the teacher signal amplification factor
zij = ( i = 1, 2,L, m; j = 1, 2,L, ni ) (28)
Aij
for NN at the ith step, εi is the output error of NN at the ith
step.
yi1 = yi ( i = 1, 2,L , m) (29)
yAi1 = Ai1 yi1 ( i = 1, 2,L , m) (30)
shown in expression (39). An output signal of ECP is shown

 j −1
 in expression (40). A teacher signal to ECP is shown in
yAij = Aij  yi − ∑ zik  ( i = 1, 2, L , m; j = 2,3,L , ni ) expression (41).
 k =1 
( )
(31)
x1( ) = x(A11) , x(A12) ,L , x(A1)nin
τ τ τ τ
ε ij = yAij − zAij ( i = 1, 2,L , m; j = 1, 2,L , ni ) (32) (35)
Aij ≠ 0 ( i = 1, 2,L, m; j = 1, 2,L, ni ) (33)

τ
(
x (i ) = Aeini xij( ) , x(Ai1) , x(Ai )2 ,L , x(Ain) in
τ τ τ τ
)
( i = 2,3,L , n; j = 1, 2, L , nin ) (36)
lim Ain = ∞
n ←∞
( i = 1, 2,L , m) (34)
xij( ) = 0
τ
where the first figure of suffix of each sign is the parallel

unit number, the second figure of it is the step number in
(τ ≤ 0; i = 2, 3,L , n; j = 1, 2,L , nin ) (37)
ECNNS, the third figure of it is the input signal number.
However, xk is the kth input signal to ECNNS. i −1
τ > 0 
xij(τ ) = x(jτ ) − ∑ zk(τ −1)  
 i = 2,3, L , n  (38)
4. Applications k =1  j = 1, 2, L , n 
 in 
4.1 Simulator
Here are two application network systems using ECNNS z(Ai ) = fi x (i
τ
( )) τ
( i = 1, 2,L , n ) (39)
and ECPNNS. For example, control signals and state signals
are necessary for input when achieving a simulator for
nonlinear plant. Furthermore, the state signals output from z( ) = xˆ (j
τ τ +1)
( j = 1, 2,L , nin ) (40)
the simulator every moment decided depending on the input
signals are recurrently needed for the state input signals. y(τ ) = x(jτ +1) ( j = 1, 2,L , nin ) (41)
When achieving the simulator with high accuracy, it is
advisable that errors are not included in output of the
simulator. Therefore, it is thought that the simulator which where xij is the input error signal at the ith step to input
works by the specified significant figure obtains a highly signal xj to ECP, Aeini is an amplification factor of the input
accurate output by the plant model using ECNNS and a error signal at the ith step, Ai is an amplification factor of the
rounding to eliminate the output error. Concretely, the teacher signal at the ith step, fi is the nin variables function
network system which the output signal of ECNNS recurs to when i is 1 or the nin + 1 variables function when i is 2 or
the input through a rounding shown in Figure 4 is thought more to show I/O relation of NN at the ith step, x̂ is the
about the simulator in case of a state signal. This is called prediction.
ECRNNS. Furthermore, the network system which output
signal vector of ECPNNS recurs to the input through 5. Computer simulations
roundings as shown in Figure 5 is thought about the
simulator in case of plural state signals. This is called 5.1 2nd-order Volterra Neuron Network
ECPRNNS. In these figures, y is the teacher signal vector,
Figure 7 shows 2nd-order Volterra neuron (2VN) in discrete-
and z is the output signal vector.
time. I/O characteristics of 2VN are shown in expressions
4.2 Nonlinear Predictor from (42) to (44).
A nonlinear predictor used for predictive coding and its
n
principle must be improved to obtain high accuracy. Here is u (τ ) = ∑ wi(τ ) xi(τ ) (42)
a means for improvement on learning capability of NN at i =1
each step in ECNNS for a nonlinear predictor used for
predictive coding. This is called ECP. Learning for a Q
nonlinear predictor using NNs is easier by strengthening of s (τ ) = ∑ σ 1 ( p )(τ ) u (τ − p )

p =0
causality between signals from past to present and a Q Q
(43)
prediction signal. Therefore, learning for NN at the first step + ∑ ∑ σ 2 ( p, q) u (τ ) (τ − p )
u (τ − q )
−h (τ )
in ECNNS is comparatively easy. However, learning for NN p=0 q= p
at the high step is difficult because it is guessed that the

causality weakens by rising of the steps. Then, NN at each z(τ ) = f ( s (τ ) ) = Atan −1 ( s (τ ) ) (44)
step in ECNNS is used as a predictor to strengthen causality th
where u is the input weighted sum, xi is the i input signal, wi
between input signals and a teacher signal of it.
is the ith connection weight, s is the input sum, D is the delay,
Moreover, learning capability of the NN can be elevated
Q is the prediction order, σ 1 is the prediction coefficient of the
by increasing the number of input signals [29]. Figure 6 is
1 st -order term corresponding to the signal obtained from
redesigning of ECNNS in Figure 2 to realize it.
between from an input of the 1st delay to an output of the Qth
Furthermore, expression (13) is changed to expressions from
delay, σ 2 is the prediction coefficient of the 2nd -order term
(35) to (38). Expression (37) shows initial conditions for
corresponding to the product of all combinations of two sig-
NNs from the second step. I/O relations of NNs in ECP are
x y x y
ECNNS ECPNNS
z z
Rounding Roundings
Figure 5. Error convergence parallel-type recurrent neuron

Figure 4. Error convergence-type recurrent neuron network
network system
system
A1 y1(τ )
x(jτ ) z1(τ −1) z1(τ ) -

x(
τ)
+ - D +
1/A1
y(A1)
τ
(τ ) -+
x1 z(Aτ1) +
Ain1
-+
NN1
y2(τ )
ε1(
τ)
+ -+
z2(
τ −1)
z2(
τ)
+ - D A2 +
x2( j) Ae
τ
in2
1/A2 y(Aτ2)
x(2 )
τ
z(Aτ2) y3(
τ)
-+
NN2
ε 2(
τ)
Ain2
A3
z3(τ −1) z3(τ )
+- D
y(Aτ3)
(τ )
x3 j Aein3 1/A3
x(3
τ)
z(Aτ3)
-+
NN3
(τ )
Ain3 ε3 n −1
(τ ) -+ y(τ )
∑ zi
i =1
z( )
τ
+
(τ )
yn
zn(
τ)
An
(τ )
xnj Aeinn 1/An y(An
τ)
(τ )
z(An)
x τ
n
-+
NNn
(τ )
Ainn εn
Figure 6. Predictor using error convergence-type neuron network system
nals included in combinations of the same signal obtained Figure 8 shows a three-layer 2VNN of one input one
from between from an input of the 1st delay to an output of output which is constructed of 2VNs. This 2VNN is used for
the Qth delay, h is the threshold, z is the output signal, f is ECP.
the output function, A is the output coefficient. wi , h , σ 1
a n d σ 2 are changed by training. 5.2 Method
y1(τ )
σ 1 (0) (
τ)
A1
x1(τ ) w1(τ ) σ 1 (1) ( )
z1(
τ −1)
z1(
τ)
τ
y(
τ)
x( +- -+
τ)
D
x2(
τ)
w2(
τ)
σ 1 (2)(τ )
(τ )
y A1
M 1/A1
s(
M τ)
u(
τ)
xn( )
τ (τ ) u (τ −1) u(τ −2) u(τ− Q) σ 1 (Q)( )
τ (τ ) (τ )
z(Aτ1)
wn
z(
z xA11 τ)
+ D D D + + （
f ･） a 11 + Rounding
-+
2VNN1
σ 2 (0, 0) (τ ) −
+ ε1(τ ) y2(τ )
σ 2 (0,1) (τ ) h (τ ) (τ )
σ 2 (0, 2) (
τ) z2
A2
M (τ ) (τ )
σ 2 (0, Q )(τ ) x21 Aein2 xA2 1/A2 y(Aτ2)
(τ ) (τ )
σ 2 (1,1) (τ ) x2 z
+
A2
-+
2VNN2
σ 2 (1, 2 ) ( )
τ
ε 2(
τ)
a 21
x(A21)
τ
M
σ 2 (1, Q )(τ )
σ 2 (2, 2) (τ )
+ Figure 9. Error convergence-type 2nd-order Volterra
M
σ 2 (2, Q ) (τ ) Predictor
M
σ 2 (Q, Q) (τ )
3.0
2.5
Voltage [mV]
Figure 7. 2nd-order Volterra neuron 2.0
1.5
1.0
0.5
Input Output
(τ )
0.00
z(Ai)
τ
xi
-0.5
2VN -1.0
0 1 2 3 4 5
Time [s]
2VN
Figure 10. Teacher signal used for training for error
Figure 8. 2nd -order Volterra neuron network
convergence-type 2nd-order Volterra predictor
In computer simulations, EC2VP constructed of 2VNNs of decided by pseudo-random numbers at the training
two steps shown in Figure 9 is trained using combinations of process a time. Gradient descent method is used for
an input signal x(τ) and a teacher signal y(τ) = x(τ+1) in the learning rule for the 2VNN.
time series pattern of one dimension in space direction. The The number of middle layer elements and filter length of
teacher signal is shown in Figure 10. This is a normal sinus the 2VNN has been decided by studying experiences of
rhythm ECG signal of MIT-BIH No.16786. This ECG ECG prediction using 2VNN.
signal is that the sampling frequency is 128 Hz, the 3) The trainings for searches are executed as a parameter to
significant figure is four-digit, the quantization step size is set a condition for the computer simulations which is the
0.005 mV, and the number of data for the input signal learning reinforcement coefficient.
without an initial input and the teacher signal is 640, 4) Averages of root mean square errors (RMSEs) obtained
respectively. Here, training for 2VNN at each step in from the searches of three times are compared.
EC2VP is completed sequentially from the first step.
At the beginning, 2VNN at the first step (2VNN1) is 2VNN achieved the minimum average of RMSEs by the
trained using combinations of an input signal x(τ) and a searches is NN at the first step in EC2VP. A part of training
teacher signal y1 (τ)=x(τ+1) in the time series pattern of one signals to 2VNN at the second step (2VNN2) in EC2VP is
dimension in space direction. Here, computer simulations an error signal obtained from a difference between an output
for the training are executed as A1 =a 11=1/3.275 according to signal of 2VNN1 and a teacher signal to EC2VP.
the following procedure from 1) to 4). Next, 2VNN2 is trained using combinations of input
signals x21(τ) and x(τ) in the time series pattern of two
1) A pair of the input signal or signals and the teacher signal dimensions in space direction and a teacher signal
is given once after 1,580 initial data are inputted into the y2(τ)=x21(τ+1) in the time series pattern of one dimension in
2VNN at the training. This process is defined as one space direction. A signal to which gain tuning which adjusts
training cycle. the maximum absolute value of error signal to 1 is
2) Table 1 shows conditions for computer simulations to performed are used for xA2(τ) and yA2(τ). Here, computer
valuate prediction accuracy for the 2VNN. Initial values simulations for the training are executed as Aein2=A2 and
of prediction coefficients of the 2VNN are decided by a 21=1/3.275 according to the above-mentioned procedure
exponential smoothing, and the other initial values are from 1) to 4).
Table 1: Conditions for computer simulations to train 2nd-order Volterra neuron network at each step
Steps The 1st step The 2nd step

Learning roule Learning roule for Volterra neuron network
Connection weights -0.3～0.3
Initial Thresholds -0.3～0.3
conditions σ1 0.7×0.3p
Prediction coefficents
σ2 0.7×0.3p×0.7×0.3q
Number of middle layer elements 4 10
Filter length 69 64
Learning Gradient-based Range 10-5～1
reinforcement method Interval 10 times
coefficients Momentum 0
Output coefficents 1
Training cycles 30,000
Processing times 3
×10-1
1.0 10
0.8
Minimum (2.60×10-2)
0.6 1
Voltage [mV] ]
0.4 N2VNN1
0.2 10-1
0
RMSE
-0.2 10-2
-0.4
N2VNN2
10-3
-0.6
-0.8 10-4
-1.0 Minimum (1.17×10-5)
-1.2 10-5
0 1 2 3 4 5
Time [s] 10-6
10-5 10-4 10-3 10-2 10-1 1
nd Learning reinforcement coefficent
Figure 11. Teacher signal for 2 -order Volterra neuron
network at the second step Figure 12. Averages of root mean square errors and their
standard deviations to learning reinforcement coefficient in
gradient descent method term
1 2VNN1
系列1 2VNN2
系列2
Evaluation function value
2VNN1 3.0
10-3 2.5
V o ltag e [ m V ]
2.0
1.5
10-6
1.0
2VNN2 0.5
10-9 0.00
0 10,000 20,000 30,000
Training cycles -0.5
-1.0
Figure 13. Relation between training cycles and average 0 1 2 3 4 5
evaluation function values at the minimum average of root Time [s]
mean square errors concerning 2nd-order Volterra neuron
network at each step Figure 14. Output signal of 2nd-order Volterra neuron
network at each step
Then, 2VNN achieved the minimum average of RMSEs 6.2 Improvement on Learning Capability
by the searches is NN at the second step in EC2VP. An Capability of discrete-time neuron network (DTNN) is
output of the 2VNN2 is restored at a level of the teacher improved if sampling frequency of signal processed with
signal to EC2V. The restored output signal is added to the DTNN is upped. However, there is a limit in such means to
output of 2VNN1 at the same time as it. Finally, the fourth improve the capability of DTNN because the limit of the
decimal place of the output after the addition is rounded off. sampling frequency is caused by restriction to operating
The obtained output is an output of EC2VP as a result. Its frequency fordigital circuit, memory capacity and data
prediction accuracy is evaluated. processing time of it.
Moreover, the lowest sampling frequency to process
efficiently data is decided by the sampling theorem. However,
5.3 Results training for DTNN must be executed in a state which the data
Results of computer simulations are shown in figures from are increased by upping the sampling frequency if excellent
11 to 14. A result of making training signal for 2VNN2 is training for DTNN is not obtained by the data. As a result,
shown in Figure 11. Averages of RMSEs and their standard data processing for DTNN becomes no efficiency. Here, it is
deviations obtained by the computer simulations for 2VNN thought that learning capability of DTNN can be improved
at each step are shown in Figure 12. Fig. 13 shows a relation without upping the sampling frequency by ECNNS.
between training cycles and average of evaluation function
values of 2VNN at each step which is recorded before 6.3 Instability
beginning to train and at the first time and every 100 times Instability of ECP is how to process components of random
of the training cycle when the minimum average of RMSEs signals which are included in inputs signal and teacher
is obtained by searches of three times. This figure shows signals of ECP. There is a means which the random signals
that prediction errors when training for 2VNN1 is saturated are processed as noises. This means eliminates these noises
can be decreased more by using 2VNN2. Moreover, gradient by filtering an input signal and a teacher signal to ECP as
of average evaluation function value of 2VNN2 at 30,000 the preprocessing. There are Fourier transform, Bayesian
cycles is surmisable like being able to train more. Figure 14 model and trend model etc. [30] as the means which can be
shows output signal of 2VNN at each step. An output signal used. However, there is also necessary information like
of EC2VP which is obtained as a result of adding output signals of nature, vital and economy etc. in the random
signals of 2VNN1 and 2VNN2, and rounding it. This signal signals. This means must be performed with attention so as
is error free at all, and is equal to the teacher signal. not to miss the information.
Thus, complete learning accomplishment capability of 6.4 Validity
EC2VP could be demonstrated. An excellent learning
capability and validity of EC2VP to normal sinus rhythm It thinks about training for 2VNN at the second step in
ECG can be confirmed. EC2VP. ECP is an improvement on its learning capability
by using NN which makes the causal relation between input
signal and teacher signal available as an internal predictor
6. Discussion in ECP to enable trainings for NNs from the second step
where the trainings becomes difficult. Furthermore, two
6.1 Accuracy Up for Output and Speed Up for input signals are used for strengthening unique concerning
Training the output signal to the input signal. Actual training also is
It is thought that a highly accurate output is obtained by shown excellent results, and the effect concerning the above-
ECNNS at a few training cycles because its error is mentioned to make easily to train is seen. It is thought that
converged by plural NNs. Here, error signal of NN at this is because a relation between the input signals and the
previous step which NN at final step uses to train is a very teacher signal is a correlation.
small signal if steps of NN are set infinitely. That is, sum of These computer simulations are insufficient as a
output signals of all NNs becomes a very highly accurate demonstration for a highly accurate predictor because
output signal because output error of NN at final step generalization capability to unlearning signals is not
becomes equal in nil compared with a teacher signal to a confirmed, though it demonstrated the complete learning
whole of ECNNS. The training cycle as the whole of ECNNS accomplishment capability of EC2VP. Therefore,
can be theoretically one time when thinking only one datum demonstrations for the generalization capability using ECGs
is trained under this condition. For example, when the of arrhythmia etc. are necessary.
probabilistic descent method is used for learning rule for
NNs in ECNNS, it is expected that a steady output signal is 7. Conclusions
obtained at any training cycles because the highly accurate
training can be executed every one datum as shown in the In this study, it was shown to obtain a highly accurate
above-mentioned. Moreover, a component ratio of output output by improving the learning capability using single
signal of NN at each step to the output signal of ECNNS to output NNs which an error of an output signal to a teacher
grow large at more former step as the training cycle increases signal does not converge at zero though it becomes smaller
is thought. It is thought that learning capability of ECNNS is than one before training under a result of executing until the
remarkable as steps of NN increases and the output accuracy training converges to all input signals with correlation
and the training speed elevate. between the teacher signal and which the inputs are
common in style connected with multi-steps. This is a
method which the error to the teacher signal of a whole of
NN system can be converged at a small value by amplifying [6] S. Uota and H. Yokoi, “A Realization of Motion
an error signal obtained from NN at each step and training Diversity of the Robotic Hand by the Hierarchical
it as a teacher signal of NN at the next step one after Motion Schema,” Technical Report of IEICE,
another. Moreover, an error to a teacher signal of NN at the NC2003-75, pp. 25-28, 2003.
first step converges at zero by infinitely setting steps of NN. [7] J. Miyoshi and H. Yokoi, “An Improvement of a Neural
This method is called ECMNNS. It explained ECNNS Network for Learning a Slip Angle of a Four-Wheel
applied this method and means to use by ECPNNS which Steering Car,” Technical Report of IEICE, NC2004
ECNNS is applied to the parallel-type NN for NN of plural -107, pp. 87-90, 2004.
outputs. Furthermore, it also explained ECRNNS and [8] S. Shigemura, T. Nishimura and H. Yokoi, “A
ECPRNNS which can be expected to use as the simulator for Method of Removing Blink Artifacts from EEG
a nonlinear plant as applications using ECNNS and Signals Using Neural Networks with Volterra Filters,”
ECPNNS. Technical Report of IEICE, MBE2004-87, pp. 57-60,
Moreover, ECP improved learning difficulty when there is 2005.
little causality between input signals and a teacher signal to [9] S. Suematsu and H. Yokoi, “A Motion Generating
NN at each step in ECNNS by structuring NN at each step in System for Multi-Fingered Myoelectric Hand,”
ECNNS as a predictor and strengthening causality between the International Congress Series 1291, pp. 257-260, 2006.
input signals and the teacher signal was designed. Using [10] S. Kobayakawa, T. Fujii and H. Yokoi, “Evaluation of
2 VNN for NN at each step i n ECP was proposed. Nonlinear Prediction Capabilities of Neuron Networks
Finally, computer simulations to train EC2VP constructed for Electrocardiogram,” Proc. of the 20th Annual
of 2VNNs of two steps were executed using a normal sinus Meeting of Biomedical Fuzzy Systems Association,
rhythm ECG signal, and prediction accuracy of EC2VP was pp. 9-12, 2007.
evaluated. As a result, learning capability obtained an output [11] S. Kobayakawa, T. Fujii and H. Yokoi, “Evaluation of
without error, that is, ECP having complete learning Prediction Capabilities of Neuron Networks Used for
accomplishment capability was demonstrated. To be a Electrocardiogram,” Proc. of the 5th International
validity which can use EC2VP as a highly accurate predictor Symp. on Management Engineering, Kitakyushu,
by this demonstration was confirmed. It can be said that Japan, pp. 156-161, 2008.
ECNNS will have an enough capability to be the leading [12] S. Kobayakawa, T. Fujii and H. Yokoi, “Nonlinear
system of means to construct NN in the future because of Prediction for ECG by 2nd-order Volterra Neuron
Network,” Journal of Biomedical Fuzzy Systems
excelling theoretically concerning accuracy up for output,
Association, Vol.11, No.2, pp. 101-111, 2009.
speed up for training and improvement on learning capability
[13] A. G. Ivakhnenko, “The Group Method of Data
of DTNN. On the other hand, an enough demonstration as a
Handling-A Rival of the Method of Stochastic
highly accurate predictor has been not able to perform Approximation,” Soviet Automatic Control, Vol.13 c/c
because generalization capability to unlearning signals is not of Avtomatika, 1, 3, pp. 43-55. 1968.
confirmed. The future work is demonstrating the [14] A. D. Back and A. C. Tsoi, “FIR and IIR Synapses, A
generalization capability of EC2VP using ECG of arrhythmia New Neural Network Architecture for Time Series
etc. Modeling,” Neural Computations, Vol.3, pp. 375-385,
1991.
[15] M. Hoshino, T. Kitamura, T. Masuda, M. Suzuki and J.
Acknowledgments Chao, “On Multilayer RBF Networks and a Novel
We wish to express our gratitude to members in our Pyramid Network,” Proc. of the Society Conf. of
laboratory who cooperate always in the academic activity. IEICE, Nagoya, Japan, p. 28, 2000.
[16] N. Kinoshita and K. Nakamura, “Two-D Spreading
References Associative Neural Network Recognizes the Shape and
Position of An Object Presented in the Two-D Space,”
[1] C. L. Giles and T. Maxwell, “Learning, Invariance and Technical Report of IEICE, NC97-166, Vol.97,
Generalization in High Order Neural Networks,” No.623-624, pp. 209-216, 1998.
Applied Optics, Vol.26, No.23, pp. 4972-4978, 1987. [17] S. Kobayakawa and H. Yokoi, “The Volterra Filter
[2] K. J. Lang and G. E. Hinton, “A Time-Delay Neural Built-in Neural Network for the Aircraft Pitch Attitude
Network Architecture for Speech Recognition,” Control,” The Lecture Proc. of the 2005 Fiscal Year
Carnegie Mellon University Computer Science Electricity Relation Institute Kyushu Branch
Technical Report, CMU-CS-88-152, pp. 1-37, 1988. Association Convention, Fukuoka, Japan, p. 429, 2005.
[3] T. Possio and F. Girosi, “Networks for Approximation [18] S. Kobayakawa and H. Yokoi, “Application to
and Learning,” Proc. of the IEEE, Vol.78, No.9, pp. Prediction Problem of Parallelized Neuron Networks in
1481-1497, 1990. the Aircraft,” Technical Report of IEICE, SANE2006-
[4] S. Iwamoto, T. Yosh ida and H. Yokoi, “Basic 119 -133, Vol.106, No.471, pp. 43- 45, 2007.
Investigation Associated with Neural Control of Biped [19] S. Kobayakawa and H. Yokoi, “Evaluation for
Walking Robot,” Technical Report of IEICE, MBE93 Prediction Capability of Parallelized Neuron
-106, pp. 23-30, 1994. Networks,” Proc. of the 8th SOFT Kyushu Chapter
[5] Y. Fujisue, E. Inohira and H. Yokoi, “Robotic Control Annual Conf., Kitakyushu, Japan, pp. 3-6, 2006.
by Volterra Network,” Technical Report of IEICE, [20] S. Kobayakawa and H. Yokoi, “Evaluation of the
NC2003-78, pp. 39-43, 2003. Learning Capability of a Parallel-type Neuron
Network,” Proc. of the First International Symp. on
Information and Computer Elements 2007, Kitakyushu,
Japan, pp. 43-47, 2007.
[21] S. Kobayakawa and H. Yokoi, “Experimental Study for
Dominance to Accuracy of Prediction Output of Authors Profile
Parallel-type Neuron Network,” Technical Report of
IEICE, NC2008-1-10, Vol.108, No.54, pp. 29-34, Shunsuke KOBAYAKAWA received the
2008. B.Eng. degree in 1986, accomplished
[22] S. Kobayakawa and H. Yokoi, “Evaluation for credits for the master's course in 1989 in
Prediction Accuracies of Parallel-type Neuron electrical engineering from Okayama
Network,” International MultiConf. of Engineers and University, Okayama, Japan, completed
Computer Scientists 2009 Proc., Hong Kong, China, auditor in faculty of engineering in 1995 and
Vol.I, pp. 156-161, 2009. received the M.Sc. degree in biological
functions and engineering in 2003 from
[23] H. Yokoi and T. Kimoto, “Multilayered Neural
Kyusyu Institute of Technology (KIT),
Networks with Intermediate Elements,” Journal of
Kitakyushu, Japan, respectively. He has
Biomedical Fuzzy Systems Association, Vol.1, No.1,
been working as a part-time lecturer at Research Course of
pp. 87-97, 1999. Telecommunication System, Subaru Professional College in 2006, a
[24] H. Sori and T. Yasuno, “Several-Hours-Ahead Wind research assistant since 2006 and a teaching assistant in 2008 at
Speed Prediction System Using Hierarchical Neural Graduate School, KIT. He also obtained Associate Professional
Network,” Journal of Signal Processing, Vol.12, No.6, Engineer in electrical and electronics engineering, First-Class
pp. 507-514, 2008. Technical Radio Operator for On-The-Ground Services, Class I
[25] S. Kobayakawa and H. Yokoi, “Proposal of Error Information Technology Engineer and Aerospace Products
Convergence-type Neuron Network System,” Inspector etc. as national qualifications in Japan. He is a student of
Presented Proc. to 2008 International Symp. on doctoral program, Department of Biological Functions and
Intelligent Informatics, Kumamoto, Japan, pp. 1-10, Engineering, Graduate School of Life Science and Systems
2008. Engineering, KIT and a director of KOBAYAKAWA Design
[26] D.E. Rumelhart, G.E. Hinton and R.J. Williams, Office at present. His present research interests include control for
“Learning Representations by Back-propagating aerospace vehicles using neuron networks. He is a member of
Errors,” Nature, Vol.323, No.6088, pp.533-536, 1986. Biomedical Fuzzy Systems Association, The Japan Society for
[27] A. Chatterjee, A. Nait-Ali and P. Siarry, “An Input- Aeronautical and Space Sciences, Information Processing Society
delay Neural-Network-Based Approach for Piecewise of Japan, The Institute of Electronics, Information and
Communication Engineers, and The Institute of Electrical and
ECG Signal Compression,” IEEE Transactions on
Electronics Engineers.
Biomedical Engineering, Vol.52, No.5, pp. 945-947,
2005.
[28] S. Kobayakawa and H. Yokoi, “Proposal of Hirokazu YOKOI received the B.Eng.
Predictive Coding Using Error Convergence-type degree in 1972, the M.Eng. degree in 1974 in
Neuron Network System,” The Proc. of the ISCA 22nd electrical engineering from Nagoya
International Conf. on Computers and Their University, Nagoya, Japan, the D.M.Sc.
Applications in Industry and Engineering, San degree in medicine in 1985 and the D.Eng.
Francisco, USA, pp. 169-174, 2009. degree in electronics in 1989 from The
[29] S. Kobayakawa and H. Yokoi, “Evaluation of University of Tokyo, Tokyo, Japan,
Learning Capabilities of BP Networks to Number of respectively. He works as a professor of
Input Signals,” Technical Report of IEICE, Department of Biological Functions and Engineering, Graduate
SANE2007-102-124, Vol.107, No.442, pp. 83-86, School of Life Science and Systems Engineering, Kyusyu Institute
of Technology, Kitakyushu, Japan at present. His present research
2008.
interests include development of ultra large-scale neurochips and
[30] M. Onodera, Y. Isu, U. Nagashima, H. Yoshida, H. their applications to biped walking robots, intelligent assistive
Hosoya and Y. Nagakawa, “Noise Filtering Using devices as well as automatic driving, and modeling of human
FFT, Bayesian Model and Trend Model for Time cognitive processes with application to human-centered information
Series Data,” The Journal of Chemical Software, equipments. He is a member of Biomedical Fuzzy Systems
Vol.5, No.3, pp. 113-127, 1999. Association, Japan Society for Fuzzy Theory and Intelligent
Informatics, The Institute of Electronics, Information and
Communication Engineers, and Human Interface Society.
Segmentation Method Based On Circular Edge

Magnitude for Classification of Textures
1
Dr. A. Nagaraja Rao , 2K. Lavanya, 3G. Aruna Kumari, 4M.N. Hima Bindu
1
Professor, Dept. of IT, Lakireddy BaliReddy College of Engg.
Mylavaram, Krishna Dt., A.P., India.
nagarajarao.a@lbrce.ac.in
2
Asst. Prof., Dept. of IT, Lakireddy BaliReddy College of Engg.
lavanya.kk2005@gmail.com
3
Associate Prof., Dept. of CSE, Vidya Jyothi Institute of Tech.,
Moinabad, RR Dist, A.P., India.
arunagullipalli@yahoo.com
4
Asst. Prof. Dept. of IT, Lakireddy BaliReddy College of Engg.
bindu.nov5@gmail.com
Abstract: Texture segmentation is one of the most important considered more useful. There are three main classes of
techniques for image analysis, understanding and image texture segmentation that belong to the above two
interpretation. The task of texture segmentation is to partition
techniques. They are cluster-based, edge-based and region-
the image into a number of regions such that each region has
the same textural properties. The present paper proposes a new based methods [7].
segmentation method which is an alternative to non-maximal
suppression based on edge magnitude. The relative edge The edge-based segmentation exploits spatial information
magnitudes are calculated by considering the spatial
by detecting the edges in an image, which correspond to
information of 5X5 mask with 3X3 circular neighborhood
masks. The central pixel under consideration in 5X5 mask is discontinuities in the homogeneity criterion for segments.
whether an edge pixel or not, can be identified by manipulating Edges in a given depth map are defined by the points where
all the edge magnitudes from circular neighboring masks. The changes in the local surface properties exceed a given
proposed method is applied on various Brodatz textures and the
experimental results shows effective segmentation which is
threshold. The local surface properties mostly used are
useful in classification of textures. surface normals, gradients, principal curvatures, or higher
order derivatives. Edge detection techniques used on texture
Keywords: Texture, Segmentation, Edge, Magnitude, Non-
Maximal Suppression, Neighborhood, Circular Mask. image could result in noisy and discontinuous edges and
therefore segmentation process becomes more complicated
1. Introduction [13]. As edge detection methods look for abrupt changes,
Texture is defined as a pattern that is repeated and is they are very sensitive to the noise in the range data.
represented on the surface of an object. Segmentation is a Moreover, as only the measurements near the edges are used
fundamental low-level operation on images and to separate to make major decisions, the available information is not
textures into a single texture type, first we need to preserve optimally utilized. In many situations the edges do not form
spatial information for each texture. A homogeneous region closed boundary curves and it can be difficult to make
refers to a group of connected pixels in the image that share correct grouping decisions there by resulting in over or
a common feature. This feature could be brightness, color, under segmentation. Some of the typical variations on the
texture, motion etc. For instance, the manual grey level edge-based segmentation techniques are reported by many
thresholding which does not provide the spatial information researchers [8, 10, 11].
for each texture [12] could generate inappropriate
segmentation result. Depending on the number of images is The organization of this paper as follows: Section 2 gives
known in advance, the techniques for image segmentation the related work and the proposed method and algorithm
can be classified as supervised [1-3] or unsupervised [4-6]. described in section 3. Experimental results of texture
Since in most of real applications the number of images is segmentation using the circular edge magnitude are shown
generally unknown, the unsupervised approach is in the section 4 and section 5 gives the conclusions.
2. Related Work
In the edge image, no zero value edge pixels are present, but
small edge values correspond to non-significant grey level
changes resulting from quantization noise, small lighting
irregularities. Simple thresholding of an edge image can be (a) (b) (c) (d)
applied to remove these small values based on an image of
edge magnitudes [14]. A problem with simple detector is the Figure 1. (a),(b),(c) and (d) represents the rotating 3x3
thickening that is evident where there should only be a masks that contain central pixel of 5x5 window in
simple boundary. This can be partially rectified if edges Horizontal(H), Vertical(V),Right Diagonal(RD) and Left
carry directional information by performing some form of Diagonal (LD) Directions.
non-maximal suppression to suppress multiple responses in
the neighborhood of single boundaries [15]. In the proposed
method, the maximum edge magnitude in 8-connectivity is Algorithm 1.
calculated leading to the edge direction. The pixel under
consideration in 3X3 window is displayed only if its grey BEGIN.
level value is greater than the maximum of the two adjacent
pixels of the edge direction. Step 1: Read the 5x5 window from original image.
3. Proposed Segmentation Method Step 2: Select circular 3X3 masks that contain the central
pixel of a 5X5 window of the image as shown in Fig.1.
Edge based segmentation is one of the common approaches Step 3: The sum of the grey levels of the circular masks are
to segment the image and it remains very important over calculated.
period of time. All edge based segmentation methods depend
on edges identified in an image by any edge detection Step 4: The difference between the two Right Diagonal, Left
techniques. Edge based segmentation techniques differ in Diagonal, and Vertical and Horizontal direction’s circular
rotating masks are computed
strategies and the necessary priori information is
incorporated into those methods. Since the edge detection is Step 5: To find edge, Maximum of the above differences is
very difficult in texture images when compared to image obtained and Central Pixel of 5X5 window is replaced by 1
segmentation, the present study proposes the new technique iff any one of the adjacent differentiating masks
which is based on relative edge magnitude in 5X5 maximum/mean/ minimum grey level is greater than or
neighborhood using circular rotating masks and the equal to Central Pixel of the current 5x5 mask, otherwise
the Central Pixel is replaced by 0.
procedure and algorithm is described as follows.
Step 6: Repeat step 1 to step 5 by 5X5 running window
A practical method that has chosen in the present approach throughout the image.
is to determine whether a central pixel is an edge pixel or
not depending on its neighboring rotating masks inside the Step 7: Display the resultant segmented image.
running window, in horizontal, vertical, right diagonal and
left diagonal directions. Note that, since the proposed END.
method checks the relative Difference of Edge Magnitude
(DEM) values of rotating masks using 5X5 window as 4. Experimental Results
shown in Fig.1 instead of 3X3 window as applied in other
methods. This will eliminate some false edge pixels on The Brodat’z Texture Images, Plastic Bubbles, Straw, Pig
seemingly smooth regions. If any one of the maximum or Skin and Raffia, are taken to apply the proposed algorithm
average or minimum of the adjacent masks of maximum and these Textures are shown in Fig.2. The segmented
DEM value of rotating masks is less than the central pixel texture images are shown, when all suppression parameters
grey level value, the pixel is identified as being not located like maximum, mean and minimum, in Fig.4, 5 and 6
for segmentation. respectively. In Fig. 3, the segmented texture images of non-
maximal suppression on 3X3 neighborhood are shown. For
The stated method is explained in the following algorithm. any texture classification, feature extraction is most
important step. The proposed algorithm segments the
textures by extracting the linear, horizontal and vertical
edge features.
segmentation and the results are shown in Fig.5 and Fig.6

The edge boundaries are extracted by non-maximal respectively. Only the strong edge features are extracted
suppression in 5X5 neighborhood using the proposed when non-average method is applied and over segmentation,
algorithm and the segmented results are shown in Fig.4. i.e. decomposing into more inner regions, is achieved when
Similarly the parameters like non-average and non- non-minimal suppression is applied.
minimum are suppressed in 5X5 neighborhood for
(a) (b)
(c) (d)
Figure 2. Original Brodatz Textures (a) Plastic Bubbles (b) Straw (c) Pig Skin (d) Raffia
(a) (b)
(c) (d)
Figure 3. Segmentation results when non-maximum is suppressed in 8-connectivity. (a) Plastic Bubbles (b) Straw (c) Pig
Skin (d) Raffia.
(a) (b)
( c) (d)
Figure 4. Segmented Images when Non-Max is considered using proposed method. (a) Plastic Bubbles (b) Straw (c) Pig
Skin (d) Raffia.
(a) (b)
(c) (d)
Figure 5. Segmentation results when non-average is suppressed in 8-connectivity. (a) Plastic Bubbles (b) Straw (c) Pig
Skin (d) Raffia
(a) (b)
(c) (d)
Figure 6. Segmented Images when non-minimum is suppressed (a) Plastic Bubbles (b) Straw (c) Pig Skin (d) Raffia.
5. Conclusions indicate the domination of linear, circular and horizontal

patterns. By the above topological patterns, one can classify
Detecting edges and then eliminating irrelevant ones and textures easily. With the proposed segmentation method the
connecting (grouping) the others are the key to a successful classification of the textures based on topological structures
edge-based segmentation. The segmentation scheme becomes more effective.
exploited spatial information by detecting edges in an image
that correspond to discontinuities in the homogeneity 6. Acknowledgements
criterion for segment. As the mask size increases in texture We would like to express our gratitude to the
segmentation by non-maximum and average suppression, management of LBR College of engineering for providing
noise reduces and the borders are more exact. The proposed facilities in the college. We are thankful to the Director, Dr.
segmentation algorithm proposed, is simple to implement L.S.S. Reddy, who has given motivation and constant
and shows better results than the existing non-maximal encouragement to complete this paper. Also we would like to
suppression algorithm. thank Dr. R. Chandrasekaran and Prof. C. Nagaraju for
their invaluable suggestions to improve the quality of this
From the Fig.4 and 5, the segmented results clearly show paper.
the local boundaries in all textures. By the technique of
rotating mask, local boundaries in all directions are made References
clearly visible. The segmentation with continuous
boundaries are more visible in Fig.4 when compared to [1] D.Dunn, W. E. Higgins and J. Wakeley, Texture
Fig.3, Fig.5 and Fig.6. The segmentation results of textures segmentation using 2-D Gabor elementary functions,
IEEE Trans. Pattern Analysis Mach. Intell. 16(2), 130- Authors Profile
149 , 1994.
[2] Y. M. Zhu and R. Goutte, Analysis and comparision of Agastyaraju NagarajaRao received the M.Sc.
space/spatial – frequency and multi scale methods for (Computer Science) Degree from S.V.
texture segmentation, Opt. Eng. 34(1), 269-282(1995). University in 1999. He received his Ph.D.
[3] M. Unser, Texture classification and segmentation using degree in Computer Science from University of
wavelet frames, IEEE Trans. Image Process. 4(11), Mysore in 2009. He is having 10 years of
teaching experience at graduate and
1549-1560, 1995.
postgraduate level. Present he has been working with LBR college
[4] J. Mao and A. K. Jain, Texture classification and of Engineering, Mylavarm, as a professor in Department of IT. He
segmentation using multiresolution simultaneous has published more than 6 international journal papers, 14 national
autogressive models, Pattern Recognition 25(2), and international conferences. His research interests include Image
173-188,1992. Processing, Pattern Recognition, Data Mining and Texture
[5] Y. Hu and T. J. Dennis, Textured image segmentation by Segmentation. He is a life member for CSI and ISCA.
context enhance clustering, IEEE Proc. Vis. Image
Signal Process. 141(6), 413-421, 1994.
[6] J. L. Chen and A. Kundu, Unsupervised texture K. Lavanya completed her B.Tech. (CSE)
segmentation using multichannel decomposition and From JNT University. Now she is pursuing her
M.Tech. (CSE) from JNT University,
hidden Markov models, IEEE Trans. Image Process.
Hyderabad. Currently she is working as
4(5), 603-619,1995. Assistant Professor in department of IT, LBR
[7] R.C. Gonzalez, and R.E.Wood,.. Digital Image College of Engineering, Mylavaram. She has
Processing., Wesley Publishing Company, pp. 458- overall 5 years of teaching experience at
461,2002. graduate level. Presently she is doing her project in image
[8] O.R.P.Bellon, A.I. Direne, and L. Silva, Edge detection processing.
to guide range image segmentation by clustering
techniques. In: International Conference on Image G.Aruna kumari received B.E. degree in
Processing (ICIP ’99), Kobe, Japan, pp. 725–729, 1999. Electronics and Communication Engineering
from Andhra University in 1995, and also in
[9] X. Jiang, A. Hoover, G. Jean-Baptiste, D. Goldgof, K.
1999 she obtained M.Tech degree in Computer
Boywer, and H. Bunke, A methodology for evaluating Science and Technology from the same
edge detection techniques for range images. In: Proc. university. She has 10 years of teaching
Asian Conf. Computer Vision, pp. 415–419, 1995. experience at graduate and postgraduate level. She is pursuing Ph.
[10] A.D. Sappa, and M. Devy, Fast range image D. (CS) from Jawaharlal Nehru Technological University,
segmentation by an edge detection strategy. In: Third Hyderabad, in the field of image processing. Her research interests
International Conference on 3-D Digital Imaging and include of image processing and pattern recognition.
Modeling, pp. 292–299, 2001.
[11] M.A. Wani and H.R. Arabnia, Parallel edge-region- M.N.Hima Bindu completed her B.Tech
(CSIT) from JNT University,Hyderabad. She is
based segmentation algorithm targeted at reconfigurable
Pursuing M.Tech. (SE) From JNT University,
multiring network. Journal of Supercomputing 25(1), Kakinada. She has been working as Assistant
pp. 43–62, 2003. Professor in department of IT, LBR College of
[12] C.E. Honeycutt and R. Plotnick , Image analysis Engineering, Mylavaram. She has 4 years of
techniques and gray-level co-occurrence matrices teaching experience at graduate level.
(GLCM) for calculating bioturbation indices and
characterizing biogenic sedimentary structures,
Computers & Geosciences 34, pp. 1461-1472, 2008.
[13] S. Zheng, J. Liu and J.W. Tian, A new efficient SVM-
based edge detection method, Pattern Recognition
Letters 25, pp. 1143-1154, 2004.
[14] A. Kundu, Mitra, A new algorithm for image edge
extraction using a statistical classifier approach. IEEE
Transactions on Pattern Analysis and machine
Intelligence, 9(4): 569-577, 1987.
[15] Milan Sonka, Vaclav Hlavac and Roger Boyle,
Image Processing,Analysis and machine vision, Second
Edition, Vikas publishing House,pp. 135-137,2001.
Design of a 3.3GHz Band Application LC VCO for

Low Phase noise and Low power Using Current
Mirror
Namrata Prasad1, R. S. Gamad2
1
Electronics & Instrumentation Engineering Department,
SGSITS, 23, Park Road, Indore, M.P., India – 452003
namrata.prasad7@gmail.com
2
Electronics & Instrumentation Engineering Department,
SGSITS, 23, Park Road, Indore, M.P., India – 452003
rsgamad@gmail.com
Abstract: In this paper a novel methodology is used to implement generate a negative resistance to cancel the loss in the LC tank. It
an LC tank Voltage Controlled Oscillator (VCO) with low power consists of a tail current mirror to provide bias current in a design.
consumption, low phase noise and low tuning range. Here a A typical resonance frequency of cross coupled VCO is given by
relationship between, the phase noise, power consumption and the [5]:
bias current is fully analyzed to improve the performance of VCO. 1
The present design architecture of VCO has employed two current Fosc =
mirrors one at the top and other at the bottom end of cross coupled 2π LC
VCO, to balance the impedance and gives exact replica of current (1)
in both the arm of current mirror circuit. The phase noise
measured is -148.59dBc/Hz at 3.3GHz carrier frequency with Where,
offset voltage of 1V. It will consume power of 5.68mW and the L is the inductor in Henry of LC tank and C is the
bandwidth is 4.31GHz, It also has the tuning range of 1.82%. From capacitance.
the simulation results we have seen that the power consumption, The passive element i.e. the on-chip spiral inductor L and the two
phase noise has reduced with current mirror. Finally we have capacitor forms the frequency tuning network and Vcon is the
compared the results with the earlier published work and got controlled voltage. In case of a single current mirror the impedance
improvement in the present results as given in table 1. is unbalanced resulting in different currents in the mirror arms
which increase the power consumption and also has a negative
Keywords: LC-VCO, current mirror, cadence, low power, low effect on the phase noise performance of the circuit. Figure 1
phase noise. shows schematic view of the earlier VCO design without using
current mirror [5]. To overcome this problem we have used current
1. Introduction mirror in the proposed VCO design.
Voltage Controlled Oscillators (VCOs’) are used in wireless

application. It is one of the most significant block for the Radio
Frequency (RF) communication system because it define all the
performance parameter of the system with the rapid development
of RF communication application in the field of cellular telephony,
codeless phone, wireless data networks, two way paging etc. the
performance parameter Low phase noise, Low power consumption
and wide tuning range are the basic requirement which are
interrelated and for the better system performance a tradeoff has to
be achieved among these crucial requirements [1].There are mainly
two confurigation of cross coupled VCO. Here, complementary
cross coupled VCO is used to achieve Low phase noise, in RF
communication phase noise is reduced by the drawback of high
power because it degrade the system integrity by reducing the
signal integrity of the output of a transceiver. Therefore by proper
analyzing the circuit with well controlled current flow a low phase
noise and a Low power consumption is obtained [2]. The paper is
described in following section 2. Briefly described the analysis and
design of cross coupled VCO, section 3 describe the proposed
VCO with the addition of current mirror at the top and bottom end
of the VCO architecture. The result is presented in section 4. And
finally the conclusion is presented in section 5.
2. Design and analysis of cross coupled VCO Figure 1. Schematic view of the earlier VCO design
The core component of the VCO consists of cross coupled PMOS
and NMOS that is (M1, M2) and (M0, M3) from VCO core to 3. Proposed VCO Design
In order to reduce the power consumption, phase noise, tuning Where,

range because they are the key parameter for the performance of a
transreceiver system. In the proposed VCO design the top and L (FM) phase noise in dBc/Hz, Fm is the frequency offset from the
bottom current controlled architecture is employed i.e. current carrier in Hz, f0 is central frequency in Hz, fc is flicker noise corner
mirror which balance the impedance in both the arm of circuit and frequency in Hz, Q is the loaded quality factor of the tuned circuit,
hence the current become the exact replica of the bias current. In F is noise factor, K is Boltzmann's constant in J/K, T is
this design a current controlled mechanism is used to reduce even temperature in K, Pav is average power at oscillator output, R is the
harmonics in the drain current which has a direct impact on the equivalent noise resistance of the varactor and Kvco is oscillator
phase noise component which results reduction in phase noise it voltage gain in Hz/V.
also achieves the negative resistance from the active devices by From equation (2), Kvco dominates the phase noise performance in
drawing minimum amount of current from the supply and reducing the modified Lesson's formula, thus phase noise performance can
the power consumed by the circuit. Here, novels current controlled be improved by reduction Kvco.
architecture is used to shift the wave form and control shape of the
output waveform by adjusting the transistor sizes for the current Maximum d. c. power dissipation= (Vsupply ) x (Ibias )
mirror. The new proposed design schematic is presented in fig. 2 (3)
.
Tuning range can be determined as follows [5]:
W0 max − W0 min
Tunningra nge % = x100 (4)
W0
Where, W0max is the maximum frequency of operation, W0min is the

minimum frequency of operation and W0 is the frequency of
operation.
4. Simulation result and discussion

This work is carried under the environment of cadence software
and schematic editor is used for design entry. In this design we
have used specter RF simulator for Simulation, by using TSMC
0.18µm technology. The design is simulated with different
architecture i.e. without current mirror, with tail current, and with
current mirror. The applied voltage is 2V at the center frequency of
3.3GHz, with the Bandwidth of 4.31GHz. We have compared our
simulation results with earlier work done and got improvement in
this reported results and are shown in the table 1. Simulated
output voltage responses of the present design are presented in fig.
3 and 4. Phase noise is given in fig. 5.
Figure 2. Schematic view of the present VCO design
To reduce phase noise and power consumption, we have used

current mirror architecture in our proposed design in cross coupled
form. Phase noise is most important parameter in VCO design
therefore; phase noise performance is optimized by using lessons’
formula [4]. Figure 3. Simulation result of the output voltage
 f f fKT 2KTRK 2vco 

L( FM) = 10log [1 + ( 0 )2 ](1 + c ) + 
 2 fmQ fm 2Pav f 2m 
(2)
Table 1: Comparison of present results with earlier Work

done
Parameters Ref. With tail This design

[6] current
mirror
only Without With
[5] current current
mirror mirror
Operating 2V 2V 2V 2V
Voltage
Technology 0.35µm 0.18µm 0.18µm 0.18µm
(CMOS) (TSMC) (UMC) (UMC) (UMC)
Power
consumption 18 15.76 8.69 5.68
(mW)
Operating
Frequency 6GHz 3.3GHz 3.3GHz 3.3GHz
Tuning 17% 3.207% 29.8% 1.82%

Range
Phase Noise
(dBc/Hz) -94 -151 69.63 -148.59
(mdB/Hz)
Figure 4. Simulation AC response of the output Voltage Bandwidth - - 1.611 4.31

(GHz)
5. Conclusion
In this paper, we have presented a novel, low phase noise, low

power Cross coupled VCO using 0.18µm UMC technology. The
proposed VCO will consume 5.68mW of power at 2V supply and
achieves the phase noise of -148.59dBc/Hz at 3.3GHz and the
tuning range is 1.82%. The result shows that the power
consumption is reduced by 10% as compared to the design with
tail current. Here we have used the technique for balanced the
impedance in the circuit for better result. Present VCO design will
become more useful where the low phase noise and low power
consumption are the main requirements. Finally present results are
compared with the earlier reported work and got improvement in
this reported results as given in table 1.
Acknowledgment
This work has been carried out in SMDP VLSI laboratory of the
Electronics and Instrumentation Engineering department of Shri G.
S. Institute of Technology and Science, Indore, India. This SMDP
VLSI project is funded by Ministry of Information and
Communication Technology, Ref. Technology (CMOS)
Government of India. Authors are thankful to the Ministry for the
facilities provided under this project.
Figure 5. Results of the Phase noise

References
[1] P. Dudulwar, K. shah, H. Le, J. Sing, “Design and Analysis
of Low Power Low Phase Noise VCO”, International
Conference Mixed Design 2006.
[2] Lin Jia, Jian-Guo Ma, Kiat Seng Yeo, Manh Anh Do, “A Novel
Methodology For The Design Of LC Tank VCO With Low
Phase Noise”, IEEE International Symposium On circuit
systems, Vol.1, 23-26 May 2005, pp. 376-379.
[3] Maria del Mar Hershenson, Ali Hajimiri and Sunderarajan S.
Mohan, Stephen P. Boyd, Thomas H. Lee, “Design and
Optimization of LC oscillator”, 1999 IEEE/ACM International
Conference on Digital Object Identifier,7-11 Nov. 1999 pp.
65-69.
[4] R. M. Weng and J. Y. Lin, “A 2.4GHz Low Phase noise
Voltage Controlled Oscillator”, Department of Electrical
Engineering, National Dong Hwa University, Taiwan, R.O.C.
PIERS Proceedings, Beijing, China, March 23-27, 2009.
[5] Namrata Prasad, R. S. Gamad and C. B. kushwah, “Design of
a 2.2-4.0 GHz Low Phase Noise and Low Power LC VCO”,
International Journal of Computer and Network Security
Vol.1, N0.3, 2009, pp. 15-18.
[6] B. Razavi, “A study of phase noise in CMOS oscillator” IEEE
Journal of Solid-State circuits Vol. 31. no.3, Mar.1996, pp.
331-343.
Authors Profile
Namrata Prasad received the B. E. Degree

in Electronics and communication
Engineering. From S.A.T.I. Vidisha in 2008
and pursuing M.Tech degree in
Microelectronics and VLSI Design from
S.G.S.I.T.S. Indore, India in 2008-2010.
Recently she is working with a project on
VCO design and analysis.
R. S. Gamad received the B. E. in

Electronics & Communication Engineering
from V. University, India in 1995 and M.E.
degrees in Digital Techniques &
Instrumentation Engineering with honors
from Rajiv Gandhi Technical University
Bhopal, India in 2003. He has been working
in teaching and research professions since
1996. He is now working as Asst. Prof. in
Department of Electronics & Instru. Engineering of S. G. S. I. T. S.
Indore, India. His interested field of research is Dynamic testing of
A/D Converter, Design of an A/D converter and communication.
New protocol to increase the reliability in WSNs

Ali HosseinAlipour1, Habib Moti Ghader 2, Mojtaba GorbanAlizadeh3
1
Islamic Azad University- Tabriz Branch, Iran
Ali.Hosseinalipour@yahoo.com
2
Islamic Azad University- Tabriz Branch – Young Research Club, Tabriz, Iran
habib_moti@yahoo.com
3
Sama organization(offilated with Islamic Azad University)-khoy Branch, Iran
Mojtaba.khoy@gmail.com
Abstract: During the recent years we observe the use of 2. Related works
wireless sensor networks in applications like phenomenon
management, Battlefield Recognition, guardianship of borders
and safe observing. Considering that all these distributions were 2.1 TinyLAP protocol
randomly and data-based protocols work properly in these
In this part TinyLAP protocol [7] which is based on our
methods. In this article we have proposed new algorithms with
automatic learning. As the results show in new algorithm proposed protocol is going to be partly explained. In
networks lifetime increases and reliability increases as well. TinyLAP protocol we allocate a Learning Automata for each
node. And they use this for routing the appropriate path
Keywords: Data-Centric protocol, Reliability, Network with their own circumstance. TinyLAP protocol includes
lifetime, Wireless Sensor Networks two levels, "Distribution" and "Routing and learning”.
Distribution level starts by the node which has a data to be
1. Introduction sent. The node makes a stack called FLOOD and sends it to
Recent improvements in complex integrations (IC) have its neighbor. This stack includes data attributes that is very
made a new generation of micro called Sensors. That from low-sized .and Neighbors by receiving this stack send to
economical aspect they are commodious and also they are their neighbors again. When the Base Station (Sink)
used in two military groups and non military [2, 3]. receives the FLOOD stack, makes another stack called
To consider having group of limitations such as battery life FEEDBACK and distributes it through the network. And
time, calculating and memory significantly they have been when nodes receive this stack add the new path to their table
predicted as non recyclable and also they live until their and distribute it. This level finishes by approaching the
powers fade away. So power is something rare for systems FEEDBACK stack for all the nodes in the network. At the
like sensor. During a special mission the correct end of this level any node has diverse paths to the central
consumption for sensors lifetime should be managed station. Each potential node for choosing the path evaluates
knowingly [1]. according to 1-2 relation. In this relation hi is the number
Considering the use of sensor networks, fault tolerance is of steps to Sink for the ith and ngh is the number of paths
something important for them. This importance especially in the routing table. Each node is armed to a learning
in uses like military and nuclear laboratory is seen automata its actions is equal with amount of paths exist in
significantly. Whereas I such environments information are that node to the base stationand possibility of choosing each
really vital, first it's necessary to deliver the information to action is equal with probability of choosing the opposite
destination. And the second is to deliver the correct path by that action in the routing table. In fact, Learning
information. Because deciding on the basis of incorrect Automata actions have one to one opposition with the paths
information in worse than not to decide. the main focus of in the routing table.
most researches in this field has been tend to fault tolerances hi
P (i) = 1 − ngh
which during them nodes are completely corrupted and less
efforts have been done for incompatible errors. Data
∑h
j =1
j
(1-2)
incompatibility errors happen due to changes in binary Routing and learning starts when the source node receives
contents when it is processing. In this article we mix three FEEDBACK stack. Source node chooses the path that has
protocols TinyLAP and FDDA. In a manner that we the maximum probability and sends the stacks with the
increased the lifetime of the network and reliability by chosen path. Middle nodes also do so and continue until
TinyLAP and other automatic learner called FDDA. And delivering the stack to central station. Each node after
also we corrected the incompatibility errors. sending the stack waits for the respond from the receiver
In the next part done works are showed and in the third part node. If they received positive answer the path receives a
proposed protocol has been explained and ultimately in parts reward. TinyLAP also uses Warn stack. Warn stack sends
4 and 5 simulation and conclusion are mentioned. when the energy of the nod is lower than 70% (for the first
time the primary energy is assumed) each I node if receives
Warn stack from j node and if j node is exist in routing table Random
α (n) Environment
of I node, penalizes the same action in the route of j node.
2.2 FDDA protocol Learning Automata

β (n)
the main focus of most researches in this field has been tend
to fault tolerances which during them nodes are completely Figure 2. Learning automata connection with environment
corrupted and less efforts have been done for incompatible [6].
errors. Data incompatibility errors happen due to changes in The environment shown by where α = {α1 , α 2 ,...,α r } is a set
binary contents when it is processing. This error is called of inputs β = {β1 , β 2 ,..., β r } is a set of outputs and
soft error [5]. This error happens when the content of d pkt c = {c1 , c 2 ,..., c r } is penalty probabilities.
stack is received by the n node similar to content of d pkt Environment can be shown by three E = {α , β , c} and set of
stack which is not sent by the n node. The incompatibility inputs α = {α1, α2 ,...,αr } is a set of outputs β = {β1 , β 2 ,..., β r } and
error can happen in rode in the manner of temporarily or c = {c1 , c2 ,..., c r } set of penalty probabilities. Whenever β
permanently. Internal problems in hardware components
set has two members, environment is type P. in such an
such as processor or memory units can appear
environment β1 = 1 is the penalty and β 2 = 0 is considered as
incompatibility errors. In contrast with corruptions of energy
resource, a node that has been involved with incompatibility reward. In an environment type Q, β set has infinite
errors still does its services correctly. But in doing some members. And in an environment type S, β set has infinite
other services they encounter errors. From few efforts that members as well. ci Is the penalty probability of αi .learner
have been done in the field of incompatibility fault tolerance automats are divided into two fixed and changing groups.
is the protocol offered by Ssu et al and his colleagues [4].
The working procedure of this protocol that we call it FDDA 3. Proposed method
is to first make two distinct routes between the source node
TinyLAP has some problems: 1) in contrary to what
and central node. And then the data in two copies and by the designers of this protocol claim, what they call Learning
two routes is being sent. In the central node datum are automata is not a Learning automata. Instead of taking
compared with each other. And if they are similar they can samples they use probabilities graph to choose the most
be admitted. Else a new route (without any shared point probable. 2) This protocol uses Warn stack to adjust the
with the rest routes) between source node and central node energy consumption. That this job can be done with the
is being made and then data in three copies and from two responding stack. 3) In TinyLAP protocol according to the
previous route and new route is being sent. To recognize the 1-2 relation, for the nodes using more than two routes to
correct data stack central node uses the quorum in vote. in base stationthe sum of probabilities becomes more than
FDDA protocol to recognize the corrupted node, the central one. (Relation 1-3) in 1-3, relation hi the number of
node sends a message for all the routes of corrupted steps to base stationfor the ith route and ngh is the
node(knowing that each node maintains returning route) number of routes to the routing table. Proposed protocol is
that with this message the error amount of corrupted route mentioned in following part that solves these problems. 4)
increases and when a node gets errors more than threshold the reliability is not offered in this part.
automatically becomes a corrupted node and has to be ngh
inactive. Central node sends a correcting message for the ngh ngh
hi
∑h j
correct routes. And the scope of errors of this route becomes ∑ P ( i ) = ∑ (1 − ngh
) = ngh −
j =1
ngh
= ngh − 1 (1-3)
zero.
i =1 i =1
∑h
j =1
j ∑h
j =1
j
FDDA Protocol has its own problems such as sending

datum from three random routes. And it can not be
optimum. And the second problem is that in FDDA after
comparing the first if they were not equal. Send from
Figure 1. data sending flow in FDDA protocol another three routes. Is main difficult this protocol no
optimal sending path and choice randomize.
2.3 Learning Automata Purpose of this protocol is to make balance in energy
consumption in network nodes. This work happens by the
A Learning automata is a conceptual mode that chooses learner automats. In fact, proposed protocols are
randomly one of its actions and applies that to its combination of three methods. This method first finds all
environement.environement evaluates the chosen action by the routes like TinyLAP or EAR. Then like the FDDA
the automata and by a make up signal sends them to method transmits the data with two routes finally these
learning automata. Learning automata update its own become compared in the Sinks. If they were equal, admits
internal situation by the chosen action and make signals. one of them as ultimate output.
and then chooses the next action. Figure 2 shows the
relation between learning automata and environment. [6]
Sink the steps, adds one unit to the routing unit that has the
least step number.
Whenever considering to a routing table, one learning
automata that its actions is equal by the number of its
routing table. In fact, there is a one to one relation between
actions of automata and routes of routing table. Probability
of choosing each action is equal with probability of the
same route in the table. Whenever learning automata
chooses a practical node, the same route with that action
Source
goes to be chosen to be sent to the Sink. If chosen action is
Figure 3. graph of proposed algorithm appropriate, the probability of selecting that increases. If
In proposed protocol each node has learning automata. not, selecting probability according to algorithm decreases.
And they act in this manner. Each node in each time by At the end of this level each node has one Learning
using the automata chooses one or two routes. If the automata and routing table to guide the data to Sink
chosen route is an appropriate route is gives reward. lese it Routing and learning level
should be punished. Purpose of this rewarding or In this level, each node that had a data to be sending sends
punishing is to increase or decrease the choosing rate. data according to its routing table. By the help of learning
Proposed protocol includes to levels “making routing automata chooses two different routes to send data stack.
tables” and “routing and learning". In level of making Data stack in addition to that data, includes primary
tables, the routing table of each node is being made. In source fields, final destination and sender node. The
level of routing and learning the source node starts to send source node quantifies the data stacks fields and then
the data from two or three diverse routes. The middle node sends the stacks on the chosen routes.
does the same job. The data tables update according to Each node that sends a data stack, if it is not that
their response stack. In this protocol each node has destination, guides that stack to the Sink by the help of
Learning automata with its routes to Sink (Sink). The routing table and learning automata. This middle node
route that is used to balance the energy. In following parts prior to send the stack makes the receiver sources number
we will explain them in details. same as its own number. Also this node makes another
Making Routing Tables stack called ACK sends is to the node that is going to
This level starts by Sink. This node makes FLOOD stack. receive data stack. ACK stack includes sender's field and
And sends this stack to all it neighbors. This stack energy stage. ACK stacks sender node, changes receivers'
contains three fields. Number of nodes sender, count of numbers to senders numbers.
steps to Sink, and energy stage of sender node. Before To reduce the transmitted messages between two nodes
distribution the FLOOD stack by the Sink, changes the and economizing in energy consumption, P parameter is
sender nodes number to its own number, changes the being introduced. Each node by choosing each route sends
number steps to zero. P number of data on that route. Also, each node while
Other nodes while receiving FLOOD stacks start to make receiving P numbers of data sends just one ACK stack to
routing tables that are like this: the sender node. This job avoids lots of ACK stacks and
If a node received just one FLOOD, adds information of reduces energy consumption.
received stack to its own routing table and makes the Every node, by receiving an ACK stack rewards the route
routing table one. that ACK stack has navigated.
If a node received more than one FLOOD stack, puts all If energy of on node that has sent the stack:
the information in its own routing table.(each record for Be lower than 50% from the average energy of first nodes
one route). Choosing probability of each from this route is in the route, choosing action gives penalty according to 3β
on the basis of relation 2-3. that β is based on 3-5 relation.
1 More than 50% and less than 80% average of energy in
numhop i energyleve li contrast with first nodes; this action gives penalty by the
∀ i i ≤ m Pi = h m + (1 − h ) m (2-3)
1
∑i = 1 numhop i
∑i =1
energyleve l i relation of 3-3.
energyleveli
In this relation, it is number of received stack. numhopi is − 0.5
avgenergy (3-3)
the number of steps for ith stack. energyleveli is the Vp = (1 + )β
amount of I stack and m is the number of received stacks. 0.3
h is the amount of constant that is a number between 1 and In relation 3-3, avenergy is the energy of primary nodes
0. This parameter is the displayer of effective ratio of steps and energyleveli is the energy stage of ACKs sender. And
in front of energy of chosen node. The more parameter β is being selected by the 5-3 relation. 0/3 is the result of
reaches to one, the more steps become effective. And as it difference between 50% and 80%. The result of relation is
reaches to Zero, energies effect will be more to all the between β and 2β.
probable routes. More than 80% and less than 100% from the average of
FLOOD stacks receiver, quantifies fields of the FLOOD primary nodes and route steps means that this action
stack and distributes them in the network. Quantifying is should be rewarded by a parameter. A is evaluated by
to introducing its own number as sender nodes number relation 5-3.
and puts its energy stage in that field. To give amount for If it is more than the energy of other primary nodes, this
action will be rewarded by parameter a.
γ * energyleveli + (max numhop − numhopi ) better and the lifetime of network increases. As you see in
α = λ1 +ψ 1 (4-3)
γ * energylevel + max numhop the Figure 4, the numbers of sent stacks have been compared
γ ( avgenergy − energyleveli ) + numhopi with the corrupted nodes. And also fault tolerance has been
β = λ2 + ψ 2 (5-3)
γ * avgenergy + max numhop added to the method. In comparing with FDDA both
In relation 5-3 and 4-3, energylevel and numhopi are methods get majority vote. But in this new method to send
energy level and ACK stack senders steps. Energylevel is uses the least energy using routes. So in contrasting with
the starting energy of node, maxnumhop the greatest FDDA the lifetime of network increases significantly. That
number of steps from the receiver node, and avgenergy is Figure 5 shows this.
the average of primary nodes energy. λ1 and λ2 are
sc ale rece iv e d irec t to

displayers of the least acceptable amount for rewarding 1
su m all d ata
and punishing parameters. Considering the difference in 0.5
scale between steps (numhop) and remaining energy 0
(energylevel) γ parameter is being selected in a way that 0 20 40 80 100 120

Failure node
140 160 180 200
balances the different scale. ψ1 and ψ2 are being selected

FDDA Proposed protocol
in a way that they don’t let a and β parameters rise more
than a specific boundary. Figure 5: Comparison of delivered stacks in contrast with
And finally, to make sure that received data is correct in corrupted nodes in proposed protocol and FDDA
Sink they are being compared. If both of the data were the
same, one of them is going to be selected as main received 5. Conclusion and future works
data. And throw away the other. If the data is not correct
the Sink sends a FEEDBACK from both routes. As soon In this article one protocol that uses aware automats to
as receiving FEEDBACK message, the node sends the find the appropriate routes to send the data stacks to
data by the third route. And when it receives them again, balance the energy consumption and correct sending of
asks for majorities vote to accept them. To receive a third datum among nodes proposed. The results of simulation
incorrect data happens rarely. In this way this algorithms showed that proposed protocol from balancing aspect
continues until receiving the correct data. But we just send among nodes and lifetime of network and reliability have
the data from 3 possible routes. the good performance than TinyLAP, EAR and FDDA.
New protocol contrary to EAR protocols does not need
4. Simulation Results local information.
We stimulated a wireless sensor network in a 100*100 Reference

space and with an equal distribution of 100 sensors
randomly by using MATLAB software. In this simulation [1]Gaurav Gupta, Mohamed Younis "Fault-Tolerant
the central node at the end of area with the co ordinations Clustering of Wireless Sensor Networks"2003 IEEE
has been put. The primary energy of sensors is 0.5 j [2]Yongxuan Lai, Hong Chen "Energy-Efficient Fault-
.duration of simulation is 1000 cycle and consumption Tolerant Mechanism for Clustered Wireless Sensor
energy is equal with table 1. Networks" 2007 IEEE.This work is supported by the
Table 2: used Radio characteristics in our simulations National Natural Science Foundation of China under
Operation Energy Dissipated
Grant.
Transmitter/Receiver Electronics Eelec=50nJ/bit [3]Ameer Ahmed Abbasi,Mohamed Younis,Saudi Arabia"A
Data Aggregation EDA=5nJ/bit/signal survey on clustering algorithms for wireless sensor
Transmit Amplifier networks" Computer Communications30(2007)2826-
Єƒs=10pJ/bit/m2 2841 WWW.ScienceDirect.com
if dmaxtoBS ≤ d0
Transmit Amplifier [4]Chessa S. and Santi P., “Comparison-based system-level
єmp=0.0013pJ/bit/m4
if dmaxtoBS ≥ d0 fault diagnosis in ad hoc networks”, in: Proceedings of
20th IEEE Symposium on Reliable Distributed Systems,
pp. 257–266, 2001.
1000000 [5]Ssu K. F., Chou C. H., Jiau H. C. and Hu W. T.,
800000
“Detection and diagnosis of data inconsistency failures
No packdt
in wireless sensor networks”, in: Proceedings of the

600000
Computer Networks, Vol 50, Issue 9, Pages 1247-1260,
400000
20 June 2006.
200000 [6] D. Chen and P. K. Varshney, "QoS support in wireless
0 sensor networks: a survey", in Proc. of International
0 100 200 300 400
No Nodes Conference on Wireless Networks (ICWN '04), pp. 227-
Propose protocol tinyLAP FDDA
233, Las Vegas, Nev., USA, June 2004.
[7] M. Ankit, M. Arpit, T. J Deepak, R. Venkateswarlu and
D.Janakiram. “TinyLAP: A Scalable learning automata-
Figure 4. Count of received data stacks by several protocol
based energy aware routing protocol for sensor
The results show that TinyLAP some times sends by three
networks”.Communicated to IEEE Wireless and
routes but has the same lifetime in network. Also there is Communications and Networking Conference to be held
fault tolerance in new method. Comparing with EAR acts in Las Vegas, NV USA. 2006.
New method to decrease probability of failure nodes

in WSNs
Ali HosseinAlipour 1, Davood KeyKhosravi 2, Abbas Mirzaei Somarin 3
1
Islamic Azad University- Tabriz Branch, Iran
Ali.Hosseinalipour@yahoo.com
2
Islamic Azad University-Osku Branch Iran
kaikhosravi2003@Yahoo.com
3
Islamic Azad University-Tabriz Branch, Iran
A.Mirzaei@iaut.ac.ir
Abstract: Clustering in wireless sensor networks is one of central station, so most of the traffic will be from them.
the crucial methods for increasing of network lifetime. There To avoid these overheads and unbalanced consumption of
are many algorithms for clustering. One of the important energy some high-energy nodes called “Gateways” are
cluster based algorithm in wireless sensor networks is
deployed in the network [2]. These sensors are used as head
LEACH algorithm. In this paper we proposed a new
clustering method for increasing of network lifetime. We
clusters due to decrease the failure probability of head
distribute several sensors with a high-energy for managing clusters. And this increases the lifetime of the network. But
the cluster head and to decrease their responsibilities in since this method takes a lot of expenditure so in this article
network. The performance of the proposed algorithm via we just use these sensors as manager for a number of head
computer simulation was evaluated and compared with other clusters. In this manner each one becomes gatewayamong
clustering algorithms. The simulation results show the high each head cluster. This method decreases both networks
performance of the proposed clustering algorithm.
lifetime and failure probability.
Keywords: Network Clustering, Nodes failure, Energy-Aware In the second part, the architecture of two networks and the
Communication, Wireless Sensor Networks relevant tasks will be explained. In the third part, the
proposed protocol has been explained and in the fourth part
1. Introduction the results of simulating and tests evaluation can be seen.
The last part involves conclusion of the article and
Recent improvements in integrated circuits (IC) have discussing about pattern of future researches.
fostered the emergence of a new generation of tiny, called
Sensors. That from economical aspect they are commodious
2. Related works
and also they are used in non military (for instance
environmental managing: temperature, pressure, tremor, System architecture for clustered sensor networks has been
etc) shown in figure 1. There just two sorts of nodes, cluster joint
To consider having group of limitations such as battery life sensors and head cluster with tolerance of energy
time, calculating and memory significantly they have been shortcoming. Joint sensors and homogeneous head clusters
predicted as non recyclable and also they live until their with a same identity have been assumed as similar. All the
powers fade away. So power is something rare for systems connections are wireless. The connection of joint nodes with
like sensor. During a special mission the correct the main station is possible only with head cluster. For
consumption for sensors lifetime should be managed sending information schedule we use TDMA (Time-
knowingly. The power of sensor can not support more than Division Multiple Access) protocol.
far connection. Therefore to transmit they need the During starting the process a unique ID, primary energy and
architecture of multi sectional. A useful way to decrease the TDMA scheduling are attributed for all the sensors and
system lifetime is to divide them to diverse clusters [2]. gateways. We suppose that the entire node are aware from
Parts of a cluster-based network sensor are base stations and others place by the GPS. In the beginning all of the
sensors. In this method sensors relay the data flow by head connective bridges are assumed in connection area. as the
clusters. The central station always stays far from where energy consumption of GPS is high , it is On at the
sensors are expanded. In this manner saving the beginning of the clustering and on the other states it is in
consumption energy and awareness of that to communicate the sleep mode. Connection scheduling among connective
with central station has various methods. Two methods of bridges first appears with head cluster when it establishes.
routing in articles have been proposed [5, 6]. These methods The central station always stays far from where the sensors
because of their route detecting and finding optimum steps are expanded. in this order, maintaining the consumption
in relation with central station have head load. In addition, energy and being aware of that in relation with central
they will have extra load on nodes that are located around station have different methods: such as LEACH (Low-
energy Adaptive Clustering Hierarchy) [1] and SEP (Stable some of the nodes have been expanded with a high-energy
Election Protocol) [7] and also two other routing method called Gateway [2] sensors are used as head clusters due to
have been explained in articles [5, 6].these methods due to decrease the failure probability of head clusters. And this
detecting the path and finding the optimum steps in relation increases the lifetime of the network but since this method
with command node have head load. In addition having takes a lot of expenditure so in this article we just use these
extra load on nodes, which is located around central station, sensors as manager for a number of head clusters.
most of the traffics will be because of them. To avoid this To do so, we expand some of nodes according to lifetime,
head loads and unstable energy consumption some of nodes space and number of exist sensors in network. While
have been expanded through the networks by a high-energy clustering we don’t need this work node. We can cluster the
that called gateway [2].these sensors act as gateway among network with the algorithms like SEP, LEACH and TEEN
clusters and central stations. And mage the entire network (Threshold-sensitive Energy-Efficient sensor Network
in cluster. Each sensor with a high-energy belongs just to Protocol).afterward the clustering is done, each head cluster
one cluster. And the connection with the central station just sends a signal to these sensors. And with these signals the
takes place through cluster Gateway. In this method failure sensors specify which cluster is appropriate to manage. And
probability decreases and networks lifetime increases. with the hypothesis of network they choose some of the
cluster to in order to manage them. And each closer is being
managed just by one of these sensors. After establishing the
network the role of these sensors as gateways between head
clusters and central stations, by the hypothesis network
chooses some of clusters to manage. And each cluster is
being controlled by just one of the sensors. After
establishing the network, the sensors have the role of a
gateway between central stations and head clusters. To be
attentive that head clusters to transmit to central stations
and data assembling and calculating in protocol consume a
great deal of energy. All the responsibility of head cluster is
given over to joint cluster sensors or Gateway. Then after
receiving data from its joint nodes without any calculating
delivers them to gateway. And its gateway that transmits
Figure 1. network style with clustering them to base station after doing necessary works and
calculations. This method can be used in two ways. One that
we spread high-energy sensors beside other sensors. And
3. Proposed method
another practical way is to put them between root station
These methods because of their route detecting and finding and head clusters. In both aspects both network lifetime
optimum steps in relation with central station have head increases and extra load eliminates from head clusters and
load. In addition, they will have extra load on nodes that are also failure probability decreases.
located around central station, so most of the traffic will be That other cluster heads don’t have connection with Sink
from them. station. And this connection is accomplished via Gateway
and these nodes with high-energy contain the rule of
Gateway. And these Gateways to lifetime termination
managing same cluster heads. But similar to LEACH
algorithm in any time period the cluster head is changing.
When the cluster node is changing, the cluster head tell to
gateway via a signal. This protocol is resumed to end of
lifetime.
4. Simulation Results
We stimulated a wireless sensor network in a 100*100
space and with an equal distribution of 100 sensors
randomly by using MATLAB software. In this simulation
the central node at the end of area with the co ordinations
has been put. And we spread 4 sensors with high power in
network. The primary energy of typical sensors is 0.5 J
and sensors with high-energy are 1.0 J. we adjust the
execution of the simulation for 1000 cycle and also
Figure 2. Multi-gateway clustered sensor network consumption energy is evaluated based on table number 1.
To avoid this extra load and unstable consumption of energy

Table 1: used Radio characteristics in our simulations [2] Gaurav Gupta, Mohamed Younis "Fault-Tolerant
Operation Energy Dissipated Clustering of Wireless Sensor Networks" 2003 IEEE
Transmitter/Receiver Eelec=50nJ/bit [3] Yongxuan Lai, Hong Chen "Energy-Efficient Fault-
Electronics Tolerant Mechanism for Clustered Wireless Sensor
Data Aggregation EDA=5nJ/bit/signal Networks". 2007 IEEE. This work is supported by the
Transmit Amplifier Єƒs=10pJ/bit/m2 National Natural Science Foundation of China under
if dmaxtoBS ≤ d0 Grant.
Transmit Amplifier єmp=0.0013pJ/bit/m [4] Ameer Ahmed Abbasi, Mohamed Younis, Saudi Arabia
if dmaxtoBS ≥ d0 4
"A survey on clustering algorithms for wireless sensor
networks" Computer Communications 30(2007)2826-
The results of simulation show that new method in 2841 WWW.ScienceDirect.com
comparison with LEACH and SEP acts better and also [5]S. Singh, M. Woo and C. S. Raghavendra, "Power-Aware
increases the networks lifetime significantly. Routing in Mobile Ad Hoc Networks", Proc. of ACM
We test this protocol and LEACH and SEP with different MOBICOM'98, Dallas, Texas, October 1998
sensors (50,100,200,300,400,500) and as seen in figure 3 [6] D. Estrin, R. Govindan, J. Heidemann, and S. Kumar.
the results show that the new method is better than exist "Scalable coordination in sensor networks" Proc. of
methods. And the lifetime of the network is more than the ACM/IEEE MobiCom 1999, Seattle, Washington,
same lifetime in LEACH and SEP. both LEACH and SEP August 1999.
die with 100 sensors when they see the first sensor and [7] Georgios Smaragdakis Ibrahim Matta Azer Bestavros”
live for another 200 time. While in the proposed protocol SEP: A Stable Election Protocol for clustered
after observing the first died sensor that itself observes heterogeneous wireless sensor networks” Technical
later than LEACH and then lives for another 300 times. Report BUCS-TR-2004
[8] Piraeus Tillapart, Sanguan Thammarojsakul, Thanachai
1400
Thumthawatworn, Pratit Santiprabhob”An Approach to
1300 Hybrid Clustering and Routing in Wireless Sensor
Networks” 2005 IEEE.
Network Lifetime
1200
?
1100 Author Profile
1000
900 Ali HosseinAlipour received the B.S.

degrees in Computer Engineering from
800 Islamic Azad University- Shabestar Branch
10 100 200 300 400 500
in 1999 & 2002 and M.S. degrees in
No.Nod Computer architecture engineering from
NEW Protocol SEP LEACH Islamic Azad University- Tabriz Branch in
2007 & 2010, respectively. My interests about researches
include WSN and schedule into multiple processes.
Figure 3. Comparing proposed algorithm with others
.
5. Conclusion and future works

The node of Gateway with a high-energy through the
sensors is used as a central manager is just a step away from
the central station. Ultimately after simulating we found out
that proposed protocol plays an indispensable role in
increasing network lifetime and could have been increased
the lifetime in comparison with SEP and LEACH.
In this article it is supposed that sensor nodes and gateways
are fixed and motionless. On the other program we will
research the mobile gateways.
Reference
[1] Kazem Sohraby, Daniel Minoli, Taieb Znati "Wireless
Sensor Networks Technology, Protocols, and
Applications" Published by John Wiley & Sons, Inc.,
Hoboken, New Jersey. Published simultaneously in
Canada. 2007.
A New Clustering Algorithm for Increasing of

Lifetime in sensor Networks
Abbas Mirzaei Somarin1, Habib Motee Ghader2, Amir Masoud Rahmani 3 and Ali Ghafari4
1
Islamic Azad University-Ardabil Branch,Ardabil, Iran
a.mirzaei@iaut.ac.ir
2
Islamic Azad University- Tabriz Branch- Young Research Club, Tabriz, Iran
Habib_Moti@yahoo.com
3 Islamic Azad University- Oloom Tahgigat Branch-, Tehran, Iran

Rahmani74@yahoo.com
4 Islamic Azad University- Tabriz Branch , Tabriz, Iran

jane.doe@email.com
Abstract: One of the crucial problems in sensor networks is can be recounting according to active nodes because in this
limitation of energy, which affects in network lifetime. There are case networks lifetime will be increased. One of the effective
many algorithms for increasing network lifetime. One of the ways in increasing of networks lifetime is clustering. In
methods in this case, is network clustering. A good clustering sensor network clustering, sensor network divided into some
increases the lifetime of network. In this paper a new clustering branches and one of the nodes have been chosen as a top
algorithm based on Learning Automata for increasing network branch. Duty of top branch is receiving information from
lifetime is proposed. The proposed algorithm based on LEACH
other branches and sends them to the Sink. In dynamic
Protocol. Performance of proposed algorithm is compared with
previous methods by implementing simulation. The obtained
clustering network can be cluster just once and nodes of
results from proposed algorithm is better than others. Results are branches never transfer to others but network nodes can be
indication of high efficiency of suggestion algorithm. the members of other branches. In this paper by using of
learning automata we will change the LEACH algorithm
Keywords: Network Clustering, Algorithm. that increases network lifetime. In the rest of this paper we
will talk about learning automata in section 2. Then in
1. Introduction section 3 LEACH algorithm will be explained. In section 4
Wireless networks are networks, which are, consist of many suggested algorithm will be introduced and in section 5 the
little nodes by low facility. These nodes, which are called a rest of dramatization had been shown.
sensor, can feel special feature such as wetness, temperature, 2. Learning Automata
and pressure in their environment and send this for their
neighbors. In other words two major facilities of these Learning automata is an abstract model that chooses
sensors are the feeling of special parameter around randomly an operation from a set of finite operations and
environment is ability of connection. In some operations then applies it on the environment to the selected operation
these nodes might be join to each other by connective cables, environment is evaluated by learning automata and then
but in most cases a wireless network is completely wireless. informs the evaluated result by the help of a reinforcement
In this network nodes are generally fixed or fixed limited signal to the learning automata. The learning automaton
motions. updates its interior situation by utilizing selected operation
Despite of fixed networks and other wireless networks, and reinforcement signal and selects the next operation
which the quality of service is completely, clear in sensor afterwards. Interconnection between learning automata and
networks there is no fixed recounting. Some of these the environment is shown in Fig.1 [1].
recounting are network proper covering, amount of active
nodes in per time, accuracy of received information in Sink Random
Environment
(central node) and the time of transferring information to
Sink. Some of these recounting such as proper covering and β (n )
α (n) Learning
amount of active nodes in per time are depended to Automata
operation and others like accuracy of received information Figure 1. Learning automata connection with
and time of transferring information Sink intended the environment [2].
feature of network. One of the sensor networks important
features is probability of devastation in some nodes specially The environment shown by where α = {α1 , α 2 ,...,α r } is a set
because of losing energy. For this reason there are so many of inputs β = {β1 , β 2 ,..., β r } is a set of outputs and
nodes in wireless networks. In this case if some of these
nodes has been destroyed others can replace them so some of c = {c1 , c 2 ,..., c r } is penalty probabilities. When β is a set
these nodes should be active but others should be inactive in of binary, so the environment is a P type. In this kind of
order not consume their energy. Therefore, quality of service environment β1 = 1 is considered as penalty and β 2 = 0 as
reward. In Q type environment, β includes numerous cluster. And the responsibility of each cluster head is
finite members and in S type β has a number of infinite gatherings the information’s from the other nodes of the
same cluster and sending of to the Sink. In LEACH protocol
members. ci is a penalty probability of operation. Learning the cluster head never be changed. And because this reason
automata is divided into two groups as stable and unstable the node that is considered as cluster head, consumes more
structure. Learning automata can be utilized in many energy. And this itself causes the declining of the life long
different ways such as: routing in communication network, of the network. In order to overcome this problem we can
face recognize, programming processes in a computer exchange dynamically the cluster head node between the
network. queue theory, access control in asynchronous members of cluster nodes. In suggested algorithm, we
transmit network, partitioning objects and finding an change the LEACH node in a way that the node of cluster
optimal structure for nerve systems [2, 7-11]. head in each period exchange dynamically between the
clusters nodes. The way of selecting the cluster head node
3. LEACH Protocol between the nodes of a cluster forms by the learning
One of the first and famous presented hierarchical protocols automata. In suggested algorithm there is considered a
to the sensor networks is LEACH protocol [3]. In this learning automata for each cluster the number of the actions
protocol, the time span of the nodes activity divides into of automata equals with the number of clusters nodes. Each
some periods. At the beginning of each period some of the action by automata corresponds with a node from the
nodes be selected haphazardly as a cluster head (CH). To do cluster. In figure 3 a cluster and its corresponded automata
this, each node produces a haphazard number between 0, 1. are showed.
Where as this number from the amount of T (n) that results
in by the use of formula (1), be less, the stated node be N2
presented as cluster head. In formula 1, p is the correlation N1
N3 N4
of the number of clusters to the total nodes of the networks, N3
r is the number of period and G the number nodes that in a1 a2
the last period of 1/p are not selected as cluster head. N5 N4
a5 N8
LA
 P a3 a4
 , if n ∈ G N5 N6 N7
1 N7
T (n) =  1 − P × ( r mod ) N8
 p N9
0, otherwise N10 N11
Relation 1. Random number generation between zero and one
Example of Sensor network
and cluster
After determining the nodes of cluster head, the other nodes Equivalence Learning
on the basis of the power of the received signal from each
cluster, decide to be accepted as a member of which cluster. automata for cluster (1)
The cluster head node divides its responsibility glen in to
some time slots (Figure 2). This time slots on the basis of
Figure 3. The assigning the cluster nodes to correspond
TDMA mechanism are cooperated between the members of
learning automata
cluster. In each time slot, the cluster head connect with on
of the members of the cluster and receives the information The Way of choosing the clustering
packs of that member. The cluster head in every some slots
After clustering in LEACH protocol, turns to the selection of
sends its received information from its members to the Sink.
cluster head to each clusters. As it mentioned, for each of
In order to the distribution of the load on different nodes
the clusters there is considered a learning automata that the
after finishing of a period, to start a new period, the cluster
number of its actions equals with the number the nodes of
head by the declared mechanism above are exchange.
the some cluster. The learning automata between its actions
choose an action that its number is more than the other.
After choosing a node as a cluster head, in order to transfer
the information of its clusters nodes by the TDMA
mechanism connects with them and again the learning
automata choose another node as the cluster. (It chooses the
node as a cluster that the corresponding action that has more
amount than the other actions). The amounts of automata in
each time are exchange by the formula 2.
eni
Figure 2. Period and Time Slice in LEACH protocol Pni =
4. Improved Protocol (LALEACH)
d ni
Relation 2. action value Relation
As mentioned in the previous section LEACH algorithm do
the act of clustering and is selected a cluster head to each
In this formula: transferring their outputs to the Sink, turns to the

exchange of cluster head node.
n : The number of cluster nodes. The way of exchanging the clusters nodes of number (1)
ni : The i th node. as the cluster head is occurs like this:
First stage: in this stage by the information of the
eni : The remain energy of ni node. number (1) cluster the exists in figure of table 6, the
number of actions of the cluster (1) automata is
d ni : The average of distance of ni node with the other appeared.
nodes of the same cluster.
p ni : The amount of i th action of automata. Node Remain Average Distance
In figure 4, period and the way of time slot in suggested No Energy AUTOMATA
protocol is showed. By comparing the figure 2 and 4 the way ACTIONS VALUES
of exchanging the node of cluster head for a cluster is n1 en = 3 J 2+ 6+8 3
1 d1 = = 5.34 pn1 = = 0.55
showed (clustering in each period is done and the exchange 3 5.43
of the cluster head node is done inside the clusters). en2 = 5 J 2 + 4 + 12 5
n2 d2 = =6 p n2 = = 0 . 83
3 6
n3 en3 = 1J 8 + 12 + 3 1
d3 = = 7.67 p n3 = = 0.13
3 7.67
n4 en4 = 2 J 6+4+3 pn =
2
= 0.46
d4 = = 4.3 4
4.3
3
Figure 6. The figure (5) network cluster number (1) nodes
information.
After determining the amount of the automata’s actions

related to the cluster (1) as table of figure 6, now it is
selected a node as a cluster head that its amount of action is
more than the others, so it is selected as the cluster head
node. After selecting the node of number (2) as a cluster
head, it declares its being as cluster head to the all nodes
Figure 4. Period and Time slice in LELEACH protocol.
exist in its cluster. The other cluster head contents of the
Figure 5 shows a sensor network. At first this network by table of figure 6m which are being up-to-date in selecting of
the LEACH algorithm is divided into three clusters. By the cluster head are being selected.
next period that a new algorithm will be occurs, the After connecting the node ( n 2 ) by all nodes of its cluster
suggested algorithm will exchange the cluster head
and transferring its information to the cluster head node, it
dynamically.
Cluster 4 Cluster 5
makes up-to-date its remained energy in table (6). After
N20 N21 N22
N23 getting up-to-date the information of the table, the node,
which its amount of action is more than the others, is
CH
CH selected as cluster head. This process continues until the
Cluster 1 N18 N19
N17 beginning of next stage.
2m N2
CH
Cluster 2 N16
N1
12m N6
5. The Result of Simulations
N15 In this section the result of suggested algorithms simulation
4m
6m
N5
is comparing and assessing with the result of previous
Cluster 6 algorithm. The results of suggested algorithm is compared
8m CH
and assessed in respect of lifetime by the algorithm of
N4 N7
N14
LEACH, HEED, Extended HEED clustering. In the tested
3m
N8 CH simulation of the sensor environment is considered
N13
N12 N25 150×150 and the radio range of the sensor environment is
N3
CH considered 30 meters. The first energy of the nodes is
N11
N9 considered 2J. Trials to some of the different sensor nodes is
N24
done by N={100, 150, 200, 250, 300, 350, 400, 450, 500}
N10
Cluster 3 the shown results is the result of min of results gained from
10 performing different simulations. The results from
Figure 5. an Example of clustered sensor network different algorithms are compared with each other respects
of lifetime that is one of the main standards of the quality of
At first after clustering by the use of LEACH algorithm, service in sensor networks. The results of assessing are
the node of cluster head is appear. After connecting the showed in Figure (7) it is clear that the lifetime of network
cluster head node with the all of its clusters node and in the suggested algorithm is more than the other ways.
[6] M. Esnaashari, M. R. Meybodi1, “A novel clustering

algorithm for wireless sensor networks using Irregular
Cellular Learning Automata”, IEEE, 27-28 Aug. 2008.
[7] Meybodi, M. R. and Beigy, H., New Class of Learning
Automata Based Scheme for Adaptation of
Backpropagation Algorithm Parameters, Proc. Of
EUFIT-98, Sep. 7-10, Achen, Germany, pp. 339-344,
1998.
[8] Oommen, B. J. and Ma, D. C. Y., Deterministic
Learning Automata Solution to the Keyboard
Optimization Problem, IEEE Trans. On Computers, Vol.
37, No. 1, pp. 2-3, 1988.
[9] Beigy, H. and Meybodi, M. R., Optimization of Topology
of neural Networks Using Learning Automata, Proc. Of
Figure 7. Comparison of suggested algorithm with others. 3th Annual Int. Computer Society of Iran Computer
Conf. CSICC-98, Tehran, Iran, pp. 417-428, 1999.
6. Conclusion [10] Beigy, H. and Meybodi, M. R."Optimization of
In this paper a new algorithm based on LEACH algorithm is Topology of Neural Networks Using Learning Automata,
suggested. The goal of suggested algorithm is the increasing Proc. Of 3th Annual Int. Computer Society of Iran
the lifetime of the sensor network. In LEACH algorithm the Computer Conf. CSICC-98, Tehran, Iran, pp. 417-428,
nodes of cluster head are selecting with out considering their 1999.
remained energy and because the nodes of cluster head [11] Hashim, A.A., Amir, S.and Mars, p. Application of
consume more energy than the other node, so LEACH Learning Automata to Data Compression, In Adaptive
algorithm in this respect that chooses haphazardly the nodes and Learning Systems, K. S. Narendra (Ed), New York:
of cluster head causes decreasing the lifetime of the Plenum Press, pp. 229-234, 1986.
network. In suggested algorithm instead of selecting
haphazardly the node of cluster head, it is selecting between
the nodes their remained energy is more than the other
nodes and also near to its neighbor nodes.
So, by this way nodes with high energy are selecting as
cluster head and increase the lifetime of network. The
effectiveness of suggested algorithm compared with cleared
clustering LEACH, HEED, Extended HEED algorithms has
better results.
Reference
[1] D. Chen and P. K. Varshney, "QoS support in wireless

sensor networks: a survey", in Proc. of International
Conference on Wireless Networks (ICWN '04), pp. 227-
233, Las Vegas, Nev., USA, June 2004.
[2] H.MotieGhader, S..Parsa, M.Hossein Nejad,
“ Application of Learning Automata for DAG Scheduling
on Homogeneous Networks", 17th Iranian Conference on
Electrical Engineering, Iran University of Science and
Technology, ICEE2009.
[3] S. M. Abolhasani, M. Meybodi, “Usage of Learning
Automata for Routing, Fault Tolerance and Topology
Control in Wireless Sensor Network”, MSC Thesis,
March 2008.
[4] W. R. Heinzelman, A. Chandrakasan, and H.
Balakrishnan, “An Application-Specific Protocol
Architecture for Wireless MicrosensorNetworks,” IEEE
Transactions on Wireless Communications, vol.1 ,no.4 ,
pp.660 –670 , October2002 .
[5] O. Younis and S. Fahmy, "Distributed Clustering in Ad-
hoc Sensor Networks: A Hybrid, Energy-Efficient
Approach", In Proc. of IEEE INFOCOM, March 2004 .
Securing Digital Information using Quasigroups

Saibal K. Pal1 and Shivam Kapoor2
1
Scientific Analysis Group, DRDO, Metcalfe House,
Civil Lines, Delhi – 110 054, India
skptech@yahoo.com
2
Department of Computer Science, University of Delhi,
Delhi – 110 007 India
shivam.dumca@gmail.com
1 2 3 4 5
Abstract: The non-associative property of quasigroups has
been recently found to be useful in many information security 5 3 1 2 4
Order 5:
applications. In particular, quasigroups have been used for 3 4 2 5 1
ensuring confidentiality and integrity of the data transmitted
4 1 5 3 2
over insecure public channels. Quasigroups operations are
computationally simple and can be efficiently used for 2 5 4 1 3
protection of voluminous media like images, audio, video and
This specific Latin square is not a reduced Latin square as
different forms of multimedia. They are also suitable for
securing data transmitted from and to mobile and miniature the rows are not arranged in order.
resource-constrained devices with limited capability. We first
describe schemes for construction of large sized quasigroups Number of Latin Squares: The total number of Latin
that can be used to ensure confidentiality without complicating squares N [3] of order n are computed using the formula:
the operations or increasing the processing of data by N(n, n) = n! (n-1)! L(n, n) (1)
introducing additional rounds. These quasigroups are generated
using isotopies, product of smaller quasigroups, affine and non- Table 1. The number of reduced Latin squares
affine mappings, theta-mappings, keyed permutations, T- n L(n , n)
functions etc. Using these concepts, design of fast encryption
schemes for different media is presented. Ensuring data 1 1
integrity is also of vital importance for many present day 2 1
applications. Schemes for generation of highly non-associative 3 1
quasigroups are described and their use for secure hashing of 4 4
data is explained. Computer implementation of these schemes 5 56
demonstrates their simplicity and power for efficiently securing 6 9408
digital information.
7 16 942 080
8 535 281 401 856
Keywords: quasigroup, isotopy, non-associativity, encryption, 9 377 597 570 964 258 816
hashing, digital media.
Here, L(n, n) is the number of reduced Latin squares of
order n. For large values of n, the number of reduced Latin
1. Introduction squares is difficult to compute & hence the total number of
1.1 Latin Square Latin squares of high order is unknown.
A Latin square [1], [2] of order n is an n x n square matrix
whose entries consist of n symbols such that each symbol 1.2 Quasigroup (QG)
appears exactly once in each row and each column.
Examples of a 4 x 4 and 5 x 5 Latin square are given below 1.2.1 Definition: A quasigroup (Q,*) [4], [5] is a set Q of
a b c d elements along with a binary operation ‘*’ having the
Order 4: b c d a following properties:
c d a b (a) For all a, b є Q, a * b є Q (Q is closed under *)
d a b c (b) For all a, b є Q, there exist unique x, y є Q so that
a * x = b and y * a = b i.e. ( (Q, *) has unique
This Latin square generated by using elements of the set {a, solubility of equations).
b, c, d} is called a reduced Latin square as the elements in Because of the unique solubility of equations, each element
the first row & the first column are in monotonically will appear exactly once in each row and exactly once in
increasing order. each column of the multiplication table of (Q,*). That is,
each row and column is a permutation of the elements of Q.
If |Q| = n, then the interior of the Cayley table for (Q, *)
forms an n x n Latin square.
Figure 2. Quasigroup used for Encryption

Examples of Quasigroups:
• The set of Integers with the subtraction operation The following convention is normally used:
( - ) form a quasigroup. Leader - Any of the symbols of Quasigroup can act as a
• The non-zero Rationals and non-zero Reals with leader. Here we have four choices: 0, 1, 2, 3 for the leader.
division operation ( ÷ ) form a Quasigroup. The leader decides which row would contribute to the
• Let Q = Z6 = {0, 1, 2, 3, 4, 5} and let x * y = (x + encryption process.
y) mod 6. Then ( Q , *) is addition modulo 6 on Z6 Key - The leader and the quasigroup itself (with the unique
and the Cayley table for ( Q ,* ) is given by arrangements of elements) may be used as keys during the
* 0 1 2 3 4 5 encryption process.
0 0 1 2 3 4 5 The basic encryption scheme [6], [7] using a leader is as
1 1 2 3 4 5 0 follows.
2 2 3 4 5 0 1
3 3 4 5 0 1 2 1.2.5 Encryption Scheme: Let the plain message be
4 4 5 0 1 2 3
represented by M = (x1, x2,…..,xn). We choose a leader L
and pass on this secret information as a key to the receiver.
5 5 0 1 2 3 4
Assuming that the quasigroup generation mechanism is
Figure 1. Cayley table for (Q ,*) available to the receiver, encryption is performed as follows
EL(x1, x2 ,…….,xn) = (y1, y2 ,……..,yn)
(Q,*) is a quasigroup because its interior is a 6 x 6 Latin where y1 = L * x1
square. This quasigroup is also a group.
and yi = yi-1 * xi (for all i > 1) (4)
1.2.2 Some Properties of QGs For example if M = (3 0 2 1 2 3 3 1), then x1 = 3, x2 = 0, x3
• Quasigroups have cancellation properties that is if ab = = 2 and so on. Then, with the Leader as L = 0, the
ac, then b = c. This follows from the uniqueness of left quasigroup given in Figure 2, and the encryption formula
division of ab or ac by a. Similarly, if ba = ca then b = c. given above, we have
• Left and right multiplication: By definition of a y1 = L * x1 = 0 * 3 = 1
quasigroup Q, the left and right multiplication operators
are defined by y2 = y1 * x2 = 1 * 0 = 0
L( x )y = xy y3 = y2 * x3 = 1 * 2 = 0 and so on.
R( x )y = yx (2)
These are the bijections from Q to itself. The inverse Thus, the encrypted message EL(M) = (1 0 0 0 3 3 3 2).
maps are given in terms of left and right division by Since there are various options for choosing a leader, the
L( x ) ˉ¹ y = x\y encryption scheme can be made stronger by using different
R( x ) ˉ¹ y = y/x (3) leaders for encryption of different blocks of the message.
• A quasigroup is not required to be associative, The Quasigroup can also be changed periodically by
commutative, or to have an identity element. permuting its rows and columns to increase the complexity
of encryption scheme.
1.2.3 Use of QGs in Coding & Cryptography
Apart from their applications in many diverse areas, 1.2.6 Decryption Scheme: For each quasigroup operation
quasigroups have been recently used as cryptographic ‘*’ we can associate a new quasigroup operation ‘o’ defined
primitives in the design of encryption schemes, for pseudo- by:
random number generation, for construction of hash x o y = z iff x * z = y (5)
functions and for design of error-detecting codes. As the The “dual” operation ‘o’ is used for decryption and the
basic operations on quasigroups are very simple, it is “inverse” quasigroup is generated using the above equation:
possible to construct fast and efficient schemes for different
applications. o 0 1 2 3
0 1 3 0 2
1.2.4 Basic Encryption and Decryption Using QGs
1 2 0 3 1
Suppose that we are given a quasigroup with the
operation * as follows: 2 3 1 2 0
3 0 2 1 3
Figure 3. Quasigroup used for Decryption
* 0 1 2 3 Decryption is carried out with the following formula:

0 2 0 3 1 DL(y1, y2 ,…….,yn ) = (x1, x2,……..,xn)
1 1 3 0 2 where x1 = L o y1
2 3 1 2 0 and xi = yi-1 o yi (for all i > 1) (6)
3 0 2 1 3
By performing calculations using (6) and the previous which is the Cartesian Product (CP) of the sets R and S.
quasigroup, the encrypted message is converted back to the Further, let ti = (ri, si), tk = (rk, sk), where i, k ∈ T; ri, rk ∈
original plain message. R; si, sk ∈ S.
DL(E L(M)) = (3 0 2 1 2 3 3 1) Noting that the quasigroup with elements belongs to the set
This scheme is computationally simple and is based on basic T cp, Qcp = < Tcp , * > is the simple product of quasigroups,
lookup operations on quasigroups. To make this scheme we can define the operation * as follows:
more powerful and practical, construction of large ti * tk = (( ri • rk ), ( si ~ sk )) (9)
quasigroups is required and is reported in Section 2. But we want to represent the simple product of quasigroups
as a quasigroup. To this end, we must convert the set Tcp
1.3 Non-Associativity of Quasigroups into the set T. There are many ways of doing such a
conversion.
In addition to the requirement of constructing large and The mapping h : tx →X, where tx ∈ T cp, X ∈ T defined by
unstructured quasigroups for cryptographic purposes, it is the function
also important to generate quasigroups with high degree of h(tx) = h(rx, sx) = n 2rx + sx (10)
non-associativity. These find applications in the design of gives one of the simplest solutions.
secure keyed hashing schemes or message authentication
codes [8]. To measure the degree of non-associativity of 2.2 Generation using Linear Mapping
quasigroups, Meyer [9] borrowed the concept based on Using this scheme, we generate two linear functions f
measuring the non-commutativity of groups. For a group and g and then elements associated with each function is
(G,o), the multiplication group is the subgroup generated by stored in one dimensional array of size equal to size of the
all the left and right multiplications of G. As the left and permutation. Now, in order to generate the ( i , j )th element
right multiplication by an element would be the same for a of the huge Latin square, the ith element of the first function
commutative group, the multiplication group would be is added to the jth element of the second function and
smaller. In contrast, for a non-commutative group, the modulus operation w.r.t the size of the Latin square to be
multiplication group would be larger. Therefore, the size of generated is applied to the addition of ith element of 1st
the multiplication group of a group indicates the degree of function and jth element of 2nd function. In this way we get a
non-commutativity of the group. A similar approach can be huge Latin square with elements in the range 0 to the size of
used to measure the degree of non-associativity of a Latin square.
quasigroup (Q,o). Here also, a larger multiplication group Let Q = Zn = {0,1…, n-1} and let the group operation be
indicates that the quasigroup has a higher degree of non- addition modulo n. Then we can create a Quasigroup (Q , ±)
associativity. from (Zn , +) by defining
f(x) = px + a (11)
2. Generation of QGs g(x) = qx + b
where p and q are relatively prime w.r.t. the order of the
Construction of large quasigroups [10], [11] from
quasigroup and a, b are positive integers less than the size of
smaller ones is an important problem for many applications.
quasigroup, and then further defining
In cryptography, larger the size of the Latin square, larger is
h( x , y ) = ( f( x ) + g( y ) ) % n (12)
the number of choices that can be made by the
For example: Let Q = Z8 = {0, 1, 2, 3, 4, 5, 6, 7} and let
communicating party and higher levels of security can be
the group operation be addition modulo 8. Then we can
provided. Moreover, generation of Latin squares based on
create a quasigroup (Q , ±) from (Z8 , +) by defining f(x) =
key permutations [12] help to regenerate the Latin square
5x + 6 and g(x) = 3x + 7, where a = 4 & b = 1 and p = 3 &
used by the sender at the receiving end with minimal
q = 5. Since we are creating (Q , ±) by linear mapping, we
information exchange. We present below different schemes
will define h(x, y) = (f(x) + g(y)) % 8. Then (Q , ±) is as
for generation of huge Latin squares.
shown below
2.1 Simple Product of Quasigroups ± 0 1 2 3 4 5 6 7
The notion of simple product of quasigroups is important 0 5 0 3 6 1 4 7 2
for cryptographic applications since, if two quasigroups are
1 2 5 0 3 6 1 4 7
given, it permits to construct a quasigroup of the order equal
to the product of the orders of these quasigroups. 2 7 2 5 0 3 6 1 4
Let Q1 = < R , • >, Q2 = < S , ~ > be two arbitrary 3 4 7 2 5 0 3 6 1
quasigroups. 4 1 4 7 2 5 0 3 6
Let the elements of two quasigroups be given by
R = {0, 1, 2, . . . , n 1 − 1}, 5 6 1 4 7 2 5 0 3
S = {0, 1, 2, . . . , n 2 − 1} 6 3 6 1 4 7 2 5 0
Then the simple product Q of these quasigroups is the 7 0 3 6 1 4 7 2 5
algebraic system Figure 4. Quasigroup using Linear Mapping
Q = Q1 × Q2 = < T , * > (7)
where T = {0, 1, . . . , (n 1n2 ) − 1} We find that (Q , ±) is not associative, because h(h(0, 2),
To define the operation * in the set T, let us first assume 4) = h(3, 4) = 0 but h(0, h(2, 4)) = h(0,3) = 6. It is not
that: commutative, because h(0,1) = 0 but h(1,0) = 2. There is no
Tcp = R × S = {t0, t1, . . . , tn1n2-1} (8) identity element and hence no inverses.
Limitations of the above method: The problem with this combination of arithmetical operations like addition,
scheme is that (Q , ±) is still quite structured. The elements subtraction, multiplication and negation together with
in each row always appear in the same order. Boolean operations like OR, XOR and NOT. This helps to
Similarly, the elements in each column are also arranged in design cryptographic schemes resistant to many present day
the same order. In addition, each row is a copy of the attacks. Example of such a mapping is
previous row, shifted one space to the right. Each column is x → x + ( x 2VC )(mod 2 n ) (14)
a copy of the previous column, shifted one space down. As
where C is a constant and V represents the logical operation
we see in above example: 1st and 2nd row are (5 2 7 4 1 6 3
OR. This turns out to be a permutation of single cycle of
0) and (0 5 2 7 4 1 6 3) respectively. Hence by determining
length 2n given certain conditions.
any of the rows, one can guess the entire Latin square.
In general, a T-function is a mapping in which the i th bit
2.3 Generation using Keyed Permutation of the output depends on 0, 1, … , ith input bits. Using a
In the previous scheme, one row could be easily derived small number of such primitive operations over n-bit (32, 64
from the other row. We remove this problem by using non- etc.) words, efficient cryptographic building blocks can be
linear mapping. In this scheme two random permutations designed.
[12] f and g are generated and then each permutation is It is interesting to note that composition of two T-
stored in one dimensional array of size equal to size of functions will also be a T-function. Because we will be using
permutation. Now, in order to generate the (i, j)th element of T-functions to create a quasigroup, it is convenient to choose
the huge Latin square, the ith element of the first k = l, so that f takes as input and produces as output k many
permutation is added to the jth element of the second binary strings of length n. In order to use a T-function f to
permutation and modulus operation with respect to the size define a quasigroup operation, we will see that f needs to be
of the Latin square to be generated is applied to the addition a permutation. Therefore, we need to know which T-
of ith element of 1st permutation and jth element of 2nd functions are invertible.
permutation. In this way we get a huge Latin square with Example: Let v : z23 * z23 → z23 be given by
elements in the range 0 to the size of Latin square.
Let Q = Zn = {0, 1…., n-1} and let the group operation be v(x , y) = x2 y2 + 3(x V y) (15)
addition modulo n. Then we can create a quasigroup (Q , ±) here V represents the Boolean OR operation and addition
from (Zn , +) by supposing and multiplication are defined for mod 23 = mod 8.
f(x) = any random permutation {6, 2, 8…} Let c = < 1 , 0 , 1 > ∈ Z23 . We define
g(x) = any random permutation {5, 3, 9…}
x o y = c + (x + y) + 2v(x , y) = 5 + x + y + 2x2y2 + 6(x V y)
then defining h( x , y ) = ( f( x ) + g( y ) ) % n (13)
(16)
In the example below, to generate a Latin square of order
8 first we will generate two permutation f and g and then Based on this, the quasigroup is created where the binary
apply h( x , y ) = ( f( x ) + g( y ) ) % 8 for x and y ranging representatives of the quasigroup elements are used for ease
from 0 to 7 of notation.
X 0 1 2 3 4 5 6 7
f(x) 2 5 0 4 6 7 1 3
g(x) 7 4 3 2 1 0 6 5 o 0 1 2 3 4 5 6 7
Then the quasigroup (Q , ±) created from f and g by using
the above equation is 0 5 4 3 2 1 0 7 6
± 0 1 2 3 4 5 6 7 1 4 7 2 5 0 3 6 1
0 1 6 5 4 3 2 0 7 2 3 2 5 4 7 6 1 0
3 2 5 4 7 6 1 0 3
1 4 1 0 7 6 5 3 2
4 1 0 7 6 5 4 3 2
2 7 4 3 2 1 0 6 5 5 0 3 6 1 4 7 2 5
3 3 0 7 6 5 4 2 1 6 7 6 1 0 3 2 5 4
4 5 2 1 0 7 6 4 3 7 6 1 0 3 2 5 4 7
5 6 3 2 1 0 7 5 4 Figure 6. The quasigroup created using a T-function v(x, y)
6 0 5 4 3 2 1 7 6
One of the disadvantages of using a T-function in creating
7 2 7 6 5 4 3 1 0
a quasigroup can be seen in the resulting structure present in
Figure 5. Quasigroup using Keyed Permutation the quasigroup. From the given example we observe that the
entries in each row and column alternate between even and
We find that (Q, ±) is not associative, because h(h(0, 2), 4) odd numbers in the quasigroup.
= h(5, 4) = 0 but h(0, h(2, 4)) = h(0, 1) = 6. It is not That is, if x o y is even, then x o (y + 1) and (x + 1) o y
commutative, because h(0, 1) = 6 but h(1, 0) = 4. There is will be odd and vice versa. It is evident that this property
no identity element and hence no inverses. Unlike the holds in any quasigroup created from a T-function. We
previous case, all the rows are independent of each others. notice that both x and y are either even or odd. Clearly, if x
2.4 Generation using T-Functions is even, then x + 1 is odd and vice versa.
T-functions, proposed by Klimov and Shamir [13], [14] Since x o y = c + x + y + 2v(x, y)
are a relatively new class of invertible mappings using a and x o (y + 1) = c + x + y + 1 + 2v(x , y + 1) (17)
o 0 1 2 3 4 5 6 7
Because 2v(. , .) is always even, the parity of xo(y+1) will
be different than the parity of xoy. 0 4 7 2 1 3 0 5 6
Other useful schemes for generation of quasigroups are 1 6 5 0 3 1 2 7 4
based on isotopies and complete mappings. In addition, if 2 0 3 6 5 7 4 1 2
we choose affine isotopies or affine complete mapping, the 3 2 1 4 7 5 6 3 0
Quasigroup created has special properties. 4 1 2 7 4 6 5 0 3
5 3 0 5 6 4 7 1 2
2.5 Theta Mapping
6 6 5 3 0 2 1 4 7
A quasigroup can also be created from a group using a
special kind of mapping called a complete mapping [9]. 7 7 4 1 2 0 3 6 5
Figure 10. Quasigroup created using the complete map Ө
Definition: Let (G, +) be a group and let i : G à G denote
the identity map on G. Ө : G à G is a complete mapping if
Ө is a bijection and i - Ө is a bijection
3. Generation of Non-associative QGs
where (i - Ө)(x) = x - Ө(x). In Section 1.3, we explained the need for generation of
highly non-associative quasigroups for cryptographic
Example: Let (G, +) = (Z9, + ) where Z9 = (0 , 1, 2, 3, 4, 5,
purposes and measurement of the non-associativity of a
6, 7, 8) and addition is performed modulo 9. Then Ө(x) = 5x
given quasigroup. We present schemes with examples for
+ 4 is a complete mapping as seen in Figure 7 because both
generation of such structures.
Ө and i - Ө are bijections
x 0 1 2 3 4 5 6 7 8 Let g : Q à Q be a transposition on a cyclic group (Q ,
Ө(x) 4 0 5 1 6 2 7 3 8 +). Then (Q , o) given by x o y = x + g(y) is a highly non-
i - Ө(x) 5 1 6 2 7 3 8 4 0 associative quasigroup. Here Mult(Q) = Sym(Q), where
Mult(Q) is the multiplication group of (Q , +) and Sym(Q)
Figure 7. A complete mapping on Z9
represent the symmetric group of (Q, +).
Creating Quasigroups using Complete Map:
Both f and g appear in Mult (Q) as (Q , o) is generated
Sade [15] suggested creating a quasigroup (Q, o) from an
from isotopy from f = id and g. g is a permutation of order 2
admissible group (Q, +) and a complete mapping Ө by
as it is a transposition.
defining
In addition, if Lx , Rx denote the left and right
x o y = Ө(x - y) + y , for x , y ∈ Q (18)
multiplication by x ε Q and Lx and Rx are in Mult (Q) for
Example all x € Q. Since for g(a)=a, we see that La ∈ Mult (Q) for a
Let (Q, +) = (Z23 , o ) and let Ө (< X2 , X1 , X0 >) is given as ∈ Q. This now implies that La is a cycle of length n = |Q|. g
< X1 XOR 1, X0 , X2 > if X2 = 0 and La together generate the entire symmetric group of size
< X1 XOR 1 , X0 XOR 1, X2 > if X2 = 1 n. Therefore, Sym(Q) = Mult(Q) and this implies that (Q ,
(19) o) is highly non-associative.
Then Ө and (i XOR Ө) are permutations as shown in the For example, let (Q , +) = (Z8 , +) and let g : Q → Q be a
following figure transposition such that g interchanges the elements 0 and 1.
x θ(x) (iXORθ )( x) Then (Q , o) created by x o y = x + g(y) is a highly non-
< 0,0,0 > < 1,0,0 > < 1,0 ,0 > associative quasigroup. Here, we have g(0) = 1 , g(1) = 0
,and g(a) = a for a >1
< 0 ,0 ,1 > < 1,1,0 > < 1,1,1 >
xoy = (x + g(y)) mod 8 (20)
< 0 ,1,0 > < 0,0,0 > < 0,1,0 > So a structured highly non-associative quasigroup can be
< 0,1,1 > < 0 ,1,0 > < 0,0 ,1 > created as
< 1,0,0 > < 1,1,1 > < 0,1,1 > o 0 1 2 3 4 5 6 7
< 1,0,1 > < 1,0,1 > < 0,0,0 > 0 1 0 2 3 4 5 6 7
< 1,1,0 > < 0,1,1 > < 1,0,1 > 1 2 1 3 4 5 6 7 0

2 3 2 4 5 6 7 0 1
< 1,1,1 > < 0 ,0 ,1 > < 1,1,0 >
Figure 8. Ө and i XOR Ө on Z23 3 4 3 5 6 7 0 1 2
4 5 4 6 7 0 1 2 3
If the elements of (Z23 , XOR) are represented as integers
corresponding to binary strings for ease of notation , then Ө 5 6 5 7 0 1 2 3 4
and i - Ө are shown in Figure 9 and the quasigroup created 6 7 6 0 1 2 3 4 5
by Ө(x XOR y) XOR y is shown in Figure 10. 7 0 7 1 2 3 4 5 6
x 0 1 2 3 4 5 6 7 Figure 11. A Structured Non-associative Quasigroup
θ ( x) 4 6 0 2 7 5 3 1 This quasigroup is highly non-commutative and non-
(i XOR θ )( x) 4 7 2 1 3 0 5 6 associative. We can see from the above example that under
Figure 9. Values of Ө and i XOR Ө on integer the operation ‘o’,
corresponding to elements of Z23 For x = 0 and y = 1 x o y = 0 and y o x = 2
For x = 1 and y = 2 x o y = 3 and y o x = 2
4. Application of QG to Encryption
Quasigroups are suitable for designing efficient as well as
secure encryption schemes due to simplicity of the basic
operation. Moreover, large and unstructured quasigroups
can be generated using one of the schemes explained in the
previous sections. Use of such quasigroups helps to improve
the security of the scheme. A basic encryption and
decryption scheme based on table lookup has been described
in Sections 1.2.4 to 1.2.6. In this scheme the leader L acts as
a key to the encryption algorithm and has to be changed
frequently. Figure 14. Encrypted Image using different Leaders and
In addition to the leader, changing the quasigroup also Quasigroups
helps to improve the security of the scheme. This can be
practically made possible [11] without having to transmit
much of information as the key to the receiver. The
5. Application of QGs to Hashing
observations below is of (a) Original Tank Image (b) 5.1 Definitions
Encrypted Image using same quasigroup but 256 different A function H( ) is said to be One Way Hash Function
leaders (one for encrypting each row) and (c) Encrypted (OWHF) if it maps an arbitrary length message M to a fixed
image using different leaders and different Quasigroups. length hash value H(M) and satisfies the following
To do away with the effects of high levels of redundancy properties [17]:
found in visual data, a new scheme has been proposed [12] • No secret information is required for operation of H( )
using lookup and shift operations so that subsequent chunks and description of H( ) is publicly known.
of similar data in the plain image are mapped into different • It is easy to compute H(M) if M is known.
blocks using the same encryption function. Other efficient
schemes using a combination of quasigroup operations
• If H(M) is given in the range of H( ), it is hard to find a
message M for given H(M), and given M and H(M), it
together with basic cryptographic primitives are also under
is hard to find a message M’(≠ M) such that H(M’) =
development.
H(M).
A function H( ) is said to be Collision Free Hash Function
(CFHF) if it maps an arbitrary length message M to a fixed
length hash value and satisfies the following property in
addition to the above three properties
• It is hard to find two distinct messages M and M’ that
hash to the same result (H(M) = H(M’)).
5.2 Construction of Hash Function Based on QG

For a quasigroup (Q, .), a hash function HQ( ) can be
defined as
H Q ( q1 q2 ... qn ) = ((...(a .q1 ).q2 ....).qn ) (21)
Figure 12. Original Tank Image
where ‘a’ is a fixed element of Q. The original
multiplication table of a quasigroup may be modified using
homotopy that permutes the rows and columns resulting in a
different multiplication table.
u.v = γ (α (u ).β ( v)) ∀u, v ∈ Q (22)
The above three permutations are generated and used for
construction of new quasigroups for hashing. It is possible to
calculate the results without storing the table, therefore,
large quasigroups may be used for design of hash functions
suitable for cryptographic applications.
Other schemes based on generation of random
quasigroups with multiple lookups have been successfully
Figure 13. Encrypted Image using same QG but 256 used for hashing. New quasigroups are constructed using
Leaders non-linear transformations involving logical OR, XOR and
circular shift operations. Computationally efficient hashing
schemes based on these simple operations have been
designed that also ensure collision & pre-image resistance.
6. Design & Implementation Doctoral Dissertation, Iowa State University Ames,

Our design and implementation work includes the Iowa, 2006, [Online] Available:
following: http://orion.math.iastate.edu/dept/thesisarchive/PHD/
(1) Generating Orthogonal Latin Square for a given KMeyerPhDSp06.pdf [Accessed: Feb. 24, 2008].
Latin Square. [10] C.Z. Koscielny, “Generating Quasigroups for
(2) Encryption and Decryption using Key-based Random Cryptographic Applications”, International Journal of
Permutation. Applied Mathematics & Computer Science, Vol. 12,
(3) Generation of Isotopes of a Latin Square. No. 4, pp. 559-569, 2002.
(4) Construction of a huge Latin Square using the Simple [11] S.K. Pal, S. Kapoor, A. Arora, R. Chaudhary, J.
product of two Latin Squares, Affine Mapping, etc. Khurana, “Design of Strong Cryptography Schemes
(5) Encryption/Decryption using Latin Square and Look- based on Latin Squares”, Proceedings of the Pre-ICM
up Table method. International Convention on Mathematical Sciences,
(6) Generation of a non-associative Quasigroup with help New Delhi, 2008.
of Cayley’s table and T-functions. [12]S.M. Hussain, N.M. Ajlouni, “Key Based Random
(7) Improving the basic encryption scheme to handle Permutation”, Journal of Computer Science, Vol. 2,
redundancy in multimedia. No. 5, pp. 419-421, 2006.
These schemes have been implemented under the [13] A. Klimov, A. Shamir, “A New Class of Invertible
Microsoft VC++ and Matlab programming environment Mappings”, CHES, LNCS-2523, 2002.
on a high-end stand-alone personal computer. [14] A. Klimov, A. Shamir, “Cryptographic Applications of
T-functions”, Selected Areas in Cryptography, SAC-
7. Conclusions 2003, LNCS-3006, Springer Verlag, pp. 248-261, 2003.
Suitability of quasigroup based structures for encryption [15] A. Sade, “Quasigroupes Automorphes par le Groupe
and hashing were established in this paper with examples, Cyclique”, Canadian Journal of Mathematics, 9,
implementations and favorable observations and results. pp. 321-335, 1957.
Construction of large and unstructured quasigroups is useful [16] S.K. Pal, Sumitra, “Development of Efficient
for design of encryption schemes with large key space. Algorithms for Quasigroup Generation and
Similarly, highly non-associative quasigroups find Encryption”, Proceedings of the 2009 IEEE
application in the design of efficient hash functions. International Advance Computing Conference, pp.
Modifications in the basic operations and combination with 2529-2534, 2009.
other cryptographic primitives in order to further improve [17] V. Snasel, A. Abraham, J. Dvorsky, P. Kromer, J.
the security would be taken up as our future work in this Platos, “Hash Function Based on Large Quasigroups”,
direction. International Conference on Computational Science
(ICSS 2009), Lousiana, USA, LNCS 5544, pp. 521-529,
2009.
References [18] S. Markovski, D. Gligoroski, V. Bakeva, “Quasigroup
[1] R. Bose, B. Manvel, Introduction to Combinatorial and Hash Functions”, Discrete Mathematics &
Theory, John Wiley & Sons, 1984. Applications, Proceedings of the 6th ICDMA, Bansko,
[2] C.F. Laywine, G.L. Mullen, Discrete Mathematics pp. 43-50, 2001.
using Latin Square, Wiley Interscience, 1998. [19] S. K. Pal, D. Bhardwaj, R. Kumar, V. Bhatia, “A New
[3] R. Alter, “How Many Latin Squares Are There?”, Cryptographic Hash Function based on Latin Squares
American Mathematical Monthly, 82, pp. 632-634, and Non-Linear Transformation”, Proceedings of the
1975. 2009 IEEE International Advance Computing
[4] S. Markovski, D. Gligoroski, V. Bakeva, “Quasigroup Conference, pp. 2529-2534, 2009.
String Processing: Part 1”, Proc. of Maced. Acad. of
Sci. and Arts for Math. and Tech. Sci., XX 1-2, pp.
13–28, 1999.
[5] S. Markovski, V. Kusakatov, “Quasigroup String Authors Profile
Processing: Part 2”, Proc. of Maced. Acad. of Sci. and
Arts for Math. and Tech. Sci., XXI, 1-2, pp. 15–32,
Saibal K. Pal received the M.S. degree in Computer Science from
2000.
University of Allahabad in 1990 and PhD from University of Delhi
[6] S. Markovski, D. Gligoroski, S. Andova, “Using in the area of Information Security. He is presently with DRDO,
Quasigroups for One-one Secure Encoding”, Proc. Delhi. His areas of interest include Cryptography, Information
VIII Conf. Logic and Computer Science, LIRA ’97, Hiding, Signal Processing and Soft Computing.
Novi Sad, pp. 157–162, 1997.
[7] S. Markovski, D. Gligoroski, B. Stojˇcevska, “Secure Shivam Kapoor is presently pursuing his Masters degree in
Two-way On-line Communication by using Quasigroup Computer Applications (MCA) from the Department of Computer
Enciphering with Almost Public key”, Novi Sad Science, University of Delhi. His areas of interest include Discrete
Journal of Mathematics, 30, No 2, 2000. Mathematics & Combinatorics, Cryptography and Design of
Algorithms.
[8] B. Schneier, Applied Cryptography (Second Edition),
John Wiley & Sons, 1996.
[9] K.A. Meyer, “A New Message Authentication Code
Based on the Non-associativity of Quasigroups”,
Column Vectorizing Algorithms for Support Vector

Machines
Chen ZhiYuan1, Dino Isa2 and Peter Blanchfield3
1
University of Nottingham, School of Computer Science,
Jalan Broga 43500 Semenyih Selangor Malaysia
eyx6czy@nottingham.edu.my
2
School of Electronic Engineering, University of Nottingham,
Jalan Broga 43500 Semenyih Selangor Malaysia
Dino.Isa@nottingham.edu.my
3
School of Computer Science, University of Nottingham,
Nottingham, NG8 1BB, UK
pxb@Cs.Nott.AC.UK
construction of the support vector machine.

Abstract: In this paper we present the vectorization method for
support vector machines in a hybrid Data Mining and Case- The rest of this paper is organized as follows: Section 2
Based Reasoning system which incorporates a vector model to presents objectives and related techniques. Section 3
help transfer textual information to numerical vector in order to describes in detail the architecture of the hybrid system.
make the real world information more adapted to the data Section 4 provides the procedure of vectorization. Section 5
mining engine. The main issue of implementing this approach is explains the conducted experiments. The conclusion is
two algorithms; the discrete vectorization algorithm and
discussed in section 6.
continuous vectorization algorithm. The basic idea of the
vectorization algorithm is to derive X value from the original
column value and where the vector value is unavailable; the 2. Objectives and Foundation
algorithm builds a vector table based on the X value by using
Our research group works on the designing of flexible and
appropriate functions. Subsequently, the vector model is
classified using a support vector machine and retrieved from the adaptable user oriented hybrid systems which aims to
case based reasoning cycle using a self organizing map. combine database technology and artificial intelligence
techniques. The preprocessing procedure related to data
Keywords: Vectorization, Support Vector Machine, Data
Mining, Artificial Intelligence, Case-Based Reasoning.
vectorization step of a classification process, going from low
level data mining processes [2] to high level artificial
1. Introduction intelligence techniques. Many domain specific system such
as user modeling systems [3] or artificial intelligence hybrid
The problem faced by traditional database technology systems have been described in literature [4] [5] [6]. Even
developer today is lack of intelligence support, while when the applied strategies are designed as generic as
artificial intelligence techniques [1] were limited in their possible, the illustration given for the system are limited to
capacity to supply and maintain large amount of factual the text document and do not develop any vectorizing
data. This paper provides a method to solve this problem. algorithm to quantitate the input raw textual data set into
From a database point of view, there was an urgent need to numeric data set.
address the problems caused by the limited intelligent Actually, to the best of our knowledge, no such complete
capabilities of database systems, in particular relational and generic vectorization process exists because of the
database systems. Such limitations implied the impossibility necessity to have an excellent know-how in the
of developing, in a pure database context, certain facilities implementation of a hybrid intelligent system. Many existed
for reasoning, problem solving, and question answering. systems have been developed on the basis of using artificial
From an artificial intelligence point of view, it was intelligence techniques to provide semantic support to a
necessary to transcend the era of the operating on numerical database system, or database techniques to aid an artificial
signals to achieve the real information management system intelligence system to deal with large amounts of
able to deal with large amounts of textual data. Our information. The key factors they concerned reside in the
approach was explicitly designed to support efficient exploitation of the equivalence between database field and
vectorization techniques by providing multiple number the knowledge representation system of artificial
resources with minimum inter-dependencies and irregular intelligence.
constraints, yet under strict artificial intelligence In our hybrid system, vector is the unique representation of
considerations. It features a table in a relational database data considering the system consistency. On the other hand,
through two types of vectorizing functions, supporting to the for both data mining process and case-based reasoning cycle
[7], vectorization and consistency are crucial. The role of 4. Vectorization

vectorization is to convert text table which stored in SQL
As can be seen from the hybrid system architecture, in order
server, into numerical vector form. Traditional vectorization
to classify individual models and domain information into
method concentrates on image object into a raster vector or
user model the support vector machine are applied.
raw line fragments. While we focus on these table column
Individual models are user information which took table
features and describe how they can be vectorized by applied format and stored in the SQL server. Domain information in
automatically approach using two kinds of vectorization the database is also sorts of tables which stored the
functions. preselected user-preferred knowledge. The support vector
In order to describe the foundation of the vectorization, the machine [10] [11] is one of AI techniques which serve as
framework of our hybrid system is simply described in the classifier in the system. The main idea of a support vector
following section. machine is to construct a hyper plane as the decision
surfaces in such a way that the margin of separation between
3. Hybrid System Architecture Overview positive and negative features is maximized. The
vectorization step is the data preprocessing for the support
The concepts of this project are as follows: vector machine which provides the numeric feature vector.
• To develop a hybrid data mining and case-based
reasoning user modeling system 3.1 Feature Type
• To combine data mining technology and artificial
For vectorization task to be as accurate as possible we
intelligence pattern classifiers as a means to construct a
predefined two type table columns or we called feature type;
Knowledge Base and to link this to the case-based
discrete columns (feature) and continuous columns (feature).
reasoning cycle in order to provide domain specific user
Discrete feature contains discrete values, in that the data
relevant information to the user in a timely manner.
represents a finite, counted number of categories. The values
• To use the self organizing map [8] in the CBR cycle in
in a discrete attribute column do not imply ordered data,
order to retrieve the most relevant information for the
even if the values are numeric; the distinct character is
user from the knowledge base.
values are clearly separated. Telephone area code is a good
Based on these concepts the architecture has been designed
example of discrete data that is numeric.
which is illustrated in Figure 1. The hybrid system contains
Continuous feature contains values that represent a
five main components:
continuous set of numeric and measurement data, and it is
• Individual models, comparable to the blackboard
possible for the data to contain an infinite number of
containing the user information from the real world.
fractional values. An income column is an example of a
• Domain database integrated the preselected domain
continuous column.
information [9].
The numeric value is not the vital factor to determine the
• A data mining engine which classified both user class feature type, but if the value is a word then it must be a
and domain information vectors. discrete feature.
• A knowledge base, containing the representation of
classified user information and combined with interested 3.2 Vectorization algorithm
domain knowledge. From the technology point of view, vectorization is an
• A problem-solving life-cycle called case-based reasoning approach modeling relationships between the data set and
cycle, assisting in retrieve reuse revise and retain the the vectorizing variable. We provide a more flexible
knowledge base. approach by allowing some of the features (columns) to be
independent and some of the features to be interdependent.
Constructing two parallel algorithms to avoid time
User interface
User ID SOM Retrieved
consuming and save a large amount of effort.
Query
RETRIEVE
Case The schema of the algorithm is specified in Figure 2 which
derives the numeric vector by implementing different
RETAIN
R functions. The schema is not exhaustive and can evolve with
Data mining engine User Model Knowledge E
SVM Base
U new data, according to user need.
S
Human E Furthermore, once the type of the column has been
expert
determined, adding a new record is quite straightforward.
Vectorization These functions are also well suited to dealing with
incomplete data. Instances with missing attributes can be
REVISE
Confirmed Proposed handled by summing or integrating the values of other
Solution Solution
Domain Individual attribute.
Database Model We represent each column as a data point in a dimensional
space, where Z is the total number of attributes (columns).
Data Mining User Model CBR The algorithm computes the vectorizing value (or
representation value) between each feature which was
Figure 1. The architecture of the system denoted by abscissa axis and the vector denoted by y-axis,
and all the feature values determine its own vectorizing
values. Once the vectorizing value list is obtained, the vector
model will be classified based on the implementation of
support vector machine so that the core of the hybrid system

the knowledge base will be constructed completely. The key computation of these two algorithms is the
vectorization value formula given in step 5 of the both table.
Formula 1:
Vdx = nd
Vdy = 1 × nd
n
Formula 2:
′
Vcx = (Vcx − AvgVcx ) MaxV
cx
−x
Vcy′ = e − e
x
, Vcy′ ∈ [−1,+1].
e x + e−x
In Formula 1, n is the weight parameter associated with the
discrete columns which is the sum of value type. Vdy is a
combination of the unit value ( 1 / n ) multiply the sequence
of the current value type ( nd ). This is a regression-like
expression [12]. Regression is used to make predictions for
numerical targets. By far the most widely used approach for
Figure 2. The schema of the vectorization algorithm
numerical prediction is regression, a statistical methodology
that was developed by Sir Frances Galeton [13]. Generally
The detailed vectorization algorithms are described in the
speaking Regression analysis methods include Linear
Table 1 and Table 2 according to discrete columns and
Regression, Nonlinear Regression. Linear Regression is
continuous columns.
widely used, owing largely to its simplicity. By applying
transformations to the variables, we can convert the
Table 1. The discrete column vectorization algorithm nonlinear model (text table column information) into a
linear one according to the requirement of the support vector
machine.
1: Let V be the representation of Vectors, D be the In order to get the negative X value and at the same time
whole set of the vector model and d be the set of discrete keep the same distance among original X value, in Formula
columns. 2 we minus average value to all x value and then get the
2: FOR each data point Z DO proportion compare with the maximum original X value,
3: Select Zd , the discrete features of all data point, after that get the new X value and by means of Hyperbolic
Tangent function [14] to map these new value into (-1, +1)
4: Compute Vd = ( Vdx , Vdy ), the corresponding value
scale.
between Z and every vector, ( Vdx , Vdy ) D. In order to explain these algorithms clearly, we show the
experiment procedure in the following section.
5: Vdx = nd , Vdy = 1 × nd ; Vdy ∈ [0,1] .
n
5. Experiments
6: END FOR
The vectorization algorithm was tested on the census-
income data set extracted from the 1994 and 1995 current
Table 2. The continuous column vectorization algorithm population surveys conducted by the U.S. Census Bureau.
The data contains 41 demographic and employment related
variables. In order to explain how to apply our approach
clearly, we choose 8 discrete columns and 8 continuous
1: Let V be the representation of Vectors, D be the set of
columns which can be found in table 3 to explain the
vector model and c be the set of continuous columns.
implementation in details. In Table 4 we list the n value of
2: FOR each data point Z DO
the discrete columns. For example the worker class n value,
3: Select Zc , the continuous features of all data point, because there are 9 kinds of worker class, so n is equal to 9.
′ ′ Parts of the experiment results implemented the proposed
4: Vc = ( Vcx , Vcy ), the corresponding value
Compute algorithms which contain 27 records are shown in figure 3.
′ ′ The input for the algorithm was given 8 discrete features
between Z and every vector, ( Vcx , Vcy ) D. and 8 features and asked to give the vectorized value as
′ output. The discrete attributes were decomposed into n
5: Vcx =（ Vcx − AvgVcx） MaxVcx , equidistances, which yielded corresponded vector value
−x
Vcy′ = e − e
x scaling to the range of (0, 1). For the continuous attribute,
, Vcy′ ∈ [−1,+1].
e x + e−x firstly the raw attribute value was transferred into the whole
6: END FOR x-axis, so that the new x value contain the negative value
and by using Hyperbolic Tangent function the vector value

was calculated. The Hyperbolic Tangent function make sure
All the experiment results was created on PC computer,
the vector value to be projected into the (-1, 1) scale which
CPU Intel(R) Core(TM) Duo CPU T2250 @ 1.73GHz 4.6
is required by support vector machine.
2.3, 2GB RAM DDR2 667 MHz, with WinXP. Program was
compiled with NetBeans 6.0.
6. Conclusions
The proposed hybrid Data Mining and Case-Based
Reasoning User modeling system is a multi purpose
platform and is characterized by three major processes. The
vectorization processing unit communicate through the raw
data set the SQL table and the output is the numeric vector,
such an approach avoid the data inconsistency usually met
in classifying documents chain when implement artificial
intelligence tools.
In this paper we built vectorization model by applying two
algorithms: The discrete vectorization algorithm and
continuous vectorization algorithm. The advantage of using
discrete algorithm is that each record in the whole table was
assigned a vector value in an easily expression calculation.
While for the continuous column we choose a relatively
Figure 3. Part of the experiment results T complicated formula that is the Hyperbolic Tangent function
to achieve the vector value.
The recommend vector value range is (0, 1) or (-1, 1) for In designing the algorithm, the key consideration is to bring
support vector machine [15]. One reason for this is to avoid up easy scientific numerical transformation. Therefore, the
vector value in great numeric ranges dominates those in formulas in the algorithm are quite basic but the impressive
smaller numeric ranges. Another reason is to avoid the part is it also provides a reasonable balance between a
numerical difficulties during the calculation. Because kernel satisfactory result and reasonable processing time. Secondly
values usually depends on the inner products of feature due to the modular structure of the algorithm it can be
vectors. For example the linear kernel and the polynomial adapted easily for application. The results of the algorithm
kernel, large vector values may cause numerical problems in the experiments labeled clean and the vector points
[16]. generated by our algorithm have a standard coverage (0, 1)
Another reason why we proposed two kinds of algorithm to and (-1, 1) which is useful in fulfilling the classification task
vectorize discrete columns and continuous columns is to by means of support vector machine for the hybrid system.
preserve the character of the column for the sake of the later
analysis.
Table 3. Parameters for experiments References
[1] S. J. Russell, P. Norvig, Artificial Intelligence A
Discrete columns Continuous columns
Modern Approach, Prentice-Hall International Inc,
class of worker* age*
1995.
education* wage per hour
[2] U. Fayyad, G. Paitetsky-Shapiro, P. Smith,
marital stat capital gains
“knowledge discovery and data mining: Towards a
sex capital losses
unifying framework”, proceedings of the International
reason for unemployment dividends from stocks
Conference on Knowledge Discovery and Data
family members under 18 person for employer
Mining, 1996, pp. 82-22.
live in this house 1 year ago weeks worked in year
[3] J. Vassileva, "A practical architecture for user
veterans benefits instance weight*
modeling in a hypermedia-based information system",
Proceedings of Fourth International Conference on
Table 4. The discrete column n value User Modeling, Hyannis, MA, August 1994, pp 15-
19.
Discrete columns n Value [4] I.V. Chepegin, L. Aroyo, P. D. Bra, “Ontology-driven
class of worker 9 User Modeling for Modular User Adaptive Systems”,
education 17 LWA, 2004, pp.17-19.
marital stat 7 [5] I. Watson, Applying Case-Based Reasoning:
sex 2 Techniques for Enterprise Systems, Morgan
reason for unemployment 6 Kaufmann Publishers, Inc., San Francisco, CA, 1997.
family members under 18 5 [6] K. Sycara, “CADET: A cased-based synthesis tool for
live in this house 1 year ago 3 engineering design”, International Journal for Expert
Veterans benefits 3 System, 4(2), 1992, pp.157-188.
[7] A. Aamodt, E. Plaza, “Case-based reasoning:

foundational issues, Methodological variations, and Peter Blanchfield is a senior tutor in the
system approaches”, AI communications, 7(1), 1994, School of Computer Science, University of
pp. 39-59. Nottingham. From September 2005 to July
2009 he was Director of the IT Institute in
[8] Kohonen, “self-organizing map using contiguity-
the School, before which he was the
constrained clustering”, Pattern Recognition Letters, Director of Computer Science and IT
1995, pp. 399–408. Division at the Malaysia Campus of the
[9] B. Hjorland, H. Albrechtsen, “Toward A New Horizon University of Nottingham. In that role he was involved in setting
in Information Science: Domain Analysis”, Journal of up the activities of the School there along with the activities of
the American Society for Information Science, 1995, what has become the Engineering Faculty on that campus.
46(6), 400-425.
[10] E. Osuna, “Support Vector Machines: Training and
Applications”, Ph.D thesis, Operations Research
Center, MIT, 1998.
[11] V.N. Vapink, Statistical Learning Theory, New
York:Wiley.
[12] D.V. Lindley, "Regression and correlation analysis,"
New Palgrave: A Dictionary of Economics, v. 4,
1987, pp. 120-23.
[13] F. Galeton,“Typical laws of heredity", Nature
15,1877, pp. 492-495, 512-514, 532-533.
[14] M.A. Abdou, A.A. Soliman, “Modified extended
tanh-function method and its application on nonlinear
physical equations”, Physics Letters A, Volume 353,
Issue 6, 15 May 2006, pp. 487-492
[15] E. Osuna, R. Freund, F. Girosi, “Improved training
algorithm for support vector machine”, IEEE Neural
Networks in Signal Processing 97,1997.
[16] C. Cortes, V. Vapnik, “Support-vector network”,
Machine Learning , 1995, pp. 273–297.
Authors Profile
Chen ZhiYuan received the B.A. in

Economics from University of HeiLongJiang
in 2001 (China). During 2006-2010, she
stayed in University of Nottingham,
Malaysia Campus to do PhD research in
imitate human experts (especially in
manufacturing and medical field) to
perceive the environment and to make
decisions which maximize the chance of success. From 2007 to
2009, she stayed in Supercapacitor Research Laboratory (SRL),
which is supported by Ministry of Science Technology and
Inovation of Malaysia to study knowledge management system for
manufacturing enviroment.
Dino Isa is a Professor in the Department of

Electrical Electronics Engineering,
University of Nottingham Malaysian
Campus. He obtained a BSEE (Hons) from
the University of Tennessee, USA in 1986
and a PhD from the University of
Nottingham, University Park Nottingham,
UK in 1991.nnnThe main aim of his research is to formulate
strategies which lead to the successful implementations of
“Intelligent Systems” in various domains.
Impact and Performance Analysis of Mobility

Models on Stressful Mobile WiMax Environments
N Vetrivelan1 Dr. A V Reddy2
1
Research Scholar, NIT, Trichy, India,caz0307@ nitt.edu
Assistant Professor, Department of Computer Applications
Periyar Maniammai University, Thanjavur, Tamilnadu, India
nvetri@yahoo.com
2
Professor, Department of Computer Applications
National Institute of Technology, Trichy, India
reddy@nitt.edu
Abstract: In this study, we have used the design and wireless Routing protocol in WiMax module has been
implementation of a new medium access control protocol with chosen and compared.
WiMax, based on the IEEE 802.16. We have compared and
analysed three different mobility models and the impact of
In Section 2, the related works have been discussed. Brief
mobility on mobile WiMax environments. We have chosen description of Mobile node and WiMax Module has been
Gauss-Markov, Manhattan Grid and Random WayPoint presented in Section 3. The Mobility models described in 4.
Mobility models with DSR routing protocol in WiMax Protocol description has been given in Section 5. The
environments. The parameter metrics Packet Delivery Fraction, evaluation methodologies have been given in Section 6. In
Routing load, Throughput and Latency have been taken into Section 7, the Simulation Parameters and Parameter Values
account. Our ns-2 simulation result shows that the functioning of WiMax V2.03 have been described. Results and
of mobility models will greatly influence the performance of Discussion presented in Section 8. The conclusion has been
WiMax environments. The result reveals that the throughput, presented in Section 9.
latency and routing load are high in Gauss Markov Mobility
model. It also shows that packet delivery fraction is high in
Manhattan Grid model. Compared to Manhattan and Gauss
2. Related Work
Markov models the random waypoint models is stable.
Tracy Camp, Jeff Boleng [1] surveyed the mobility models
that are used in the simulations of Ad hoc networks. Authors
Keywords: Mobility, DSR, WiMax, MAC and Simulation
described several mobility models that represent mobile
1. Introduction nodes whose movements are independent of each other (i.e,
entity mobility models) and several mobility models that
represent mobile nodes whose movements are dependent on
WiMax [3] is the short form of the Worldwide
each other ( i.e. group mobility models.) This paper presents
Interoperability for Microwave Access. Typically, fixed
a number of mobility models in order to offer researchers
WiMax networks have a higher-gain directional antenna
more informed choices when they are deciding upon a
installed near the client which results in greatly increased
mobility model to use in their performance evaluations.
range and throughput. Mobile WiMax networks are usually
Illustrated how the performance results of an ad hoc network
made of indoor customer premises equipments (CPE) such
protocol drastically change as a result of changing the
as desktop modems, compared to directional antennas but
mobility model simulated.
they are more portable.
Per Johansson, Tony Larsson compared three routing
The mobility model [1] is designed to describe the protocols for wireless mobile ad hoc network. They have
movement pattern of mobile users, and how their location, done simulation on a scenario where nodes move randomly.
velocity and acceleration change over time. Since mobility Furthermore, three realistic scenarios were introduced to test
patterns plays a significant role in determining the protocol the protocols in more specialized contexts. In most
performance, it is desirable for mobility models to emulate simulations the reactive protocols (AODV,DSR) performed
the movement pattern significantly better than DSDV. At moderate traffic load
of targeted real life applications in a reasonable way. We DSR performed better than AODV for all tested mobility
have provided a categorization for various mobility models values, while AODV performed better than DSR at higher
onto several classes based on their specific mobility traffic loads.
characteristics. For some mobility models, the movement of
the WiMax node is likely to be affected by its movement Leonardo Betancur [5] performed experiments for WiMax
history. The authors are aware that this performance Channel-Phy Model in ns-2. Their work has been on a novel
comparison of mobility scenarios has not attempted in proposal for PHY layer and propagation model that allowed
WiMax Environments or IEEE 802.16 module. That is faster and more detailed the link level execution. They
why, the performance of mobility scenarios using DSR described the development process for model and the
implementation details in ns-2 for PMP topologies. They Jenhui Chen, Chih-Chieh Wang, [4] presented detailed
described the development of physical layer model based on design and implementation of WiMax module based on the
the IEEE 802.16 standard also known as WiMax using ns-2. IEEE 802.16 broadband wireless access networks (BWANs)
Their work presented a statistical equivalent model for ns-2, or WiMax module with the point-to-multipoint (PMP) mode
which consider the channel effects in WiMax networks for for the ns-2. They implemented modules comprised
upper layer simulations. Through simulation they reached fundamental functions of the service-specific convergence
the conclusion that their model reduced computational effort sub layer (CS), the MAC Common part sub layer (CPS), and
in link level simulations, without using complex bit-bit the PHY layer. A simple call admission control (CAC)
simulations also it is necessary to analyze the performance mechanism and the scheduler have also been included in
of WiMax networks and found more realistic coverage this module.
areas.
PKT
Sent
Application Routing Agent
Traffic Generating Agent

Link Layer ARP
Queue
Link Layer
ok Queue
Not ok MAC
decides
OK ?
P ROPOSED WIMAX
Radio Propagation
Network Interface Model
Propagatio Network
n Model Interface
Channel Channel
Figure 1. Conversion of Mobile node in to WiMax node

3. Conversion of Mobile Node in to WiMax two lanes for each direction (North and South direction for
Node vertical streets, East and West for horizontal streets). The
WiMax node is allowed to move along the grid of horizontal
Each mobile node has made use of a routing agent for the and vertical streets on the map. At an intersection of a
purpose of calculating routes to other nodes in the ad-hoc horizontal and vertical street, the WiMax node can turn left,
network. Packets were sent from the application and were right or go straight. This choice is probabilistic: the
received by the routing agent. The agent decided a path that probability of moving on the same street is 0.5, the
the packet must travel in order to reach its destination and probability of turning left is 0.25 and probability of turning
stamped it with this information. It then sent the packet right is 0.25. The velocity of a mobile node at a time slot is
down to the link layer. The link layer level used an Address dependent on its velocity at the previous time slot. Also, a
Resolution Protocol to decide the hardware addresses of node’s velocity is restricted by the velocity of the node
neighboring nodes and map IP addresses to their correct preceding it on the same lane of the street.
interfaces. When this information was known, the packet
was sent down to the interface queue and awaited a signal 4.3 Random Way Point
from the Multiple Access Control (MAC) protocol. When
the MAC layer decided it was ok to send it on to the The Random Waypoint Mobility model includes pause
channel. The propagation model used the transmit and times between changes in direction and/or speed. An
received stamps to determine the power with which the WiMax [3] node begins by staying in one location for a
interface would receive the packet. certain period of time (i.e pause time) Once this time
expires, the WiMax node chooses a random destination in
the simulation area and speed that is uniformly distributed
A mobile node has been converted to WiMax node as between minimum and maximum speed. The node then
discussed subsequently. The 802.16 based WiMax module travels toward the newly chosen destination at the selected
[3][4] consisted of Mac 802_16, Common and Queue have speed. Upon arrival, the node pauses for a specified time
been in accordance with the specifications of the IEEE period before starting the process again. This is a memory-
802.16-2004 standard and based on the ns-2 version 2.29. less mobility pattern because it retains no knowledge
An Object oriented programming language C++ were concerning its past locations and speed values. The current
developed for classes. The relationship between WiMax speed and direction of a node is independent of its past
module and ns-2 modules was represented in the stack of the speed and direction. This characteristic can generate
ns-2 is as shown in Fig.1. It consists of the type of objects unrealistic movements such as sudden stops and sharp turns.
for the traffic generating agent (TGA), the link layer (LL),
the interface queue (IFQ), the designed MAC layer (WiMax 5. Protocol description
module), and the PHY layer (Channel). The implemented
module comprised fundamental function of the service- 5.1 Dynamic Source Routing
specific convergence sublayer (CS), the MAC Common part
sublayer (CPS) and the PHY layer. A simple call admission The key distinguishing feature of DSR [10] is the use of
control mechanism and the scheduler have also been source routing. That is, the sender knows the complete hop-
included in this module by-hop route to the destination. These routes are stored in a
route cache. The data packets carry the source route in the
packet header. When a node in the ad hoc network attempts
4. Mobility Models to send a data packet to a destination for which it does not
already know the route, it uses a route discovery process to
4.1 Gauss Markov Model dynamically determine such a route. Route discovery works
by flooding the network with route request (RREQ) packets.
This model [1] was designed to adapt to different levels of Each node receiving an RREQ rebroadcasts it, unless it is
randomness. Initially each node is assigned a current speed the destination or it has a route to the destination in its route
and direction. At fixed intervals of time, movement occurs cache. Such a node replies to the RREQ with a route reply
by updating the speed and direction of each node. (RREP) packet that is routed back to the original source.
Specifically, the value of speed and direction at the nth RREQ and RREP packets are also source routed. The RREQ
instance is calculated based upon the value of speed and builds up the path traversed across the network. The RREP
direction at the (n-1)th instance. The main advantages of this routes itself back to the source by traversing this path
model are that it eliminates the sudden stops, sharp turns backward. The route carried back by the RREP packet is
present in Random way point mobility model and is close to cached at the source for future use. If any link on a source
being realistic. route is broken, the source node is notified using a route
error (RERR) packet. The source removes any route using
4.2 Manhattan Mobility Model this link form its cache. A new route discovery process must
be initiated by the source if this route is still needed. DSR
The Manhattan model [1] is used to emulate the movement makes very aggressive use of source routing and route
pattern of mobile nodes on streets defined by maps. The caching.
Maps are used in this model too. The map is composed of a
number of horizontal and vertical streets. Each street has
6. Evaluation Methodology 7.1 Simulation Model
To evaluate the three mobility models in WiMax, we used In this section, the network simulation was implemented
four performance metrics to compare and analyse the using the NS-2 simulation tool. The Network Simulator NS-2
realistic movements was a discrete event simulator. For simulation Scenario and
network topology creation it used OTCL (Object Tool
6.1 Packet Delivery Fraction Command Language). To create new objects, protocols and
routing algorithm or to modify them in NS-2, C++ source
The ratio of number of data packets successfully delivered to code used. The WiMax module consisted of the type of
the destination, generated by CBR Sources. objects for the traffic generating agent (TGA), the link layer
(LL), the interface queue (IFQ), the designed MAC layer
PDF=(Received Packets/Sent Packets)*100
(WiMax module), and the PHY layer. The simulations were
6.2 Routing Overhead conducted on Due Core processor at speed 3.0 GHz, 1 GB
RAM running Gygwin Environment.
It is an important metric for measuring scalability of a
protocol. The number of routing packet transmitted per data Simulation Mobility Model
packet delivered at destination. Each hop wise transmission
of a routing packet is counted as one transmission. Table 3 : Simulation Parameter of Mobility Models
Routing load = Packets sent/Received packet

Manhattan Random
Gauss Markov
6.3 Throughput Grid waypoint
x=1000.0 x=1000.0 x=1000.0
These are the number of packets sent from the source and the y=1000.0 y=1000.0 y=1000.0
number of packets received at the destination Duration Duration Duration
=300.0 =300.0 =300.0
6.4 Latency Update
Update
Frequency=2. Dim=3
The time, it takes for a packet to cross a network connection Dist=5.0
5
from sender to receiver
Maxspeed=4. TurnProb Minspeed
0 =0.5 =1.0
7. Simulation parameters
AngleStdDev SpeedChang Maxspeed
0.39269909 eProb=0.2 =4.0
Table 1: Simulation Parameter Values SpeedStdDev MinSpeed Maxpause
Routing Protocol DSR =0.5 =1.0 =60.0
MAC Layer IEEE 802.16 MeanSpeed
=1.0
Number of Mobile
100,200,300,400 SpeedStd
WiMax Nodes
Dev=0.2
Gauss Markov
Mobility Model Manhattan Grid
Random Waypoint
Transmission 8. Results and Discussion
Bandwidth of each 20
line We have used ns-2 with WiMax to compare the
Simulation time 300Secs performance of the mobility models. Four sets of results have
been presented vide 100,200,300 and 400 WiMax nodes. In
Traffic Type Constant Bit Rate
all stressful situations with communication channel 2,4,6,8
Antenna Type Omni Antenna
and 10 have been chosen. The routes of packets are
accomplished with the DSR.
Table 2 : Important Parameter Values of WiMax V2.03
Bandwidth 20MHz
Service flow Scheduling Type 11.13.11
SS_MAC_Address 48
Transaction_ID 16
Downlink frequency 32
Uplink channel_ID 8
30 400 Nodes
50 100 Nodes
40 20
F
30 D
PDF
P
20 10
10
0
0
2 4 6 8 10 2 4 6 8 10
Communication Channel
GAUSS-MARKOV
GAUSS-MARKOV MANHATTAN GRID
RANDOM WAY POINT
MANHATTAN GRID
RANDOM WAY POINT Figure 5. Mobility Models - PDF (%) with varied
communication channel in 400 WiMax Environments
Figure 2. Mobility Models – PDF (%) with varied
communication channel in 100 WiMax Environments As shown in the Figures 2,3,4,5, with respect to the PDF, of
all 100,200,300 and 400 WiMax nodes situation Manhattan
Grid Model out performs where as in the Random waypoint
there is decreasing trend. In the Manhattan Grid Model the
20 200 Nodes nodes have been chosen at a random destination either in
15
the horizontal or vertical directions. Due to the more
restricted movement of the nodes in the network which
PDF
10
leads to slightly lesser number of broken links and
5 subsequently lower the chance of getting a stray route error
message. At the same time it is found out that Gauss
0
2 4 6 8 10
Markov model shows steadiness in performance.
GAUSS-MARKOV
MANHATTAN GRID 30
RANDOM WAY POINT 100 Nodes
d
a
o20
Figure 3. Mobility Models – PDF (%) with varied L
g
communication channel in 200 WiMax Environments n
ti10
u
o
R
0
2 4 6 8 10
20 300 Nodes
Communicatio n Channel
15 GAUSS-MARKOV
MANHATTAN GRID
PDF
RANDOM WAY POINT

10
5 Figure 6. Mobility Models - Routing Load (packets) with

varied Communication channel in 100 WiMax
0
Environments
2 4 6 8 10
Communication channel
GAUSS-MARKOV 35 200 Nodes
MANHATTAN GRID 30
RANDOM WAY POINT
Routing Load
25
Figure 4. Mobility Models – PDF (%) with varied 20
communication channel in 300 WiMax Environments 15
10
5
0
2 4 6 8 10
GAUSS-MARKOV
MANHATTAN GRID
RANDOM WAY POINT
Figure 7. Mobility Models - Routing Load (packets) with

varied communication channel in 200 WiMax
Environments
50 700
300 Nodes
600
200 Nodes
40
Routing Load
500
Throughput
30
400
20 300
10 200
0 100
2 4 6 8 10 0
Comminucation Channel 2 4 6 8 10
GAUSS-MARKOV Communication Channel
MANHATTAN GRID
GAUSS-MARKOV
RANDOM WAY POINT MANHATTAN GRID
RANDOM WAY POINT
Figure 8. Mobility Models - Routing Load (packets) with
varied communication channel in 300 WiMax Figure 11. Mobility Models - Throughput (packets) with
Environments varied communication channel in 200 WiMax
Environments
80 400 Nodes
700
600
Routing Load
60 300 Nodes
Throughput
500
40 400
300
20 200
100
0 0
2 4 6 8 10 2 4 6 8 10
Communication Channel Communication Channel
GAUSS-MARKOV
MANHATTAN GRID GAUSS-MARKOV
RANDOM WAY POINT MANHATTAN GRID
RANDOM WAY POINT
Figure 9. Mobility Models - Routing Load (packets) with Figure 12. Mobility Models – Throughput (packets) with
varied communication channel in 400 WiMax varied communication channel in 300 WiMax
Environment Environments
As shown in the Figures 6,7,8,9, in all the four WiMax 700

600
400 Nodes
situations, as far as the routing load is concerned the
Throughput
500
Manhattan model outperforms compared to other two 400
models. It is more likely for the route error messages to be 300
dropped by some intermediate nodes before they actually 200
reach the intended source node. So, the source node, which 100
0
is unaware of the loss of route error messages, attributes any
2 4 6 8 10
genuine timeouts due to broken links. This mobility model Communication Channel
is choosing random destination either in the horizontal or GAUSS-MARKOV
vertical directions and also more restricted movement of the MANHATTAN GRID
RANDOM WAY POINT
nodes.
Figure 13. Mobility Models – Throughput (packets) with
700 varied Communication channel in 400 WiMax
100 Nodes
600 Environments
Throughput
500
400
As shown in the Figures 10,11,12,13, in all 100,200,300
300
200
and 400 WiMax nodes Gauss Markov model gives high
100 throughput whereas the Random waypoint and Manhattan
0 models perform with slight decreasing trend compared to
2 4 6 8 10
Gauss Markov model. This is because the Gauss Markov
GAUSS-MARKOV Mobility Model can eliminate the sudden stops and sharp
MANHATTAN GRID turns encountered. This Gauss Markov is a more realistic
RANDOM WAY POINT
mobility model when compared with the Random Waypoint
Figure 10. Mobility Models – Throughput (packet) with model. Because of this the chances of getting unrelated route
varied Communication channel in 100 WiMax error messages is comparatively less in the case of Gauss
Environments Markov model.
As shown in the Figures 14,15,16,17, in all 100,200,300

0.02 100 Nodes
and 400 nodes Gauss Markov model gives high latency
0.015 whereas the Random waypoint and Manhattan models
Latency
0.01 perform with slight decreasing trend compared to Gauss

0.005 Markov model. This Gauss Markov is a more realistic
0 mobility model when compared with the Random Waypoint
2 4 6 8 10 model. Because of this, the chances of getting unrelated
Communication channel route error messages is lower in the case of Gauss Markov
GAUSS-MARKOV model.
MANHATTAN GRID
RANDOM WAY POINT
Figure 14. Mobility Models-Latency (time) with varied 9. Conclusion

In this paper, the three mobility models have been taken up
and their performance have been compared with different
0.02 200 Nodes stressful WiMax environments with ns-2. The performance
0.015 metric of PDF, Routing load, Throughput and Latency for
Latency
0.01 DSR protocols under the different simulation environment

0.005 with varying communication channel has been computed.
0
As a result, through simulation, it reveals that the
throughput, latency and routing load are high in Gauss
2 4 6 8 10
Communication Channel Markov Mobility model. This is due to a more realistic
mobility model. Because of this the chances of getting
GAUSS-MARKOV unrelated route error messages is lower in the case of Gauss
MANHATTAN GRID
RANDOM WAY POINT
Markov Model when compared Random Waypoint mobility
Model. It also shows that packet delivery fraction is high in
Figure 15. Mobility Models- Latency (time) with varied Manhattan Grid model. This is due to the more restricted
communication channel in 200 WiMax Environments movement of the nodes in the network, which leads to
slightly lesser number of broken links and subsequently
lowering the chances of getting a stray route error message.
0.025 300 Nodes Compared to Manhattan and Gauss Markov models the
0.02 Random waypoint model is stable.
Latency
0.015
0.01 References
0.005
0 [1] T.Camp, J.Boleng, V.Davies, “ A Survey of Mobility
2 4 6 8 10 Models for Ad Hoc Network Research”, in Wireless
Communication Channel Communication and (WCMC): Special issue on Mobile
GAUSS-MARKOV Ad Hoc Networking: Research, Trends and
MANHATTAN GRID
RANDOM WAY POINT
Appications, Vol.2, no 5, pp. 483-502, 2002.
[2] C. Bettsltter, G. Resta, P.Santi, “ The Node Distribution
Figure 16. Mobility Models-Latency (time) with varied
of the Random Waypoint Mobility Model for Wireless
Ad Hoc Networks”, IEEE Transactions on Mobile
Computing, July - September 2003, pp 257-269
0.06 400 Nodes [3] Chen, J, Wang, C., Tsai, F., Chang, C., Liu, S.,Guo,J.,
0.04
Lien, W., Sum, J., Hung, C., “The Design and
Implementation of WiMax Module for ns-2 Simulator”,
0.02 ACM Valuetools 2006,Pisa,Italy,ACM Press, New
York (2006)
0
2 4 6 8 10 [4] Jenhui Chen, Chih-Chieh Wang,“The Design and
Communication channel Implementation of WiMax Module for ns-2
GAUSS-M A R KOV
Simulator”,WNS2-06.
M A N H A T T A N GR ID
R AN D OM WA Y P OIN T
[5] Cicconetti,C., Erta, A., Lenzini,L.,Mingozzi, E.
“Performance Evaluation of the Mesh Election
Figure 17. Mobility Models-Latency (time) with varied
Procedure of IEEE 802.16/WiMax”,
MSWiM’07,October 22-26,2007,Chania
[6] Leonardo Betancur, Roberto C. Hincaple, Roberto

Bustamante, “WiMax Channel-Phy Model in Network Authors Profile
Simulator 2”, Oct 10, 2006.
[7] Shiang Ming Huang, “NCTUns simulation tool for N Vetrivelan Research Scholar,
WiMax modeling”, WICON 2007, Oct 22-24, 2007. Department of Computer Applications,
National Institute of Technolgoy, Trichy.
[8] Bohnert, T.M.,Jakubiak, J.,Katz, M.,Koucheryavy, India. Working as an Assistant Professor
Y.,Monteiro,E., Borcoci,E, “On Evaluating a WiMax in Periyar Maniammai University,
Access Network for Isolated Reserch and Data Thanjavur, India With 15 Years of
Networks Using NS-2”, Springer-Verlag Berlin Engineering Collegiate experience.
Presented five International Papers and
Heidelberg 2007.
published two papers in the refereed
[9] Cao,M., Raghunathan,V and Kumar, P.R A tractable journals. Presented paper in IAENG International Conference at
algorithm for fair and efficient uplink scheduling of Hong Kong. Established WiMax Broad Band connection with Six
multi-hop Wimax mesh networks. Proc. WiMesh 2006, rural villages in and around Thanjavur district under PURA.
Reston(VA),USA, Sep.25,2006,pp 101-108.
Dr. A V Reddy received Ph.D in II.Sc
[10] Elizabeth M.Royer and C.K Toh “A Review of Current Bangalore, India. Working as a Professor,
Routing Protocols for Ad Hoc Mobile Wireless Department of Computer Applications,
Networks,” IEEE Personeal Communications, April National Institute of Technology, Trichy,
1999,pp 46-55. Tamilnadu, India with 25 Years of
Academic experience. Published Six
[11] S.Azad,A Rahman and F.Anwar,”A Performance International journal Papers in refereed
Comparison of Proactive and Reactive Routing journal and also ten International papers
Protocols of Mobile Ad-hoc NET work(MANET),” presented.
Journal of Engineering and Applied Sciences 2(5),2007,
pp 891-896.
[12] Perkins C.E and Royer.E.M, “Ad Hoc On-demand
Distance Vector Routing” In Proceedings of the 2nd
IEEE Workshop on Mobile Computing Systems and
Applications,New Orleans,LA,February 1999,pp.46-55
[13] Network Simulator 2 (NS-2),
http://www.isi.edu/nsnam/ns/
[14] NDSL WiMax Module for ns2 Simulator,
http://nds1.csie.cgu.edu.tw/wimax_ns2.php
[15] The WiMax forum.available at
http://www.wimaxforum.org/home/.
[16] The WiMax module for ns-2.available at
http://ndsl.csie.cgu.edu.tw/wimax_ns2.php.
[17] BonMotion Scenario Generation.
www.informatik.unibonn.de/IV /BonnMotion/
Partial Aggregation for Multidimensional

Online Analytical Processing Structures
Naeem Akhtar Khan1 and Abdul Aziz2
1
Faculty of Information Technology,
University of Central Punjab, Lahore, Pakistan
naeemkhan@ucp.edu.pk
2
aziz@ucp.edu.pk
Abstract: Partial pre-computation for OLAP (On-Line- HOLAP, etc., where all physical implementation determines
Analytic Processing) databases has become an important the advantages and disadvantages of storage access an
research area in recent years. Partial pre-aggregation is analysis capabilities and also determines any possible
implemented to speed up the response time of queries that are extensions in future to the model. In the models quoted
posed for the array-like decision support interface, subject to the above, the two most common in practice are the
different constraint that all pre-computed aggregates must fit Multidimensional On-line Analytic Processing (MOLAP)
into storage of a pre-determined calculated size. The target
model and the Relational On-line Analytic Processing
query workload contains all base and aggregate cells that are
stored in a multidimensional structure (i.e. cube). These queries
(ROLAP) model. The main advantage of ROLAP, which
are in fact range queries pre-defined by users for the support of depends on relational database (RDB) technology, is that the
decision makers. The query workload of an OLAP scenario is database technology is well standardized (e.g., SQL2) and is
the set of queries expected by the users. Most of the published readily available too. This permits for the implementation of
research only deals with the optimization for the workload of a physical system, based on readily available technology and
views in the context of ROLAP (Relational OLAP). Many open standards. As this technology is well studied and
researchers have criticized partial-computation schemes, researched, there are mechanisms which allow for
optimized for views that lack of support to ad-hoc querying. The transactions and authorization schemes, thus allowing for
other main aspect is that a view may be too large for pre- multi-user systems with the ability to update the data as
computation that calculate very small answers. In this paper, we required.
study the problems of partial pre-computation for point queries,
The main disadvantage of this technology is that the query
which are best for MOLAP (Multidimensional OLAP)
language as it exists (SQL) is not so sufficiently powerful or
environment. We introduce multidimensional approach for
efficiency of the cover-based query processing and the flexible enough to support true OLAP features [1].
effectiveness of PC Cubes. Furthermore, there is an impedance difficulty, that the
results returned, tables, always required to be converted to
Keywords: OLAP, MOLAP, Data-Cubes, Data warehouse. another form before further programming abilities can be
performed. The main advantage of MOLAP, which depends
1. Introduction on generally proprietary multi-dimensional (MDD) database
technology, is based on the disadvantages of ROLAP and is
A data warehouse (DW) is centralized repository of the major reason for its creation. MOLAP queries are quite
summarized data with the main purpose of exploring the powerful and flexible in terms of OLAP processing. The
relationship between independent, dimensions, static physical model further closely matches the
variables and dependent, dynamic, variables facts or multidimensional model, and the impedance issue is
measures. There is a trend within the data warehousing remedied within a vendor’s side. However, there are
community towards the separation of the requirements for disadvantages of the MOLAP physical model: 1st) There is
preparation and storage necessary for analyzing the no real standard for MOLAP; 2nd) there are no off-the-shelf
accumulated data and the requirements for the exploration MDD databases per se; 3rd) there are scalability problems;
of the data with the necessary tools and functionality and 4th) there are problems with authorizations and
required [1]. In terms of the storage necessities, a transactions. As the physical implementation ultimately
convergent tendency is towards a multi-dimensional determines the abilities of the system, it would be advised to
hypercube model [2]. On the other hand in terms of analysis find a technology that combines and maximizes the
and the tools required for On-Line Analytic Processing advantages of both ROLAP and MOLAP while at the same
(OLAP), there is a trend towards standardizing this as well; time minimizing the dis-advantages.
e.g., the efficient OLAP Council’s Multi-Dimensional Online Analytical Processing (OLAP) has become a basic
Application Programmers Interface (MD-API). Although component of modern decision support systems. As claimed
the trends are for separating the storage from the analysis, in [3] introduced the data-cube, a relational operator as well
the actual physical implementation of DW/OLAP systems as model used for computing summary views of data that
reconnects them. This is an evident from the parade of can, in turn, significantly improve the response time of core
acronyms used today, e.g., MOLAP, ROLAP, DOLAP, OLAP operations such as roll-up, drill down, and slice and
dice, drill down approach is reflected in Figure 1. Typically approaches. The main issue, as outlined in [6] is the
developed on top of relational data warehouses these indexing problem for the fastest execution of OLAP queries.
summary views are formed by aggregating values across The main problem for ROLAP is that it does not offer an
different attribute combinations. For a d-dimensional input immediate and fast index for OLAP queries. Many well-
set R, there are 2d probable group bys. A data-cube as well known vendors have chosen the sacrifice scalability for
as a lattice which is often used for representing the inherent performance. Query performance issue is discussed for
relationships between group-bys [5], there are two standard ROLAP and proposed a novel, distributed multi-
data-cube representations: ROLAP (set of relational tables) dimensional ROLAP efficient indexing scheme [7]. They
and MOLAP (multi-dimensional OLAP). The main showed that the ROLAP advantage for high scalability can
differences between ROLAP and MOLAP architectures are be maintained, while at the same time providing a rapid
shown in Table 1 [4]. index for OLAP queries. They proposed a distributed
indexing efficient scheme which is a combination of packed
R-trees with distributed disk striping and Hilbert curve
based data ordering. Their method requires very meager
communication volume between processors and works in
very low bandwidth connectivity multi-processor
environments such as Beowulf type processor clusters or
workstation farms. There is no requirement of a shared disk
and scales well with respect to the number of processors
used, and for further improving the scalability of ROLAP
with respect to the size and dimension of the data set (which
is already better than MOLAP’s scalability), they extend
their indexing scheme to the partial cube case.
The large number of group-bys, 2d , is a major problem in
practice for any data-cube scheme. They considered the case
Figure 1. Hierarchies in data
where they do not wish to build (materialize) all group-bys,
The array-based structure, MOLAP (Multi-dimensional but only a subset. For example, a user definitely wants to
OLAP), has the advantage that native arrays provide an only materialize those group-bys that are frequently used,
immediate form of indexing for queries of cube. Research thereby saving disk space and time for the cube
has shown, however, that MOLAP has scalability problems construction. The problem was to find a best way to answer
[6]. For example, high-dimension data-cubes represent effectively those less frequent OLAP queries which required
tremendously sparse spaces that are not easily adapted to the group-bys that had not yet been materialized. Solving this
MOLAP model. Hybrid indexing schemes are normally problem they presented an indexing scheme, based on
used, significantly diminishing the power of the model. “surrogate group-bys”, which answers such queries
effectively. Their experiments showed that their distributed
Table 1: ROLAP Vs MOLAP query engine is almost as efficient on “virtual” group-bys as
it is on ones that actually exist. In summary, they claimed
Feature ROLAP MOLAP that their method provides a framework for distributed high
performance indexing for ROLAP cubes with the following
Usage Variable Good performance properties [7].
performance
In practical, it shows lower communication volume, fully
Relational engine Multidimensional engine adapted to external memory. There is no requirement of
Storage Tables/tuples Proprietary arrays shared disk, maintainable, incrementally; it is efficient for
and Access
SQL access Lack of a standard spatial searches in various dimensions, scalable with respect
language language
to data sizes, dimensions, and number of processors. They
Third party tools Sparse data compression
Database
implemented their distributed multi-dimensional ROLAP
Size Easy updating Difficult updating indexing scheme in STL and MPI, C++ and tested it on a 17
Large space for 2% index space
indexes node Beowulf cluster (a frontend and 16 compute nodes).
Gigabyte-Terabyte Gigabyte While easily extendable for sharing everything multi-
processors, their algorithms performed well on these low-
Moreover, since MOLAP requires to be integrated with cost commodity-based systems. Their experiments showed
standard relational databases, middleware of some form that for RCUBE index construction and updating, close to
must be employed for handling the conversion between optimal speed has been achieved. A RCUBE index having
relational and array-based data representations. The fully materialized data cube of ≈640 million rows (17 Giga-
efficiency of relational model, ROLAP (Relational OLAP), bytes) on a 16 processor cluster can be generated in just
does not suffer by such restrictions. In standard relational within 1 minute. Their method for distributed query
tables its summary records are stored directly without any resolution also exhibited good speedup achieving, for
need for data conversion. ROLAP table based data example, a speedup of 13.28 on just 16 processors. For
representation does not pose scalability problems. Yet, many distributed query resolution in partial data-cubes, their
current commercial well-known systems use the MOLAP experiments showed that searches against absent (i.e. non-
materialized) group-bys can typically be easily resolved at various dimensions that characterize the facts. For example,
only a small additional cost. Their results demonstrated that in a area of retail, typical indicators are price and amount of
it is possible to build a ROLAP data-cube that is scalable a purchase, dimensions being location, product, customer
and tightly integrated with the standard relational database and time. A dimension is mostly organized in hierarchy, for
approach and, at the same time, provide an efficient index to example the location dimension can be aggregated in city,
OLAP queries. division, province, country.
The "star schema" molds the data as a simple cube, where
2. DW and OLAP Technologies hierarchical relationship in a dimension is not explicit but is
rather encapsulated in attributes; the model of star schema
The prominent definition of Data Warehouse is a "subject- with dimensions is reflected in Figure 2.
oriented, integrated, nonvolatile and time-variant collection
of data in support of management's decisions" [8]. A data
warehouse is a well organized single site repository of
information collected from different sources. In data
warehouse, information is organized around major subjects
and is modeled so as fast access to summarize data. "OLAP"
as discussed in [9],[10] refers to analysis functionalities
generally used for exploring the data. Data warehouse has
become a most important topic in the commercial world as
well as in the researcher’s community. Data warehouse
technology has been mostly used in business world, in
finance or retail areas for example. The main concern is to
Figure 2. Star Schema
take benefits from the massive amount of data that relies in
operational databases. According to [11], the data-modeling The dimension tables are normalized by “snowflake schema”,
paradigm for a data warehouse must fulfill the requirements and make it possible for explicitly representing the
that are absolutely different from the data models in OLTP hierarchies by separately identifying dimension in its
environments, the main comparison between OLTP and different granularities. At last, when multiple fact tables are
required, the "fact constellation" or "galaxy schema" model
OLAP environment is reflected in Table 2.
allows the design of collection of stars. OLAP architectures
Table 2: OLTP Vs OLAP adopt a multi-tier architecture as reflected in Figure 3 where
first tier is a warehouse server, implemented by a relational
Feature OLTP OLAP DBMS. Data of interest must be extracted from OLTP
Amount of data systems (operational legacy databases), extracted, cleaned
retrieved Small Large and transformed by ETL (Extraction, Transformation,
per transaction Loading) tools before going to load in the warehouse.
Level of data Detailed Aggregated
Views Pre-defined User-defined

Age of data Current (60- Historical 5-10 years and
90 days) also current
Typical write Update, Bulk insert, almost no
operation insert, delete deletion
Tables Flat tables Multi-Dimensional tables Figure 3. How does an OLAP piece fit together
This process aims to consolidate heterogeneous schema
Number of Large Low-Med
users (structure heterogeneity, semantic heterogeneity) and for
Data High (24 hrs, Low-Med reducing data in order to make it conform to the data
availability 7 days) warehouse model (by implementing aggregation, dis-
cretization functions). Then the data warehouse holds high
Database size Med (GB- High (TB – PB) quality, efficient historical and homogeneous data. The
TB)
second tier is a data mart, a data mart handles data received
Query Requires Already “optimized”
Optimizing experience
from the data warehouse, which is reduced for a selected
subject or category.
The data model of the data warehouse must be simple for The main focus of data marts is to isolate data of interest for
the decision maker to understand and for write queries, and a smaller scope or department, thus permitting the focusing
on optimization needs for this data and increase more
must get maximum efficiency from queries. Data warehouse
security control. However this intermediate tier of data mart
models are called hyper-cubes or multidimensional models
is optional and not mandatory. The OLAP server is
and have been prescribed by [12]. The Models are designed
implemented at 3rd level. It optimizes and calculates the
for representing measurable indicators or facts and the hypercube, i.e. the set of fact values for all the relevant tuples
of instances for different dimensions (also called embers). In parameters. The co-impact of these parameters was not also
order for optimizing accesses against the data, query results discussed. Furthermore, for getting the accurate reliability
are advance calculated in the form of aggregates. OLAP value, some models are so complicated that it is difficult to
operators allow materializing different views of the figure out the best value of each parameter.
hypercube, allowing interactive queries from decision makers In [22] worked for designing a reliable large-scale data
and analysis of the data. Common OLAP operations include warehouse or storage system and presented a new object-
drilldown, roll-up, slice and dice, rotate. The fourth tier is an based-repairing Markov model, which induces many key
OLAP client, which provides a user interface with various challenges. One problem was to figure out some basic
reporting tools, analysis tools and data mining tools for system parameters, such as the number of nodes, the total
obtaining results. Software solutions are existed for a number of stored objects and the bandwidth of switch and
traditional use. the node. For designing a reliable system with the optimal
system parameters, compared with previous work, their
3. Distributed Index Construction for ROLAP approach makes a significant contribution in two aspects.
Firstly, they presented a new object-based Markov model for
Different methods have been proposed and recommended quantifying the impact of key system parameters on the
for building ROLAP data-cubes [13], [14], [15] but there are system reliability in three replica placement strategies. They
only very few results available for the indexing of such compared their model with previous complex models; this
cubes. For sequential query processing, [16] proposed an object-based compact model not only turns to be easier for
indexing model, which is composed of a collection of b- solving because of its smaller state transition matrix, but
trees. While adequate for low-dimensional data-cubes, b- also leads to more integrative and practical efficient results.
trees are inappropriate for higher dimensions in that (a) Secondly, they proposed a two-step analyzing process.
multiple, redundant attribute orderings are required to The first is to find out the comparatively precise optimal
support arbitrary user queries (b) their performance value of a system parameter by independently analyzing its
deteriorates rapidly with increased dimensionality. impact on the system reliability when other system
In [17] proposed the cube-tree, an indexing model which is parameters are fixed. The second is to figure out the best
based on the concept of a packed R-tree [18]. In the possible combination of these parameters by analyzing their
dimension of parallel query processing, a typical approach integrated and complex impacts on the system reliability
used by current commercial systems like ORACLE 9i RAC while all of them are tuned. Their analysis results showed
for improving throughput by distributing a stream of that the optimal values do exist and have simple formulas.
incoming queries over multiple processors and having each They presented a new efficient object-based repair model
processor answer a subset of queries. But this type of an and by analyzing this model, they worked out the individual
approach provides no speedup for each individual query. optimal value of parameters and their optimal combination.
For OLAP queries, which are time consuming, the The results obtained by them can provide the engineers with
parallelization of each query is important for the scalability direct instructions to design reliable systems.
of the entire OLAP system. With respect to the
parallelization for general purpose environments of R-tree
4. Related Work
queries, a number of researchers have presented solutions.
As claimed in [19] Koudas, Faloutsos and Kamel presented Here we introduce the lattice view that depicts the
a Master R-tree structure that employs a centralized index relationships between existing views, some algorithms of the
and a collection of distributed data files. Schnitzer and pre-computation of views, answering query using the
Leutenegger’s Master-Client R-tree [20] improves upon the materialized views and view selection.
earlier model by partitioning the central index into a smaller
4.1 The View Lattice Framework
master index as well as a set of associated client indexes.
While offering considerable performance advantages in The CUBE BY in [23] has resulted in the computation of
generic indexing environments, neither approach is well- views which related to SQL queries which has been grouped
suited for OLAP environment systems. In addition to the on all possible combinations of the dimension attributes.
sequential problems on the main server node, both utilize Those views are usually denoted by the grouping attributes,
partitioning methods that can lead to the localization of e.g. T, L, P, LT, PT, PL, PLT, and All for the example of
searches. Furthermore, neither approach provides the database with three attributes Time(T), Location(L) and
methods for incremental updates. Product(P). As claimed by [24] that a lattice is used to depict
Data reliability is major issue in data warehouses and many the relationship between views. An aggregate view is
solutions have been proposed for its solution. Replication is represented by each node in the lattice. In the lattice view
a widely used mechanism for protecting permanent data loss edge existed form node i to node j, view j can be computed
while replica placement will significantly impact data from view i to view j which contains one attribute less than
reliability. Several replica placement policies for different view i. In this case, view i is called, parent view of j. in this
objectives, have been proposed and deployed in real-world situation there is a basic view on which every view is
systems, like RANDOM in GFS, PTN in RAID, Q-rot in dependent. The complete aggregation view “ALL” can be
[16], [13], [21] have analyzed how the system parameters, computed from any other view of lattice.
such as object size, system capacity, disk and switch
bandwidth, could affect the system’s reliability. However,
they only focused on the rough trend about their impact but
did not illustrate the correct optimal values of these
4.3 View Selection

The main issue in full computation of views is storage space;
so many researchers study this problem and recommended
partial-computation. In [31] efficient approach was proposed
for choosing a set of views for materialization under a
situation of limited storage space. They introduced a linear
cost model; this algorithm assumes that the cost of answering
a query is related to the size of view from which the respond
of the query can easily be computed. This linear cost model
was verified by the experiments. In a greedy view selection
algorithm which tries to decide which aggregate view is best
for minimizing the query cost. This greedy view selection
Figure 4. Lattice View algorithm first chooses the base view. Materializing one more
view can allow some queries which can be easily answered
4.2 Pre-computation of aggregates by a smaller metalized view, in this way query cost can be
In Figure 4 lattice views are shown, for example database reduced. Their proposed algorithm chooses the view which
which have three dimensions, Time, Location and Product produces the more reduction in query cost. Research proved
which are reflected as T, L and P respectively. A view is that the benefit of the aggregate views selected by this
labeled by the dimension’s name, where it is aggregated on algorithm in no worse than (0.63-f) times the benefit of an
view PLT is the basic view while PL is parent view. optimized selection. The same problem was discussed in [32]
A lot of aggregation is involved on OLAP queries in data and another view selection algorithm PBS (Pick by Size) was
warehouse. Performance can greatly be improved by the pre- proposed. The main difference was that PBS selects the
computation of aggregates. Many researchers have developed views solely based on the size of the views. In each turn, the
pre-computation algorithms for efficient computing all view with the smaller size from unselected views is chosen
possible views, which are so called view materialization. In until the total size of the selected views reaches the allocated
[25] and [26] different efficient view materialization space limit.
algorithms have been proposed, e.g., Overlap, Pipesoft and 4.4 Query Processing
PipeHash, which incorporate many efficient optimization
In [33], the traditional query optimization algorithms were
techniques, such as using the data sorting in a particular
generalized to optimize the query in the presence of
order for computing all views that are prefixes in that order,
materialized views. In [34] proposed some techniques for
computing a view from its smallest previously computed
rewriting a given SQL query, such that it uses one or more
parent, and computing the views with most common prefixes
than one materialized view. They also proposed a semantic
in a pipelined technique.
approach for determining whether the information existing in
In [27] there are some efforts have been made for studying,
a view is sufficient for answering a query. Another query re-
how the skewed data may affect the pre-computation of
writing method was suggested in [35], this technique can
aggregates and an approach for dynamically manage the
utilize the views having different granularities, aggregation
memory usage is recommended. A comparison has been made
granularities and selection regions. Generally, for an OLAP
in [28], difference between the view materialization in
query, their can be many equivalent re-writings using
MOLAP and ROLAP further an array-based pre-computation
different materialized cubes/views in different ways. Their
efficient algorithm for MOLAP is proposed. In this algorithm
execution cost is different from one another. An efficient
the partitions of views is stored in main memory array and
algorithm is also proposed in [35] for determining the set of
further overlaps the computation of different views while
materialized views used in query re-writing.
using minimum storage memory for each view.
In [29] author examined the pre-computation on compressed
MOLAP database and some algorithms for the computation 5. Recommendations for an Efficient Partial
of view without de-compression is proposed. In [30] pre-Aggregation
suggested another pre-computation algorithm for MOLAP.
A completely full pre-computation, where all possible
One distinct feature in these algorithms was that the
aggregates have been pre-computed, can provide the best and
aggregation cells are managed in the similar way as the
most efficient query performance, but in our point of view
source data. Primary cells and aggregation cells are stored
this approach in not recommended for the following reasons:
together single data structure, further proposed one
• A full pre-computation requires a great storage space
multidimensional array, which allows them for quickly
for aggregates, so it often exceeds the available space.
accessing. In this algorithm pre-computation considers the
• This technique is not based on cost effective use of
points in multidimensional space. The algorithm examines
resources-beneficial, in [32], it is mentioned that the
the coordinates of the cells and relies on the relationships
gains from pre-computation outweigh the cost of
among cells for determining how to perform efficient
further disk space after some level of pre-
aggregation. A graph-theoretical model is employed for
computation.
ensuring the correctness of the summation computation.
• Maintenance cost is increased.
• It takes long load time.
• A full pre-computation is not suitable as very sparse queries. It has most excellent overall performance among
cube (wastage of space) for high cardinality. variations of member selection algorithms.
However in the situation of fully pre-computation
environment, to overcome the storage limitations it is References
advisable that cube may be partitioned. One logical cube of
data should be spread across multiple physical cubes on [1] Thomsen, Erik, OLAP Solutions: Building
distinct servers. Divide and Concur approach helps alleviate Multidimensional Information Systems, John Wiley
the scalability limitations of full pre-computation approach. and Sons, 1997.
The other approach, which is partial pre-computation of [2] Agrawal, R., Gupta, A., Sarawagi, S., “Modeling
aggregates, can resolve problems of fully aggregation. The Multidimensional Databases”, Proceedings of the 13th
main objective of a partial pre-computation technique is to International Conference on Data Engineering, pp.
select a certain amount of aggregates for computing before 232-243, 1997.
querying time, in this way query answering time can be [3] J. Gray, A. Bosworth, A. Layman, and H. Pirahesh,
optimized. But there are two major issues about the “Data cube: A relational aggregation operator
optimization process: generalizing group-by, cross-tab, and sub-totals”,
• How many partial pre-computed aggregates should be Proceeding of the 12th International Conference On
computed? Data Engineering, pages 152–159, 1996.
• What kind of queries is optimized for pre- [4] Aejandro A. Vaisman, “Data Warehousing, OLAP, and
computation strategy? Materialized Views”, A Survey Technical Report
The first question depends upon the storage available, we TR015-98, University of Buenos Aires, Computer
recommends that 40% of all possible aggregates should be Science Department, 1998.
computed in advance. Few years back it was recommended [5] V. Harinarayan, A. Rajaraman, and J. Ullman,
by Microsoft, as the vendor of one of the popular OLAP “Implementing data cubes”, Proceedings of the 1996
systems that 20% of all possible aggregates should be ACM SIGMOD Conference, pages 205–216, 1996.
computed in advance, but now the technology has improved [6] S. Agarwal, R. Agrawal, P. Deshpande, A. Gupta, J.
and storage capacity can be achieved by very low cost. The Naughton, R. Ramakrishnan, and S.Sarawagi, “On
best answer of second question is that these queries should be the computation of multidimensional aggregates”,
those that most expected by the decision maker of OLAP Proceedings of the 22nd International VLDB
application. Conference, pages 506–521, 1996.
This type of answer in not so easy because it varies from user [7] F. Dehne, T. Eavis and A. Rau-Chaplin, “Parallel
to user and application to application, for this purpose and Multi-Dimensional ROLAP Indexing”, Proceedings
understanding the question systematically, one needs to of the 3rd IEEE/ACM International Symposium on
precisely categorize the chunk of queries as an object for Cluster Computing and the Grid, (CCGRID’03)
optimization, this type set of queries or expected queries is 2003.
often called ”query workload”. Pattern of expected queries [8] W.H. Inmon Building the Data Warehouse 3rd Edition,
can be obtained from other same type of case studies and Eds.Wiley and Sons, 1996.
navigational data analysis task. There are many algorithms for [9] S. Chaudhuri, U. Dayal “An Overview of Data
partial pre-computation which have already been discussed. Warehousing and Olap Technolog”, SIGMOD Record
For optimized and efficient processing of OLAP queries, 26(1), 1997.
most commonly used approach is to store the results of [10] P. Vassiliadis P., T. Sellis, “A Survey of Logical
frequently issued queries by decision makers in to summary Models for OLAP Databases”, SIGMOD Record
tables, and further makes use of them for evaluating other Volume 28, Number 1, March, 1999.
queries, this approach is best. In our point of view PBS (Pick [11] R. Kimball, The Data Warehouse Toolkit, J.Wiley
by size) algorithm is so fast and best as it will facilitate and Sons, Inc, 1996.
database administrators for determining the points, where [12] L. Cabibbo and R. Torlone, “A Logical Approach to
diminishing returns outweigh the cost of the additional Multidimensional Databases”, Proceedings of the 6th
storage space. This algorithm also shows how much space International Conference on Extending Database
should be allocated for pre-computation. Technology (EDBT'98), Valencia, Spain, 1998.
[13] V. Harinarayan, A. Rajaraman, and J. Ullman,
“Implementing data cubes” Proceedings of the 1996
6. Conclusion
ACM SIGMOD Conference, pages 205–216, 1996.
The partial pre-computation is most popular research area in [14] K. Ross and D. Srivastava, “Fast computation of
OLAP environment. Most of the published papers for partial sparse data cubes”, Proceedings of the 23rd VLDB
pre-computation are about optimized performance of views as Conference, pages 116–125, 1997.
the query workload. Practically users do not care about the [15] S. Sarawagi, R. Agrawal, and A.Gupta, “On
processing overhead and the time used in determining the computing the data cube”, Technical Report RJ10026,
output of the given query, when planning to implement IBM Almaden Research Center, San Jose, California,
partial pre-computation strategy. For implementation of point 1996.
queries, the processing overhead is most important fact of [16] H. Gupta, V. Harinarayan, A. Rajaraman, and J.
consideration. PBS approach is efficient, and selection by Ullman, “Index selection for olap”, Proceeding of the
PBS is much effective because PC Cube generated by the 13th International Conference on Data Engineering,
PBS leads to shorter time required for answering the point pages 208–219,1997.
[17] N. Roussopoulos, Y. Kotidis, and M. Roussopolis, [31] V.Harinarayan, A.Rajaraman, and J.D.Ullman,
“Cubetree: Organization of the bulk incremental up- “Impleminting Data Cubes Efficienty”, In
dates on the data cube”, Proceedings of the 1997 Proceedings of SIGMOD, pages 205-206, 1996.
ACM SIG-MOD Conference, pages 89–99, 1997. [32] A.Shukla, P.Deshpande, and J.Naughton,
[18] N. Roussopolis and D. Leifker, “Direct spatial search “Materialized View Selection for Multidimensional
on pictorial databases using packed r-trees”, Datasets”, In Proceedings on 24th VLDB Conference,
Proceedings of the 1985 ACM SIGMOD Conference, New York, 1998.
pages 17–31, 1985. [33] S.Chaudhuri,R.Krishnamurthy, S.Potamianos, and
[19] N. Koudas, C. Faloutsos, and I. Kamel, “De- K.Shim., “Optimizing Quereis with Materialized
clustering spatial databases on multi- computer Views”, In Proceedings of the 11th IEEE International
architecture”, In Proceedings of Extended Database Conference on Data Engineering, pages 190-200,
Technologies, pages 592–614, 1996. 1995.
[20] B. Schnitzer and S. Leutenegger, “Master-client r- [34] D.Srivastava, S.Dar, H.V.Jagadish, A.Y.Levy.,
trees: a new parallel architecture”, 11th “Answering Queries with Aggregation using views”,
International Conf-erence of Scientific and In Proceedings of 22nd VLDB Conference, Bombay,
Statistical Database Management, pages 68–77, 1999. India, pages 318-329, 1996.
[21] I. Kamel and C. Faloutsos, “On packing r-trees”, [35] C.S.Park, M.H.Kim and Y.J.Lee., “Finding an
Proceedings of the Second International Conference Efficient Rewriting of OLAP Queries Using
on Information and Knowledge Management, pages Materialized Views in Data Warehouses”, Decision
490–499, 1993. Support Systems, vol.32, No.4, pages 379-399,2002.
[22] K.Du, Z.Hu, H.Wang,Y.Chen, S.Yang and Z.Yuan,
“Reliability Design for Large Scale Data
Warehouses”, Journal of Computing, Vol.3, No.10,
pp 78-85 October 2008. Authors Profile
[23] J.Gray, A.Bosworth, A.Layman, H.Pirahesh., “Data
Cube: A Relational Aggregation Operator Naeem Akhtar Khan received the B.S. degree in Computer
Generalizing Group-By,Cross-Tabs, and Sub-Totals”, Science from Allama Iqbal Open University, Islamabad, Pakistan
in 2005 and M.S. degree in Computer Science from University of
In Proceedings of International Conference on Data
Agriculture, Faisalabad, Pakistan in 2008. He is currently pursuing
Engineering (ICE’96), New Orleans, February,1996. Ph.D. (Computer Science) degree in University of Central, Punjab,
[24] V. Harinarayan, A.Rajaraman, and J.D.Ullman, Lahore, Pakistan. His research interests include large-scale data
“Implementing Data Cubes Efficiently”, In management, data reliability, Data Mining, MOLAP.
Proceedings of SIGMOD, pages 205-216,1996.
[25] Sameet Agarwal, Rakesh Agarwal, Prasad M. Dr. Abdul Aziz did his M.Sc. from University of the Punjab,
Deshpandre, Asnish Gupta, Jeffrey F.Naughton, Pakistan in 1989; M.Phil and Ph.D in Computer Science from
Ragnu Ramakrishnan, Sunita Sarawagi, “On the University of East Anglia, UK. He secured many honors and
awards during his academic career from various institutions. He is
Computation of Multidimensional Aggregates”, In
currently working as full Professor at the University of Central
Proceedings of the 22nd VLDB Conference, Bombay, Punjab, Lahore, Pakistan. He is the founder and Chair of Data
India, pages 506-521,1996. Mining Research Group at UCP. Dr. Aziz has delivered lectures in
[26] P.M.Deshpande, S.Agarwal, J.F.Naughton, and many universities as guest speaker. He has published large number
R.Ramakrishnan, “Computation of Multidimensional of research papers in different refereed international journals and
Aggregates”, Technical Report 1314, University of conferences. His research interests include Knowledge Discovery
Wisconsin-Madison, 1996. in Databases (KDD) - Data Mining, Pattern Recognition, Data
[27] Yu,J.X., Hongjun Lu., “Hash in Place with Memory Warehousing and Machine Learning.
Shifting: Datacube Computation Revisited”, In He is member of editorial board for various well known journals
Proceedings of 15th International Conference on Data and international conferences including IEEE publications. (e-mail:
Engineering. Page: 254 March, 1999. aziz@ucp.edu.pk).
[28] Y.Zhao, P.M. Deshpande, and J.F.Naughton., “An
Array-based Algorithm for Simulataneous
Multidimensional Aggregates”, In Proceedings of
ACM SIGMOD, pages 159-170, 1997.
[29] Li,J.,Rotem, D., Srivastava, J., “Aggregation
Algorithm for very large compressed Data
Warehouses”, In Proceedings of 25th very large
Database (VLDB) Conference. Edinburgh, Scotland,
1999.
[30] Woshun Luk., “ADODA: A Desktop Online Data
Analyzer”, In 7th International Conference on
Database Systems for Advanced Applications
(FASFAA,01), Hong Kong, China, April,2001.
Browser Extensible Secured Hash Password

Authentication for Multiple Websites
1
T.S.Thangavel and 2Dr. A. Krishnan
1
AP/ Dept. of M.Sc(IT), K.S.Rangasamy College of Technology, Tiruchengode 637215,Tamilnadu,India.
tsthangavel123@yahoo.in
2
Dean, K.S.Rangasamy College of Technology, Tiruchengode 637215,Tamilnadu,India.
a_krishnan26@hotmail.com
Abstract: The techniques such as secured socket layer (SSL) with A password generator can be part of a password
client-side certificates are well known in the security research manager. When a password policy enforces complex rules, it
community, most commercial web sites rely on a relatively weak can be easier to use a password generator based on that set
form of password authentication, the browser simply sends a of rules than to manually create passwords. In situations
user ’s plaintext password to a remote web server, often using SSL. where the attacker can obtain an encrypted version of the
Even when used over an encrypted connection, this form of password, such testing can be performed rapidly enough so
password authentication is vulnerable to attack. In common
that a few million trial passwords can be checked in a matter
password attacks, hackers exploit the fact that web users often use
the same password at many different sites. This allows hackers to of seconds. The function rand presents another problem. All
break into a low security site that simply stores pseudo-random number generators have an internal memory
username/passwords in the clear and use the retrieved passwords or state. The size of that state determines the maximum
at a high security site. Recently, some collisions have been number of different values it can produce, an n-bit state can
exposed for a variety of cryptographic hash functions including produce at most 2n different values. On many systems rand
some of the most widely used today. Many other hash functions has a 31 or 32 bit state, which is already a significant
using similar constructions can however still be considered security limitation.
secure. Nevertheless, this has drawn attention on the need for new
hash function designs. This work developed an improved secure
hash function, whose security is directly related to the syndrome The main cryptographic hash function design in use today
decoding problem from the theory of error-correcting codes. The iterates a so called compression function according to
proposal design and develop a user interface, and implementation Merkle’s [12] and Damgard’s[13] constructions. Classical
of a browser extension, password hash, that strengthens web compression functions are very fast but, in general, cannot
password authentication. Providing customized passwords, can be proven secure. However, provable security may be
reduce the threat of password attacks with no server changes and achieved with compression functions designed following
little or no change to the user experience. The proposed techniques public key principles, at the cost of being less efficient. This
are designed to transparently provide novice users with the has been done for instance by Damgard,
benefits of password practices that are otherwise only feasible for
security experts. Experimentation are done with Internet Explorer where he designed a hash function based on the
and Fire fox implementations and report the result of initial user. Knapsack problem. Accordingly, this function has been
broken by Granboulan and Joux,[10] using lattice reduction
Keywords: password authentication, secured hash, multi- algorithms. The present paper contributes to the hash
website password, pseudo random, phishing, cryptographic function family by designing functions based on the
password syndrome decoding problem, which is immune to lattice
reduction based attacks.
1. Introduction
Unlike most other public key cryptosystems, the
A random password generator is software program encryption function of the McEliece cryptosystem is nearly
or hardware device that takes input from a random or as fast as a symmetric cipher. Using this function with a
pseudo-random number generator and automatically random matrix instead of the usual parity check matrix of a
generates a password. Random passwords can be generated Goppa code, a provably secure one-way function has been
manually, using simple sources of randomness such as dice constructed since there is no trapdoor, its security can be
or coins, or they can be generated using a computer. While readily related to the difficulty of syndrome decoding.
there are many examples of "random" password generator
programs available on the Internet, generating randomness The purpose of this paper is to improve updated
can be tricky and many programs do not generate random parameters for the hash function. Our paper analyzes
characters in a way that ensures strong security. A common asymptotical behavior of their attack. We shall establish that
recommendation is to use open source security tools where this attack is exponential, such that the design for the hash
possible, since they allow independent checks on the quality function is sound.
of the methods used. Note that simply generating a
password at random does not ensure the password is a
strong password, because it is possible, although highly
unlikely, to generate an easily guessed or cracked password.
2. Literature Review information, but instead, will look like a cloud formation.
Moreover, chaotic systems may be biased by outside sources
Computer applications may require random which may be predictable. As an example, a radio signal can
numbers in many contexts. Random numbers can be used to be affected by a strong external signal, or the behavior of a
simulate natural or artificial phenomena in computer noisy diode can be changed by the surrounding temperature.
simulations, many algorithms that require randomness have All of the above problems arise because the behavior of a
been developed that outperform deterministic algorithms for chaotic system may not be completely random. More
the same problem, and random numbers can be used to specifically, an adversary observing or wishing to affect the
generate or verify passwords for cryptography-based random number source can take advantage of certain
computer security systems. The present invention relates to localities that may be inherent in chaotic systems. These
the use of random numbers in such security systems, called localities can occur either in space or time.
as cryptographic applications. Specifically, the present Finally, a number of existing applications including
invention pertains to generating a random number in a Mozilla Firefox provide convenient password management
secure manner for such cryptographic applications. In the by storing the user’s web passwords on disk, encrypted
context of cryptographic applications[1], there may be an under some master password. When the user tries to log in
hostile trespasser or agent, who desires to infiltrate the to a site, the application asks for the master password and
security of cryptographic security system in order to gain then releases the user’s password for that site. Thus, the user
access to sensitive, confidential, or valuable information need only remember the master password. The main
contained therein. For example, banks often encrypt their drawback compared to PwdHash is that the user can only
transactions and accounts. use the web on the machine that stores his passwords. On
the plus side, password management systems do provide
In order to ensure the utmost security, it is essential stronger protection against dictionary attacks when the user
that the security system implements a method for generating chooses a unique, high entropy password for each site.
a random number that appears completely random. In this However, many users may fail to do this.
manner, a completely random password or cryptographic
key presents no opening or prior knowledge that can be 3. Methodology
exploited by an hostile agent.[2]
Random password generators normally output a
Many prior art methods exist for generating
string of symbols of specified length. These can be
random numbers. These prior art methods typically involve
individual characters from some character set, syllables
the use of some type of chaotic system. A chaotic system is
designed to form pronounceable passwords, or words from
one with a state that changes over time in a largely
some word list to form a passphrase. The program can be
unpredictable manner. To use the chaotic system[4] to
customized to ensure the resulting password complies with
generate a random number, there is some means of
the local password policy, say by always producing a mix of
converting the state of the system into a sequence of bits
letters, numbers and special characters. The strength of a
(i.e., a binary number). In the past, chaotic systems were
random password can be calculated by computing the
based on various sources, such as the sound of radio static,
information entropy of the random process that produced it.
the output of a noisy diode, output of a Geiger counter, or
If each symbol in the password is produced independently,
even the motion of clouds. These chaotic systems can be
the entropy is just given by the formula
converted to produce binary numbers by using standard
techniques.
For instance, a pseudo-random binary string can be where N is the number of possible symbols and L is the
generated from the digital recording of static noise via a number of symbols in the password. The function log2 is the
digital microphone. Alternatively, a noisy diode can be base-2 logarithm. H is measured in bits. An eight character
sampled at a suitable frequency and converted into a digital password of single case letters and digits would have 41 bits
signal, or a picture of an area of the sky can be taken and of entropy (8 x 5.17). Thus a password generated using a
subsequently scanned and digitized. These resulting binary 32-bit generator has a maximum entropy of 32 bits,
strings that are generated over time are generally random in regardless of the number of characters the password
nature. However, there are several problems associated with contains.
simply using a chaotic system as a source of random
numbers.[3] First, chaotic systems can be completely or 3.1 Secure Hashing
partially predicted over small amounts of time. For example, The proposed methodology of the secure hash
the position of clouds in some area of the sky at some time password system contains one-way hash functions that can
can be used to achieve reasonably accurate predictions of the process a message to produce a condensed representation
position of clouds in the same area a short time into the called a message digest. This algorithm enables the
future. determination of a message’s integrity, any change to the
message will, with a very high probability, results in a
Furthermore, the behavior of chaotic systems [6] can be far different message digest. This property is useful in the
from completely random. For instance, a digitized picture of generation and verification of digital signatures and
a cloud formation will not look like a picture of random
message authentication codes, and in the generation of Password hashing is implemented naively inside a
random numbers. browser with rudimentary knowledge of HTML form
components. Forms begin with a tag <form action=URL>
The algorithm is described in two stages, that tells the browser where the form is to be submitted, and
preprocessing and hash computation. Preprocessing involves HTML password fields are tagged using <input
padding a message, parsing the padded message into m-bit type=“password”>. The naive browser extension listens for
blocks, and setting initialization values to be used in the blur events, which fire when focus leaves a field. When the
hash computation. The hash computation generates a blur event occurs, the extension replaces the contents of the
message schedule from the padded message and uses that field with the hashed value, using the form action attribute.
schedule, along with functions, constants, and word Thus, after the user enters a password into a form, the clear
operations to iteratively generate a series of hash values. text password is replaced by a hashed version.
The final hash value generated by the hash computation is
used to determine the message digest. The goal, however, is to defend against web
The design principle of hash functions is iterating a scripting attacks with minimal change to the user
compression function (here denoted F), which takes as input experience. For this leverage the browser extension as a
s bits and returns r bits (with s > r). The resulting function protective but largely transparent intermediary between the
is then chained to operate on strings of arbitrary length(Fig user and the web application. All input can be first
1). The validity of such a design has been established and its monitored and secured by the browser extension before the
security is proven not worse than the security of the web application is aware that the user is interacting with it.
compression function. This requires a mechanism by which users can notify
password hash browser extension that they are about to enter
a password. Password hash can then take steps to protect the
password as it is being entered. A distributed hash table is
introduced to handle the browser utility replicas of the
multiple users across hash authentication mode.
4.1 Distribute Hash Table
The distributed hash table provides incremental

scalability of throughput and data capacity as more nodes
Fig 1: Iterative hash function structure
are added to the cluster. To achieve this, we horizontally
3.2 Compression Hash Function Algorithm partition tables to spread operations and data across bricks.
Each brick thus stores some number of partitions of each
table in the system, and when new nodes are added to the
cluster, this partitioning is altered so that data is spread onto
the new node. Because of our workload assumptions, this
horizontal partitioning evenly spreads both load and data
across the cluster.
The core of the compression function is a random Given that the data in the hash table is spread
binary matrix H of size r × n. The parameters for the hash across multiple nodes, if any of those nodes fail, then a
function are n the number of columns of H, r the number of portion of the hash table will become unavailable. For this
rows of H and the size in bits of the function output, and w reason, each partition in the hash table is replicated on more
the number of columns of H added at each round. than one cluster node. The set of replicas for a partition
form a replica group; all replicas in the group are kept
4. System Model strictly coherent with each other. Any replica can be used to
service a get(), but all replicas must be updated during a
The system model concerned with attacks on the put() or remove(). If a node fails, the data from its
extension that originate on malicious phishing sites. partitions is available on the surviving members of the
Password hashing is computed using a Pseudo Random partitions' replica groups. Replica group membership is thus
Function (PRF) as follows: dynamic; when a node fails, all of its replicas are removed
from their replica groups. When a node joins the cluster, it
hash(pwd,dom) = PRFpwd(dom) may be added to the replica groups of some partitions.
where the user’s password pwd is used as the PRF key and The illustration below describe the steps taken to
the remote site’s domain name dom or some variant is used discover the set of replica groups which serve as the backing
as the input to the PRF. The hash value is then encoded as a store for a specific hash table key. The key is used to
string that satisfies the site’s password encoding rules, under traverse the DP map tries and retrieve the name of the key's
control of a configuration file used by the browser extension. replica group. The replica group name is then used looked
up in the RG map to find the group's current membership.
We do have a checkpoint mechanism in our When the password-key or password-prefix is

distributed hash table that allows us to force the on-disk detected, the browser extension determines whether the
image of all partitions to be consistent, the disk images can active element is a password field. If it is not a password
then be backed up for disaster recovery. This checkpoint field, the user is warned that it is not safe to enter his
mechanism is extremely heavy weight, however; during the password. If it is a password field, the extension intercepts
check pointing of a hash table, no state-changing operations all keystrokes of printable characters until the focus leaves
are allowed. We currently rely on system administrators to the field. The keystrokes are canceled and replaced with
decide when to initiate checkpoints. simulated keystrokes corresponding to the “mask”
characters. The system implementation of secured hash
password authentication is accomplished through following
process. The client utility is in a web browser, generating
hash password as shown in Fig. 2.
5.2 Result and Discussion
The proposed hash based multi-site pseudo random

password mechanism shows proposal considers N number of
times that the user U might authenticate before re-
registration is required. This suggests that high values of N
are desirable. The host H has to store R hash function values
at the server. This implies that to reduce the storage
requirements, it is desirable to have a low value of R.
However, N/2R is the average number of hash function
computations that U has to do for every authentication
session. Thus, it is desirable to have a high value of R. The
parameter R therefore represents a tradeoff between
computational requirements of the user U and the storage
requirements of the host H. This implies that the value of N
and R are best selected by the system administrator keeping
in mind the system requirements. We believe that given the
5. Experimental Result and Discussion current state of storage technologies, the storage
requirement is significantly less important than the
5.1 Experimental Implementation computational requirement. Major improvement over the
previous cryptographic method is the significant reduction
In the proposed hash based password in computational requirements per authentication session
authentication, a user can change her password at a given and increase in the number of logins before re-initialization.
site without changing her password at other sites. In fact,
the recommended method for using password hash is to
choose a small number of strong, distinct passwords, one for
every security level (e.g. one password for all financial sites,
one password for all news sites, etc). The password hash
extension ensures that a break-in at one financial site will
not expose the user’s password at all other banks.
The system implemented the prototype as a browser
helper object for Internet Explorer. The extension registers
three new objects i.e., an entry in the Tools menu (to access
extension options), an optional new toolbar, and the
password protection service itself. Internet Explorer support
COM event sinks that enable Browser Helper Objects to
react to website events. Use these sinks to detect focus
entering and leaving password fields, drag and drop events,
paste events and double click events. The DHTML event
model used by Internet Explorer allows page elements to
react to these events before they “bubble” up to the extension
at the top level. Since extension must handle keystroke Fig 2: Client side hash password generation
events before scripts on the page, we intercept keystrokes
using a low-level Windows keyboard hook.
Table 1: Effectiveness of Proposed Hash Based Pseudorandom Password Authentication Over Existing Cryptographic
Password
Resistance to Web browser Web browser Computational Storage Communication
eaves compatibility compatibility no. efficiency capacity effectiveness
dropping of rounds for
authentication
Existing Feasible False Low Low High False
cryptographic
password
authentication
Proposed hash Highly True High High Low true
based feasible
pseudorandom
password
authentication
Regarding the computation evaluation the host [4] Muxiang Zhang, “Analysis of the SPEKE password-
verifies the proposed hash password sent by user by authenticated key exchange protocol,” IEEE
computing just a single hash function and one comparison Communications Letters, Vol. 8, No. 1, pp. 63-65, January
with the stored last one time password. For the investigation 2004.
of communication factor the host sends the user a hash
value and an integer t. The user returns only a single hash [5] Z. Zhao, Z. Dong, Y. Wang, “Security analysis of a
value. The resultant of the proposed hash based pseudo password-based authentication protocol proposed to IEEE
random password authentication and cryptographic 1363,” Theoretical Computer Science, Vol. 352, No. 1, pp.
password authentication are listed in the below Table 1. 280–287, 2006.
6. Conclusion [6] R. Sekar, V. N. Venkatakrishnan, S. Basu, S. Bhatkar,

and D. C. DuVarney. Model carrying code: A practical
The paper proposed a provably secure hash approach for safe execution of untrusted applications. In
functions based password authentication scheme. This ACM Symposium on Operating Systems Principles (SOSP),
construction provides features such as both the block size of 2003.
the hash function and the output size are completely
scalable. The password hashing method is extremely simple, [7] P. Vogt, F. Nentwich, N. Jovanovic, E. Kirda, C.
rather than send the user’s clear text password to a remote Kruegel, , and G. Vigna. Cross site scripting prevention
site, it sends a hash value derived from the user’s password, with dynamic data tainting and static analysis. In Network
and the site domain name. Password Hash captures all user and Distributed System Security Symposium (NDSS), San
input to a password field and sends hash (pwd, dom) to the Diego 2007.
remote site. The hash is implemented using a Pseudo
Random Function keyed by the password. Since the hash [8] A. Perrig, R. Canetti, D. Song, and D. Tygar, "Eficient
output is tailored to meet server password requirements, the Authentication and Signing of Multicast Streams over Lossy
resulting hashed password is handled normally at the server; Channels," Proc. of IEEE Security and Privacy Symposium
no server modifications are required. This technique deters S & P 2000, May 2000.
password phishing since the password received at a phishing
site is not useful at any other domain. The proposed model [9] O. Hallaraker and G. Vigna. Detecting Malicious
implements the password hashing as a secure and JavaScript Code in Mozilla. In Proceedings of the IEEE
transparent extension to modern browsers. International Conference on Engineering of Complex
Computer Systems (ICECCS), pages 85–94, Shanghai,
References China, June 2005.
[1] N. Chou, R. Ledesma, Y. Teraguchi, and J. Mitchell, “ [10] Antoine Joux. Multicollisions in iterated hash
Client-side defense against web based identity theft “, In functions. Application to cascaded construction. In
Proceedings of Network and Distributed Systems Security Advances in Cryptology - CRYPTO '04 Proceedings,
(NDSS), 2004. Lecture Notes in Computer Science, Vol. 3152, M. Franklin,
ed, Springer-Verlag, 2004, pp. 306-316.
[2] J. A. Halderman, B.Waters, and E. Felten “A convenient
method for securely managing passwords” To appear in [11] Yevgeniy Dodis, Thomas Ristenpart, Thomas
Proceedings of the 14th International World Wide Web Shrimpton. Salvaging Merkle Damgård for Practical
Conference (WWW 2005), 2005. Applications. Preliminary version in Advances in
Cryptology - EUROCRYPT '09 Proceedings, Lecture Notes
[3] F. Hao, P. Zielinski, “A 2-round anonymous veto in Computer Science Vol. 5479, A. Joux, ed, Springer-
protocol,” Proceedings of the 14th International Workshop Verlag, 2009, pp. 371-388.
on Security Protocols, SPW’06, Cambridge, UK, May 2006.
[12] R.C. Merkle. A Certified Digital Signature. In

Advances in Cryptology - CRYPTO '89 Proceedings,
Lecture Notes in Computer Science Vol. 435, G. Brassard,
[13] I. Damgård. A Design Principle for Hash Functions. In

Advances in Cryptology - CRYPTO '89 Proceedings,
Lecture Notes in Computer Science Vol. 435, G. Brassard,
Authors Profile
T.S.Thangavel received the Bsc degree in

Computer Science (Bharathiyar University)
in 1991 and the Msc degree in computer
science (Bharathidasan University) in 1993
and the Mphil degree in Computer Science
(Bharathidasan University) in 2003. He is
pursuing the PhD degree in department of
science and humanities (Anna University).
He is working as an assistant professor in MCA department
at K. S. Rangasamy College of Technology, Tiruchengode
Dr. A. Krishnan received his Ph.D

degree in Electrical Engineering from
IIT, Kanpur. He is now working as an
Academic Dean at K. S. Rangasamy
College of Technology, Tiruchengode
and research guide at Anna University
Chennai. His research interest includes Control system,
Digital Filters, Power Electronics, Digital Signal processing,
Communication Networks. He has been published more
than 176 technical papers at various National/ International
Conference and Journals.
Detection of Masses in Digital Mammograms

J.Subhash Chandra bose1, Marcus Karnan2, and R.Sivakumar2
1
Research Scholar, Anna University,
Coimbatore, India
2
Tamilnadu College of Engineering
Coimbatore, India
drmkarnan@Gmail.Com
nipple and accounts for more than 75% of breast cancers

Abstract: Mammography is at present the best available
technique for early detection of breast cancer. The most common 20% of lobular carcinoma begins in the milk-secreting
breast abnormalities that may indicate breast cancer are masses glands of the breast but otherwise fairly similar in its
and calcifications. The challenge is to quickly and accurately behavior to ductal carcinoma; 5% of other varieties of breast
overcome the development of breast cancer which affects more cancer can arise from the skin, fat, connective tissues and
and more women through the world. Microcalcifications appear other cells present in the breast.
in a mammogram as fine, granular clusters, which are often
difficult to identify in a raw mammogram. Mammogram is one 2. Over View of CAD System
of the best technologies currently being used for diagnosing
breast cancer. Breast cancer is diagnosed at advanced stages Detection of microcalcification is performed in two steps:
with the help of the mammogram image. In this paper an preprocessing and segmentation, the global appearance
intelligent system is designed to diagnose breast cancer through (brightness, contrast, etc.) of the mammogram images may
mammograms, using image processing techniques along with differ, usually due to variations in the recording procedure
intelligent optimization tools such as GA and PSO. The [1, 3, 15]. Initially the film artifacts and x-ray labels are
suspicious region is extracted or segmented using two different removed from the mammogram images and median filter is
approaches such as asymmetry approach and Markov Random
applied to remove the high frequency components (i.e.
Field (MRF) hybrid with Particle Swarm Optimization (PSO)
noise) from the image. Then the mammogram images are
algorithm. 161 pairs of digitized mammograms obtained from
the Mammography Image Analysis Society (MIAS) database are normalized to avoid differences in brightness between the
used to design the proposed diagnosing system. mammograms caused by the recording procedure. And to
increase the reliability of segmentation, the pectoral muscle
Keywords: Breast boarder, nipple identification, Genetic
region is removed from the breast region. The enhanced
Algorithm (GA), Particle Swarm Optimization (PSO), Asymmetry,
Texture Segmentation, Receiver Operating Characteristics (ROC).
images are used for segmentation.
In this paper the suspicious region is extracted or segmented

1. Introduction using two different approaches such as asymmetry approach
In the clinical practice of reading and interpreting medical [11] and Markov Random Field (MRF) hybrid with Particle
images, radiologists often refer to and compare the similar Swarm Optimization (PSO) algorithm. In case of asymmetry
cases with verified diagnostic results in their decision approach, the suspicious regions on digital mammograms
making of detecting and diagnosing suspicious lesions or are segmented based on the asymmetries between
diseases. Microcalcification is one of the keys for early corresponding regions in the left and right breast images.
detection of breast cancer. Cancer involves the Due to the recording procedure the size and shape of the
uncontrolled growth of abnormal cells that have mutated corresponding mammograms does not match. So the
from normal tissues. This growth can kill when these cells mammogram images must be aligned prior to subtraction.
prevent the normal functioning of vital organs or spread The breast border and the nipple points can be used as
throughout the body damaging essential systems. The term reference points for alignment of mammograms. In this
benign refers to a condition, tumor or growth that is not paper the breast border is detected using Genetic Algorithm
cancerous. This means that it does not spread to other parts and the nipple position is identified using a novel method
of the body or invade and destroy nearby tissue. Benign called Particle Swarm Optimization (PSO) algorithm. Using
tumors usually grow slowly. In general, benign tumor or the border points and nipple position as references the
condition is not harmful. However, this is not always the mammogram images are aligned and subtracted to extract
case. If a benign tumor is big enough, its size and weight the suspicious region [10,11].
can press on nearby blood vessels, nerves, organs or
otherwise cause problems. Breast cancer, also known as In the next Texture segmentation technique, a novel
carcinoma, is a malignant growth that begins in the tissues method, Markov Random Field (MRF) hybrid with Particle
of the breast. Swarm Optimization (PSO) algorithm is used to segment
the microcalcifications from the mammogram image.
There are several types of breast cancer. Ductal carcinoma Initially, a unique label is assigned for similar patterns in
begins in the cells lining the ducts that bring milk to the the mammogram image. The MRF based image
segmentation method is a process seeking the optimal breast. Fig 2 shows directions of mammogram capturing to
labeling of the image pixels. The optimum label is which X-ray. Depending on the viewpoint of the X-rays, the
minimizes the MAP estimate. To optimize this MRF based images are classified into different categories, Cranio-
segmentation, Particle Swarm Optimization (PSO) Caudal view and Medio-Lateral Oblique view. Fig 3 and 4
algorithm is implemented to compute the adaptive optimal show view of the images.
threshold value. [12, 13, 14].
.
Figure 2. physically the viewpoints’ directions of

mammogram capturing to X-ray
Figure 1. Flow diagram for mammogram image

Preprocessing and Segmentation.
A Receiver Operating Characteristics (ROC) analysis is

performed to evaluate the classification performances of the
proposed approaches [5]. The area under the ROC curve Az
value is used as a measure of the classification performance.
A higher Az indicates better classification performance
because a larger value of True Positive (TP) is achieved at
each value of False Positive (FP). The proposed algorithms
and the techniques are tested on 161 pairs of digitized
mammograms from Mammography Image Analysis Society
(MIAS) database. The Figs. 1 shows the overview of the
work.
2.1 Mammography Figure 3. Cranio-Caudal view

Medical imaging in general and mammography in
particular, lets physicians evaluate a specific area of the
body of a patient which might be externally visible. Medical
imaging as one of the most important medical developments
of the past thousand years, basically due to the fact that it
provides physicians with physiology and functionality of
organs and cells inside human bodies. Among the different
imaging modalities used for breast cancer detection,
mammography remains the key screening tool for the
detection of breast abnormalities.
In a recent study, the proportion of breast tumors that were

detected in Vermont (US) by screening mammography
increased from 2% during 1974 - 1984 to 36% during 1995
- 1999. However, it is also well known that expert
radiologists can miss a significant portion of abnormalities. Figure 4. Medio-Lateral Oblique view.
In addition, a large number of mammographic abnormalities
turn out to be benign after biopsy. Mammograms capture the It is important to notice that in the MLO views there is one
low energy X-rays which passes through a compressed region corresponding to a portion of the pectoral muscle
which may be present in the left or the right upper corner of
the image. Moreover, some annotations and labels can

appear in the images. Figure 5 and 6 shows the normal left
and right mammogram image. 2.3 Mass Shapes Mass Margins
The shape and margin of a mass are strong signs of their

malignancy or benignancy degree. A distortion in the
normal breast architecture (architectural distortion) refers to
a derangement or disruption of the normal arrangement of
the tissue strands of the breast resulting in a radiating or
haphazard pattern without an associated visible centre. This
includes speculations radiating from a point, and focal
retraction or distortion of the edge of the parenchyma.
Micro-calcifications are tiny calcifications that range from
50 to several hundred microns in diameter, which usually
appear in clusters. In these cases, they are analyzed
according to their size, shape, number, and distribution. The
general rule is that larger, round or oval shaped
calcifications with uniform size have a higher probability of
being associated with a benign process, whereas smaller,
Figure 5. Right Mammogram of a woman irregular, polymorphic, branching calcifications
heterogeneous in size and morphology are more often
associated with a malignant process.
A breast mass, on the other hand, is a localized swelling,

protuberance, or lump in the breast, which usually is
described by its location, size, shape, margin characteristics,
and any other associated findings (i.e. architectural
distortion, X-ray attenuation). Depending on morphologic
criteria, the likelihood of malignancy can be established.
Normally, a benign process is associated with the presence
of circular or oval shapes, while, in contrast, speculated
masses are more probable to Circular Shape Lobular Shape
Speculated Shape Circumscribed Margin Well Defined
Margin Ill Defined Margin. The last one has an increased
probability to be malignant be the sign of a malign process.
The margin refers to the border of a mass, and it should be
Figure 6. Left Mammogram of a woman examined carefully because it is one of the most important
criteria in determining whether the mass is the result of a
2.2 Mammographic Abnormalities benign or malign process. Radiologists classify the margin
among five classes:
There is a large number of types of abnormalities that can be
present in a breast. Among those, signs of breast cancer are • Circumscribed margins, which are well defined and
normally associated with: sharply demarcated with an abrupt transition between the
ü Asymmetry between images of left and right breasts. lesion and the surrounding tissue.
ü Distortion of the normal architecture of the breast • Obscured margins, which are hidden by superimposed or
tissue. adjacent normal tissue.
ü Presence of micro-calcifications in the breast. • Micro-lobulated margins, which have small undulating
circles along the edge of the mass.
ü Presence of masses in the breast.
• Ill-defined margins, which are poorly defined and
It is generally accepted that the essential characteristic of a scattered.
high-quality mammogram is the ability to visualize these
four features. Both breasts are usually considered as almost • Speculated margins, which are marked by radiating thin
symmetric structures. While exact mirror images are not to lines.
be expected when comparing them (usually the first practice
of expert physicians), the tissue patterns within each breast The probability to find a malignancy mass is normally
should be similarly distributed. An asymmetric area may be ordered according to this classification. The more ill-defined
indicative of a developing mass or a variation of normal and speculated the margin, the higher the probability to be
breast tissue. associated with a malignant process. It should be clear that
these morphological aspects can be very subtle and difficult 4. Need for Computer Aided Detection Breast
to diagnose, even for an expert radiologist. Cancer
The idea of computer systems aiding radiologists to detect
breast cancer is not recent. However, the nowadays rapid
3. Digital Mammography development of full digital mammographic systems has
being accompanied by the natural increase of such systems.
As a consequence of the actual digital revolution, traditional A Computer-Aided System (CAD) is a set of automatic or
film-based hospitals are themselves converting to digital semiautomatic tools developed to assist radiologists in the
hospitals, where patient medical records, chart information, detection and evaluation of mammographic images.
and test results are easily available electronically for
physicians from anywhere in the hospital. In that sense, full- The Need for Computer Aided Detection Breast cancer is
digital mammography is gaining importance compared to the most common cancer among women in the United
the nowadays still conventional film-screen mammography, States, other than skin cancer. It is the second leading cause
due to the fact that digital acquisition, digital storage, and of cancer death in women, after lung cancer. The American
digital display processes may be separated and individually Cancer Society estimates that 182,460 women in the United
optimized. States will be found to have invasive breast cancer in 2008.
About 40,480 women will die from the disease this year. In
3.1 Image Acquisition the US, breast cancer is the most common form of cancer
among women and is the second leading cause of cancer
Mammographic Image Analysis Society (MIAS) deaths, after lung cancer. Women in the U.S. have about a 1
All mammograms used in this work are from a mini in 8 lifetime risk of developing invasive breast cancer.
Mammographic database provided by Mammographic
Image Analysis Society (MIAS), which includes 23 cases Incidence of breast cancer in India is on the rise and is
with 28 MCs . The Mammographic Image Analysis Society rapidly becoming the number one cancer in females pushing
(MIAS) Mini Mammographic Database from the Royal the cervical cancer to the second spot. The seriousness of the
Marsden Hospital in London was used in this study. It situation is apparent after going through recent data from
contains 322 images (Medio-Lateral Oblique (MLO)) Indian Council of Medical Research (ICMR).
representing 161 bilateral pairs. The database is divided into
seven categories. The rise is being documented mainly in the metros, but it
can be safely said that many cases in rural areas go
These include normal image pairs and abnormal pairs unnoticed. It is reported that one in 22 women in India is
containing microcalcifications, circumscribed masses, likely to suffer from breast cancer during her lifetime, while
spiculated lesions, ill-defined masses, architectural the figure is definitely more in America with one in eight
distortion and asymmetric densities. Each mammogram being a victim of this deadly cancer. The problem with
from the database is a 1024 x1024 pixels and with a spatial preventing breast cancer is that there is no one cause that
resolution of 200µm/pixel. The odd number cases represent can be pinpointed as being the culprit. Of course screening
the left breast mammogram while the even number cases for the presence of BRCA1 and BRCA2 mutations is
represent the corresponding right breast mammogram. available though it must be admitted of being of little use in
The database lists the film and provides appropriate details the Indian context. It is here that the task of spreading the
as follows: awareness of the prevalence of this cancer and advising
ü 1st column: MIAS database reference number. women on undertaking self-breast examination comes into
the picture. Health officials must try and talk about this
ü 2nd column : Character of background tissue (Fatty, condition so that women have a say in their own health.
Fatty-glandular, or Dense- glandular) Finally, there are procedures like mammography and Fine
Needle Aspiration Cytology (FNAC) and biopsy that need to
ü 3rd column : Class of abnormality present be widely publicized so that women are aware of exactly
(Calcification, Well-defined/circumscribed what they are letting themselves in for.
ü masses, Spiculated masses, Other/ill-defined masses,
Architectural distortion, Asymmetry, or Normal) Early detection of breast cancer increases the survival rate
and increases the treatment options. Screening
ü 4th column: Severity of abnormality (Benign or mammography, or x-ray imaging of the breast, is currently
Malignant) the most effective tool for early detection of breast cancer [6,
7]. Screening mammography examinations are performed
ü 5th and 6th columns : x,y image coordinates of on asymptomatic woman to detect early, clinically
center of abnormality. unsuspected breast cancer. Radiologists visually search
mammograms for specific abnormalities. Some of the
ü 7th column ; Approximate radius (in pixels) of a important signs of breast cancer that radiologists look for
circle enclosing the abnormality. are clusters of micro calcifications, masses, and architectural
distortions.
5. Pre processing and Segmentation than asymmetry method and existing techniques.
Performance of each test is characterized in terms of its
Computer-Aided Detection (CAD) systems have been ability to identify true positives while rejecting false
developed to aid radiologists in detecting mammographic positives using Receiver Operating Characteristic (ROC)
lesions that may indicate the presence of breast cancer. Analysis [4]. The area under the ROC curve is an important
These systems act only as a second reader and the final criterion for evaluating diagnostic performance. Usually it is
decision is made by the radiologist. Recent studies have also referred as the AZ index. The AZ value of ROC curve is just
shown that CAD systems, when used as an aid, have the area under the ROC curve [19, 20].
improved radiologists accuracy of detection of breast cancer
[3, 7, 23, 24]. Computer-Aided Diagnosis (CAD) systems
6. Experiments and Results
for aiding in the decision between follow-up and biopsy are
still in development. It is important to realize that Ultimately, the effectiveness of the proposed technique is
mammographic image analysis is an extremely challenging determined by the extent to which potential abnormalities
task for a number of reasons. can be extracted from corresponding mammograms based
on analysis of their asymmetry image. The Mammographic
The mammogram images were taken from the MIAS Image Analysis Society (MIAS) Database is used to evaluate
database. Initially the X-ray labels and the film artifacts are the technique. All 161 MIAS image pairs were used in this
removed from the mammogram images using gradient- paper. A randomly selected set of 20 bilateral pairs drawn
based tracking algorithm. And the median filter is applied to from the pairs with spiculated and circumscribed lesions
remove the noise from the mammogram images [4, 8, 9, was used for developing the algorithm and for guiding
18]. To apply median filter, the intensity value of every parameter setting. One of the training circumscribed cases
pixel is replaced with the median value of the neighborhood also had an asymmetric density. The remaining abnormal
pixels with the window size of 3×3. Due to the recording and the normal image pairs were used to measure
procedure the brightness between the mammograms may performance. The true positive detection rate and the
vary. In order to reduce the variation, and achieve number of false positive detection rate at various thresholds
computational consistency, the images are normalized, by of the asymmetry images are used to measure the
mapping all mammograms into a fixed intensities range. algorithm’s performance.
In the next step, the pectoral region is removed from the These rates are represented using Receiver Operating
breast region to increase the reliability of the segmentation. Characteristic (ROC) curves. True Positive (TP) and False
It is achieved by applying histogram-based thresholding. Positive (FP) rates are calculated at 20 different thresholds
The enhanced mammogram images are evaluated using selected on asymmetry image pixels to generate an ROC
signal-to-noise ratio [21,22]. In the next step, the suspicious curve. A region extracted in the asymmetry image, which
regions are extracted from the enhanced mammogram overlaps with a true abnormality as provided in the ground
image using two different approaches such as asymmetry truth of the image, is called a true positive detection. An
and texture segmentation. Asymmetry is based on the overlap means that at least 80% of the region extracted lies
asymmetry between the corresponding left and right image. within the circle indicating a true abnormality as determined
In this paper, the breast border and the nipple position are by MIAS database.
considered as reference points for mammogram alignment.
For example, the mammogram mdb239.pgm, the spatial
The Genetic Algorithm (GA) is used for breast border coordinate position of the suspicious region, x and y are
detection and for nipple identification Particle Swarm 567, 808 respectively, and the radius is 25 pixels. The
Optimization (PSO) algorithm is applied. Once the resultant asymmetry image contains the suspicious region],
mammogram images are aligned they can be subtracted to with the radius of 25 pixels. Compared to the MIAS
extract the suspicious region. In case of texture information on mdb239.pgm, results from the proposed
segmentation, Markov Random Field (MRF) is applied to method overlaps 99% of the specified region and this image
label the image pixels. For labeling, a kernel is extracted for is classified as true positive image. Suppose the overlap is
each pixel, kernel is a window of neighborhood pixels with less than 80% of the specified region, and then the image is
the size of 5×5. A unique label is assigned to the kernels considered as false positive image.
having similar patterns [17]. A pattern matrix is maintained
to store the dissimilar patterns in the image. For each
patterns in the pattern matrix, the posterior energy function
value is calculated. The challenge of finding the MAP
estimate of the segmentation is search for the optimum
label, which minimizes the posterior energy function. In this
paper a new effective approach, PSO is applied for the
minimization of the energy function.
The segmentation from both the methods is compared with

the MIAS information, by adaptive thresholding the
segmented image using various operating points. The
statistical results show that the MRF-PSO performs better
Original image segmented image
Figure 7. Truth Image of mammogram mdb239.pgm Figure 9. Segmented Image using MRF – PSO
Detection Ratio: The area under the ROC curve (Az value) is
In the previous methods such as Sallam and Bowyer, Lau
an important criterion for evaluating diagnostic performance
and Bischof have taken the overlap region of only 40% are
. The AZ value of ROC curve should be computed by
considered as true positive[11, 16]. But in this paper, the
normalizing the area under the ROC curve by the range of
true positive is considered only at 80% of overlap occurs.
the abscissa. The value of AZ is 1.0 when the diagnostic
All other regions extracted by the algorithm are labeled as
detection has perfect performance, which means that TP rate
false positives. Figure 5,6,7 shows the ROC curves
is 100% and FP rate is 0%.
generated on the full test set, using 20 operating points.
The Az value for the proposed MRF PSO algorithm is 0.983.
In general, it is expected that the true positive detection rate
The Table 1 shows the comparison of classification rate
in an ROC curve will continue to increase or remain
between the previous works and the proposed method.
constant as the number of false positives increase. In this
Figure 10, 11 and 12: shows the ROC curve – GA, ROC
case the true positive rate actually drops at certain points. If
curve – PSO and ROC curve – GA and PSO. Fig 13 shows
the threshold value is low true detections may become
the bar chart of Comparison of Classification rate.
merged with false positive regions.
Fig 7 shows the Truth Image of mammogram mdb239.pgm.

Figure 8 shows the Asymmetry Image using PSO. Figure 9
shows the extracting suspicious region from background
tissue using MRF – PSO technique.
Figure 10. shows the ROC curve – GA
Figure 8. Asymmetry Image using PSO

Figure 11. shows the ROC curve – PSO
Figure 13. Comparison of Classification Rate
7. Conclusion
In this paper the suspicious region is extracted or segmented
using two different approaches such as asymmetry approach
and Markov Random Field (MRF) hybrid with Particle
Swarm Optimization (PSO) algorithm. In case of asymmetry
approach, the suspicious regions on digital mammograms
are segmented based on the asymmetries between
corresponding regions in the left and right breast images.
The breast border is detected using Genetic Algorithm and
the nipple position is identified using a method called
Particle Swarm Optimization (PSO) algorithm. In the
texture segmentation technique, Markov Random Field
Figure 12. shows the ROC curves generated on the full test (MRF) hybrid with Particle Swarm Optimization (PSO)
set for GA and PSO algorithm is used to segment the microcalcifications from
the mammogram image. To optimize this MRF based
Table 1: Comparison of Classification Rate segmentation, Particle Swarm Optimization (PSO)
algorithm is implemented. A Receiver Operating
Sl. Authors Methods Classificat Characteristics (ROC) analysis is performed to evaluate the
No. ion Rate classification performances of the proposed approaches. The
1 Lau and Bischof, Asymmetry 85.00% approach using MRF-PSO based segmentation was superior
1991 [11] Measures to the other methods. The overall performance and the
2 Sallam and Unwarping 86.60% results show that the particle Swarm Optimization
Bowyer,1999 [16] Technique algorithm performs better than other methods
3 Ferrari and Directional 74.40% comparatively. Fig 14 shows the Snapshoot for the detailed
Rangayyan, 2001 Filtering with result of the Detection of Microcalcification in
[2] Gabor wavelets mammograms using MATLAB 7.
4 Thangavel and MRF-ACO 94.80%
Karnan,2005 [24]
5 Bilateral 94.60%
The proposed Subtraction
Metaheuristic using PSO
6 Approach MRF-PSO 98.30%
Segmentation
[11] Lau, T.K and Bischof,W., “Automated detection of

breast tumors using the asymmetry approach,” Comput.
Biomed. Res. 24, 273-295(1991).
[12] M. Karnan, R. Sivakumar, M. Almelumangai, K.
Selvanayagi and T. Logeswari,” Hybrid Particle Swarm
Optimization for Automatically Detect the Breast
Border and Nipple position to Identify the Suspicious
Regions on Digital Mammograms Based on
Asymmetries”, International Journal of Soft Computing
3 (3): 220-223, 2008
[13] Karnan and K.Thangavel, Automatic Detection
of the Breast Border and Nipple
Position on Digital Mammograms Using
Genetic Algorithm, International Journal on
Computer Methods and Programs in
Biomedicine (Elsvier). VOL 87, pp.12-20 2007
[14] M.Karnan, K.Thangavel, “Weight Updating in BPN
Network Using Ant Colony Optimization for
Classification of Microcalcifications in Mammograms,
Figure 14. Snapshoot for the Detection of Masses in International Journal of Computing and Applications,
mammograms Vol:2,no.2, pp 95-109, 2007
[15] Naga R. Mudigonda, Rangaraj M Rangayyan and Leo
Desautel.J.E, ”Gradient & Texture analysis for the
References
classification of Mammographic Masses. IEEE trans.
[1] Dhawan,P, Buelloni,G and Gordon,R. “Enhancement of on MI, vol. 19, no. 10, Oct. 2000 pp. 1032 ‘ 1042.
mammgraphic features by optimal adaptive [16] Sallam, M.Y., and Bowyer, K.W.: “Registration and
neighborhood image processing,” IEEEtans. Acoust, difference analysis of corresponding mammogram
Speech, Signal processing. images,” Medical Image Analysis, vol. 3, no. 2, pp:
[2] Ferrari, R.J., Rangayyan, R.M., Desautels, J.E.L., 103-118, 1999
Borges, R.A., and Frere, A.F.: “Analysis of Asymmetry [17] Stephan Olariu, Albert Y. Zomaya, “Handbook of
in Mammograms via Directional Filtering With Gabor Bioinspired Algorithms and Applications” CHAPMAN
Wavelets,” IEEE Transactions on Medical Imaging, & HALL/CRC COMPUTER and INFORMATION
vol. 20, no. 9, pp: 953–964, 2001. SCIENCE SERIES, 2006
[3] Guido M. te Brake and Nico Karssemeijer,”Single and [18] K.Thangavel and M.Karnan, “CAD system for
multiscale detection of masses in digital Preprocessing and Enhancement of Digital
mammograms., IEEE Transactions on Medical Mammograms,” International Journal on Graphics
Imaging, vol. 18, No. 7, July 1999, pp. 628-638. Vision and Image Processing, vol. 9, no. 9, pp: 69-74,
[4] Gonzalez, R.C and Wintz,P, Digital Image 2006.
Processing(Add-Wesley, Reading, 1987. [19] K.Thangavel, M.Karnan, P.Jaganathan, A.
[5] J. A. Hanley, and B. J. McNeil. The meaning and use Pethalakshmi, R. Sivakumar, “Computer-Aided
of the area under a receiver operating characteristic Diagnosis: Automatic detection of microcalcifications
(ROC), curve. Radiology, 143:29–36, 1982. in Mammography Images using Soft Computing”,
[6] Harvey.J.E, Fajardo.L.L, & Inis.G.A, ”Previous Lecturer Notes in Engineering and Computer Science
mammograms in patients with impalpable breast IMECS Hong Kongm PP 280-286 June, 2006.
carcinoma: Rctrospective vs. blinded interpretation. [20] M.Karnan, K.Thangavel, R.Sivakumar, “Ant Colony
AJR, vol. 161, PP. 1167-1172, 1993 optimization algorithm for Feature Selection and
[7] Hult.I.W, Astley.S.M, & Boggis.C.R.M,. Prompting as Classification of Microcalcifications in Mammograms,
an aid to diagnosis in mammography; in Digital IEEE International Conference on Advanced
Mammography,. A. G. Gala, S. M. Astley, D. R. Computing and Communications, 2006. ,IEEE press,
Dance, & A. Y. Cairns, Eds. Amsterdam. The pp: 298-303,2006
Netherlands : Elsevier, 1994, pp. 389-398. [21] K.Thangavel, M.Karnan, R. Siva Kumar, and A. Kaja
[8] Jain, A.K,” fundamentals of digital image processing. Mohideen. “Automatic Detection of Microcalcification
1995. in Mammograms-A Review,” International Journal on
[9] Jain, A.K, Duin, R.P.W, and Mao, J. Statistical Graphics Vision and Image Processing, vol. 5, no. 5,
pattern recognition: A review. IEEE Transactions on pp: 31-61, 2005.
Pattern Analysis and Machine Intelligence, 22(1):4–37, [22] K.Thangavel and M.Karnan. “Computer Aided
2000. Diagnosis in Digital Mammograms: Detection of
[10] Jong kook kim and Hyun wook park,”Statistical Microcalcifications by Meta Heuristic Algorithms,”
textural features for detection of micro calcification in International Journal on Graphics Vision and Image
digitized mammogram”,IEEE trans.on medi. Imaging, Processing, vol. 7, no. 7, pp: 41-55, 2005.
vol.no.18, no.3, mar 1999.
[23] K.Thangavel and M.Karnan, “Automatic Detection of India.His area of interests includes medical image
Asymmetries in Mammograms Using Genetic processing, artificial intelligence, neural network, and fuzzy
Algorithm,” International Journal on Artificial logic
Intelligence and Machine Learning, vol. 5, no. 3, pp:
55-62, 2005.
[24] K.Thangavel, M.Karnan, R.Siva Kumar and
A.Kajamohideen, “Segmentation and Classification of
Microcalcification in Mammograms Using the Ant
Colony System,” International Journal on Artificial
Intelligence and Machine Learning, vol. 5, no. 3, pp:
29-40, 2005.
Authors Profile
J.Subashchandra bose Received the

Master of Computer Science and
Engineering Degree from Anna University
Chennai, Tamil Nadu, India, Bachelor of
Electrical and Electronics Engineering
Degree from Anna University Chennai,
Tamil Nadu, India, Currently he is
working as Assistant Professor,
Department of Computer Science & Engineering,
Hindusthan College of Engineering and Technology, Tamil
Nadu, India, and doing part-time paper in the Department
of computer Science and Engineering, Anna University-
Coimbatore, Tamil Nadu, India. His area of interests
includes medical image processing, artificial intelligence,
neural network, and fuzzy logic
Marcus Karnan received the BE

Degree in Electrical and Electronics
Engineering from Government College
of Technology,Bharathiar University,
India. Received the ME Degree in
Computer Science and Engineering
from Government College of
Engineering ,Manonmaniam Sundaranar University in
2000. Received the PhD degree in CSE from Gandhigram
Rural University, India in 2007, Currently he is working as
Professor, Department of Computer Science & Engineering
Department, Tamilnadu College of Engineering, India. He
has been in teaching since 1998 and has more than eleven
years in industrial and paper experience. His area of
interests includes medical image processing, artificial
intelligence, neural network, genetic algorithm, pattern
recognition and fuzzy logic
Sivakumar Ramakrishnan, Received

the Master of Computer Science and
Engineering Degree from Computer
Science and Engineering Department,
from Government College of
Engineering, Manonmaniam
Sundaranar University, Tamil Nadu,
India, in 2000. Currently he is working
as Assistant Professor, Department of Computer Science &
Engineering, Tamilnadu College of Engineering, Tamil
Nadu, India. And doing part-time paper in the Department
of computer Science, Bharathiar University, Tamil Nadu,
A Dynamic Trust-Based Context-Aware

Authentication Framework with Privacy Preserving
Abdallah MHAMED1, Pierre E. ABI-CHAR1, Bachar EL-HASSAN2 Mounir MOKHTARI1
1
Laboratory of Reseaux et Service De Telecommunication
(RST), Telecom SudParis (ex. INT), 9 Rue Charles Fourier, Evry, France
{Pierre.abi_char, abdallah.mhamed, mounir.mokhtari}@it-sudparis.eu
2
Laboratory of Electronic Systems, Telecommunication and
Networking (LASTRE), Faculty of Engineering, Branch 1, Al Arz
Street, El Kobbeh, Tripoli, Lebanon.
Bachar_elhassan@ul.edu.lb
trust denotes the grounds for confidence that a system will

Abstract: As ubiquitous technologies ingrain themselves
further into our lives, rapid progress has been made in context- meet its security objectives. The development of mobile
aware computing. Context-aware environments are set to communications technologies and ubiquitous computing
become a reality. However, major challenges remain to be paradigm and the convergence of m-healthcare, m-business,
addressed including privacy, authentication, access control, and m-entertainment and m-education services have raised the
trust. These security challenges have to be non-intrusive, urgency of dealing with privacy threats (i.e. personal
intelligent, and able to adapt to the rapidly changing contexts of
information, etc.). These threats are caused by the detection
users. Context-aware environments are expected to make these
challenges more accurate and to consider them in place from the of personal sensitive information such as location,
start, so that a mutual trust relationship can be formed between preferences, and activities about individuals through sensors
entities. It is therefore, a key challenge in a ubiquitous network or invisible computing devices gathering collating data and
society to design an effective privacy preserving authentication deriving user context. Moreover, the ubiquitous computing
and access control framework that adequately meet security environment is characterized by people constantly moving,
requirements posed by the context-aware service paradigm in and engaged in numerous activities simultaneously.
pervasive computing environment. In this paper, we propose a
Therefore, we proposed an authentication and access control
security framework that integrates context-awareness to perform
authentication and access control approach in a very flexible agent framework for context-aware services. Our
and scalable model that is both context-aware and privacy framework’s objectives are to provide the most suitable
preserving. Moreover, we show how our framework can be security scheme on the basis of context, such as users'
integrated with trust management. In this paper, we focus on location and profiles, and to protect personal information
introducing an anonymous authentication and access control such as user location, user's ID, etc. This paper provides us a
scheme to secure interactions between users and services in scheme to protect privacy of users and to maintain the
ubiquitous environments. The architecture focuses on the
authentication of users who request access to the resources of
flexibility for users while using available service in
smart environment system through static devices (i.e. smart card, ubiquitous environments. The ultimate goal is anonymity
RFID, etc.), or dynamic devices (i.e. PDA, mobile phones, etc.). which keeps the users anonymously interacting with the
services, through that, preserving context privacy of users.
Keywords: Context-Aware, Authentication, Access Control,
Smart Spaces, Privacy Control, Fuzzy Logic, Trust Management, And also it keeps confidentiality and integrity on
Risk Assessment, Quality of Privacy. communication channels. The proposed schemes is at
application level without relying on any underlying system
1. Introduction infrastructure such as “light house” or “Mist router” in [6].
This scheme possesses many desirable security properties,
The growing evolution of Information and Communication such as anonymity, nonlinkability, trust management, etc.
Technology (ICT) systems towards more pervasive and The rest of this paper is as follows. Context-aware
ubiquitous infrastructures contribute significantly to the definition and usage, authentication and access control
deployment of services anywhere, at anytime and for characteristics and their privacy effects, and trust
anyone. To provide personalized services in such management definition and properties are outlined in
infrastructures, we should consider both user's privacy and Section 2. Section 3 provides an outline for the
security requirements and context-awareness environment. mathematical backgrounds needed for our protocol process.
Security, Privacy and Trust in pervasive computing are Section 4 provides a summary regarding relevant related
currently hot issues in digital information technology area. work. Our proposed agent framework, its process
Security is used to describe techniques that control who may descriptions, and security discussion are introduced in
use or modify private data and context information, privacy Section 5, 6, and 7 respectively. Finally, the paper future
is viewed as the ability of an entity to determine whether, work and conclusion are described in Section 8.
when, and to whom information is to be released and finally
2. Pervasive Computing Paradigm individuals to check on and correct any information held
about them for use in authentication. Context-aware services
In this section we briefly introduce some assumptions,
should be able to trust context data provided to them from
concepts, and values that constitute a real way for viewing
these various sources and to respond to changes. The
the necessity to a novel scheme.
dynamic nature of a context-aware environment necessitates
2.1 Context-Aware the need for a very active and flexible authentication
Context-Aware computing is an emerging computing mechanism that allows members across different domains to
paradigm that tries to exploit information about the context identify and communicate with each other with a reasonable
of its users to provide new or improved services. [2] Have level of trust. More generally, systems architects' developers
defined context as: any information that can be used to should focus more on reconciling authentication and privacy
characterize the situation of an entity. An entity is a person, goals when designing, developing, and deploying systems.
place, or object that is considered relevant to the interaction Understanding security needs and developing appropriate
between a user and an application, including the user and threat models are keys for determining whether and what
applications themselves. The use of context information authentication are necessary and what kind is needed.
gives a number of advantages in communication as According to [1], [3] the context-aware authentication
motivated before. Context-awareness is an enabling service has to hold the following distinguishing properties:
technology to build helpers that are disappearing from the
users’ perception. This allows creating new classes of Context-Awareness: A context-aware service has to use
services. The combination of several context values provides
context data to provide relevant services to users. The
a very powerful mechanism to determine the current
security system adapts itself to match with the dynamism of
situation.
context information. It also has to be able to prune its
services accordingly to changes in context data, such as
2.2 Authentication and Access Control changes in time, location, activity, etc. Therefore, it is
The title Authentication systems are used for security critical to check the authenticity and integrity of the context
purposes to verify the authenticity of one or more parties or data from context-providers.
entities during a transaction. Most traditional authentication Autonomy: The context-aware service should involve the
methods either do not scale well in massively distributed last human intervention possible. The security may
environments, with hundreds or thousands of embedded improvise new policies based on the available or new
devices like smart spaces, or they are inconvenient for users context data.
roaming around within smart environments. In addition, Scalability: The authentication service has to be capable
authentication in smart environments can not use a one- of bootstrapping trust and authentication across
size-fits-all approach, as authentication requirements differ heterogeneous domains.
greatly among different spaces and different applications Flexibility: In an open, massively distributed, pervasive
and contexts within the same smart space. In general, users computing system, using different means of authentication
must be able to authenticate with other entities with a varied should be made possible, and it does not have to be
level of confidence, in a transparent, convenient, and private constrained to a specific format. Therefore, the system has
manner. The concept of context-aware authentication and to be able to provide a great level of customization to each
access control is: (1) Collect and recognize the user’s individual.
current context, and (2) Generate and control a secure user Privacy-Preserving: In a context-aware environment,
environment based on the current context. Generally, the there will be thousands of sensors recording every type of
context includes user’s location and services, present important information about users. They will silently track
information, environmental information (temperature, user's location, preferences, and activities in the
loudness, and brightness), terminal attributes, and network environment. Therefore, protecting privacy of the user is
status (QoS), etc important, and there has to be a provision to protect it
against abuse.
Anonymity: The real identity of a user should never
2.2.1 Privacy Effects
be revealed from the communications exchanged between
An inherent tension exists between authentication and the user and a server unless it is intentionally disclosed by
privacy because the act of authentication often involves the user. Different communication sessions between the
some disclosure or confirmation of personal information. same user and service should not be linkable. Different
System designers sometimes fail to consider the myriad devices of user should not be linkable.
impact that authentication affects privacy. When designing Context privacy: Except users want to disclose their
an authentication system, selecting one for use, or context information (location, time, preference, name of
developing policies for one, we should authenticate only for services, etc), no one should know about such information
necessary (well-defined purposes), minimize the scope of the even system administrator or service providers they interact
data collected, articulate what entities will have access to the with.
collected data, articulate what kinds of access to and use of Confidentiality and integrity: System should provide
the data will be allowed, and finally provide means for protection measures on the communication channels while
users are interacting with services in order to protect and Koblitz [19]. The elliptic curves which are based on the
sensitive information from eavesdroppers. elliptic curve discrete logarithm problem over a finite field
Nonlinkability: Ideally, nonlinkability means that, for have some advantages than other systems: the key size can
both insiders (i.e., service) and outsiders, 1) neither of them be much smaller than the other schemes since only
could ascribe any session to a particular user, and 2) neither exponential-time attacks have been known so far if the curve
of them could link two different sessions to the same user. is carefully chosen [20], and the elliptic curve discrete
logarithms might be still intractable even if factoring and
the multiplicative group discrete logarithm are broken. In
2.3 Trust Management
this paper we use an elliptic curve E defined over a finite
Trust in pervasive computing is a complex subject relating
to belief in the honesty, trustfulness, competence, and field F p . The elliptic curve parameters to be selected
reliability of an entity. In the context of pervasive [21,22] are:
computing, trust is usually specified in terms of a 1 -Two field elements a and b ∈ F p , which define the
relationship between a resource or service requester and a
resource or service provider [1]. To trust pervasive equation of the elliptic curve E over Fp
computing systems, we must be able to manage the privacy,
(i.e., y2 = x3 + ax + b ) in case p ≥ 4, where
confidentiality, availability, and controlled access to digital
information as it flows through the system. Trust forms the 4a + 27b ≠ 0 .
3 2
basis for allowing a requester to use services or manipulate 2 -Two field elements x p and y p in F p which define a
resources owned by a service provider. Also, it may
influence a requester's decision to use a service or resource finite pointP ( x p , y p ) of prime order in E ( F p ) ( P is not
from a provider. So trust is an important factor in the equal to O , where O denotes the point at infinity).
decision-making process. 3 -The order n of the point P .
For trust establishment in the pervasive computing The Elliptic Curve domain parameter can be verified to
environments, the mobility and uncertainty of the systems meet the following requirements [21] and [22]. In order to
and clients need more dynamic and flexible trust strategies. avoid the Pollard-rho [23] and Pohling-Hellman algorithms
In addition to the traditional trust strategies such as access for the elliptic curve discrete logarithm problem, it is
control and PKI, other trust strategies are proposed and used necessary that the number of F p -rational points on E ,
for trust establishment and management in pervasive
computing environments [1]. These trust strategies are: denoted by # E ( F p ) , be divisible by a sufficiently large
Trust Negotiation: Is needed when system does not have prime n . To avoid the reduction algorithms of Menezes,
the client information and there is no third party to consult Okamoto [29] and Vanstone [24] and Frey and Ruck [25],
with on the trustworthiness of the client. In this case, it is the curve should be non-supersingular (i.e., p should not
only reasonable and practical for the client and system to
build their trust relationship by disclosing their credentials devide ( p + 1−# E ( F p )) . To avoid the attack of Semaev
gradually to meet the access control policies of each other. [26] on F p -anomalous curves, the curve should not be F p -
Trust Delegation: Is needed when one entity in the system
trusts the client and can assign its rights to the clients. anomalous (i.e., # E ( F p ) ≠ p ).
Trust Based on Reputation: Is used when the system can
derive the clients' trustworthiness from the client's behavior 3.4 Bilinear Pairing:
records. Because the system may need to collect the clients' This section briefly describes the bilinear pairing, the BDHP
reputation from other peer systems, the trust level of the and CDHP assumptions. Let G1 and G2 denote two groups
network and the peers systems are taken into account when
of prime q , where G1 is an additive group that consists of
deciding the trust reputation of the clients.
Trust Based on Context and Ontology: Can be use when points on an elliptic curve, and G2 is a multiplicative group
clients and the systems may have the smart sensing devices. of a finite field. A bilinear pairing is a computable bilinear
This ontology information can help the system to determine map between two groups, which could be the modified Weil
the trust levels of its clients or assign them trust rights in the pairing or the modified Tate pairing [27]-[28]. For our
given context. proposed architecture within this paper, we let e denote a
general bilinear map e : G1 × G2 → G2 which has the
3. Mathematical Backgrounds: following four properties:
In this section we briefly introduce some mathematical 1-Bilinear: if P , Q, R ∈ G1 and a ∈ Z q* ,
backgrounds necessary for the description of our scheme.
e( P + Q, R) = e( P , R).e(Q, R)
3.3 Elliptic Curve Cryptography, (ECC): e( P , Q + R) = e( P , Q ).e( P , R) and
Many researchers have examined elliptic curve
cryptosystems, which were firstly proposed by Miller [18] e( aP , Q) = e( P , aQ) = e( P , Q ) a .
2 -Non-degenerate: There exists P , Q ∈ G1 , such that Fire04[47] No Yes N.A N.A N.A No Yes Yes No
Dim04[45] No N.A N.A N.A Yes No Yes Yes Yes
e( P , Q) ≠ 1 . Dim05[46] No No N.A N.A N.A No No Yes Yes
Yuan06[43] No Yes N.A N.A N.A N.A Yes Yes No
3 -Computability: There exist efficient algorithms to
Yuan06[44] No Yes N.A N.A N.A N.A Yes Yes No
compute e( P , Q) for all P , Q ∈ G1 . Ries07[42] No Yes N.A N.A N.A N.A Yes Yes Yes
Xu07[41] No No N.A N.A N.A N.A No Yes Yes
4 -Alternative: e( P , Q) = e(Q, P ) −1 . Uddin08[39] No Yes N.A N.A N.A N.A Yes Yes No
Mohan08[40] No Yes N.A N.A N.A N.A Yes Yes No
Definition 1 -The bilinear Diffie-Hellman problem
The comparison is done based on privacy and security
(BHDP) for a bilinear pairing is defined as follows: Given
related features. The following comparison cover these
P , aP , bP , cP ∈ G1 , where a , b, c are random numbers features includes Trust Management (TM), Context-
from Z q* , compute e( P , P ) abc ∈ G1 . BDHP assumption: Awareness (CA), Mutual Authentication (MA), User
Context Privacy (UCP), Non-Linkability (NL), Data
The BDHP problem is assumed to be hard, that is, there is Confidentiality and Integrity (DCI), Differentiated Service
no polynomial time algorithm to solve BDHP problem with Access Control (DS), Level of Anonymity (LA), Quality of
non-negligible probability. Privacy (QoP), and Risk Awareness (RA).
Definition 2 -The computational Diffie-Hellman problem
(CDHP) is defined as follows: Given P , aP , bP ∈ G1 , Closely Related Work: Authors, in [11], have defined a
where a, b are random numbers from Z * model that uses contextual attributes to achieve an approach
q, compute
to authentication that is better suited for dynamic, mobile
abP ∈ G1 . CDHP assumption: There exists no algorithm computing environments. They examined the use of trusted
running in polynomial time, which can solve the CDHP platforms to provide assurances for these contextual
problem with non-negligible probability. attributes. Although authors claimed that their model
provides a seamless and flexible user experience that can
4. Related Work: protect privacy and reduce administrative overhead, it does
not provides trust and reasoning and there no mention about
how to protect privacy (i.e, user, attributes, and data
Basic Related Work: Recently, many papers have been
privacy). Marc Langheinrich [10], introduces a privacy
published to address mechanisms designed against security,
awareness system that allows data collectors to both
privacy threats, and trust in pervasive computing
announce and implement data usage policies. The
environments. However, most of these designs fall in the
announced data collections of each services and their
scope of establishing a general security framework
policies is delegated by a mobile privacy assistant to a
identifying general security and privacy requirements. Some
personal privacy proxy residing on the platform, which
of these efforts focused on designing security infrastructures
interact with corresponding service proxies and inquires
to protect users' personal information such as Mix-Network
their privacy policies (Privacy Beacon). Corner et al. [12]
architecture, Mist system, Aware Home Architecture, Solar,
describe Transient Authentication as a means of
etc. Others focused on designing identity management
authenticating users with devices through a small, short-
approach. Some efforts focused on providing privacy control
ranged wireless communications token. This research is
through integrating privacy preferences (P3P), policies and
limited to the use of location-based context (i.e., proximity)
context-aware systems. Various trust management strategies
as an attribute in authentication.
including, trust negotiations and trust establishments, have
A similar approach is taken by Glynos et al. [14] where
been proposed to prevent unauthorized disclosure of any
they combined traditional authentication with a limited set
relevant information that can be used for inferring sensitive
of contextual information used to identify users. Other
credentials. Based on most important relevant schemes, a
similar approaches were taken by [52]-[53] where they also
full exhaustive comparison study of the most important
have used a limited set of attributes to perform
features is provided in [1] and summarized in Table 1.
authentication process. However, we have presented a more
generic approach that allows any attributes to be used for
Table 1: Protocols Security Features Comparison (P:
authentication. Creese et al. [13] present a general overview
Partially, H: High, M: Medium, N.A: Not Available)
MA UCP NL LA DCI DS CA TM RA of security requirements for authentication in pervasive
Mist[6] P. N.A Yes H. Yes No No No No computing and discuss how traditional authentication does
Aware H[9] Yes Yes N.A N.A Yes No No No No not fit these requirements. Although they discuss
Solar[8] N.A No N.A N.A N.A N.A No No No authentication of entities using attributes, they did not
PawS[10] P. Yes N.A H. Yes No Yes No No
Jend02[7] No No No M. No Yes No No No
present a framework for authentication as we have done. In
He04[5] Yes Yes No M. No No Yes No No [15], Authors present a service provision mechanism which
Ren05[48] Yes Yes P. H. Yes Yes Yes No No can enable effective service provision based on semantic
Ren06[49] Yes Yes P. H. No Yes Yes No No similarity measure with the combination of user profiles and
Kim07[4] Yes Yes Yes H. Yes Yes Yes No No
Yes Yes Yes H. Yes Yes Yes No No
situation context in WLAN enabled environment. The paper
Ren07[50]
suggests the combination of user profiles and contextual
information to provide a more pervasive service experience interact with one another. A high-level overview of these
in smart assistive environments with mobile device. Behzad logical components and how they interact is given in
et al. [16] propose a framework to construct a context-aware following figure 2.
authentication system. Although the framework is flexible
and privacy preserving, it is not context-aware user
authentication and does not support user trustworthiness
evaluation neither user role assignment. Moreover, the
framework is designed to be applicable to Ad-Hoc network,
does not provide users a way to control attributes, and not
suitable for static environments where users may be holding
RFID tags only. In [17], authors propose an authentication
scheme for a mobile ubiquitous environment, in which the
trustworthiness of a user’s device is authenticated
anonymously to a remote Service Provider (verifier), during
the service discovery process. However, the scheme does not
provide support for contextual information, and does not
support fuzzy private matching.
5. Toward A New Solution:

Here, we outline our proposed authentication-based privacy
enhancing infrastructure. Our framework is based on a
Figure 2. A High Overview of the Framework
privacy control layer, a context-aware authentication broker,
a context-aware Access Control broker and the use of
Our model is based on contextual information obtained
attributes-based private set intersection and trust evaluation
from a distributed network of sensors. In the following we
engines. Our framework is a layered architecture that
will detail the functionality of these components.
discriminates service providers (context consumers), privacy
control process, authentication process, access control
process, service receivers (context producers) and the
6. Context-Based Authentication Scheme
borders that separate these layers. The figure below (Figure
1) shows the process of granting access to resources with the The dynamic nature of a context-aware environment
help of user and attributes. Attributes can contain identity necessitates the need for a very active, flexible
and other contextual information (i.e user's profile). authentication mechanism that allows users to securely
authenticate and access services with a reasonable level of
trust and while privacy is preserved. Our framework consists
of the following layers: A Privacy Control Layer (Layer 1)
for providing users a way for controlling privacy over the
reveal of their personal and contextual information. An
access layer, (Layer 2) which combines authentication
process (SubLayer 2.1) and access control (SubLayer 2.2)
both at context-aware level. The authentication process
(Figure 3) contains a trust process (Figure 4 ) where the
trustworthiness parameters value are computed in order to
provide access to users, and contains a private set
Figure 1. Context-Aware Framework intersection process (PSI). In the following sections, we
detail the functionality of these processes and describe how
In our framework, we design an integration scenario they interact with one another. In this section, we present
where mobile subjects (i.e users) carrying embedded devices the access process architecture scheme. Figure 3 shows the
(i.e., smart phones, PDA, etc.) receive pervasive services authentication process architecture. The purpose of access
according to their identity and real-time context information process is to provide authentication and access control
environments. The cornerstone of our framework is the according to user's profile and environment (attributes-based
flexibility to provide authentication and access control for authentication and access-control) and then to establish a
independent and dependent (With a special need) people secure communication link between entities, whilst
both at context level and where privacy is preserved. preserving the privacy of users. Moreover, we will introduce
Moreover, our framework provides a distributed context-aware based user trustworthiness and role's required
infrastructure that allows the tracking of the context in a trustworthiness and show how to improve user assignment
real-time manner. In the following sections, we detail the and role activation.
functionality of these components and describe how they
Our framework is composed of various mechanisms that public key [30] as PU = xu .PPub = xu .s.G , where xu ∈ Zq*
altogether yield a flexible, scalable context-aware based
is generated on user's or entity's behavior.
authentication. In our model, confidence and trust are
defined based on each user's contextual information. First,
In addition, we define a role as a set of pair of public and
we introduce the system parameters initialization used for
private keys belonging to the role. Each role is represented
the protocol process. Next, we state the different phases
upon which the scheme is based. Finally, we describe the as r = ( r pub , r priv ) . When a role ri is added to the system,
operation of the architecture. the TKGC picks a random rpki as ri 's private key and
sets RPKi = rpki .G as ri 's public key. To assign the role
ri to a user with an identity ID , the TKGC check the
user ID , computes QID = H (ID) , and generates the
user's assigned key KID ri corresponding to ri with
KID ri = rpki .Q ( ID) and where rpki is the ri 's private
key.
Finally, the TKGC sends Si , Pi , Z and the set of
Q = {QSP , QPSI , QTE } to the user via a secure channel.
The User-Based Authentication Engine UBAE manages
an stores, for each user U i with an ED , a record pair
consisting of Qi , Si , s1 , s2 , where (s1 , s 2 ) are the
prover’s secret.
The Table 2, below, shows the mathematical parameters that
are used in our proposed framework.
Figure 3. The Authentication Architecture Process
Table 2: Mathematical notations
Index Explanation
6.1 The Scheme: The trusted key generation center
TKGC
Our infrastructure involves a context-based authentication
G1 An additive group with prime order q
process, a context-based access control process, a Trusted
Key Generation Center ( TKGC ), embedded devices G2 A multiplicative group with prime order q
EDs , Service Providers ( SP ), Inference engines IEs , G A generator of G1
and users denoted by ( U i ). The
TKGC chooses two primes
PPub The public key of TKGC, PPub = s.G
order group G1 and G2 of prime order q . q is a prime
s Chosen from Z q* by TKGC, s and kept secret
which is large enough to make solving discrete logarithm
problem in G1 and G2 infeasible. The TKGC chooses IDi ∈ {0,1}
*
IDi The identity of the user i,
G as a generator of G1 , chooses Map-To-Point/Curve Si The long term private key of user i, 1 ≤ i ≤ n
function H and chooses e where e is the bilinear pairing Qi The long term public key of user i,
map. The TKGC compute PTKGC = s.G , where s ∈ Z q
*
Qi = H ( IDi ) where H is a Map function
is the TKGC 's private master key and keep s secret. We H1 , H 2 Hash functions
define each user as U i = ( ID, AKra ) , where ID is a user H A map to curve algorithm where an ID is
identity information and AKra is a set of assigned keys mapped into a point on G1
corresponding to the roles assigned to the user defined as
e A bilinear pairing map
{ }
AKra = K IDr1 ,...., KIDrn . For each user U i to be
p, q Large prime numbers, where p = 2.q + 1
registered, the TKGC calculates Qi , where Qi is user's P, Q Random points over elliptic curve
partial public key with Qi = H ( IDi ) , determines U i 's a, b Random generated private keys
partial private key Si = s.Qi and calculates QSP , QPSI E Non-super singular elliptic curve
and QTE which are the framework entities' partial public B E ( F p ) with order q
keys. Moreover, the TKGC calculates a user's or an entity's
x(Q) x coordinate of point Q
In the following, we will propose our model to achieve advertise its services to any user within a certain acceptable
attribute-based authentication. In our architecture, end-users proximity. The advertised service announcement contains
can interact with the infrastructure (e.g. walking into a the following: A Universal Resource Locator ( URL ), that
room, entering the subway system using smart phone, PDA, could allow a client to locate and access the advertised
etc). The infrastructure provides a set of resources generally access. Authentication Requirements ( AR ), allowing
tied to different geographical areas, such as printers, clients to package their access request with the necessary
surveillance cameras, campus-based location tracking authentication credentials and contextual information. The
functionality, and so on. These resources are all modeled as exchange of traffic between the Service Provider ( SP ), the
services that can be automatically discovered based on
user U i , and inference engines is based on an extension for
different relevant mechanisms which are out of our band.
Our Authentication scheme involves two distinct phases: the our previous work [32]. For the SP to construct and send
Join Phase, and the Mutual Authentication Phase. We will the authenticated services advertisement message, he will be
describe the various interactions that take place between the performing the following: The SP starts the protocol by
entities described in our logical system model. We refer our
generating two fresh random nonce r1 and r2 ∈ Zn , then
readers to (Figure 2) for a comprehensive high level
overview of our framework model. he calculates the point X where X = r1 × P1 + r2 × P2 .
Next, SP constructs the service advertisement message as
Join Phase: The purpose of this phase is to automatically in (2):
provide services to users through a context-based provision
Adv = (Qsp , ( srv1 , srv2 ,...., srvn ), X ) (2)
process. In our attributes-based authentication, we aim to
have a service provision framework that combines user's Where {srv1 , srv2 ,..., srvi }represent the set of available
profiles and contextual information to select appropriate suitable context-aware services defined in the first phase
services to the end users from thousands of desultory (Join~Phase). Finally, the service provider encrypts and
services. In order to achieve our contributions, we firstly
sends the Adv message to the embedded device ED , as
have adopted the framework proposed by Qin et al. [31] that
given in (3):
automatically provide appropriate services to the right
person with the right form with the relevant consideration of
EKe (Qsp , (srv1 , srv2 ,...., srvn ), X ,URL) (3)
contextual information in smart environment. Moreover, we In our framework and hereafter, any two entities denoted by
took the assumption that the proposed protocol in [31] is X and Y, can directly compute a partial private shared key
extended to add two new context type fields which will be between them without exchanging any previous message.
executed during the provision process. The first context type Based on the one's own partial private key and the other
is related to users with special needs equipped with a body party's partial public key, they can directly compute the
network sensor. This context type is collected by a BNS share key as follows. We denote their partial private
adapter and translated to the provision protocol in order to
key/public key by Sx = s.Qx , where Qx = H1 ( IDx ) and
be proceeded. The second context type is related to a Meta
classification process which will be helping in well selecting by S y = s.Qy , where Qy = H1 ( ID y ) . The nodes X and
services. Once the service provider (SP) has initiated the
context-aware service provision process, we can go a step
Y then compute Kx / y = e(Sx , Qy ) and
forward to start the Authentication Phase. K y / x = e(Qx , S y ) , respectively. And finally the private
Authentication Phase: Service discovery typically shared key will be Ke where:
involves the exchange of service advertisement and service [
K X / Y = H 2 ( K x / y ) = H 2 e ( Qx , Q y ) s ]
reply messages between the user and service provider. To
avoid increasing the communication overheads, we = H 2 ( KY / X ) = Ke
incorporate our extended previous authentication
mechanism into these messages. In other words, service This approach is very efficient in terms of communication
discovery and authentication can take place concurrently. and computation and this feature makes it very attractive to
We now examine how these messages are constructed to the environments where the entities capabilities are limited.
achieve our aim of attributes-based authentication.
⇒ Within The Second Round (From : ED → SP ) :
⇒ Within The First Round (From : SP → ED) : After receiving the advertised service announcement, the
Our Attributes-based authentication model will start with a client Ci decrypt the message and retrieve the credentials.
service provider engine advertising available context-aware Suppose that the client is interested in an advertised service
services to the end user, clients Ci , as indicated in (1). srvi , (i.e, request access to perform an operation O on
SP Advertise Context Aware Services Ci (1) service srvi from the service provider), he will be
For example, a location-based service allow providers to
performing the following: As srvi is a context-based
resource, Ci is promoted to present not only identity activities, etc. The validity of these data could be verified by
credentials but also all the required contextual information introducing Context Trustworthy Engine, CTE in the
framework. This is the role of the authentication broker,
and bundle them with the access request that is sent to SPi .
using the CTE , to validate these data before starting the
In our attribute-based authentication model, authentication authentication process. After receiving relevant reply
requirements are dynamic and can vary dramatically from
message from the PCL , the user agent retrieves the set of
one access to the next. Moreover, we must expect that some
contextual information received from the attribute
attributes will be generated by the user while others by the
platform. Our model provides the client with a full control provider(s) through the PCL , and performs the following:
over the reveal of the personal information. The option of The queried ED selects the role or the corresponding set of
collecting contextual information attributes from the roles denoted by SR = {r1 , r2 ,...., rh }. Generates the
platform is done by using a Privacy Control Layer PCL .
message Q and calculates the signature Sig Q on Q with
In order to retrieve needed attributes to fulfill the access
request, the user issues a service request which is handled by Q = Si SR p and where per is the permission that the
er
the user agent. The user agent does not directly invoke the
service. Instead, it retrieves the privacy policy of the service, user wants to enforce. The Sig Q is denoted by U ,V . In
without revealing any information about the user. The user addition, ED generates two fresh random nonces f and a ,
agent compares the service's policy to the user's preferences
and performs the following: where f ∈R Z and a ∈ Z
t
2
*
q, and calculates TED , where
TED = a .G . For a static context-less system, the user
Based on the user’s preferences:
computes ( Rx , Tx ) , where ( Rx , Tx ) is the signature pair
1-If there is a preference rule that accepts the privacy over the user's private key Si . This ( Rx , Tx ) will be
policy, then: - Extract the context-dependent preferences
from the user’s extended preferences document. - Store an replacing the couple U ,V in equation (6) for the protocol
association between the user, the service and the user’s process run. Finally, the client will package all the collected
context-dependent privacy preferences in the platform and attributes encrypted (i.e., user's profile and environment's
finally a request for contextual information is issued to attributes) with needed information in order to be sent to the
the PCL . service provider for authentication process. Let assume that
a user U i has received the request set of context-data D
2-If there is no accepting rule, or there is a rule that
indicates that the user should be alerted, and then the from the PCL . Therefore, the set A , given in (5), denotes
service will not be invoked. The user is prompted for further all the attributes that user U a may present to set her rules
evaluation of the policy. in the authentication process.
Whenever a request for contextual information arrives at

{ }
A = ACi , AAPi = {a1 ,..., a i , b1 ,..., b j }
the privacy control layer, the PCL should perform the = {ca1 ,..., ca i , ca i +1 ,..., ca j } ⊆ D (5)
following actions: D is the reference set that contains all the attributes a user
may hold or the context data received and ca represent the
1- Check for an association record between the service context data collected. Finally, the client packages the final
that is requesting the contextual information and the user required set of context and attributes that the service
about whom information is requested. If this association provider may use for authentication process and construct
does not exist, try to contact the user’s agent and ask it to the message as described in (6):
store an association record.
Ci EKe (QCi , (servi ), EKU / PSI ( A)),TED, f , U ,V SPi (6)
2-Retrieve and evaluate the context-dependent Hereafter, these attributes are mapped into integer numbers
preferences referenced in the association:
a: If the context -dependent preferences evaluate to ‘true’,
cai for i = 1,2,3,...., l that is ca1 is a number
then retrieve the requested information from the context representing Name, ca 2 is a number representing Location,
interpreter and return the information to the user agent. and so on. Our model is very flexible in that the service
b: If the context-dependent preferences evaluate to ‘false’, provider engine may accept or refuse a subset of attributes in
then refuse the request for contextual information. Acorresponding to different level of confidence. If the user
can present all attributes in A required by service provider
When the PCL is introduced in the infrastructure, the in order to access for identification, a full confidence will be
access request itself is altered to include information that achieved, otherwise the confidence level will be depending
was provided by the PCL . Context-Aware providers will both on PSI 's reasoning process and on the user's
publicize to their users information such as positions, roles, requirements by computing user's trustworthiness and role's
required trustworthiness. In the next section, we will

PSI QQCTE , Verification Re sults CTEi (8)
demonstrate how our scheme could be combined with Timed
Fuzzy Logic [33] in order to set a threshold under
uncertainty and to account for changes in context-data. Once PSI determines the verification process of these
attributes provided on behalf of the client, it passes the
⇒ Within The Third Round (From : SP → IEs) : authentication credentials and attributes to the relevant
The service provider now has an authentication package, engines that will complete the processing of the client's
containing the requested context attributes that was provided access request. Each engine will start it's own process as
by the client. The first step requires the SP to decrypt the follows:
encrypted message and retrieve the data in order to
Description of the PSI Engine: One new component that
determine the source and authenticity of these attributes
will be added to our architecture is the Private Set
provided by both U i and AP , and later on to complete Intersection Engine ( PSIE ). PSI are cryptographic
authentication process. Once the service provider has techniques allowing two or more parties, each holding a set
retrieved the data set from equation (6), the authentication of inputs, to jointly identify the intersection of their inputs
process will be performed as follows: The service provider sets (i.e, shared context), without leaking any information
send the encrypted set A where A = {ca1 , ca 2 ,..., ca j } to about credentials that each entity might have. Nevertheless,
B = { U ,V , f } to both entities, the prover and the verifier, need to protect

both PSIE and TE engines, and send their credentials from each other. Moreover, any entity
UBAE engine. waiting to be authenticated by a server has to establish
enough confidence in it and be able to present the required
The service provider's platform is composed of the two attributes. Therefore, the conditions that the server sets for
main brokers. The authentication process and the access authentication become extremely valuable, as they
control process. Each of these brokers contains different determine the reasoning mechanisms in the authentication
relevant engines that they interacted altogether provide a protocol. To keep a high level of security, the server needs to
flexible, a scalable context-aware authentication framework. keep those attributes private. For this purpose, we make use
For the authentication broker, we have the following of the Private Set Intersection ( PSI ). Once the PSIE
engines: a Private Set Interaction Engine ( PSIE ), a Trust receives and extract/decrypt the set A of attributes and
Engine ( TE ), and a User-Based Authentication Engine upon the sender request's selected srvi , the PSIE will
( UBAE ). We also have an Identity Based Encryption
initializes a PSI protocol over the two sets A and Ssi .
Engine ( IBEE ) that will be responsible for setting a shared
secret key for secure future communications. This IBE { }
Ssi = Ssrvi 1 , Ssrvi 2 ,..., Ssrvi j represent the needed set of
protocol will be interacting with the PSI in order to
contextual information defined by the service srvi
calculate the shared secret key. Moreover, the PSI engine
will be interacting with the CTE engine to accomplish the
administrator deployment. The Ssi set reside on a Services
attributes verification process. Therefore, our authentication Proxy Server ( SPS ), and the PSI protocol will be
process decision will be based on the output of these several initialized between PSI engine and SPS . There are many
engines. The description of these engines and their PSI protocols in the literature. We can adopt the one that
interacting process will be explained in the coming section. was chosen by [33], [34] since it has a provision for
approximate matching, referred to as Fuzzy Private Match.
⇒ Within The Fourth Round (From: IEs → SP) : The PSI engine performs two kinds of tasks: First, it gives
Upon receiving the encrypted messages from the service a level of confidence when a user is on an authentication
provider, the PSI start the attributes verification process. process. It makes use of authentication contextual
To verify the source of AP ’s attributes, we have introduced information to assign the confidence level. Second, it
evaluates a Fuzzy Logic Matching protocol queries from
the Context Trustworthy Engine ( CTE ) which is
applications about whether a certain entity is allowed to
responsible for verifying all attributes provided by APs
access a certain resources. It makes use of applications
and other contextual information provided by the client (i.e., specific contextual information, the credentials of the entity,
case of an RFID or a smart card and a client with special and entity's contextual information to decide whether an
need). The interactions (7) and (8) show the entity is authenticated and has access to resources. For
PSI requesting the CTE to verify the validity of the convenient readerships, we urge our readers who want to go
attributes ca i . deeper in the theory of Fuzzy Private Matching Protocol and
getting acquainted with the principles of PSI theory to refer
PSI EKe (QSPi ,(ca1,ca2 ,...,caj )) CTEi (7) to [3], [33]. Moreover, the PSI engine will be also
interacting with the Identity Based Encryption protocol to
calculate the secret shared key. This step will be discussed
in the following paragraph:

The user assignment UA level is performed based on the
Description of The Identity-Based Encryption Protocol trust level UT in comparison with the trust level RT .
The IBE removes the need to set and exchange certificates However, our trust model is based on the trust policies, the
as the message can be encrypted based on the identity of the environment contextual information, and the user’s roles
entities. The identity can be defined as a location, name, permissions. As a cognitive process, trust is complex and
email address, time, etc. or a combination of them. The fuzzy. That is, for a special context, we can not easily make
combination of them could be refereed to the context data. a decision about whether to trust an entity or distrust it.
For convenient readership, we urge our readers to refer to Therefore, Our Trust evaluation engine is adopted as a
[35]. In the following, we will describe the details of how combination from [36] [37] where trust model is provided
PSI interacts with IBE protocol in order to calculate the by integrating trust into a fuzzy logic-based trusted decision
shared secret key. From the PSI, let AI S Si be the upon building the trustworthiness's prediction. For
convenient readership of this work, we will briefly describe
intersection set of A and SSi defined above:
the trust model process here: Trust establishment can be
AI SSi = {d1 , d 2 , d 3 ,...., d i } (9) thought as a process that identifies or verifies the principal's
claim against the trust evidence. Trust evidence,
Tev = {Tc , Tce , Tbr } , are further classified into the following
where d i denotes the context that are shared between the
user and the service provider. Finally, the IBE will calculate
categories: credentials, the context of the environments, and
and send TSP to PSI engine with TSP = (∑ d i ).G .
behavior records. We define UT as the trust level of the
user by using a function F , as given in (10):
Description of the Trust Engine: Another new
component that will be added to our architecture is the Trust
UT = Fres (Tp , Tev ) = F (Tp , Tc , Tce , Tbr ) (10)
Engine TE . To trust pervasive computing, we must be able where Fres is the function of the trust level of the client to
to manage privacy, confidentiality, availability, and
access the resource and Tp is the set of trust policies for the
controlled access to digital information as it flows through
the systems. In the following, we will describe the Trust resources. In our definition, the trust level of the user, UT ,
process design architecture shown in Figure 4. Our ultimate for accessing the resource in the system is determined by
goal is to provide a trust model that is flexible and adaptive evaluating the trust evidence against the trust policies for
for various application scenarios and environments. This the resource and the user assignment, UA, is evaluated
approach could be solved using the concept of fuzzy-based based on UT in comparison with RT . For simplicity, we
trustworthiness.
will consider Tev = Tattributes = Ta and finally
In this section, a dynamic trust model is formally
introduced to incorporate trust strategies in order to first UT = F (Ta ) . F , UT and RT could be calculated using
build up the user's and role's required trustworthiness level the formal mathematical equations from [36]. Once these
and than the User Assignment UA trustworthy value. There parameters are calculated, the trust decision modular will
are several ways and approaches to design trust models. The evaluate the user assignment UA based on UT in
component-based approach is chosen for our model design comparison with RT , and will package the final result in
because it can be implemented in a distributed way and be order to be sent to the Authentication Process Decision.
extended easily and transparently (i.e, to include later the
Risk Assessment Engine). During a real-time trust Description of the UBA Engine: Moreover, upon
management process in pervasive computing environments,
the trust information may be from different resources at any receiving the encrypted signature pair message EKe U ,V
time. Therefore, our adopted trust model is designed to be from the service provider, the UBA engine will decrypt the
able to evaluate the trust information concurrently. Using message, and then verify the signature pair. If it is valid,
this approach, the trust engine derives the level then the UBA engine accepts, and the pair
trustworthiness of a user UT and role's required
(s1 , s2 ) associated with the authenticated ED is extracted
trustworthiness RT by using user’s attributes and roles
from the database server, and encrypted using the Weil-
permission, respectively.
Pairing-based encryption algorithm. Finally, the user based
authentication engine packages the encrypted message
EKe ( s1 , s2 ) with the evaluated result in order to be sent
to the authentication process decision broker.
The authentication process decision will take the decision

based on its different engines evaluation and package the
final output result and send it encrypted to the service
Figure 4. The Trust/Risk Process Architecture provider.
⇒ Within The Fifth Round (From : SP → ED) :

Upon receiving the message from the authentication
process decision broker, the service provider first decrypts
the message and then evaluates the output. If the result is
false, he will deny access request to resources, otherwise,
if true (i.e., the user is authenticated, the user
trustworthiness parameters are acceptable, and the
confidence level is acceptable), the service provider
extracts the pair EKe ( s1 , s2 ) and then computes:
yi = ( ri + ( f × si ))(mod n) (11)
For i = 1,2 and starting packaging the following data
( yi with i = 1,2 ) in order to be sent later to the ED .
Meanwhile, as the final decision will be evaluated based on
both authentication and access control process decision
brokers, the user's access request is also subject to context-
Figure 5. The Extended Access Control Process
aware access control rules which will be discussed in the
following:
In the following, we describe needed rules definitions for
a dynamic context-aware access control infrastructure to
Context-Based Access Control Process: A key challenge
fulfill the framework's security requirements:
in ubiquitous environment is the design of an effective
active access control schemes [36] that can adequately meet
Rule Definition 1: Dynamic Adjustment. In our
the security challenges represented by the system's ability to
approach, we believe that any pervasive model should
capture security relevant contextual information, such as
dynamically adjust role assignments and permission
time, location, user's profile, or environmental state
assignments based on presented context information.
available at the time the access request are made and to
Therefore, we consider DRBAC concept [38] where each
incorporate these information in its access control process.
user is assigned a set of roles and the context information is
We specify and integrate our own context-aware access
used to decide which role is active at a time. User will
control rules definitions to further enhance the security of
access the resource with the active role. Moreover, each role
our proposed authentication-based framework scheme.
is assigned a set of permission, where the context
Moreover, the context directly affects the level of trust
information will be used to decide which permission is
associated with a user, and hence the authorizations granted
active for that role. The systems-based context for resources
to him. Therefore, we introduce the user trustworthiness and
should be taken into consideration, and the security policy
role's required trustworthiness parameters into the design
for the resources should be able to define a permission
the context-based access control by incorporating them
transition for a current role.
within the development of the context constraints.
Conditions on the access control to solve the semantic
Rule Definition 2: Context Type. A context type is
problem is to check the trust engine parameters UT and
defined as a property related to every participant in a
UA if they satisfy the condition, the user will be subject to service. In simple scenario, context type may be a concrete
authorization rules and policies based on the available property familiar in everyday life, such as time or location,
presented attributes. We believe that the introduction for the etc. However, in a more complex scenario, we believe that
rules definitions is necessary for providing an adequate context type should be extended to describe more attributes
authorization decision for any Service Access Request and such as user's capability and/or willingness (i.e, case of
to accomplish a secure authentication process. Figure 5 People with special need equipped with a hidden body
shows our extended access control scheme with the rules
network sensor). We define such context type by CTc .
definitions.
Therefore, based on a complete users' context type CTi , we
can define that each resource ri has its own context set
CSri which is defined as follow:
CSri = {CT1 , CT2 ,...., CTc ,...., CTn } (12)
In any access control design to be integrated within our
framework, we define two sets of context types, passive and
active sets. While the authentication process will be subject
to only the active set, the access control decision will be
subject to the two sets. under RCi (that is, when all CTs in constraint CC are
Rule Definition 3: Context Constraint. We define our replaced with their available presented values in RCi , then
context constraint as a regular expression that is capable the resulted Boolean expression is true).
of specifying any complex context related constraint to
introduce all kinds of security requirements. In general a ⇒ Finally, the service provider evaluate the final access
context set is defined as: Context request decision (the one from authentication broker and
Constraint := CC := Clause1 ∪ Clause2 ..... ∪ Clausei the other from the access control broker) and packages the
where results with the relevant data ( TSP , yi for i = 1,2 ) and send
Clause := Condition 1 ∩ condition 2 ... ∩ condition j it to the user with the embedded device. The ED
computes:
and Condition := CT OP VALUE ,
(∑ ( yi × Pi ) + f × Z ) (15)
where CT ∈ CS , OP is a logical operator in the set
(∑ ( yi × Pi ) + f × Z ) is equals to
{<, ≤, >, ≥, ≠, =}, and VALUE is a specific value of
and then checks that if
CT . Therefore, we suggest that should be extended to X , if so the ED accepts and extract the shared secret key
in order to be used for encrypting future communications,
accommodate user trustworthiness UT and user
else rejects.
assignment trustworthiness UAas a new clause. The new
context constraint will be:
After the above messages, TED and TSP are exchanged,
CC := Clause1 ∪ Clause2 ..... ∪ Clausenew (13)
the service provider and the user can agree and compute the
Where secret shared key:
Clausenew = ((UT ≥ VALUE) ∩ (UA ≥ VALUE)) (14) KSP / ED = e(QED , PED ) b .e( xsp , Ssp , TED ) (16)
As an illustration, suppose we have a context set
and
CS = Time, Location, Authentication Level and we have a
partial security rule such as a patient data can be accessed KED / SP = e(QSP , PSP ) a .e( xed , Sed , TSP ) (17)
from within the hospital between 8am and 5pm with a trust respectively. We denote by K = KSP / ED = KED / SP , the key
level of a password; otherwise a higher level of trust is shared between the entities. To ensure forward security, we
required.
can use the new shared key Kh after applying a hash
Rule Definition 4: Authorization Policy. We define an function to K . Once the protocol run completes
authorization policy as a quadruple AP = S, P , O , CC successfully, both parties may use Kh to encrypt subsequent
where S is the subject in this policy, which could be a user session traffic in order to create a confidential
or a set of roles, P the mode of operation defined by communication channel. In the following we will present a
{READ, APPEND, DELETE, UPDATE, WRITE}, O is a brief verification regarding the similarity of the shared key
data object and CC is a context constraint defined equations:
according to Definition 3. KSP / ED = e(QED , PED ) b .e( xsp , Ssp , TED )
= e(QED , xed .s.G )b .e( xsp , Ssp , a.G )
Rule Definition 5: Resource_Access_Request, denoted
by RAR , is defined as a quadruple = e( xed .s.QED , b.G ) .e( xsp , s.QSP , a .G )
RAR = U i , Pi , Oi , RCi where U i ∈UserSet , = e( xed .Sed , TSP ) .e(QSP , PSP ) a
Pi ∈ PermissionSeT , O is the data object requested, and = KED / SP
context RC is a runtime context set of values for every
context type in the context set CS . RCi is defined 7. Protocol Analysis:
according to Definition 2 and captured dynamically at the Security Analysis: Our proposed architecture is
time of the access request. considered to provide privacy and anonymity for users. In
the following, we evaluate our architecture regarding the
Dynamic Context Evaluation: Finally, the access control security and privacy requirements.
decision for any service access request
Mutual Authentication: Considering the fact that the
RAR = U i , Pi , Oi , RCi is granted only if there exists
digital signature pair (U , V ) , created by the ED , is verified
an authorization policy AP = S, P , O , CC , such that by the Back-end server. Considering that the pair ( s1 , s 2 ) ,
U i ∈ S , Pi , Oi = P , O , and CC evaluated to true sent by the back-end server UBAE , is recalculated by the
service provider under ( y1 , y2 ) and verified by the ED . provider and the back-end server, and by verifying the pair
Therefore, our proposed architecture guarantees the secure ( y1 , y2 ) by the legitimate user or ED . Thus, our scheme
mutual authentication between the embedded device protects against unauthorized service provider.
ED and the back-end server.
Protocol Correctness: We can choose one of many
Passive attack: Suppose an attacker performs a passive identity-based signature scheme to compute the Sig Q .
attack, then the session will terminate with both legitimates
Therefore, we will adopt the signature scheme that was used
parties accepting. That is, the two parties successfully
identify themselves to each other. And regarding the fact by [51]. To compute the Sig Q , the user selects random
that the exchanges messages between the service provider
r ∈Z *
q, computes U = r .QID , computes h = H (Q,U ) ,
and the ED are generated from random nonce which are
generated with every new session, so it is infeasible that an ,computes KSR = (∑ ih=1 K IDri ) , and finally computes
attacker computes any useful information including the IDi V = (r + h) KSR . The validity of Sig Q can be
of a user U i . Therefore the architecture resists against the accomplished by verifying if
passive attack. e( P , V) = ? e( PAR ,U + hQID ) . The Proof is given below:
Man in the middle attack (or active attack): Suppose that
e( PAR ,U + hQID ) = e(∑ ik=1 Pi , rQID + hQID )
an attacker intercepts X and replaces it with X ' , the
attacker then receives f and (U , V ) from the ED . He = e(∑ ik=1 s.P , (r + h)QID )
would like to replace the pair with (U ' ,V ' ) , as before. = e( P , (r + h ) ∑ ik=1 si QID )
However, and unfortunately for the attacker, he can not = e( P , (r + h) SIDAR )
compute the value of the new pair because he does not know
the users credentials and parameters and because the
= e( P , V )
transmitted messages are meaningless. Therefore the
proposed scheme thwarts the man in-the-middle attack. As a summary, a context-based access is developed and it
can thus be granted to both known and unknown agents.
Perfect forward secrecy: Each run of the protocol The integration for the IEs engines, the extension for the
context-based access control definitions, and the
computes a unique X , a unique Signature pair (U , V ) and
development of IBE engine form the core of our context-
a unique pair ( y1 , y2 ) . In addition the transmitted messages based authentication framework where every request is
are meaningless as they are generated for each new session authenticated and filtered in order to remove any
using new random nonce. Thus, the architecture is secure unauthorized actions. After filtration the service provider
against perfect forward secrecy. can evaluate the request and create an appropriate response
depending on the contextual information.
Data Confidentiality: Since our architecture provides
secure mutual authentication between the ED and the 8. Conclusion:
system and since the information transmitted between
In this paper, we have proposed a usable dynamic
the ED and system is meaningless, thus, our architecture authentication framework for pervasive computing
provide data confidentiality and the user privacy on data is environments. We hope that our work can raise interests on
strongly protected. the design of authenticated-base framework dedicated for
areas with specials needs. From the above arguments, we
ED Anonymity and Location Privacy: During the can see that our context-aware access control rules provide
authentication processes, a signature algorithm is used to an effective second-line defense against attacks on
produce the signature pair (U , V ) . The pair (U , V ) and f authentication. We have identified security and privacy
that are transmitted between the ED and SP are threats that may arise during access services in a pervasive
randomized and anonymous since they are updated for each computing environment; we also derived corresponding
read attempt. Thus, our architecture provides user security and privacy requirements. We have presented our
anonymity and location privacy is not compromised. attributes-based authentication scheme, using Trusted
Computing Functionality, which preserves user privacy. The
Unauthorized SP Detection: Our Proposed architecture is scheme also satisfies all the identified security requirements.
based on the insecure communication channel between To a user and service provider, security and privacy are both
desirable. However, they are potentially conflicting
SP and UBA engine. The unauthorized SP ' is detected
requirements, and it is challenging to achieve them both.
and prevented by the back-end server UBAEID using the However, this is achieved by our attributes-based
weil pairing based encryption algorithm between the service authentication scheme presented here, enabling secure
access to services while privacy is preserved. In the coming Dartmouth Computer Science Technical Report
future work, we are working on introducing risk assessment, TR2002-422, February 28, 2002.
quality of privacy within our framework. Moreover, we are [9] M., Covington, W., Long, S., Srinivasan, A., Dey, M.,
working on an implementation for the proposed model into Ahamad and D.G., Abowd, “Securing Context-Aware
the platform at the lab of Telecom SudParis (ex. INT). Applications Using Environments Roles”. In
Proceeding of the sixth ACM Symposium on Access
Controls Models and Technologies, pp. 10-20, 2001.
9. Acknowledgment:
[10] L. Marc, “A Privacy Awareness System for
The authors would like to thank Telecom SudParis and Ubiquitous Computing Environments”. In Proceeding
Lebanese University Laboratories’ staff for their of the 4th International Conference on Ubiquitous
contributions and support. I would like to thank everyone for Computing, (UbiComp) pp. 237-245, 2002.
his help, guidance, advice as well as his enthusiasm and [11] J.M. Covington, M. Sastry, and D.J. Manohar,
many valuable contributions to this work. Their suggestions “Attribute-Based Authentication Model for Dynamic
and observations were extremely helpful throughout this Mobile Environments”. In Proceeding of the Third
paper. Without their input, I would not have been able to International conference on Security in Pervasive
complete this work. Computing (SPC), York, UK, pp. 227-242, 2006.
[12] M.D. Corner, and B.D. Noble, “Protecting
Applications with Transient Authentication”. In
Proceeding of the First International Conference on
References Mobile Systems, Applications and Services, pp. 57-
[1] P. ABI-CHAR, A. Mhamed, B. EL-Hassan and M. 70, 2003.
Mokhtari, “Controlling Trust and Privacy in Context- [13] S.J. Creese, M.H. Goldsmith, and B.R. Zakiuddin,
Aware Environments, State of Art and Future “Authentication in Pervasive Computing”. In
Directions”, In Trust Modeling and management in Proceeding of the First International Conference on
Digital Environments: From Social Concept to Security in Pervasive Computing (SPC), 2003.
System Development. Book Edited by Nokia Research [14] D. Glynos, P. Kotzanikolaou, and C. Douligeris,
Center, Finland, January 2010. “Preventing Impersonation Attacks in MANET with
[2] K. Dey and D. Abowd, “Towards a better Multi-Factor Authentication”. In Proceeding of the
understanding of context and context-awareness. The Third International Symposium on Modeling and
CHI 2000 Workshop on the What, Who, when, and Optimization in Mobile Ad-hoc, And Wireless
How of Context-Awareness”, The Hague, Networks, pp. 59-64, 2005.
Netherlands, Apr. [15] W. Qin, Z. Daqing, M. Mounir, S. Yuanchun, and D.
2000,ftp://ftp.cc.gatech.edu/pub/gvu/tr/1999/92- Kejun, “Combining User Profiles and Situation
22.pdf. Context for Spontaneous Service Provision in Smart
[3] B. Malek, A. Miri, and A. Karmouch, “A Framework Assistive Environments”. In Proceeding of the 5th
for Context-Aware Authentication”, In 2008 IET 4th international conference on Ubiquitous Intelligence
International Conference on Intelligent and Computing, pp. 187-200, 2008.
Environments. IEEE Computer Society Press, pp. 1-8, [16] M. Behzad, M. Ali, and K. Ahmed, “A Framework
2008. for Context-Aware Authentication”. In Proceeding of
[4] J., Kim, Z. Kim, and K. Kim, “A lightweight Privacy 2008 IET 4th International Conference on Intelligent
Preserving Authentication and Access Control Environments, pp. 1-8, 2008.
Scheme for Ubiquitous Environment”, In Proceeding [17] L. Adrian, and M.J. Chris, “Ninja: Non Identity
of the 10th international Conference on Information Based, Privacy Preserving Authentication for
Security and Cryptography, ICISC07, pp. 37-48, Ubiquitous Environments”. In Proceeding of 9th
2007. International Conference on Ubiquitous Computing,
[5] Q. He, L. Wu, and P. Khosla, “for Personal Control (UbiComp2007), Springer LNCS 4717, pp. 73-90,
over Mobile Location Privacy”, In IEEE Commun. 2007.
Mag. vol. 42, no. 5, pp. 130-136, 2004. [18] V. Miller, “Uses of elliptic curves in cryptography”.
[6] J. Al-Muhtadi, R. Campell, A. Kapadia, M. Mickunas, In Proceeding of Crypto '85, Santa Barbara, pp. 417 -
and S.Yi, “Routing Through the Mist: Privacy 426. 1986.
Preserving Communication In Ubiquitous Computing [19] N. Koblitz, “Elliptic Curve cryptosystems”.
Environments”. In Proceedings of the International Mathematics of Computation, vol. 48. pp. 203 - 209,
Conference of Distributed Computing Systems 1987.
(ICDCS02), pp. 65-74, 2002. [20] N. Koblitz, “CM-Curves with Good Cryptography
[7] U. Jendricke, M. Kreutzer, and A. Zugenmair, Properties”. In Proc. of Crypto' 91, Santa Barbara,
“Pervasive Privacy with Identity Management”. In USA, 1992.
Proceeding of. 1st Workshop Security, UbiComp [21] L. Law, A. Menezes, M. Qu, J. Solinas and S.
2002. Vanstane, “An efficient Protocol for Authenticated
[8] K., Minami, K., and D., Kotz, “Controlling Access to Key Agreement”. Technical report CORR98-05,
Pervasive Information in the "Solar" system”. Department of CO, University of Waterloo, 1998.
[22] L. Law, A. Menezes, M. Qu, J. Solinas and S. Control”. In Proceeding of the IEEE/ACS
Vanstane, “An efficient Protocol for Authenticated International Conference on Computer Systems and
Key Agreement”. Designs, Codes and Cryptography, Applications (AICCSA), pp. 807-814, 2007.
vol. 28, pp. 119-134, 2003. [37] D., Xiu, and Z., Liu, “A Dynamic Trust Model for
[23] J. Pollard, “Monte Carlo methods for index Pervasive Computing Environments. A Research
computation mod p”. Mathematics of Computation, Paper}, Research Supported by the NSF0406325,
vol. 32, pp. 918-924, 1978. coitweb.uncc.edu/~zhliu/Research/Papers/asc.pdf,
[24] A. Menezes, T. Okamoto and S. Vanstane, “Reducing 2004.
elliptic curve logarithms in a finite field”. IEEE [38] G. Zhang, and P. Manish, “Context-Aware Dynamic
Transactions on Information Theory, vol. 39, pp. Access Control for Pervasive Applications”. In
1639-1646, 1993. Proceedings of the Communication Networks and
[25] G. Frey and H. Ruck, “A remark concerning m- Distributed Systems Modeling and Simulation
divisibility and the discrete logarithm in the divisor Conference, (CNDS'04), USA, 2004.
class group of curves}, Mathematics of Computation, [39] M. Uddin, M. Zulkernine, and S. Ahamed, “CAT: a
vol. 62, pp. 865-874, 1994. context-aware trust model for open and dynamic
[26] I. Semaev, “Evaluation of Discrete logarithms in a systems}, In Proceedings of the 2008 ACM
group of p-torsion points of an elliptic curve in symposium on Applied computing, pp. 2024-2029,
Characteristic p”. Mathematics of Computation, vol. 2008.
67, pp. 353-356, 1998. [40] A., Mohan, and M., Blough, “Attribute Trust-a
[27] D. Boneh and M. Franklin, “Identity-based Framework for Evaluating Trust in Aggregated
encryption from the Weil Pairing”. Advanced in Attributes via a Reputation System”. In proceeding of
CRYPTO2001, LNCS 2139, pp. 213-229, 2001. the 6th Annual Conference on Privacy, Security and
[28] G. Frey, M. Muller and H. Ruck, “The Tate Pairing Trust. IEEE Computer Society Press, pp. 201-212,
and the discrete logarithm applied to elliptic curve 2008.
cryptosystem”. IEEE Transaction on Information [41] W., Xu, T., Xin, and G., Lu, “A Trust Framework for
Theory, Vol. 45, No.5, pp. 1717-1719, 1999. Pervasive Computing Environments”. In the
[29] D. R. Stinson, “Cryptography Theory and Practice”. proceeding of the International Conference on
In Proceeding of Chapman and Hall/CRC, Third Wireless Communications, Networking and Mobile
Edition, pages: 353-438, 2006. Computing (WiCom07), IEEE Society Press, pp.
[30] S. Wang, Z. Cao, and H. Bao, “Efficient 2222-2225, 2007.
Certificateless Authentication and Key Agreement [42] S., Ries, “Certain Trust: A Trust Model for Users and
(CL-AK) for Grid Computing”. In Proceeding of the Agents”. In Proceeding of the 22nd Annual ACM
International Journal of Network Security, vol.7, Symposium on Applied Computing, ACM Press, pp.
No.3, pp. 342-347, 2008. 1599-1604, 2007.
[31] W. Qin, Z. Daqing, M. Mounir, S. Yuanchun, and D. [43] W., Yuan, D., Guan, S., Lee, Y., Lee, and H., Lee,
Kejun, “Combining User Profiles and Situation “Filtering out unfair recommendations Finding for
Context for Spontaneous Service Provision in Smart trust model in ubiquitous environments”. In
Assistive Environments”. In Proceeding of the 5th proceeding of the second International Conference on
international conference on Ubiquitous Intelligence Information Systems Security (ICISS06), pp. 258-
and Computing, pp. 187-200, 2008. 263, 2006.
[32] P. ABI-CHAR, M. Mokhtari, A. Mhamed and B.~EL [44] W., Yuan, D., Guan, S., Lee, Y., Lee, and A.,
-Hassan, “Towards a Robust Privacy and Anonymity Gavrilov, “Finding Reliable Recommendations for
Preserving Architecture for Ubiquitous Computing”. trust model”. In proceeding of the 7th International
In Proc. of the Third International Conference on Conference on Web Information Systems Engineering
Risks and Security of Internet and Systems (WISE06), pp. 375-386, 2006.
(CRISIS08). Tozeur, Tunisia, IEEE Computer Society [45] N., Dimmock, A., Belokosztolski, D., Eyers, J.,
Press, October 28-30, pp. 125-132, 2008. Bacon, D., Ingram, and K., Moody, “Using Trust and
[33] J.F. Michael, N. Kobbi, P. Benny, “Efficient Private Risk in Role-Based Access Control Policies”. In
Matching and Set Intersection”. In Proceeding of the Proceedings of the 9th ACM Symposium on Access
Advances in Cryptology-Eurocrypt'04', vol. 3027 of Control Models and technologies. USA, ACM Press,
lectures Notes in Computer Science, Springer-Verlag, pp. 156-162, 2004
pages 1-19, 2004. [46] N., Dimmock, J., Bacon, D., Ingram, and Moody,
[34] M. Behzad, M. Ali, and K. Ahmed, “A Framework “Risk Models for Trust-Based Access Control
for Context-Aware Authentication”. In Proceeding of (TBAC)”. In Proceedings of the 3rd Annual
2008 IET 4th International Conference on Intelligent Conference on Trust Management, France, pp. 364-
Environments, pp. 1-8, 2008. 371, 2005.
[35] B. Dan, and F. Matthew, “Identity-Based Encryption [47] T., Huynh, N., Jennings, and N., Shadbolt, “FIRE:
from the Weil Pairing”. In Proceeding of the SIAM An integrated trust and reputation model for open
Journal of Computing, 32(3):586-615, 2003. multi-agent systems”. In Proceeding of the 16th
[36] T. Hassan, A. Morteza, and J. Rasool, “Trust-Based European Conference on Artificial Intelligence,
User-Role Assignment in Role-Based Access Spain, pp. 18-22, 2004.
[48] K. Ren, and W. Lou, “Privacy Enhanced Access

Control in Ubiquitous Computing Environments”. In
Proceeding of the 2nd International Conference of
Broadband Networks, Vol. 1, pp. 356-365, 2005.
[49] K., Ren, and W., Lou, “Privacy-Enhanced, Attack-
Resilient Access Control in Pervasive Computing
Environments with Optional Context Authentication
Capability” In Springer Science LLC, Mobile Netw
Appl-12:79-92, 2007.
[50] K., Ren, W., Lou, K., Kim, and R., Deng, “A Novel
Privacy Preserving Authentication and Access
Control Scheme for Pervasive Computing
Environments”. In IEEE Transactions on Vehicular
Technology, Vol. 55, no. 4, pp. 1373-1384, 2007.
[51] J. Wang, J. Yu, D. Li, X. Bai, and Z. Jia, “Combining
User Authentication With Role-Based Authorization
Based on Identity-Based Signature”. In Proceeding of
International Conference on Computational
Intelligence and Security, CIS, pp.847-857, 2006.
[52] B. Jakob, K. Rasmus, P. Michael, “Context-aware
user authentication: Supporting proximity-based login
in pervasive computing”. In Proceeding of
International conference on ubiquitous computing
No5, Seattle WA, vol. 2864, pp. 107-123, 2003.
[53] C. Wang, L. Fang, and Q. Wang, “Zero-Knowledge-
Based User Authentication Technique in Context-
aware System”. In Proceeding of the International
Conference on Multimedia and Ubiquitous
Engineering, pp. 874-879, 2007.
Integration of Metaheuristic Algorithms for

Minimum Spanning Tree
R.Thamilselvan1, Dr.P.Balasubramanie2
1
Assistant Professor, Department of Computer Science and Engineering,
Kongu Engineering College, Perundurai, Erode 638 052, Tamilnadu, India
r_thamilselvan@yahoo.co.in
2
Professor, Department of Computer Science and Engineering,
Kongu Engineering College, Perundurai, Erode 638 052, Tamilnadu, India
pbalu_20032001@yahoo.co.in
minimum spanning tree algorithm [10] is shown in table 1

Abstract: This paper presents a new algorithm based on
integrating Tabu Search (TS), Simulated Annealing (SA) and
Genetic Algorithms (GA) for solving minimum spanning tree A
problem. This proposed integration method is a general kind of 7
solution method that orchestrates the interaction between local 2 2
improvement procedures and higher level strategies to create a 5 4 D 5
process that is capable of escaping from local optima and B
O T
performing a robust search of a feasible region. This paper 4 1 3 1
applies Tabu Search, Simulated Annealing and Genetic 7
algorithm for minimum spanning tree problem and compares the C 4
results obtained by each. With the implementation of our E
approach the minimum spanning tree problem is solved with
feasible solution.
Figure 1. Network with nodes are connected
Keywords: Tabu Search, Simulated Annealing, Genetic
Algorithm, Minimum Spanning Tree. Table 1: Algorithm for Minimum Spanning Tree
1. Introduction 1. Select a node arbitrarily, and then connect it.

2. Identify the unconnected node that is closest to a
There are two cases in minimum spanning tree [12], an connected node, and then connect these two nodes.
undirected and connected network is being considered, Repeat this step until all nodes have been connected.
where the given information includes some measure of the 3. Tie breaking: Ties for the nearest distinct node or
positive length (distance, cost, time etc.) associated with the closest unconnected node may be broken
each link. The problem has to choosing a set of links that arbitrarily, and the algorithm must still yield an
have the shortest total length among all sets of links that optimal solution.
satisfy a certain property. For the minimum spanning tree
problem, the required property is that the chosen links must To apply the algorithm in Table 1 to the Network in
provide a path between each pair of nodes. The minimum Figure 1. Keep the arbitrarily select node O to start. The
spanning tree problem can be summarized as follows. unconnected node closest to node O is node A. Connect
node A to node O. Repeat the algorithm, then we will get
i. There are set of nodes of a network but not the links. the spanning tree in Figure 2.
Instead, you are given the potential links and the
positive length for each if it is inserted into the
network. A
ii. To design the network by inserting enough links to
satisfy the requirement that there be a path between 2 2
D 5
every pair of nodes. B
iii. The objective is to satisfy this requirement in a way O T
1 3 1
that minimizes the total length of the links inserted
into the network.
C
E
A network with n nodes required (n-1) links to provide a
path between each pair of nodes. No extra links should be Figure 2. Minimum Spanning Tree
used, since this would needlessly increase the total length of
the chosen links. The links need to be chosen in such a way The minimum spanning tree problem is the one
that the resulting network forms a spanning tree. The basic problem that falls into the broad category of network design.
2. Related Work constraints must be observed when choosing the links to

include in the network.
2.1 Tabu Search
Constraint 1: Link AD can be included only if link DE also
Tabu Search [1] [2] is a widely used metaheuristic that included.
uses some common sense ideas to enable the search process
to escape from a local optimum. Any application of tabu Constraint 2: At most one of the three links AD, CD and
search includes as a subroutine a local search procedure that AB can be included.
seems appropriate for the problem being addressed. The To take the constraint into account is to charge a huge
process begins by using this procedure as a local penalty, such as charge a penalty of 100 if constraint 1 is
improvement procedure in the usual way to find a local violated and charge a penalty of 100 if two of the three links
optimum. A key strategy of tabu search is that is then specified in constraint 2 are included.
continues the search by allowing non improving moves to
the best solutions in the neighborhood of the local optimum.
Once the point is reached where better solutions can be B
found in the neighborhood of the current trial solution, the
local improvement procedure is reapplied to find a new local 30 30
optimum. The danger with this approach is that after
moving away from a local optimum, the process will cycle 10 5
A C E
right back to the same local optimum. A tabu list records
these forbidden moves, which are referred to as tabu moves.
25
2.1.1 Basic Tabu Search Algorithm 15 40
Initialization: Start with a feasible initial trial solution D
Iteration: Use an appropriate local search procedure to

Figure 3. Data for a Minimum Spanning Tree before
define the feasible moves into the local neighborhood [5] of
choosing the links to be included in the network
the current trial solution. Eliminate from consideration any
move on the current tabu list unless that move would result B
in a better solution than the best trial solution found so far.
Determine which of the remaining moves provides the best 30 30
solution. Adopt this solution as the next trial solution,
regardless of whether it is better or worse than the current 10 5
A C E
trial solution. Update the tabu list to forbid cycling back to
what has been the current trial solution. If the tabu list
already had been full, delete the oldest member of the tabu 25
list to provide more flexibility for future moves. 15 D 40
Stopping rule: Use some stopping criterion, such as a Figure 4. Optimal solution for a Minimum Spanning
fixed number of iterations [7], a fixed amount of CPU time, Tree
or a fixed number of consecutive iterations without an
improvement in the best objective function value. Also stop The Figure 4 shows the desired minimum spanning tree,
at any iteration where there are no feasible moves into the where the dark lines represent the links that have been
neighborhood of the current trial solution. Accept the best inserted into the network with a total cost of 50. This
trial solution found on any iteration as the final solution. optimal solution [8] is obtained by the greedy algorithm.
The method to conduct tabu search is follows.
The basic idea behind tabu search is that, adding short-term
memory to local search, improves its ability to locate optimal Local Search Procedure: At each iteration, choose the best
solutions. It is an is an iterative search that starts from some immediate neighbor [6] of the current trial solution that is
initial feasible solution [8] and attempts to determine the best not ruled out by its tabu status.
solution in the manner of a hill-climbing algorithm.
Neighborhood structure: An immediate neighbor of the
2.1.2 A Minimum Spanning Tree Problem with current trial solution is one that is reached by adding a
Constraints single link and then deleting one of the other links in the
Figure 3 shows a network with five nodes, where the cycle that is formed by the addition of this link.
dashed lines represent the potential links that could be Form of tabu moves: List the links that should not be
inserted into the network and the number next to each deleted
dashed line represents the cost associated with inserting that
particular link. Thus, the problem is to determine which Addition of tabu move: At each iteration, after choosing
four of these links should be inserted into the network to the link to be added to the network, also add this link to the
minimize the total cost of these links. The following tabu list.
Maximum size of tabu list: Two Add link BE to the network shown in Figure 5, the
automatically place this added link on the tabu list. Delete
Stooping rule: Stop after three consecutive iterations
link AB from the network shown in Figure 5. Various
without an improvement in the best objective function value.
options for adding a link and deleting a link in iteration 2
In Figure 4 both constraints are violates, so that penalties of are shown in Table 3.
100 need to be imposed twice. Therefore the total cost of
Table 3: The options for adding a link and deleting another
this solution is
link in iteration 2
Cost = 20+10+5+15+200 = 250
Add Delete Cost
Iteration 1: There are three options for adding a link to the AD DE (Tabu move)
network in Figure 4 are BE, CD and DE. If BE were to be AD CD 85+100=185
chosen, the cycle formed, so the three options for deleting a AD AC 80+100=180
link would be CE, AC and AB. If CE were to be deleted, the BE CE 100+0=100
change in the cost would be 30-5=25 with no change in the BE AC 95+0=95
constraint penalties, so the total cost would increase from BE AB 85+0=85 (Minimum)
250 to 275. So that iteration 1 is followed as per Table 2. CD DE 60+100=160
CD DE 95+100=195
Table 2: The options for adding a link and deleting another
link in iteration 1 Iteration 3: The following are the decisions are made in
this iteration.
Add Delete Cost
BE CE 75+200=275 Add a link CD to the network shown in Figure 6, and then
BE AC 70+200=270 automatically place this added link on the tabu list. Delete
BE AB 60+100=160 link DE from the network shown in Figure 6. Various
CD AD 60+100=160 options for adding a link and deleting a link in iteration 3
CD AC 65+300=365 are shown in Table 4.
DE CE 85+100=185
DE AC 80+100=180 B
DE AD 75+0=75 (Minimum)
30 30
The final output of the iteration 1 in a network is shown in
Figure 5. 10 5
A C E
B
25
30 30 15 40
D
A 10 C 5 E
Figure 7. Modified network after Iteration 3 with optimal
solution cost=70
25
15 40 Table 4: The options for adding a link and deleting another
D link in iteration 3
Figure 5. Modified network after Iteration 1 with new Add Delete Cost
cost=75
Iteration 2: In the second iteration the following decisions AB BE (Tabu move)
are made. AB CE 100+0=100
AB AC 95+0=95
B AD DE 60+100=160
AD CE 95+0=95
30 30 AD AC 90+0=90
CD DE 70+0=70 (optimal solution
10 5 CD CE 105+0=105
A C E
With a well designed tabu search algorithm, the best trial
solution found in Figure 7 after the algorithm has run a
25
modest number of iterations is likely to be a good feasible
15 40
D solution. It might even be an optimal solution [1], [2] but no
such guarantee can be given. Selecting a stopping rule that
Figure 6. Modified network after Iteration 2 with new provides a relatively long run of the algorithm increases the
cost=85 chance of reaching the global optimum.
2.2 Simulated Annealing happens to inherit the better features of the parents is
Simulated annealing [1]-[5] is another widely used slightly more likely to survive into adulthood and then
metaheuritic that enables the search process to escape from become a parent who passes on some of these features to the
a local optimum. Each iteration of the simulated annealing next generation. The population tends to improve slowly
search process moves from the current trial solution to an over time by this process. A second factor that contributes to
immediate neighbor in the local neighborhood of this this process is a random, low-level mutation rate in the
solution, just as for tabu search. However, the difference DNA of the chromosomes. Thus, a mutation occasionally
from tabu search lies in how an immediate neighbor is occurs that changes the features of a chromosome that a
selected to be the next trial solution. Let child inherits from a parent.
Zc = objective function value for the current trial solution 2.3.1 Basic Genetic Algorithm
Zn = objective function value for the current candidate to be Initialization: Start with an initial population of feasible
the next trial solution. trial solutions, perhaps by generation then randomly.
T = a parameter that measures the tendency to accept the
Iteration: Use a random process [5] that is biased toward
current candidate to be the next trial solution if this
the more fit members of the current population to select
candidate is not an improvement on the current trial
some of the members to become parents. Pair up the parents
solution.
randomly and then have each pair of parents give birth to
The rule for selecting which immediate neighbor will be the two children whose features are a random mixture [10] of
next trial solution is the following. the features of the parents, except for occasional mutations.
Evaluate the fitness for each new member in the new
Move Selection rule: Among all the immediate neighbors
population.
of the current trial solution [9], select one randomly to
become the current candidate to be the next trial solution. Stopping rule: Use some stopping rule, such as a fixed
Assuming the objective is maximization of the objective number of iterations, a fixed amount of CPU time, or a fixed
function, accept or reject this candidate to be the next trial number of consecutive iterations without any improvement
solution as follows: in the best trial solution found so far. Use the best trial
solution fond on any iteration as the final solution.
If Zn ≤ Zc, always accept this candidate
If Zn < Zc, accept the candidate with the following 3. Proposed Work
probability
The new algorithm based on integrating genetic
Zn − Zc algorithms, tabu search and simulated annealing methods to
Prof {acceptance} = e where x =
x
T solve the minimum spanning tree problem. The core of the
proposed algorithm is based on tabu search algorithms.
Genetic algorithm is used to generate new population
2.2.1 Basic Simulated Annealing Algorithm members in the reproduction phase of the tabu search
Initialization: Start with a feasible initial trial solution. algorithm. Simulated annealing method is used to accelerate
the convergence of the tabu search algorithm by applying
Iteration: Use to move selection rule to select the next the simulated annealing test for all the population members.
trial solution. A new implementation of the tabu search algorithm is
Check the temperature schedule: When the desired introduced. In the genetic algorithm part of the proposed
number of iterations has been performed at the current value algorithm, a simple short-term memory procedure is used to
of T, decrease T to the next value in the temperature counter the danger of entrapment at a local optimum, and
schedule and resume performing iterations at this next the premature convergence of the tabu search algorithm. A
value. simple cooling schedule has been implemented to apply the
simulated annealing test in the algorithm [3]-[5].
Stopping rule: When the desired number of iterations
have been performed at the smallest value of T in the 3.1 Basic Integration Algorithm
temperature schedule as final solution. Initialization: Start with a feasible trial solution.
2.3 Genetic Algorithms Iteration: Use an appropriate local search procedure to
Genetic algorithms [1]-[3] provide a third type of define the feasible moves into the local neighborhood [6] of
metaheuristic that is quite different from first two. This type the current trial solution. Determine which of the remaining
tends to be particularly effective at exploring various parts moves provides the best solution. At the same time Use the
of the feasible region and gradually evolving toward the best move selection rule to select the next trial solution. For each
feasible solutions. The modern field of genetics provides a iteration invoke the algorithm to test the cycles in a graph.
further explanation of this process of evolution and the
Stopping rule: If a pre-defined termination condition
natural selection involved in the survival of the fittest. In
is satisfied, output Best Solution and exit.
any species that reproduces by sexual reproduction, each
offspring inherits some of the chromosomes from each of 3.1.1 Algorithm for testing Cycles
the two parents, where the genes within the chromosomes
determine the individual features of the child. A child who
Step 1: From the combination of edges and incidence Step 3: Test whether at least three nodes of degree more
matrix, obtain degree of each node contributed by the edges than one? If not go to step 5.
under consideration. Step 4: Delete pendant edges, if exists of n-1 edges and
Step 2: Test whether at least two nodes of degree one? If modify the degree of the modes accordingly and go to step 2.
not, go to step 6. Otherwise continue. Otherwise go to step 6.
Table 6: Comparisons of TS, GA, SA and Integration Algorithm for Spanning Tree
Graph TS GTA SA Integration Algorithm
size (No.
of Time(s) Cost Iteration Time(s) Cost Iteration Time(s) Cost Iteration Time(s) Cost Iteration
Nodes)
5 50 75 3 52 75 3 55 75 4 50 75 2
7 55 80 3 54 80 4 55 80 4 48 80 2
9 58 95 3 57 90 3 60 90 4 52 90 2
11 65 100 4 68 100 3 75 110 5 60 95 1
13 70 120 4 75 120 4 78 120 5 60 115 1
Figure 8. Comparisons of TS, GA, SA and Integration Algorithm for Spanning Tree
Step 5: Edge combinations are tree. The final output of the iteration 1 in a network is shown in
Step 6: Stop. Figure 9. Well defined tabu search algorithm [2], the best
trial solution found after the algorithm has run a modest
Iteration 1: Instead of charge a penalty for the constraint,
the constraint should be verified for every change in the number of iterations is likely to be a good feasible solution
graph. There are three options for adding a link to the [8]. But the integration algorithm found the feasible solution
network in Figure 4 are CD, DE and DE. If CD were to be in only little iteration from the starting.
chosen, the cycle formed, so the two options for deleting a
link would be AD and AC. If AD is added then AB should B
be deleted according to the second constraint. So that for
connecting B, BE should be added. So that iteration 1 is 30 30
followed as per Table 5.
10 5
Table 5: The options for adding a link and deleting another A C E
link in iteration 1
25
Add Delete Cost 15 40
D
BE CE 75
BE AC 70
Figure 9. Modified network after Iteration 1 with new
BE AB 60
Constraints violated. cost=70
CD AD 60
CD AC 65
CD AB 70 4. Experiments and Results
DE CE 85 Five testing graphs of the size 5, 7,9,11 and 13 nodes. All
Constraints violated.
DE AC 80 the five graphs are generating a parse tree using tabu search,
genetic algorithm, simulated annealing and integration
algorithm. All the algorithm were stopped when the solutions
did not change over a fixed number of iterations. The relative [7] K. Aggarwal and R. D. Kent, “An Adaptive
performance between TS, GTA, SA and Integration algorithm Generalized Scheduler for Grid Applications”, in
for the minimum spanning tree problem is given in Table 6 Proc. of the 19th Annual International Symposium on
and Figure 8. Five trials were taken on each graph and the High Performance Computing Systems and
average value of the cost and time in milliseconds are Applications (HPCS’05), pp.15-18, Guelph, Ontario
tabulated in each case. It is evident from the table that Canada, May 2005.
Integration algorithm performed very well as the problem size [8] Anant Oonsivilai, Wichai Srisuruk, Boonuruang
increased and showed superliner speedups at higher problem Marungsri and Thanatchai Kulworawanichpong,
sizes. “Tabu Search Approach to Solve Routing Issues in
Communication Networks”, World Academy of
5. Conclusion and Future Work Science, Engineering and Technology, pp. 1174-
In this paper, we have implemented an integration of 1177, 2009.
all the three metaheuristics [5] algorithms for minimum [9] D.Janaki Ram, T.H.Steenivas and K.Ganapathy
spanning tree problem. In this approach, a pre-defined Subramaniam, “Parallel Simulated Annealing
number of starting solutions are chosen from widely Algorithms”, International journal of Parallel and
separated regions in the sample space, and used in Distributed Computing, Vol.37,pp. 207-212, 1996.
local search procedures to obtain a set of locally optimal [10] D.Janakiram, “Grid Computing”, A Research
solutions. These locally optimal solutions [9] [10] are Monograph, Tata McGraw-Hill Publishing Company
then examined to provide an idea about the probability Limited, 2005.
of each vertex being included in an optimal solution [11] Bernard Chazelle, “A Minimum Spanning Tree
to the instance. Using these ideas, the neighborhood Algorithm with Inverse Ackermann Type
of each solution is searched in a probabilistic manner, Complexity”, Journal of the ACM, 47(6), pp. 1028-
such that the moves excluding vertices that have a 1047, 2000.
high probability of being in an optimal solution or of [12] Sanjay Kumar Pal, “Renovation of Minimum
including vertices that have a low probability of being Spanning Tree Algorithms of Weighted Graph”,
in an optimal solution are discouraged. Experiment Journal of ACM Ubiquity, Vol.9, Issue 7, pp. 1-5,
result shows that the integration of metaheuristics 2008.
provides a feasible solution [11] if the numbers of nodes [13] Greening, Daniel R., “Parallel Simulated Annealing
Techniques”, Physica D, Vol.42, pp. 293-306,1990.
are increased in a graph. In future based on this result
[14] Van Laarhoven, P.J.M., E.H.L., AArts, and Jan Karel
this may be extended for feasible dynamic routing in a
Lenstra, “Job Shop Scheduling by Simulated Annealing”,
network. Operation Research, Vol. 40, pp. 113-125, 1992.
References Authors Profile

[1] R.Thamilselvan and Dr.P. Balasubramanie, “A Genetic
Algorithm with a Tabu Search (GTA) for Travelling R.Thamilselvan is an Assistant Professor in the department of
Salesman Problem”. International Journal of Recent computer Science and Engineering, Kongu Engineering College,
Trends in Engineering, Issue. 1, Vol. 1, pp. 607-610, Perundurai, Tamilnadu India. He has completed his M.E Computer
June 2009. Science and Engineering in 2005 under Anna University Chennai.
[2] R.Thamilselvan and Dr.P. Balasubramanie, He has completed 8 years of teaching service. He has published 3
papers in national conference and 2 paper in International Journal.
“Integrating Genetic Algorithm, Tabu Search
He was the recipient of the Best Faculty award during the year
Approach for Job Shop Scheduling” International 2007-2008. His area of interest includes Grid Computing, Parallel
Journal of Computer Science and Information Processing, and Distributed Computing. He has organized 2
Security, Vol. 2, No.1, pp. 134-139, 2009 national level seminar sponsored by AICTE, New Delhi.
[3] Jingjing Zhang, Yaohui Jin and Wsisheng Hu, “A
Genetic Algorithm for Smooth Scheduling in Slotted Dr.P.Balasubramanie is a Professor in the department of
computer Science and Engineering, Kongu Engineering College,
WDM Network”, International Journal of Information
Perundurai, Tamilnadu India. He was awarded junior research
Technology, Vo. 12, No. 6, pp.26-34, 2006. Fellowship by Council of Scientific and Industrial Research
[4] S.Jayalakshmi and S.P. Rajagopalan, “Modular (CSIR) in 1990 and he has completed his Ph.D degree in 1990
Simulated Annealing in Classical Job Shop under Anna University in 1996. He has also qualified for the state
Scheduling”, International Journal of Information Level Eligibility test for Lectureship in 1990. He has completed 13
Technology, Vol. 6, No.2, pp. 222-226, 2007. years of teaching service. He has published more than 60 articles
[5] N.K.Cauvery and Dr. K.V Viswanatha, “Routing in in International/National Journals. He has authored six books with
Dynamic Network using Ants and Genetic the reputed publishers. He was the recipient of the Best Faculty
Algorithm”, International Journal of Computer award for consecutively two years. He is also the recipient of the
CTS Best Faculty Award- 2008. He has guided 3 Ph.D scholars
Science and Network Security, Vol. 9, No. 3, pp. 194-
and 23 research scholars are working under his guidance. His area
200, March 2009. of interest includes Image processing, data mining, networking and
[6] Van Laarhoven, P.J.M., E.H.L., AArts, and Jan Karel so on. He is a member of Board of studies, Anna University
Lenstra, “Job Shop Scheduling by Simulated Coimbatore. He has organized several seminar/workshops.
Annealing”, Operation Research, Vol. 40, pp. 113-
125, 1992.
Forward Collision and Delay Avoidance Using

Energy Efficient Collision Sets in WSN
Bhushan N. Mahajan 1, Dr. R. V. Dharaskar 2 and Dr. V. M. Thakare 3
1
Computer Science Department
GHRCE, Nagpur
anupcus123@gmail.com
2
HOD , Computer Science Department
GHRCE, Nagpur.
rvdharaskar@gmail.com
3
SGB Amaravati University.
vilthakare@rediffmail.com
its information flow through other mobile nodes. This

Abstract: This paper present battery-efficient system design, and
outlines collision and delay avoidance using scheduling method. implies that mobile nodes in Ad Hoc networks bear routing
Batteries power electronic systems and maximizing battery life functionality so that they can act both as routers and hosts.
needs understanding of the capabilities and limitations of the These networks provide mobile users with everywhere
batteries. The amount of energy that can be supplied by a given communication capacity and information access regardless
battery varies significantly. It depend on how the energy is drawn. of location. The Ad Hoc networks can be seen in to two
Researchers are attempting to develop new battery driven categories whether dynamically changing their position or
approaches to system design. not, once create communication link. These are wireless
Keywords: Collision , Delay , Energy level , power , Battery sensor networks and Mobile Ad Hoc networks (MANETs).
power , Sleep mode , Ideal mode. Wireless sensor networks’ mobile nodes are deployed in
large number on small area. Once the nodes are deployed,
1. Introduction they are static. In Mobile Ad Hoc networks the nodes can
dynamically change their position . An Ad Hoc network can
Battery powered electronic systems, and the integrated be used in an area where infrastructures for mobile
circuits within them, account for a large and rapidly communication are not available, probably due to high
growing revenue segment for the computer, electronics, and deployment costs or disaster destruction. The typical
semiconductor industries. We know “battery gap”, between application of Ad Hoc networks includes battle field
trends in processor power consumption , and improvements communication, emergency relief and extension of the
in battery capacity . Bridging this gap is a challenge that coverage area of cellular networks.
system designers must face for the future. Low power design
techniques are successful in reducing the energy that is Ad-hoc routing algorithms broadly can be categorized into
drawn from the battery, and hence improve battery life to pro-active and on-demand routing . The on-demand routing
some extent. But, truly maximizing battery life requires an algorithms initiate to find out the suitable route when a
understanding of both the source of energy and the system route is requested . The pro-active routing algorithm
that consumes it. [1] exchanges routing information periodically and generates
the routing table in advance of route request . [3] These
2. Related work protocols select the routes based on the metrics of minimum
hop count. Ability to forward packets is depend on Battery
This energy efficient routing protocol is modified version of power . Overall network lifetime is depend on battery power
the Ad-hoc on demand distance vector (AODV) protocol by Design and manufacturing of less energy consume
taking into consideration the results of the pre-simulation, components of mobile nodes such processors, memory and
the existing feature of AODV for implementation of the OS power management strategies is used to reduce non-
design . An Ad Hoc network Networks that do not require a communication energy consumption.
pre-established infrastructure . An Ad Hoc network does not During communication, energy is consumed in either
guarantee that a mobile node can directly communicate with inactive state of communication or active communication
destinations all the time. So, there is independence of any states. The energy consumption of active communication is
fixed infrastructure or centralized administration. [2]An Ad more significant than the others for high traffic
Hoc network is capable of operating autonomously. It is environment.
completely self-organizing and self-configuring. It has Energy efficient routing protocols can be designed to
multi-hop capability. A mobile node, which lies outside the formulate energy efficient active communications . Energy
transmission of its specific destination, would need to relay efficient virtual topology can be designed to formulate
energy efficient active communications . Minimum hops , Existing wake-up schemes

Minimum message exchange , Minimum collision , Nodes have to remain awaken when they are not receiving
Minimum retransmission strategies can be designed to any data. Nodes have to listen idle channel . Network nodes
formulate energy efficient active communications . Result is cannot easily know exactly when events happen. The delay
Energy efficient active communications. It will prolong the exist between receiving wake up message and actual waking
network life time . The network life time is defined as the up of node. That delay is called as wake-up latency. Node
time when a node runs out of its own battery power for the wakes up itself when it senses an communication from other
first time . [4] Active communication energy efficient nodes. Then, it wakes up the MCU and RF transceiver.
routing protocols can be categorized into two categories:
• Transmission power control approach Time-based wake-up mechanisms
• Load distribution approach. It require each node wake up periodically to listen to the
For protocols that belong to the former category, the active radio channel .It is time-based wake-up mechanisms. Low
power sleeping nodes wake up at the same time periodically
communication energy can be reduced by adjusting each
to communicate. Data are transferred from sensor nodes to
node’s radio power just enough to reach the receiving node
the sink through a multi-hop communication paradigm [3].
but not more than that. This transmission power control
By choosing a good time synchronized sleep and wake-up
approach can be extended to determine the optimal routing
mechanism the network may save much power consumption.
path that minimizes the total transmission energy required
Drawbacks to this mechanism is the high quality clock
to deliver data packets to the destination. The specific goal synchronization requirement. It makes it hard to be
of the load distribution approach is to balance the energy. implemented in large WSN. [6 , 11] If the packet arrives at
How active transmission energy can be reduced? . Answer node at the end of the listening cycle, then wake up call will
is by adjusting each node’s radio power just enough to reach be delayed while the node sleeps. It is wake up latency .
the receiving node but not more than that , by determining
the optimal routing path , by determining the task Asynchronous wake-up mechanisms
distribution approach to balance the energy usage of all Each node follows its own wake-up schedule in idle states, as
mobile nodes . by Selecting a route with underutilized nodes long as the wake-up intervals among neighbors overlap. It do
rather than the shortest route, by Selecting energy levels to not require time synchronization among the different nodes
minimizes transmission energy to deliver data packets to in the network. Energy harvesting for sensor is still in its
the destination. early stages, and is gaining momentum in the research
All above approaches prolong the network life time in community [11]. Drawbacks to this mechanism are this
different ways. [5] Two approaches are mutually strategy can lead to large packet delays within the network
contradicting each other with some extent. .It cannot achieve the same level of power savings that
Advantages- synchronous approach can.
• Increased the life time of the node.
• Increased packet delivery fraction On-demand wake-up mechanism
Possible solution is to design an on-demand wake-up In this
• Reduced the Variance of the nodes residual battery Energy.
mechanism , Here out-band signal is used to wake up
• Minimized energy consumed per packet.
sleeping nodes in an on-demand manner. For example, a
The purpose of energy-aware routing protocols is to
transceiver is woken up by a special wake-up signal from a
maximize the network lifetime. The energy-efficient routing
terminal that wants to communicate with a sleeping
protocols should consider energy consumption from the
transceiver. By this way, a transceiver is woken up on
viewpoints of both the network and the node level. demand instead of periodical wake-up so that the power
Why task distribution approach is required from node point consumption can be dramatically reduced .
of view ?
Nodes have limited remaining battery energy. Nodes can Span [11] is a connectivity-driven protocol that adaptively
perform every kind of task and while working they consume elects “coordinators” of all nodes in the network.
energy. But equal consumption should occur among all Coordinators stay awake continuously and perform multihop
nodes. So, work should not be assigned to same set of nodes routing, while the other nodes stay in sleeping mode and
again and again. It may lead to over consumption of those periodically check if there is a need to wake up and become a
nodes battery power . Constant amount of power of all nodes coordinator. The protocol achieves the following four goals.
should get utilized. It is called as balancing .The way for First, it ensures that there is always a sufficient number of
coordinators so that every node is in the transmission range
achieving balancing is –
of at least one coordinator. Second, to spread energy
• Equal distribution of work. consumption as uniformly as possible among network nodes
Span rotates the coordinators. Third, it tries to minimize the
• Selecting different routing path at different time instance. number of coordinators (to increase the network lifetime)
• Changing routing path frequently. while avoiding a performance degradation in terms of
network capacity and message latency. Fourth, it elect
• Same routing path will not be repeatedly used again and coordinators in a decentralized way by using only local
again. information.
• Same set of nodes will not get over utilized again and
STEM (Sparse Topology and Energy Management) [11]
again
It uses two different radios for wakeup signals and data Conditional Max-Min Battery Capacity Routing
packet transmissions, respectively. The wakeup radio is not a (CMMBCR):
low power radio (to avoid problems associated with different If there are nodes that have more battery power than
transmission ranges). Therefore, an asynchronous duty cycle threshold power, it applies MTPR to the nodes. Otherwise, it
scheme is used on the wakeup radio as well. Each node mimics MMBCR. When battery power is plentiful, it
periodically turns on its wakeup radio for Tactive every T minimizes the total energy consumption like MTPR, and the
duration. When a source node (initiator) has to communicate other case it considers the nodes with lower energy like
with a neighboring node (target), it sends a stream of MMBCR. When the current drawn is sufficiently large,
periodic beacons on the wakeup channel. As soon as the the rate of diffusion fails to keep up with the rate at which
target node receives a beacon it sends back a wakeup ions are consumed at the cathode. As a result, the
acknowledgement, and turns on its data radio. [7] concentration of positively charged ions decreases near the
If we select longer-hop routes, which spend more energy end cathode and increases near the anode, degrading the
to end. battery’s output voltage. However, if the battery is allowed
Minimum total power routing (MTPR): to idle for a period of time, the concentration gradient
S is the set containing all the possible routes. Rth route is decreases (due to diffusion), and charge recovery takes place
selected . Transmission power for route R is P . at the cathode. As a result, the capacity and lifetime of the
R battery increase..
This approach may select the route that includes one or 3. Algorithm
more mobile node with least energy level or min-hop
routing . it makes no effort to use energy evenly among the Nodelist={all nodes as per pre-defined index};
nodes . This leads to “die” of the first node sooner and it
causes partition of the network early. Hierarchy _ I = {set of nodes in level I} ;
Power aware localized routing: Hierarchy _II = {set of nodes in level II} ;
It is assumed that the power neededα for transmission and Hierarchy _III = {set of nodes in level III} ;
reception is a linear function of d where d is distance Hierarchy _IV = {set of nodes in level IV} ;
between the two neighboring nodes and α a parameter that
depends on the physical environment. [8] The authors make Threshold = x ; // pre-set
use of GPS position information to transmit packets with the
Send_Nodelist = nodelist;
minimum required transmit energy. Two drawbacks
Rev_Nodelist = nodelist;
are,GPS cannot provide useful information about the
SendDoneNodelist={set all zero} ;
physical environment and the second is that the power
RecvDoneNodelist={set all zero} ;
dissipation overhead of the GPS device is an additional
power draw on the battery source of the mobile node. do
Minimum Battery Cost Routing (MBCR): {
It tries to use battery power evenly by using a cost function slot = new Slot();
which is inversely proportional to residual battery power. // set flags here .
One possible choice for the cost function of a node i is given // flag may be Fwd or Rev
as
check_direction (node);
bi is the residual battery energy of node i. the total cost of Fwd_send_collision_set = null;
the route is defined as the sum of costs of nodes that are the Fwd_recv_collision_set = null;
components of the route, and MBCR selects a route with Rev_send_collision_set = null;
minimum total cost. [9] Rev_recv_collision_set = null;
The Min-Max Battery Cost Routing (MMBCR):
for each node in nodelist
{
It selects the route with the minimum path cost among node->u = check_ Hierarchy (node) ;
possible routes. Because this metric takes into account the node->v = check_ Hierarchy _Energy_Level (node) ;
remaining energy level of individual nodes instead of the node->w = check_incoming_packet_seq_no (node) ;
node->x = check_rec_slots(node) ;
total energy, the energy of each node can be evenly used.
node->y = check_send_slots(node) ;
The limitation of this algorithm is that since there is no
node->z = check_traffic_over (node) ;
guarantee that paths with the minimum hop-count or with
the minimum total power are selected. It can select paths
if (
that results in much higher power dissipation in order to (node not in Fwd_send_collision_set)
send traffic from a source to destination nodes. This feature & & (node->dest not in Fwd_recv_collision_set)
actually leads to in shorter network lifetime because in & & ( node->z not “over crowding” )
essence the average energy consumption per delivered & & ( node->y not “0” )
packet of user data has been increased. [10] & & ( node->x == “0” )
& & ( node->w == “No duplication” )
& & ( node->w == “Above_threshold” ) power for the first time. If a node stops its operation, it can
) result in network partitioning and interrupt communication.
{ We will see why transmission power control approach from
node point of view .The nodes with lower power should
// 7à 6 . 7=send , 6 =Recv avoid transmission up to large distance. The node should
balance the energy usage among the neighbor nodes . It
slot[node] = ‘‘SEND’’; minimizes the total transmission energy required to deliver
slot[node?dest] = ‘‘RECV’’; data packet to the destination .[15] We will see why
transmission power control approach at network level . The
// Nodes those will cause collision best route in network is that one which require minimum
total transmission power . The network lifetime is the time
Fwd_send_collision_set.add(
when a node runs out of its own battery power for the first
node->dest->neighbors.flag Fwd );
time. [16] If a node stops its operation, it can result in
Fwd_send_collision_set.add( network partitioning and interrupt communication.
node->dest.flag.Fwd); Minimizing the total energy consumption tends to favor the
route that consumes minimum energy .
// Nodes those will face collision 4. Experiment
Fwd_recv_collision_set.add(
Node.flag Fwd); Now we will concentrate on delay avoidance and collision
avoidance . Here is a topology and related routing table .
Fwd_recv_collision_set.add( Node 0 is a cluster head .Routing table contain list of all
node->neighbors.flag Fwd); neighbors and their send slot and receive slots . we will
define a schedule for each node so that path wake up timing
// Nodes those will cause collision will get decided [11].
Rev_send_collision_set.add( 5
node->dest->neighbors.flag Rev ); 7
6
4 2
Rev_send_collision_set.add( 9
8
node->dest.flag.Rev); 0 10 11
3 1
// Nodes those will face collision
13
Rev_recv_collision_set.add( 14
12
node->neighbors.flag Rev);
change_priority_algo(node->dest->recv); 15
SendDoneNodelist.addtoLastPos(node); Figure 1 16
RecvDoneNodelist.addtoLastPos(node?Dest);
Send_Nodelist.Remove(node); Node Neighbors
Recv_Nodelist.Remove(node?Dest); 6 2, 5, 7, 8
7 6, 9
If ( decide_(collision , threshold ) > 1 ) 8 0, 6, 10
{ 9 7, 11
Node_List = Recv_Nodelist.
} Our aim is to reach up to node 0 .
else
{ Node 1-hop Neighbors Dest Hops upto node O
Node_List = Send_Nodelist.
} 6 2, 5, 7, 8 8 2
7 6, 9 6 3
8 0, 6, 10 0 1
} // if
9 7, 11 11 3
} // for
Node Neighbors Dest Hops Recv Send
frame.add(slot);
6 2, 5, 7, 8 8 2 1 1
7 6, 9 6 3 0 1
} while (slot.length > 0) 8 0, 6, 10 0 1 1 1
// do 9 7, 11 11 3 0 1
We will see why task distribution approach from network Now algorithm will decide slots for receive and send .
point of view. The best route in network is that one which
require minimum total transmission power . The network Node/ slot 1 2 3 4 5 6 7
lifetime is the time when a node runs out of its own battery
6 R S
7 S
Figure 3.
The more improved topology construction is suggested
below . It will include delay avoidance strategy . Few other Five levels of power should be decided at each node.
strategies that can be adopted are also suggested below . Level 1 = sleep mode
Three levels of power should be decided at each node. Level 2 = forward data
Level 1 = Nodes those are at one hop . Level 3 = receive data
Level 2 = Nodes those are at two hop . Level 4 = send data to neighbor.
Level 3 = Nodes those are at three hop . Level 5 = send data up to cluster head .
Level 6 = send busy tone .
As shown in figure 2, we can create topology. Node 0 is
currently acting as cluster head . Its task can be handed over Three types of sleep periods should be created
to any other node which is at one hop distance from node 0 . Level 1 = data to be send after x unit of time .
Node in square boundary can act as cluster heads of their Level 2 = data to be stored after x unit of time .
own clusters. They can communicate to other cluster head Level 3 = data to be forwarded after x unit of time .
by using node 0 using two hops . Node 3,5,1,16 are still Node list = { 17,18,19,20,21,22,23,24,25,26,27,28,
available for forming new cluster . 29,30,31,32,33,34,35,36,37,38,39,40 }
Start from 17 .
17->41 ,
[ 17,1] = ”S”
[ 41,1] = ”R”
Node One hop neighbors
16 0.7,15,17,18,19,20
17 41,42,16
Node Choose Dest Hops Send Recv
16 0 1 4 4
17 16 2 4 4
Figure 2.
Cluster head can communicate with node 0 in two hops. Divide transmission of 17 in two plus two
Scheduling information can be stored at node 0. Task of slots by blocking alternate reverse paths .
node is to dispatch packet up to cluster head in two hops. It
is now node0 task to forward packet up to destination in Fwd_send_collision_set = { 16, 0.7,15 }
two hops only . Hierarchical addressing can be also used . // new nodes // They cause collision
Probability of send , probability of receive , probability of // 16 and its neighbor
idle , probability of Sleep are depend on four important
factors .Data forward rate , data arrival rate , delay or Fwd_recv_collision_set = { 17,41,42,16 }
waiting time , time spend in wake up . // existing nodes // They face collision
// 17 and its neighbor
Division of task can be designed at different level
Level 0 = Scheduling information Rev_send_collision_set = { 17,18,19,20 }
Level 1 = Routing information . // new nodes
Level 3= Energy, collision set information // They cause collision
Rev_recv_collision_set = { 17,41,42,16 }
// existing nodes // They face collision
Now we will assign a last priority to 7 in

SendDoneNodelist . we will assign a last priority to 6 in
RecvDoneNodelist. So , the collisions those were likely to
occur will get avoided .and superior priority will be given to
rest of the nodes of Fwd_send_collision set nodes.
5. Power related parameters

We can use multi channel sender initiated random access

mac layer protocols - Divide available bandwidth into The end-to-end delay, is the sum of the transmission delay,
multiple channel .Instead of switch off mode , use low the access delay, the queuing delay , the propagation delay.
power switch mode . They can be turned on using timer Or The delay is proportional to the number of hops N. Delay
using wake-up packets. Mac layer can decide turn off/on of is affected by number of hops times the period Tf. The
module. [12] transmission at each hop is delayed until the next receiving
sensor node wakes up. [14 ]Small distance between the
Various advantages of using this scheme are –
wireless sensor nodes makes the propagation delay small.
• Minimum delay Large distance between the wireless sensor nodes makes the
• Energy level selection propagation delay large. Say (N - 1)intermediate forwarders
• Schedule distribution are there between sender node and receiver node .[11]
• Collision level
• Selection of bandwidth and speed. Average delay
• Fast input and fast output
• Allocate function to individual nodes
• Achieve division of work. The delay in the S-MAC protocol
When a new packet is generated, the sensor node must wait
until the next TDMA frame. [13] It transmits the Path-WU
message to wake-up all the nodes. Since the packet is
generated randomly, the average time until the next frame is The delay in the adaptive listening
Tf/2. [11]
The delay in TDMA
tcs = Access time

ttx = Transmission delay.
Here , Power as a function of the arrival rate for the same N = No. of nodes.
delay. Delay as a function of the arrival rate, for the same T f = No. of hops times.
power consumption. The minimum values for Ti and Tslot T f/2. = Average time until the next frame is generated.
are limited by the available hardware and by the reception When Tf increase , end to end delay increase . When Tf
sensing time. we consider Ti = Tslot. [11] decrease , nodes must become active more often . It result in
increased power consumption. End to end delay is affected
by term which is proportional to T f . This is because,
transmission at each hop is delayed until the next receiving
sensor node wakes up . Transmission must wait until the
next wake up time . [11]
7 . Conclusion
L = average size of packets .
. = data packet arrival rate. The study of scheduling algorithm helps to achieve the
. = data packet forward rate. balance between energy saving and delay. By sensing energy
n = the number of timeslots the sensor from the radio signals, the node provides a interrupt to the
MCU. The MCU detects the preamble and wakes up RF
node listens transceiver when wake-up signal is indeed to wake itself up.
The analysis results show that, based on the scheduling,
nodes can be woken up timely when it is necessary to wake
Tf = the length of the total period
Tslot = the length of each TDMA timeslot. up with the aid of extremely low power.
n * (Tslot/Tf) = The time a sensor node spends in periodical
wakeups .
LWait. = The average of waiting time Reference
[1] V. Raghunathan, C. Schurghers, S. Park, M. Srivastava,
Based on the calculation , the balance energy level is “Energy-aware Wireless Microsensor Networks”, IEEE
calculated . The balance energy is calculated and it is used Signal Processing Magazine, March 2002, pp. 40-50.
to decide the remaining lifetime of battery of node . [2] G. Pottie, W. Kaiser, “Wireless Integrated Network
Sensors, Communication of ACM, Vol. 43, N. 5, pp. 51-
6. Delay related parameter 58, May 2000.
[3] I. Akyildiz, W. Su, Y. Sankarasubramaniam and E. GHRCE , Nagpur university , India . He is a professional
Cayirci, “Wireless Sensor Networks: a Survey”, software developer . He is now working on energy and power
Computer Networks , Volume 38, N. 4, March 2002. management topics and various schedule development strategies in
[4] IEEE Pervasive Computing, “Energy Harvesting and WSN . He has a special interest in topology modeling of ad-hoc
Conservation”, Vol. 4, Issue 1, Jan-Mar. 2005. network i.e. wireless sensor network , wireless mesh network and
[5] A. Kansal, J. Hsu, V. Raghunathan, J. Friedman, and MANET . He has simulated various network scenario using ns2
network simulator software and other programming languages.
M.B. Srivastava, UCLA Energy-Harvesting Project,
http://nesl.ee.ucla.edu/projects/heliomote.
[6] L. Doherty, B.A. Warneke, B.E. Boser, K.S.J. Pister,
“Energy and Performance Considerations for Smart Dr. Rajiv Dharaskar is presently
working as Professor at PG Department of
Dust,” International Journal of Parallel Distributed
Computer Science and Engineering, G.H.
Systems and Networks, Volume 4, Number 3, 2001, pp. Raisoni College of Engineering, Nagpur. He
121-133. is Ph.D. in Computer Science &
[7] J.A. Paradiso and T. Starnar, “Energy Scavenging for Engineering in the Faculty of Engineering &
Mobile and Wireless Electronics”, IEEE Pervasive Technology, M.Tech. in Computers, P.G.
Computing, Vol. 4, Issue 1, pp.18-26, Jan-Mar. 2005. Dip., M.Phil., and M.Sc. He is having 24
[8] J. Krikke, “Sunrise for Energy Harvesting Products”, years of teaching and 18 years of R&D experience in the field of
IEEE Pervasive Computing, Vol. 4, Issue 1, pp. 4-8, Jan- Computers & IT. He is approved PhD guide for Computer
Mar. 2005. Engineering and Science for Nagpur and Amravati University and
[9] M. Philipose, J.R. Smith, B. Jiang, A. Mamishev, S. Roy, 22 research scholars are perusing Ph.D. degree under his guidance.
and K. Sundara-Rajan, “Battery-Free Wireless He is an author of number books on Programming Languages..
Identification and Sensing”, IEEE Pervasive Computing,
Vol. 4, Issue 1, pp. 37-45, Jan-Mar. 2005. Dr V M Thakare is Professor and Head
[10] C. Schurgers, V. Tsiatsis, M. B. Srivastava, “STEM: of PG department of computer Science and
Topology Management for Energy Efficient Sensor Engg in SGB Amravati University
Networks”, IEEE Aerospace Conference '02, Big Sky, Amravati, Maharastra (India) and has
completed ME in Advance Electronics and
MT, March 10-15, 2002.
[11] Energy efficiency in wireless sensor networks using sleep mode Ph.D. In computer Science/Engg. His Area
TDMA scheduling Nikolaos A. Pantazis , Dimitrios J. Vergados , of Research are Robotics and Artificial
Dimitrios D. Vergados , Christos Douligeris . Intelligence, Information Technology. He is Recognized Giude for
computer science and computer engineering in this University and
[12] Y. Xu, J. Heidemann, D. Estrin, “Geography-informed In other universities also. He has also received received national
Energy Conservation for Ad Hoc”, Proc. ACM level Award for excellent paper award. More than 10 candidates
MobiCom 2001, pp. 70 – 84. Rome, 2001. are working for Ph D Under his supervision. He has Published
[13] P. Santi, “Topology Control in Wireless Ad Hoc and and presented more than 115 papers at National and international
Sensor Networks, ACM Computing Survey, Vol. 37, n. level. He has worked on various national level bodies like
2, p. 164-194, June 2005. AICTE/UGC and also worked on various bodies of other
[14] A. Cerpa, D. Estrin, “Ascent: Adaptive Self- universities. He is presently member of BOS, RRC, BUTR of this
Configuring Sensor Network Topologies, Proc. IEEE university and also chairman and Member of various committees
INFOCOM 2002. of this university .
[15] B. Chen, K. Jamieson, H. Balakrishnan, R. Morris. “Span: An
Energy-Efficient Coordination Algorithm for Topology Maintenance in
Ad Hoc Wireless Networks”, ACM Wireless Networks, Vol. 8, N. 5,
September 2002.
[16] F. Koushanfar, N. Taft and M. Potkonjak, “Sleeping

Coordination for Comprehensive Sensing Using Isotonic
Regression and Domatic Partitions”, Proc. of Infocom
2006, Barcelona, Spain, April 2006.
Authors Profile
Bhushan N. Mahajan received
Diploma in Mechanical Engineering
[DME] in 1999 , A.M.I.E. [CSE]
Engineering degree in 2007 , BCA
degree in 2007 , qualified GATE 2008
and MCA degree in 2009 . He is perusing
Master of Engineering degree [M.E.] in
WCC in Computer Science department at
Efficient Code Dissemination

Reprogramming Protocol for WSN
R.P.Shaikh 1, Dr. V.M. Thakare2, Dr. R.V. Dharaskar3
1
Computer Science Department
GHRCE, Nagpur
rpshaikh@gmail.com
2
GHRCE, Nagpur.
rvdharaskar@gmail.com
3
SGB Amaravati University.
vilthakare@rediffmail.com
Abstract: Network reprogramming is a way of reprogramming can disseminate efficiently a large data object from one node
wireless sensor nodes by disseminating the program code over to many other nodes over an entire wireless sensor network.
radio for uploading new code or for changing the functionality The three important steps for code dissemination
of existing code.. Existing reprogramming protocols, such as protocols is: advertisement of available software, selection
Deluge,achieve this by bundling the reprogramming protocol of a source, and reliable download to the target,which may
and the application as one program image, hence it increases then become a source in turn (Figure1).
the overall size of the image which is transferred through the
network. This increases both time and energy required for
network reprogramming. A proposed protocol divides the code
image into application and reprogramming support. It pre-
installs the reprogramming protocol as one image and the
application program equipped with the ability to listen to new
code updates as the second image that mitigates the above
problem.
Keyword : Wireless Sensor Network, Sensor, Wireless

reprogramming, code dissemination.
1. Introduction Figure 1. Three Way handshake for code distribution

A wireless sensor network is expected to consist of a
potentially large number of low-cost, low-power, and Thus, reprogramming sensor nodes, i.e. changing
multifunctional sensor nodes that communicate over short the software running on sensor nodes after deployment, is
distances through wireless links. Due to their potential to necessary for sensor networks. A scheme is required to
provide fine-grained sensing and actuation at a reasonable wirelessly reprogram the nodes The scenario poses many
cost, wireless sensor networks are considered ideal challenges, of them being energy, bandwidth and
candidates for a wide range of applications, such as industry reprogramming. Requirements and Properties of Code
monitoring, data acquisition in hazardous environments, Distribution are :
and military operations. It is desirable and sometimes 1. The complete image, starting from specific points in the
necessary to reprogram sensor nodes through wireless links network, must reach all the nodes. This is a requirement.
after they are deployed, due to, for example, the need of We do not consider the extended problem of reaching only
removing bugs and adding new functionalities. The process a subset of the nodes.
of propagating a new code image to the nodes in a network 2. If the image cannot fit into a single packet, it must be
is commonly referred to as code dissemination.
Traditionally, reprogramming was done manually. placed in stable storage until the transfer is complete, at
Therefore, nodes were reprogrammed one by one. However,
as the size of sensor nodes becomes larger and larger this which point the node can be safely reprogrammed. This is
technique is not very efficient. What is more, it might be
impossible to collect all the nodes from the field and then to also a required property.
reprogram them. Hence, reprogramming needs to be
accomplished without physical contact with the nodes. A 3. The lifetime of the network should not be severely
reliable data dissemination protocol is to be implemented affected by the distribution operation. This is a desirable
which takes under consideration the previous factors and property.
4. The memory and storage requirements of the mechanism then starts the data transfer. Missing segments are requested
should not be very high since that would limit the available directly from the sender, which prioritises these over further
space for the normal application. This property is also data transmissions. Once a node has received an entire
desirable. image, it becomes a sender in turn. If a sender receives no
subscribe messages, it transfers the new image to program
2. Relatedwork memory from EPROM, and reboots with the new code.
Sliding window acknowledgements reduce power
Code dissemination protocols have been developed to consumption (reduced EEPROM reads) at the cost of
propagate new code images using the wireless network reduced out-of-order message tolerance. There is no support
formed by the sensor nodes. Data dissemination in wireless for rate control, or suppressing multiple senders (apart from
networks, retransmission of broadcasts can lead to the link statistics).
broadcast storm problem, where redundancy, contention, Trickle[6] runs under TinyOS/Mate – it acts as a service
and collisions impair performance and reliability. Scalable to continuously propagate code updates throughout the
Reliable Multicast (SRM) is a reliable multicast mechanism network. Periodically (gossiping interval τ) using the
built for wired networks [15], using communication maintenance algorithm every node broadcasts a code
suppression techniques to minimize network congestion and summary (‘metadata’) if it has not overheard a certain
request implosion at the server. SPIN-RL is an epidemic number of neighbours transmit the same information. If a
algorithm designed for broadcast networks that makes use recipient detects the need for an update (either in the sender
of a three phase (advertisement-request-data) handshaking or in the receiver) then it brings everyone nearby up to date
protocol between nodes to disseminate data [16]. The by broadcasting the needed code. Trickle dynamically
epidemic property is important since WSNs experience high regulates the per-node, Trickle-related traffic to a particular
loss rates, asymmetric connectivity, and transient links due rate (rx + tx), thus adjusting automatically to the local
to node failures and repopulation. However, their results network density. This scales well, even with packet loss
show control message redundancy at over 95% as it only taken into account. A listen-only period is used to minimise
considers the suppression of redundant request messages, the short-listen problem (where de synchronised nodes may
and SPIN-RL does not perform as well as naive flooding for cause redundant transmissions due to a shift in their timer
lossy network models. phases). The CSMA hidden-terminal problem does not lead
The earliest network reprogramming protocol XNP[17] to excessive misbehaviour by Trickle, as long as the traffic
only operated over a single hop and did not provide rate is kept low. By dynamically changing the gossip
incremental updates of the code image. . A special Boot interval, Trickle can propagate changes rapidly, while using
Loader must be resident in a reserved section of program less network bandwidth when there are no known changes.
memory, and the xnp protocol module must be wired into an Programs fit into a single TinyOS packet.
application (to allow for subsequent XNP updates). A host
PC application xnp loads the image, via a base station mote 3. System Models
running TOSBase (this acts as a serial-to-wireless bridge) to
one (mote-id specific) or many (group-id specific) nodes The conventional reprogramming protocol system model for
within direct radio range of the base. The image is sent in sensor networks is depicted in figure 2, in which the code
capsules, one per packet; there is a fixed time delay between images are propagated from base station to every sensor
packet transmissions. In unicast mode, XNP checks delivery node in the network
for each capsule; in broadcast mode, missing packets are
handled, after the full image download has completed, using
a follow-up query request (nodes respond with a list of
missing capsules). The program is loaded into external
(nonprogram) memory. Applications are halted during the
program download. When a reboot command is issued (via
the xnp host program), then the boot loader is called: this
copies the program from external to program memory, and
then jumps to the start of the new program.
MOAP [4] is a multi-hop, over-the-air code distribution
mechanism. It uses store-and-forward, providing a ‘ripple’
pattern of updates; lost segments are identified by the
receiver using a sliding window, and are re-requested using
a unicast message to prevent duplication; a keep alive timer
is used to recover from unanswered unicast retransmission
requests – when it expires a broadcast request is sent. The
basestation broadcasts publish messages advertising the
version number of the new code. Receiving nodes check this Figure 2. Reprogramming model for sensor network
against their own version number, and can request the
update with subscribe messages. A link-statistics
mechanism is used to try to avoid unreliable links. After 4. The three substantially more sophisticated
waiting a period to receive all subscriptions, the sender protocols :
object as a set of fixed-size pages, a key feature needed for

4.1 MNP spatial multiplexing. Deluge is based on protocol Trickle , a
protocol designed for manipulating code updates in sensor
The design goal of MNP[8] is to choose a local source of networks. Deluge's basinality (borrowed from Trickle) is the
the code which can satisfy the maximum number of nodes. suppression and dynamic adjustment of the broadcast rate
They provide energy savings by turning off the radio of non- so as to limit the transmitted messages among n. Deluge
sender nodes. MNP is targeted at MICA2 motes running uses an epidemic protocol for efficient advertisement of
TinyOS and uses the XNP boot loader along with a code meta data and spatial multiplexing for efficient
dedicated network protocol to provide multi-hop, in-network propagation of code images. Deluge is generally accepted as
programming. The MNP protocol operates in 4 phases: the state of the art for code dissemination in wireless sensor
1. Advertisement/Request, where sources advertise the new networks, and has been included in recent TinyOS
version of the code, and all interested nodes make requests. distributions
Nodes listen to both advertisements and requests, and decide Deluge is a data dissemination protocol and algorithm for
whether to start forwarding code or not (this acts as a propagating large amounts of data throughout a WSN using
suppression scheme to avoid network overload); incremental upgrades for enhanced performance. It is
2. Forward/Download, where a source broadcasts a particularly aimed at disseminating software image updates,
StartDownload message to prepare the receivers, and then identified by incremental version numbers, for network
sends the program code a packet at a time (in packet-sized reprogramming. The program image is split into fixed size
segments) to the receivers to be stored in external memory pages that can be ‘reasonably’ buffered in RAM, and each
(EEPROM) – there is no ack, the receiver keeps a linked-list page is split into fixed size packets so that a packet can be
of missing segments in EEPROM to save RAM space; sent without fragmentation by the TinyOS network stack. A
3. Query/Update, where the source broadcasts a Query to all bit vector of pages received can be sent in a single TinyOS
its receivers, which respond by unicast by asking for the network packet. Nodes broadcast advertisements containing
missing packets (segments) – these are then rebroadcast by a version number and a bit vector of the associated pages
the source node, and then another Query is broadcast until received, using a variable period based on updating activity.
there are no requests for missing packets. The receivers, If a node determines that it needs to upgrade part of its
having received the full image, now become source nodes image to match a newer version, then, after listening to
and start advertising the new program; further advertisements for a time, it sends a request to the
4. Reboot, entered when a source received no requests in selected neighbour for the lowest page number required, and
response to an advertisement, where the new program the packets required within that page. After listening for
image is transferred to program memory, and the node further requests, the sender selects a page, and broadcasts
reboots with the new code. A node sends a download every requested packet in that page. When a node receives
request to all senders, this assists in sender selection, and the last packet required to complete a page, it broadcasts an
also allows the hidden terminal effect to be reduced (as advertisement before requesting further pages – this
other potential senders can overhead this request). The enhances parallelisation (‘spatial multiplexing’) of the
sender selection algorithm attempts to allow only one active update within the network (as the node can now issue
sender in a particular neighborhood. further requests in parallel with responding to requests from
other nodes). The protocol keeps the state data to a fixed
4.2 FRESHET size, independent of the number of neighbours. There are no
ACK’s or NACK’s – requesters either request new pages, or
Freshet[10] is different in aggressively optimizing the re-request missing packets from a previous page. There is
energy consumption for reprogramming. It introduces a new no global co-ordination to select senders; heuristics are used
phase called blitzkrieg when the code update is started from to try and elect relatively remote senders in order to
the base node. During the blitzkrieg phase, information minimise radio network contention. Incremental updating is
about the code and topology (primarily the number of hops supported through the use of Complete Advertisements
a node is away from the wave front where the code is at) which indicate which pages in an image have changed since
propagates through the network rapidly. Using the topology the previous version; requesters can then request just the
information each node estimates when the code will arrive changed pages. Future versions of Deluge are expected to
in its vicinity and the three way handshake will be initiated address the following issues: control message suppression,
– the distribution phase. Each node can go to sleep in running updates concurrently with applications, explicitly
between the blitzkrieg phase and the distribution phase reducing energy consumption, and support for multiple
thereby saving energy. Freshet also optimizes the energy types and versions of images.
consumption by exponentially reducing the meta-data rate
during conditions of stability in the network when no new 5. Proposed Protocol
code is being introduced, called the quiescent phase.
Each protocol discuss above transfers the image of the entire
4.3 DELUGE reprogramming protocol together with the minimally
necessary part.The researchers have found that it is difficult
Deluge[6] is a density-aware protocol with epidemic to improve over Deluge the rate of transfer of data over the
behavior that can help propagate the code reliable over wireless link. Hence to optimize what needs to be
unpredictable network conditions. It represents the data transferred, keeping the basic mode of transfer the same as
in Deluge ,transfer just what is needed, in other words, the •Improve the protocol for a new node to get image from
application code (or the code of the updates to the network
application).This idea gives rise to our proposed protocol.It •It optimizes the steady state energy expenditure by
transfers close to the minimally required image size by switching from a push-based mechanism where periodically
segmenting the total program image into an application node sends advertisements to pull based mechanism where
image and the reprogramming image.(Application image newly inserted node request for the code.
refer to the user application , reprogramming image refer to •In Freshet to save energy the sleeping time of node is to be
protocol component for protocol, such as MNP ,Deluge or estimated prior and this estimation if often found inaccurate
Freshet) The benefit of our protocol shows up in fewer due to variability of the wireless channel however stream
number of bytes transferred over the wireless medium protocol achieve this goal by rebooting the node from
leading to increased energy savings and reduced delay for Stream-RS only when new node arrives at one of its
reprogramming neighbors thus the user application running on the node can
put the node to sleep till the time to reboot comes. This
5.1 Protocol Description opportunistic sleeping feature conserve energy in resource
constrained sensored network.
An application is modified by linking it to a small •In Deluge, once a node’s reprogramming is over, it keeps
component called Application Support (AS) while on advertising the code image it has hence radio resources
Reprogramming Support (RS) is pre-installed in each node. are continuously used in the steady state but in stream
Overall, design principle is to limit the size of AS and ,Stream-AS does not advertise the data it has .
providing it the facility to switch to RS when triggered by a
code update related message.
Consider that initially all nodes have RS as image 0 and 5.3 Evaluation Results
the application with AS as image 1 Each node is executing
the image 1 code. The node that initiates the
reprogramming is attached to a computer through the serial With the help of reprogramming using the ns-2 simulator
port and is called the base node. Following is the we have to evaluate the message inter-arrival period and
description of how Stream works when a new user compared it with the total energy consumption of the sensor
application, again with the Stream-AS component added to nodes. Indeed our aim is to compare our proposed protocol
it, has to be injected into the network. with the known Deluge protocol [6] for wireless sensor
1.Reboot procedure takes place as follows: network and obtain the result and graph as displayed in
a. The base node executing image 1 initiates the process by Table I.and fig 3. Main objective is observe that the energy
generating a command to reboot from image 0. It broadcasts consumption has also been reduced because of the reduction
the reboot command to its one hop neighbors and itself in the overall size of the program image that needs to be
reboots from image 0. transferred over the wireless medium which may increase
b. When a node running the user application receives the the time and energy required for reprogramming the sensor
reboot command, it rebroadcasts the reboot command and nodes. Thus fewer number of bytes transferred over the
itself reboots from image 0. wireless medium leading to increased energy savings and
2.When all nodes receives reboot command they all start reduced delay for reprogramming
running RS. Then the new user application is injected into
the network using RS.
Table 1: Time Taken for Download Phase
3.Reprogramming of entire network stars using three way
handshake as discussed above. Each node maintains a set S
Code Size Download
containing the node ids of the nodes from which it has
time
received the requests for code.
Case 1 45.2 KB 112 sec
4. Once the node downloads the new code completely, it
performs a single-hop broadcast of an ACK indicating it has Case 2 54.3 KB 120 sec
completed downloading. Case 3 67.8 KB 135 sec
5. When a node receives the ACK from a node, it removes Case 4 75.7 KB 139 sec
the id of from its set S. Case 5 80.2 KB 141 sec
6. When the set S is empty and all the images are complete
and after sometime entire network is reprogrammed and
nodes will reboot from apllcation support.
5.2 Advantages :
•Reduce transmitted bit over wireless medium leading to

increased energy savings and reduced delay for
reprogramming
•Reduce programming time, energy costs and program
memory
sensor net (ref. energy-aware MAC layers, and energy-

aware Routing).
5.In order to support the various possible patterns in which
software updates may be received, and to support any
requirements for backwards and forwards version
compatibility, tighter control over the order of node
activation will be required.
6. There are a number of aspects of this which are not
directly related to software updating, but the key ones which
are related are: checking the downloaded software before
activation (integrity, version mismatches, platform
mismatches) and dynamically checking the operation of the
downloaded software after is has been activated. It is likely
that further advances will be necessary in this area, probably
using techniques from autonomic computing, to increase the
robustness of software updates.
7. There is a need for tools to monitor the ‘version’ state of
Figure 3. Message inter arrival period (SEC) a WSN and report status and problems to an operator/user.
These will be able to use existing techniques for fusing data
Other Parameters of implementation on ns-2 is as shown in to reduce the overhead, and for tracking update-related
in Table II faults.
8. The normal issues of: key-distribution, authentication,
secrecy, integrity, and authorization needing to be
Table 2: Parameters for Implementation on ns2 addressed. Results from existing WSN security research will
be needed, along with other work specific to the software
update problem.
9. The protocols used need to be energy-aware, so that the
current energy-state of both individual nodes and the entire
network can be taken into account during an update.
10. Recovering from faulty updates methods are required
before execution and during execution has been done
6. Conclusion
This paper examines the challenges of incorporating
scalable bandwidth management scheme and reducing the
reprogramming time, the number of bytes transferred, the
energy expended, and the usage of program memory for
wireless reprogramming in WSN environment with brief
There are a number of open research problems common to description of some existing proposals that specifically
all the classes: address this problem .
In future analysis of parameters as shown in table I &
1.Before initiating an update, it would be invaluable to be tableII by Simulation experiments to show the increasing
able run a model (such as a simulation) to advantages of proposed protocol over Deluge with larger
determine/analyze reprogramming time and the energy cost network sizes. Certain issues were not addressed in this
of different update options on the current network work, like the security issue, reliability etc. If an
configuration, and thus allow the user to make informed acknowledgement/code segment lost in a particular hop of
tradeoffs against energy. For example: which is the best multihop network due wireless medium constraints, then the
injection strategy for the current configuration? What size nodes which are in that hop have to take some necessary
reduction technique will result in the quickest update? etc. actions to achieve reliability.
2. There is not yet a definitive answer to the best way to
reduce the size of updates, with modular, differences-based,
References
and compression all showing promise in different
circumstances.
[1] P. Levis, N. Patel, S. Shenker, D. Culler, “Trickle: a
3. There are a number of different injection strategies in
selfregulating algorithm for code propagation &
use: simulating a base station, sending the update to a base
maintenance in wireless sensor network,” in:
station for dissemination, sending the update to a number of
Proceedings of the First USENIX/ACM Symposium on
seed nodes for further dissemination, and sending the
Networked Systems Design and Implementation “ (NSDI
update individually to each node.
2004) .
4. It is likely that an energy aware approach will have to be
taken in order to respond to current energy patterns in a
[2] “Remote Incremental Linking for Energy-Efficient [15] Beutel, J., Dyer, M., Meier, Ringwald, Thiele: Next-
Reprogramming of Sensor Networks”,Joel Koshy & Raju Generation Deployment Support for Sensor Networks.
Pandey University of California, Davis, California TIK-Report No: 207. Computer Engineering and
95616, USA. Networks Lab, Swiss Federal Institute of
[3] J.W. Hui, D. Culler, The dynamic behavior of a data Technology(ETH), Zurich (2004)
dissemination protocol for network programming at [16] S. K. Kasera, G. Hjálmtýsson, D. F. Towsley, and J.
scale, in: The Proceedings of the Second International F.Kurose. “Scalable reliable multicast using multiple
Conference on Embedded Networked Sensor Systems, multicast channels. IEEE/ACM Transactions on
Baltimore, MD,USA, 2004, pp. 81–94. Networking, 8(3):294–310, 2000.”
[4] T. Stathopoulos, J. Heidemann, D. Estrin, A remote [17] J.Kulik, W. R. Heinzelman, and H.Balakrishnan.”
code update mechanism for wireless sensor networks, Negotiation-based protocols for disseminating
Technical Report CENS Technical Report 30, 2003. information in wireless sensor networks. Wireless
[5] Synapse: A Network Reprogramming Protocol for Networks,8(2-3):169–185, 2002.”
Wireless Sensor Networks using Fountain Codes” [18] Crossbow Tech Inc.,” Mote In-Network Programming”
Michele Rossi, Giovanni Zanca, Luca Stabellini User Ref, http://www.tinyos.net/tinyos 1.x/doc/Xnp.pdf,
Riccardo Crepaldi, Albert F. Harris III and Michele 2003
Zorzi Dept. of Information Engineering, University of [19] Q.Wang, Y.Y. Zhu., L. Cheng, “Reprogramming
Padova, 35131 Padova, Italy. wireless sensor networks: challenges and approaches”,
[6] R.K. Panta, I. Khalil, S. Bagchi, “Stream: low IEEE Networks, 2006, 20(3): 48.
overhead wireless reprogramming for sensor networks, [20] A. Chlipala, J. Hui, and G. Tolle. Deluge: Data
in: Proceedings of the 26th IEEE International dissemination for network reprogramming at scale.
Conference on Computer Communications (INFOCOM),
May 2007, pp. 928–936.” Authors Profile
[7] M. D. Krasniewski, S. Bagchi, C-L. Yang, W. J.
Chappell, “Energy efficient, On-demand
Reprogramming of Large-scale Sensor Networks,”
Submitted to IEEE Transactions on Mobile Computing Dr V M Thakare is Professor
(TMC). Available as Purdue ECE Technical Report TR- and Head of PG department of
ECE-06-02, 2006. computer Science and Engg in
[8] S.S.Kulkarni and L. Wang, “MNP: Multihop Network SGB Amravati University
Reprogramming Service for Sensor Networks,” in IEEE Amravati, Maharastra (India) and
ICDCS, Columbus, Ohio, USA, Jun. 2005. has completed ME in Advance
[9] Efficient wireless reprogramming through reduced Electronics and Ph.D. In computer
bandwidth usage and opportunistic sleeping,” Rajesh Science/Engg. His Area of
Krishna Panta , Saurabh Bagchi , Issa M. Khalil a Research are Robotics and Artificial Intelligence,
Dependable Computing Systems Lab, School of Information Technology. He is Recognized Giude for
Electrical and Computer Engineering, Purdue University computer science and computer engineering in this
[10] N. Reijers, K. Langendoen, “Efficient code distribution University and In other universities also. He has also
in wireless sensor networks, in:” Proceedings of the received received national level Award for excellent paper
Second ACM International Conference on Wireless award. More than 10 candidates are working for Ph D
Sensor Networks and Applications (WSNA), 2003, pp. Under his supervision. He has Published and presented
60–67. more than 115 papers at National and international level.
[11] J. Koshy, R. Pandey, “Remote incremental linking for He has worked on various national level bodies like
energy efficient reprogramming of sensor networks” in: AICTE/UGC and also worked on various bodies of other
Proceedings of the Second European Workshop on universities. He is presently member of BOS, RRC, BUTR
Wireless Sensor Networks (EWSN), 2005, pp. 354–365 of this university and also chairman and Member of various
[12] “Updating Software in Wireless Sensor Networks: A committees of this university .
Survey” S. Brown, Dept. of Computer Science, National
University of Ireland, Maynooth C.J. Sreenan, Mobile & Dr. Rajiv Dharaskar is presently
Internet Systems Laboratory, Dept. of Computer Science, working as Professor at PG
University College Cork, Ireland Technical Report Department of Computer Science and
UCCCS- 2006-13-07 Engineering, GH Raisoni College of
[13] Shen,, Srisathapornphat, Jaikaeo: “Sensor Information Engineering, Nagpur. He is Ph.D. in
Networking Architecture and Applications”. In: Proc. of Computer Science & Engineering in
the International Workshop on Pervasive Computing, the Faculty of Engineering &
Toronto, Canada, August. IEEE(2004) 52-59 Technology, M.Tech. in Computers,
[14] Stann, F., Heidemann, “RMST: Reliable Data P.G. Dip., M.Phil., and M.Sc. He is
Transport in Sensor Networks”. In: Proc. of the 1st IEEE having 24 years of teaching and 18
Intl. Workshop on Sensor Network Applications and years of R&D experience in the field of Computers & IT. He
Protocols. IEEE (2003) 102-112 is approved PhD guide for Computer Engineering and
Science for Nagpur and Amravati University and 22
research scholars are perusing Ph.D. degree under his

guidance. He is an author of number books on Programming
Languages.
Riyaz Shaikh received B.E degree in

Computer Technology from Nagpur
University.Joined as MIS INCHARGE
in Govt Project at ZP.Before that she
also worked as Lecturer in Polytechnic
and MCA college. Presently perusing
Master of Engineering degree in
Wireless communication and
Computing branch under Computer Science department at
GHRCE , Nagpur university , India . Her area of interest
are wireless adhoc and sensor network.
Performance analysis of Dynamic load Balancing

Techniques for Parallel and Distributed Systems
Z.Khan1, R.Singh2, J. Alam3 and R.Kumar3
1
Department of Computer Science and Engineering, IIET, Bareilly India
zubair.762001@gmail.com
2
Department of Computer Science & IT, IET, R.U. Bareilly India
r.singh@mjpru.ac.in
3
University Women’s Polytechnic, Faculty of Engineering & Technology, AMU Aligarh India
jahangir.uk786@yahoo.co.uk
SMCA, Thapar University- Patiala India

rakumar@thapar.edu
process with inter process communication (IPC). The load

Abstract: The DAANNS (Diffusion algorithm asynchronous
nearest neighborhood) proposed in this paper is an allocation and then balancing it among the processor is an
asynchronous algorithm which finds the unbalanced nodes in a essential phase in parallel and distributed systems.
network of processors automatically, and makes them balanced This paper addresses the problem of load balancing in
in a way that load difference among the processors is 0.5 units. parallel and distributed systems. Here each processor has its
In this paper the DAANNS algorithm has been evaluated by own identity and processor pass messages through inter
comparing it with RID (Receiver Initiated Diffusion) algorithm
process communication between the tasks they are running.
across a range of network topologies including ring, hypercube,
and torus where the no of nodes have been varied from 8 to 128. A peer to peer interconnection network is used for
All the experiments were performed on Intel parallel compiler. communication. Many commercial parallel computers fall
After simulation we have noticed that the DAANNS performed under this category including Intel Paragon, The Thinking
very well as compared to RID. Machine, CM-5, IBMSP2, Origin 2000, and CrayT3D/T3E.
In these systems to achieve load balancing a processor relies
Keywords: RID (receiver initial diffusion), DAANNS, Load on nearest neighborhood methods in which the processor in
balancing, IPC.
question tries to transfer (accept) load (tasks) to (from) all of
its neighbors simultaneously. The strategies proposed in
1. Introduction [2],[3],[4],[5],[6] assume that the load of a processor is a
real number. Such assumption may be successful for parallel
The load balancing problem in parallel and distributed
systems deals with how to distribute the work load programs which enjoy large grain parallelism. Under the
(computational tasks) among the available processors so more realistic cases when parallel programs have medium
or small grain parallelism these strategies may fail.
that each processor has equal or nearly equal load. Load
This paper proposes an algorithm called Diffusion
balancing is of two types - static load balancing and
Algorithm Asynchronous Nearest Neighborhood Strategy
dynamic load balancing. In static load balancing the load
(DAANNS) which is capable of handling the indivisible
balancing is done prior to execution and is done only once.
Static load balancing is also called static mapping and it is tasks and therefore targeted for the parallel programs which
exploit medium or small grain parallelism.
quite effective for those computations which are of
The rest of the paper is organized as fallows. Section 2, 3
predictive runtime behavior. For this type of load balancing
and 4 describe the DAANNS (diffusion algorithm
the computational time is non deterministic i.e. the load
asynchronous nearest neighborhood strategy) and the
balancing problem is NP Complete. Performing load
balancing in the beginning of execution is not sufficient for methodology used. In section 5 simulation results have
certain types of problems like parallel data applications. In been presented and section 6 concludes the paper.
such cases load balancing is performed again and again or
periodically during run time when the resources required by 2. The DAANNS strategy:
a particular part of a problem are available in the system. DAANNS is a diffusion algorithm asynchronous nearest
Other examples of using this type of load balancing include Neighborhood strategy based on RID (receiver initiated
protein structure prediction, seismological data analysis, diffusion) proposed by Willebeck Le Mairet et. al. in [5].
meteorological data analysis and high quality animation RID is a highly distributed local approach which makes use
generation. of near neighbor load information in which under loaded
A task or load is a set of modules and module execute on processor request load from heavily loaded near neighbors.
one of the processing node and communicate with other The balancing process is initiated by any processor whose
load drops below a pre specified threshold (LLOW). Second Table 1: Algorithm DAANNS (Diffusion algorithm
upon receipt of a load request a processor will full fill the asynchronous nearest neighborhood)
request only up to an amount equal to half of its current load
(this reduces the effect of aging of data upon which the
request was based). Finally in the receiver initiated
approach the under loaded processor in the system take on
the majority of the load balancing over head. The RID
algorithm can be adapted to the integer work load model
using floor and ceiling functions. However this integer
approach may fail to guaranteed a global balanced situation
although the load of each processor may differ by only one
unit at most from that of its neighbor. The global load
balance may be very poor because of overhead.
The DAANNS is developed to solve this problem detecting
the unbalanced domain and it performs local exchange of
load among the processors to achieve a global balancing
state (where the maximum load difference between two
processor is 0.5 units i.e. the threshold value). The
DAANNS algorithm is explained in table 1 where load of a
processor P at a moment (m) is defined as and
represent the
global load vector at a time m.
Each processor executes the same group of operations, at
every iteration, of the load balancing. Now each processor
sends its load information to all its neighbors and receives
the load information from all its neighbors. Then first it
computes the average load in the domain.
(1)
It also computes its load value . If the processor

load is below the average load by more than a threshold
value i.e. 0.5 units or further, it proceeds to load balancing
(2)
Under load 4. Methodology:
That is a processor is under loaded processor when the In this section we have discussed an important mechanism
value is non negative ( otherwise the process used in most dynamic load balancing schemes the load
update strategy. Many dynamic load balancing strategies
is overloaded and the value will be negative value
like the one proposed here make the balancing decisions
. based on the load levels of a subset of processor in the
An under loaded processor performs load system. This subset may include anything from a single
neighbor to all processors in the system. The degree of
balancing by requesting proportionate amount of load from knowledge may vary from one strategy to another. This
over loaded neighbors. And each neighbor k is assign a quality of information governs the intelligence of the load
weight . balancing decisions. The quality of information depends on
if (3) three primary factors:
otherwise (4)
a. The accuracy of processor load estimates
These weights are summed to determine the total surplus
b. The aging of information due to communication latency
(5) of inter connection network. And the destination of load
The amount of load requested by processor from neighbor information and finally
is computed as c. The frequency of load messages updates.
(6) The first factor is application dependent and may involve a

tradeoff between the quality of the estimate and the
complexity of estimation process. The second factor is
3. Algorithm: dependent on machine architecture and the load balancing
strategy. The third factor is the interval between updated • Varying 75% from global load average for all P
messages. It may be computed as a function of time or
function of load level. The differentiation of performance
between RID and DAANNS is because of following reasons:
4.1 Implementation: In theory both approaches should • Varying 100% from global load average for all P
yield similar results, practical implementation issues
however distinguish these approaches from one another.
The RID strategy receive local update message from
neighbor. While DAANNS is a superset of RID and Here 25% variation patterns indicate that processors have
performs well. equal load. Otherwise 100% variation shows that there is
differences of load among the processors while 50% and
4.2 Stability: both schemes RID and DAANNS strategies 75% refer to intermediate situation among processors.
make load balancing decision based on load status of their Now the pathological distribution is classified as fallows
neighbors. This load information suffers from aging • All load on single processor
processes. • 25% of idle processor
• 50% of idle processor
4.3 Overhead: In minimizing total execution time it is • 75% of idle processor
beneficial to spare overloaded processor burden of load
Simulation runs till termination and we get the variance
balancing responsibilities. The extent of the overhead is
with respect to load average obtained by both strategies for
dependent on the task granularity, and may become
all distribution used in our experiment. Results of
significant if tasks are small.
experiment have shown in table 2 and table 3, and
graphically in fig 1 and fig 2 respectively.
5. Simulation/results and analysis: Table2: Variance obtain on an average by DAANNS
In this section the simulation of both algorithms is Likely Pathological
presented. We have compared RID and DAANNS algorithm distribution(Variance) distribution(Variance)
No. of Hyperc Toru Ring Hypercu Torus Ring
with respect to their stability and efficiency. The stability (or
Proc. ube s be
balance quality) measures the ability of an algorithm to
8 0.02 0.10 0.25 0.04 0.24 0.36
coerce any initial load distribution into an equilibrium state,
16 0.25 0.23 1.00 0.25 0.25 1.00
i.e., to reach the global uniform distribution state. The
32 0.25 0.24 5.21 0.23 0.26 5.21
efficiency measure is reflected by the time incurred in the 64 0.25 0.35 19.20 0.33 0.45 23.04
load communication steps and the number of balancing 128 0.25 0.41 24 0.33 0.79 37.24
steps required by the algorithm to drive an initial workload
distribution into a stable distribution. Table 2: Variance obtain on an average by RID
To check the aspects we implemented these algorithms on
Likely distribution(Variance) Pathological
the hypercube, torus and ring processor topologies. The distribution(Variance)
communication network sizes in terms of number of No. of Hypercub Torus Ring Hypercu Torus Ring
Proc. e be
processors were 8, 32, 64 and 128. The initial load 8 7.24 5.76 28.09 1.44 4.41 23.04
distribution is denoted as and the total work load is 16 26.01 32.39 31.36 22.09 19.36 161.29
32 33.64 33.64 118.8 70.56 127.69 524.41
denoted as L. So we can evaluate priori on the expected 1
64 37.21 54.75 342.2 70.12 184.96 1011.24
final load at each processor i.e. average load └L/n┘ or 5
┌L/n┐, where n is the size of topology. 128 40.96 67.24 482.4 84.64 207.36 1755.61
1
In our experiment the problem size was L=2520 and initial
load distribution were likely distribution and pathological
distribution. The parameters used in likely distribution were
as fallow.
• Varying 25% from global load average for all P
• Varying 50% from global load average for all P

Fig 5.1Vafriance of Torus and hypercube Topology by From our experiment we have noticed that the DAANNS
DAANNS performed better than RID.
References
[1] M. Willebeek-LeMair, A. P. Reeves, “Strategies for
Dynamic Load Balancing on Highly Parallel Computers”,
IEEE Transactions on Parallel and Distributed Systems,
vol. 4, No. 9, September 1993, pp. 979-993
[2] R. Subramain, I. D. Scherson, “An Analysis of
Diffusive Load-Balancing”, In Proceedings of 6th ACM
Symposium on Parallel Algorithms and Architectures,
1994
[3] V. Kumar, A. Y. Grama and N. R. Vempaty, “Scalable
load balancing techniques for parallel computers”, J. of
Fig
Par. And Distrib. Comput., 22(1), 1994, pp. 60-79.
5.2 Variance of Ring Topology by DAANNS
[4] S. H. Hosseini, B. Litow, M. Malkawi, J. McPherson,
and K. Vairavan, “Analysis of a Graph Coloring Based
Distributed Load Balancing Algorithm”, Journal of
Parallel and Distributed Computing 10, 1990, pp. 160-
166.
[5] G.C.Fox, M.A. Johnson, G.A. Lyzenga,S.W. Otto,
J.K.Salmon and D.W. Walkeer, Solving Problems on
Concurrent Processors, vol. 1, Prentice-Hall, 1998.
[6] A. Cortés, A. Ripoll, M.A.Senar , F. Cedó and E. Luque,
“On the convergence of SID and DASUD load-balancing
algorithms”, Technical Report, UAB, 1998
[7] A. Cortés, A. Ripoll, M. A. Senar and E. Luque,
“Dynamic Load Balancing Strategy for Scalable Parallel
Systems”, PARCO’97, 1997.
[8] C. Z. Xu and F. C. M. Lau, Load Balancing Parallel
Computers - Theory and Practice, Kluwer Academic
Publishers, 1997
Fig 5.3Variance of ring torus and hypercube by RID
Authors Profile
6. Analysis: As we know that stability reflects the ability
Zubair Khan received his Bachelor Degree
of an algorithm in bounding any initial load distribution
in Science and Master of Computer
into an equilibrium state i.e. to reach the global uniform Application Degree from MJP Rohilkhand
distribution state. The variance of load with respect to the University Bareilly, India in 1996 and 2001
average load has been obtained by both the strategies for all respectively and also Master in Technology
the distribution used in our experiment. As it can be seen (M.Tech) Degree in Computer Science and
DAANNS achieve very low variations for all topologies Engineering from Uttar Pardesh Technical University in the year
(always less than 0.5 for hypercube and torus) and less than 2008. He is currently pursuing his P.hd in computer science and
25 in ring. In contrast RID exhibits a high variation for all Information Technology from MJP Rohilkhand University
Bareilly, UP India. He also worked as a senior lecturer in JAZAN
cases. Now we can conclude that on one hand DAANNS
University Kingdom of Saudi Arbia . He is also servicing as
achieve a small maximum difference than RID on average Reader in the Department Of Computer Science and Engineering
on the other hand all the processor have a better final state Invertis Institute of Technology Bareilly, India. His area of
i.e. system is in its optimal state . interest include data mining and warehousing, parallel systems
and computer communication networks. He is an author/ co-
7. Conclusion: author of more than 15 international and national publication in
In this paper we have compared two algorithms namely journals and conference proceedings.
DAANNS and RID for dynamic load balancing in parallel
R. Singh received the B. Engg. Degree in
systems. This comparison has been carried out by Electronics Engineering from M.I.T. Aurangabad,
considering a large set of load distribution which includes India in 1991 and the Ph. D. in
different amount of load imbalances. These distributions Computer Science and Engineering from
have been applied on ring, hypercube and torus topologies Lucknow University/Institute of Engineering &
and the no of processor ranges from 8 to 128. The Technology Lucknow, India. He is a doctoral
experiment was conduct on INTEL parallel computer for investigator at MJP Rohilkhand University, Bareilly and U.P.
analyzing the balance degree achieved by both algorithms. Technical University, Lucknow and a visiting Associate Professor
at various Technical Collages/Universities in India. After a
number of years with the research and development wing of
different industries at various positions, he joined as a faculty for
the Department of Computer Science & Information Technology,
M0-JP Rohilkhand University, Bareilly (India) in Dec 1997 and is
currently working as a Head of the Department, CS & IT. . He had
also worked as a Professor & Director at R. B. Institute of
Engineering & Technology, Bareilly (India). His Research
interests are in the area of Routing issues in the Wired and
Wireless Network, QoS provisioning for Service Level
Agreements in IP networks, Software Architectures with
admission control schemes for Real time communication over the
Internet, Acoustic communications in Under-water Sensor
Network and issues related to Data mining Techniques.
Jahangir Alam graduated in science from Meerut

University, Meerut in the year 1992 and received
Master degree in Computer Science and
Applications from Aligarh Muslim University,
Aligarh in the year 1995 and Master in Technology
(M.Tech) Degree in Computer Engineering from
JRN Rajasthan Vidyapeeth University, Udaipur, Rajasthan, in the
year 2008. He is currently working towards his Ph.D. in Computer
Engineering from Thapar University, Patiala. He is also serving as
an Assistant Professor in University Women’s Polytechnic, Faculty
of Engineering and Technology at Aligarh Muslim University,
Aligarh. His areas of interest include Interconnection Networks,
Scheduling and Load Balancing, Computer Communication
Networks and Databases. He has authored/ co-authored over 10
publications in journals/conference proceedings.
Rajesh Kumar received his bachelor degree in

science from Gurukul Kangri University, Haridwar,
Utrakhand India and his Master of Science in
Mathematics, Master of Philosophy in Computer
Science and Ph.D degrees from IIT Roorkee in the
year 1988,1990 and 1995 respectively. He has also
done several skill oriented practical courses in computer science
from different Institutes/ Universities. Presently he is serving as an
Associate .Professor in School of Mathematics and Computer
Application at Thapar University Patiala, Punjab India. He is also
discharging the duties of Head Computer Centre at
ThaparUniversity,Patiala. He has published several papers in
National and International Journals/Conferences and has guided a
large no of students on their MCA projects. His areas of interest
include fracture mechanics, image processing, S/W Engineering
and parallel systems. A number of Ph.D theses are being guided by
him at Thapar University and at some other Indian universities.
Effect of error concealment methods on UMTS

radio network capacity and coverage planning
Dr. Bhumin H. Pathak1, Dr. Geoff Childs2 and Dr. Maaruf Ali3
1
Airvana Inc., Chelmsford, USA, bhumin.pathak@gmail.com
2
School of Technology at Oxford Brooks University, Oxford, UK, gnchilds@brookes.ac.uk
3
School of Technology at Oxford Brooks University, Oxford, UK, mali@brookes.ac.uk
In this case as an alternative to FEC, an optimized

Abstract: The radio spectrum is a precious resource and
careful utilization of it requires optimization of all the processes automatic repeat request (ARQ) scheme with selective
involved in data delivery. It is understood that achieving repeat and dynamic frame referencing provides better
maximum coverage and capacity with the required QoS is of the performance.
utmost importance for network operators to maximize revenue
generation. Whilst current methods of video compression Correcting or compensating for errors in a compressed
accelerate transmission by reducing the number of bits to be video stream is complicated by the fact that bit-energy
transmitted over the network, they have the unfortunate trade-
concentration in the compressed video is not uniformly
off of increasing signal sensitivity to transmission errors. In this
paper we present an approach to transmit MPEG-4 coded video distributed, especially with most motion-compensated
in a way that optimizes the radio resources while maintaining interframe predictive codecs like MPEG-4. The encoded bit-
the required received video quality. This approach thereby stream has a high degree of hierarchical structure and
provides increased capacity for network operators. Two different dependencies which impact on error correction techniques.
methods used for this purpose are selective retransmission of The problem is compounded by the fact that it is also not
erroneous parts in frames and dynamic changes in the reference feasible to apply a generalized error-resilience scheme to the
frames for predictive coding. Network operators can still provide
the required video quality with implementation of these two
video stream as this may impact on standardized parameters
methods despite the low signal-to-interference ratios and can in the different layers of the UMTS protocol stack. For
therefore increase the number of users in the network. We show example when overhead or redundancy is added to the
here with the help of simulation results the performance existing video standard for FEC implementation, the
enhancements these methods can provide, for various channel modified video stream can become incompatible with the
propagation environments. Comparison is performed using a standard and the subsequent data format may not be
standard PSNR metric and also using the VQMp metric for
decoded by all standard video decoders. It is therefore
subjective evaluation.
important that any overhead or redundancy be added in a
Keywords: MPEG-4, UMTS, NEWPRED, UEP, network way which does not make the modified video bit-stream
coverage and optimization. incompatible with the standard.
With reference to the above mentioned constraints it is vital

1. Introduction to design an error-protection mechanism that exploits the
hierarchical structure of the inter-frame video compression
Future wireless mobile communication is expected to algorithm, is compatible with the standardized wireless
provide a wide range of services including real-time video communication system, and preserves the integration of the
with acceptable quality. Unlike wired networks, wireless standard bitstream..
networks struggle to provide high bandwidth for such
services which therefore require highly compressed video In this paper we present one such scheme which uses
transmission. With high compression comes high sensitivity unequal error protection (UEP) in conjunction with ARQ to
to transmission errors. Highly compressed video bitstreams exploit different modes at the radio link layer (RLC) Error!
like MPEG-4 Error! Reference source not found. can lose Reference source not found. of the UMTS architecture.
a considerable amount of information with the introduction This scheme is coupled with a dynamic frame referencing
of just a few errors. In this context, it is important to have a (NEWPRED) Error! Reference source not found. scheme
video codec with a repertoire of efficient error-resilience which exploits frame dependencies and interframe intervals.
tools. Transmission errors in a mobile environment can vary The techniques discussed in this paper are relevant and
from single bit errors to large burst errors or even applicable to a wide variety of interframe video coding
intermittent loss of the connection. The widely varying schemes. As an illustrative example, the video compression
nature of the wireless channel limits the use of classical standard, MPEG-4, is used throughout.
forward error correction (FEC) methods which may require The rest of the paper is organized into the following
a large amount of redundant data to overcome bursty errors. sections. Section-2 gives an introduction to the general
principles of the predictive interframe MPEG-4 codec, the same pixel in the previous frame. However, when there is a
frame hierarchies and dependencies involved. Section-3 motion, assuming the movement in the picture is only a
discusses the general UMTS network and protocol shift of object position, then a pixel in the previous frame,
architecture. Section-4 classifies transmission errors into displaced by a motion vector, is used.
different classes and introduces the UEP and NEWPRED Assigning a motion vector to each pixel is very costly.
protection schemes.. Section-5 details the UEP scheme Instead, a group of pixels are motion compensated, such
while Section-6 details NEWPRED. In Section-7 the that the motion vector overhead per pixel can be very small.
simulation scenarios and results are presented. All the In a standard codec a block of 16 × 16 pixels, known as a
Macroblock (MB), are motion estimated and compensated.
achieved results are then discussed from the network
It should be noted that motion estimation is only carried out
operators’ point of view in Section-8. Finally the conclusion
on the luminance parts of the pictures. A scaled version of
is drawn at the end in Section-9. the same motion vector is used for compensation of
chrominance blocks, depending on the picture format. Every
MB is either interframe or intraframe coded. The decision
2. Motion-compensated predictive coding on the type of MB depends on the coding technique. Every
MB is divided into 8 × 8 luminance and chrominance pixel
blocks. Each block is then transformed via the DCT –
2.1 General Principles Discrete Cosine Transform. There are four luminance
Image and video data compression refers to a process in blocks in each MB, but the number of chrominance blocks
which the amount of data used to represent images and depends on the color resolution. Then after quantization,
video is reduced to meet a bit rate requirement, while the variable length coding (entropy coding) is applied before the
quality of the reconstructed image or video satisfies a actual channel transmission.
requirement for a certain application. This needs to be
undertaken while ensuring the complexity of computation
involved is affordable for the application and end-devices. 2.2 Frame types
The statistical analysis of video signals indicates that there An entire video frame sequence is divided into a Group of
is a strong correlation both between successive picture Pictures (GOP) to assist random access into the frame
frames and within the picture elements themselves. sequence and to add better error-resilience. The first coded
Theoretically decorrelation of these signals can lead to frame in the group is an I-frame. It is followed by an
bandwidth compression without significantly affecting arrangement for P and B frames. The GOP length is
image or video resolution. Moreover, the insensitivity of the normally defined as the distance between two consecutive I-
human visual system to loss of certain spatio-temporal frames as shown in Figure-2. The I-frame (Intra-coded)
visual information can be exploited for further bitrate frame is coded using information only from itself. A
reduction. Hence, subjectively lossy compression techniques Predictive-coded (P) frame is coded using motion
can be used to reduce video bitrates while maintaining an compensated prediction from a past reference frame(s).
acceptable video quality. While a Bidirectionally predictive-coded (B) frame is a
frame which is coded using motion and texture compensated
Figure 1 shows the general block diagram of a generic prediction from a past and future reference frames.
interframe video codec Error! Reference source not
found..
Figure 2. Frame structures in a generic interframe video

Figure 1. Generic interframe video codec codec.
In interframe predictive coding the difference between It is important to note here that the frame interval between
pixels in the current frame and their prediction values from two consecutive I-frames and between two consecutive P-
the previous frame are coded and transmitted. At the frames has a significant effect on the received video quality
receiving end after decoding the error signal of each pixel, it as well as on the transmission bitrate. It is actually a trade-
is added to a similar prediction value to reconstruct the off between error-resilience capabilities and the required
picture. The better the predictor, the smaller the error operational bitrate.
signal, and hence the transmission bit rate. If the video A major disadvantage of this coding scheme is that
scene is static, a good prediction for the current pixel is the transmission errors occurring in a frame which is used as a
reference frame for other P or B frames causes errors to The radio interface protocols are needed to set-up,
propagate into the following video sequence. This reconfigure and release the RAB. The radio interface is
propagation continues until an intra-refresh is applied. divided into three protocol layers:
In the presented work this hierarchy of frame types is Physical layer (L1)
exploited. It should be obvious from the above discussion Data Link layer (L2)
that the I-frame which acts as a reference frame for the Network layer (L3)
entire GOP needs to be protected the most from
transmission errors with perhaps weaker protection for P- The data link layer is split further into the Medium Access
frames. Since no other frame depends on B-frames, errors Control (MAC) Error! Reference source not found.,
occurring in B-frames affect just a single frame and do not Radio Link Control (RLC) Error! Reference source not
propagate into the video sequence. found., Packet Data Convergence Protocol (PDCP) Error!
Reference source not found. and Broadcast/Multicast
As an illustrative example, the video compression standard, Control (BMC) Error! Reference source not found.. Layer
MPEG-4, is used throughout. The rest of the paper is 3 and RLC are divided into Control (C) and User (U) planes
organized into the following sections. Section-2 gives as shown in Figure-3.
introduction of general principles of predictive interframe
C-plane U-plane
MPEG-4 codec, frame hierarchies and dependencies signalling information
involved. Section-3 discusses the general UMTS network
L3
and protocol architecture. Section-4 classifies transmission RRC control
errors into different classes and introduces the UEP and Radio
NEWPRED protection schemes. Section-5 introduces UEP Bearers
control
while Section-6 introduces NEWPRED. In Section-7 error PDCP L2/PDCP
PDCP
detection method is described. Section-8 discusses video
quality measurement techniques used in the paper. Section- control
9 provides information on header compression protocol used BMC L2/BMC

control
control
in UMTS architecture. Section-10 introduces various radio

propagation environments used in simulations. In section-
11 simulation scenarios and results are presented. All the RLC
RLC
RLC
RLC
L2/RLC
achieved results are then discussed from the network RLC
RLC
RLC
RLC
operators’ point of view in Section-12. Finally the Logical

conclusion is drawn at the end in Section-13. channels
MAC L2/MAC
Transport
channels
PHY
3. UMTS network and protocol architecture L1
Figure 3. Radio interface protocol structure for UMTS

3.1 Brief introduction
The service access point (SAP) between the MAC and
UMTS is a very complex communication system with a physical layers are provided by the transport channels
wide range of protocols working at different layers of its (TrcCHs). While the SAP between RLC and MAC sub-
network architecture Error! Reference source not found.. layers are provided by the logical channels (LcCHs). We
UMTS has been designed to support a wide range of will explain the RLC in further details as the services
applications with different Quality of Service (QoS) profiles provided by different RLC modes are exploited in order to
Error! Reference source not found.. It is essential to provide different levels of UEP for video transmission.
understand the overall architecture of the system before we
consider details of transmission quality for a given
application. 3.2 RLC modes
UMTS can be briefly divided into two major functional
Different services provided by RLC sub-layers includes,
groupings: Access Stratum (AS) and Non-Access Stratum
Transparent Mode (TM) data transfer, Unacknowledged
(NAS). The AS is the functional grouping of protocols
Mode (UM) data transfer, Acknowledged Mode (AM) data
specific to the access techniques while NAS aims at
transfer, maintenance of QoS as defined by upper layers and
different aspects for different types of network connections.
notification of unrecoverable error Error! Reference
The Radio Access Bearer (RAB) is a service provided by the
source not found.. TM data transfer transmits upper layer
AS to the NAS in order to transfer data between the user
PDUs without adding any protocol information, possibly
equipment (UE) and core network (CN). It uses different
including segmentation/reassembly functionality. It ignores
radio interface protocols at the Uu interface, those are
any errors in received PDUs and just passes them onto the
layered into three major parts as shown in Figure 3.
upper layer for further processing. UM data transfer
transmits upper layer PDUs without guaranteeing delivery
to the peer entity. PDUs which are received with errors are
discarded without any notice to the upper layer or to the propagation is an indispensable component of any error
peer entity. AM data transfer transmits upper layer PDUs robust video communication system. Errors introduced
with guaranteed delivery to the peer entity. Error-free during transmission can lead to frame mismatch between
delivery is ensured by means of retransmission. For this the encoder and the decoder, which can persist until the
service, both in-sequence and out-of-sequence delivery are next intra refresh occurs. Where an upstream data channel
supported. exists from the decoder to the encoder, NEWPRED or
demand intra refresh can be used. NEWPRED is a
As mentioned before, in the presented scheme unequal error technique in which the reference frame for interframe
protection is applied to the different types of video frames coding is replaced adaptively according to the upstream
transmitted. The I-frame which has the most crucial data is messaging from the decoder. NEWPRED uses upstream
always transmitted using the AM data transfer mode of messages to indicate which segments are erroneously
service to ensure error-free delivery. decoded. On receipt of this upstream message the encoder
will subsequently use only the correctly decoded part of the
prediction in an inter-frame coding scheme. This prevents
4. Error classification temporal error propagation without the insertion of intra
coded MBs (Macro Blocks) and improves the video quality
From the discussion about the interframe predictive coding in noisy multipath environments.
hierarchical structure of MPEG-4 it is clear that various
parts of the bit stream have different levels of importance. The following section explains the concept in more detail.
We have classified errors into three main classes based on
their impact on the coded hierarchy:
• Most critical errors – Class-A errors

• Less critical errors – Class-B errors
• Least critical errors – Class-C errors
Class-A errors contain errors which can significantly

degrade the received video quality. This class includes
errors introduced in the video sequence header, I-frame
headers and I-frame data. Class-B, includes errors in P-
frame headers and P-frame data while Class-C includes
errors in B-frame headers and B-frame data.
In the following sections we discuss two different methods
to deal with Class-A and Class-B errors separately as
described
4.1 Unequal Error Protection (UEP) for Class-A errors Figure 4. Error propagation due to interframe decoding
dependencies
As mentioned before, the I-frame has relatively more
importance than the P and B frames. Hence an I-frame data When a raw video sequence is encoded utilizing MPEG-4,
needs to be transmitted with higher protection using the each of the raw video frames is categorized according to the
proposed UEP scheme. This is achieved by using a different way in which predictive encoding references are used. An
mode of transmission on the RLC layer of UMTS. The I- Intra-coded (I) frame is coded using information only from
frames are transmitted using AM while other frames are itself. A Predictive-coded (P) frame is coded using motion
transmitted using TM. If any error is introduced during compensated prediction from a past reference frame(s).
radio transmission, the specific PDU of the I-frame is While a Bidirectional predictive-coded (B) frame is a frame
retransmitted using AM. This guaranties error-free delivery which is coded using motion and texture compensated
of the I-frames. prediction from a past and future reference frames. A
disadvantage of this coding scheme is that transmission
errors occurring in a frame which is used as a reference
4.2 NEWPRED for Class-B errors frame for other P or B frames, causes errors to propagate
into the video sequence. This propagation continues until an
The MPEG-4 ISO/IEC 14496 (Part-2) Error! Reference intra-refresh is applied. In the example shown in Figure-4,
source not found. standard provides error robustness and an error occurred in frame P3 which acts as a reference
resilience capabilities to allow accessing of image or video frame for P4, subsequent P-frames and B-frames (B5, B6
information over a wide range of storage and transmission etc), until the next intra-refresh frame (I2) occurs.
media. The error resilience tools developed for this part of
ISO/IEC 14496 can be divided into three major categories: Where the transmission error has damaged crucial parts of
synchronization, data recovery and error concealment. The the bit-streams such as a frame header, the decoder may be
NEWPRED feature falls into the category of error unable to decode the frame which it then drops. If this
concealment procedures. Recovery from temporal error dropped frame is a P-frame, none of the frames that are
subsequently coded with reference to this dropped P-frame

can be decoded. So in effect all subsequent frames until the
next intra-refresh are dropped. This situation can seriously
degrade the received video quality leading to the occurrence
of often long sequences of static or frozen frames.
If through the use of an upstream message the encoder is

made aware of errors in the particular P-frame (P3), the Figure 5. CRC insertion
encoder can change the reference frame for the next P-frame
(P4) to the previous one which was received correctly (P2). This method adds an extra 16 bits of overhead to each frame
P-frames and B-frames after P4 then refer to the correctly but the performance improvements in video quality with
decoded P4, rather than the faulty P3 frame. The technique NEWPRED implementation coupled with CRC error
therefore reduces error propagation and frame loss detection, justifies this overhead.
occurring from dropped P-frames. The discussion in the previous sections described the
proposed methods for error-protection for different parts of
This method can significantly improve the performance of the video stream. The following section describes the
the received video quality. simulation scenarios used to test the implementation of
To implement the NEWPRED feature, both the encoder and these methods.
decoder need buffer memories for the reference frames. The
required buffer memory depends on the strategy of the The entire end-to-end transmission and reception system
reference frame selection by the encoder, and transmission scenario is developed using different simulation tools
delay between the encoder and decoder. In this paper results ranging from simple C programming to using SPW 4.2 by
for two different schemes are reported. CoWare and OPNET 10.5A.
5. Error detection 6. Video clips and video quality measurement

To implement NEWPRED it is important to identify errors techniques used
at the frame level at the decoding end. Mapping of errors
For evaluation purposes three standard video test sequences
identified by the lower layer to the application layer with the
were used, these being: Mother-Daughter, Highway and
precision of a single video frame or video packet often
Foreman. Each of these clips is of 650 frames in length of
results in a complicated process consuming a considerable
QCIF (176 × 144) resolution and is encoded with the
amount of processing resource and introduces severe
standard MPEG-4 codec at 10 fps.
processing delays. Insertion of CRC bits in the standard
MPEG-4 bit-stream at frame level provides a simpler
The objective video quality is measured by the PSNR (Peak
solution to this problem. With the insertion of extra bits
Signal to Noise Ratio) as defined by ANSI T1.801.03-1996
which are not defined as part of the standard video encoded
Error! Reference source not found.. One more
sequence, would normally result in the production of an
sophisticated model which is developed and tested by ITS
incompatible bit-stream which standard decoders will not be
(Institute for Telecommunication Sciences) is the Video
able to decode. But as mentioned in Error! Reference
Quality Metric (VQM) Error! Reference source not
source not found. this would not be the case if these bits are
found.. VQM has been extensively tested on subjective data
inserted at a particular place of the standard MPEG-4 bit-
sets and has significantly proven correlation with subjective
stream Error! Reference source not found..
assessment. One of the many models developed as a part of
While decoding, the decoder is aware of the total number of
this utility is Peak-Signal-to-Noise-Ratio VQM (VQMP).
macroblocks (MB) in each frame. It starts searching for a
This model is optimized for use on low-bit rate channels
new video frame header after decoding these macroblocks.
and is also used in this paper for the near subjective analysis
It ignores everything between the last marcoblock of the
of the received video sequence.
frame and the next frame header as padding. If generated
CRC bits are inserted at this place as shown in Figure-5,
Typical VQMP is given by:
after the last macroblock and before the next header, this
should preserve the compatibility of the bit-stream with 1
standard MPEG-4. Such insertion of CRC bits does not VQMp = 0.1701×( PSNR − 25.6675 )
affect the normal operation of any standard MPEG-4
1+ e
decoder. Also because the inserted CRC only consists of the
16 bits, generated using the polynomial G16 defined for the
MAC layer of the UMTS architecture, it is not possible for it
The higher the PSNR value the better is the objective quality
to emulate any start code sequences.
whilst the lower the VQMP value the better is the subjective
quality of the video sequence.
7. Upper layer protocol overheads and PDCP Once these PDUs are mapped onto different transport
header compression channels, they are transmitted using the WCDMA air
interface. The physical layer of WCDMA is simulated using
Each lower layer defined in the UMTS protocol stack the SPW tool by CoWare which models an environment to
provides services to the upper layer at defined Service generate error patterns for various types of channel
Access Points (SAPs) Error! Reference source not found.. propagation conditions defined by the 3GPP standards. A 64
These protocols add a header part to the video frame kbps downlink data channel and 2.5 kbps control channel
payload to exchange information with peer entities. were used for this UMTS simulation. These two channels
Depending upon protocol configurations and the size of the were multiplexed and transmitted over the WCDMA air-
video frames, these headers can be attached to each video interface.
frame or multiple video frames can be used as a single
payload as defined by RFC-3016 Error! Reference source The transmission time interval, transmission block size,
not found.. As many of the successive headers contain a transmission block set size, CRC attachment, channel
huge amount of redundant data, header compression is coding, rate matching and inter-leaving parameters were
applied in the form of the Packet Data Convergence configured for both channels compliant with the 3GPP TS
Protocol (PDCP) Error! Reference source not found.. 34.108 specification Error! Reference source not found..
With PDCP compression, higher layer protocol headers like The typical parameter set for reference RABs (Radio Access
RTP, UDP and IP headers are compressed into one single Barriers) and SABs (Signaling Access Barriers) and
PDCP header. In the presented simulation, header relevant combinations of them are presented in this
attachments, compression and header removal was achieved standard.
using the C programming language. Figure-6 shows a
typical structure of the video frame payload and header The different channel propagation conditions used in the
before it is submitted to the RLC layer for further simulation were static, multi-path fading, moving and birth-
processing. death propagation. These channel conditions are described
in some details in the following section.
8. Propagation conditions
Four different standardized propagation conditions – static,
multi-path fading, moving and birth-death were used to
generate different error patterns. The typical parameter sets
Figure 6. PDCP header compression for conformance testing as mentioned in 3GPP TS 25.101
Error! Reference source not found. is used for the radio
Once the PDCP compression is achieved this packet is then interface configuration. A common set of parameters for all
submitted to the RLC layer for further processing. The RLC kinds of environment is listed in Table 1, while any specific
layer can be configured in any one of the transparent, parameters to the environment are mentioned in the
unacknowledged or acknowledged modes of transmission. respective sections.
The RLC then submits the PDU to the lower layers, where Table 3. Common set of parameters
the MAC layer and physical layer procedures are applied as Interference -60 dB
appropriate.
Received signal / Noise (SNR) -3.0 dB
In the presented simulation, each video frame coming from
the application layer is mapped to the RTP layer. Each
frame is mapped into a separate RTP packet after addition AWGN noise 4 *10-9 watts
of the above mentioned RTP, UDP, IP headers with PDCP
compression, and if required, RLC headers. Eb/No (Overall) 6.01 dB
For TM RLC service, the RLC SDU cannot be larger than
the RLC PDU size. For this reason, if the video frame size is
BER (Bit Error Rate) 0.001
larger than the RLC payload size for TM, then the frame is
fragmented into different RTP packets. Other protocol
headers are then added to these packets separately. Each Data Rate (Downlink) 64 kbps
RLC SDU is then either of the same size as the RLC PDU
or smaller.
For AM RLC service, an entire frame can be considered as 8.1 Static propagation condition
one RLC SDU regardless of the size. For this reason
protocol headers are added to each video frame. This RLC As defined in 3GPP TS 25.101 V6.3.0, the propagation
SDU is then fragmented into different RLC PDUs at the condition for a static performance measurement is an
RLC layer. Additive White Gaussian Noise (AWGN) environment. No
fading and multi-paths exist for this propagation model.
Table 2, lists the received values of BLER and FER for this Generated error patterns are applied to the data transmitted
propagation condition. from the RLC layer. Different RLC modes are simulated
using the C programming language.
Table 4. Parameters for static propagation conditions
BLER (Block Error Rate) 0.034
9. Simulation Results
FER (Frame Error Rate) 0.0923 As mentioned before VQMP is used as the quality
measurement metric in this simulation. Table 6 presents the
VQMP values obtained during the simulations.
8.2 Multi-path fading propagation conditions The following conventions are used in Table 6.
Multi-path fading normally follows a Rayleigh fading Video clips names:

pattern. In this simulation Case-2 as mentioned by TS - Mother and Daughter – MD
25.101 is used with frequency band-1 (2112.5 MHz) and the - Highway – HW
number of paths set to 3 with relative delay between each - Foreman – FM
paths of 0, 976 and 20000 ns and a mean power of 0 dB for
all three paths. The delay model used in this case is fixed. VQMP without UEP and without NEWPRED – Results A
The vehicle speed is configured to be 3 km/h. The received VQMP with UEP and without NEWPRED – Results B
values of BLER and FER are given in Table 3. VQMP without UEP and with NEWPRED – Results C
VQMP with UEP and with NEWPRED – Results D
Note the lower the VQMP value the better the subjective
Table 5. Parameters for multi-path fading propagation
image quality
conditions
Table 8. Simulation results
FER (Frame Error Rate) 0.0225 Video Results Results Results Results
Clip A B C D
8.3 Moving propagation conditions Static Environment

MD 0.70 0.46 0.45 0.25
The dynamic propagation conditions for this environment HW 0.64 0.36 0.32 0.19
for the test of the baseband performance is a non-fading FM 0.85 0.83 0.74 0.68
channel model with two taps as described by 3GPP TS Multi-path Environment
25.101. One of the taps, Path-0 is static, and other, Path-,1is MD 0.27 0.18 0.18 0.16
moving,. Both taps have equal strengths and phases but
HW 0.37 0.21 0.18 0.13
unequal time difference exists between them.
FM 0.62 0.53 0.47 0.34
The received values of BLER and FER are given in Table 4. Moving Environment
MD 0.63 0.45 0.44 0.23
Table 6. Parameters for moving propagation conditions HW 0.55 0.35 0.31 0.26
FM 0.83 0.70 0.67 0.52
Birth-Death Environment
FER (Frame Error Rate) 0.088 MD 0.58 0.32 0.35 0.27
HW 0.49 0.40 0.31 0.22
FM 0.85 0.72 0.68 0.45
8.4 Birth-Death propagation conditions
The following observations can be made on the above
These conditions are similar to the Moving propagation results. The received VQMP values for all video sequences
condition except, in this case both taps are moving. The are improved by implementation of the UEP and
positions of paths appear randomly and are selected with an NEPWRED methods. It can be observed that in most cases
equal probability rate. Table 5 lists the received values of the VQMP values show greater improvement with
BLER and FER for this propagation condition. implementation of the NEWPRED method than the
improvements obtained with the UEP method. Errors in the
Table 7. Parameters for birth-death propagation conditions wireless environment occur in bursts and are random in
BLER (Block Error Rate) 0.037 nature. Due to the time varying nature of the wireless
environment it is not possible to predict the exact location of
the error. The NEWPRED method provides protection to
FER (Frame Error Rate) 0.0851
the P-frames while UEP is aimed at protecting the I-frames.
As P-frames are much more frequent than I-frames in the
encoded video sequence, P-frames are more susceptible to
these bursty randomly distributed errors. This explains why
the NEWPRED implementation gives better performance

than the UEP implementation. Combined implementation of Now in second run, the Eb/No value is decreased by 1 dB to
the UEP and NEWPRED methods always outperforms a 3.6 dB and the total throughput per cell is presented in
single method implementation and gives the best quality of graphical form in Error! Reference source not found..
received video. The same observation is made in terms of
performance assessment using the VQMP score in most
cases. Objective assessment of VQMP scores is highly
dependent on bit-by-bit comparison which does not take into
consideration the number of frozen frames. In most cases
frozen frames give better correlation with original frames
compared to frames with propagation errors.
10. Network capacity and coverage

improvements
As shown by the simulation results above, implementation
of UEP and NEWPRED enhances the received video
quality. This implies that the same video quality or QoS can
be attained at lower Eb/No if the two proposed methods are
Figure 8. Throughput in downlink per cell at Eb/N0 – 3.6
utilized. A reduced Eb/No value have a direct impact on
dB
radio network coverage and capacity planning for UMTS
networks Error! Reference source not found., Error!
As can be clearly compared from Error! Reference source
Reference source not found.. A number of variable
not found. to Error! Reference source not found., the
parameters determine the resultant received video quality.
decrease in the Eb/No value by 1 dB results in a significant
The nature of the video clip, the encoding parameters,
increase in the total throughput per cell in the downlink
channel propagation environment and many more variables
direction.
can have varying effects on the overall video quality. For
these reasons it is not easy to quantify exactly how much
reduction in Eb/No value would be achieved using the 11. Conclusions
techniques described above. For the simulation purpose, a As can be seen from the simulation results, implementation
reduction of 1 dB is assumed. Network capacity and of UEP and NEWPRED results in significant improvements
coverage enhancements are quantified assuming 1 dB on the received video quality. Improvements achieved using
reduction in the required Eb/No value for video service. these methods provides extra margin for network operators
to increase capacity. With implementation of these error
The basic radio network used for this simulation is shown in concealment methods the same video quality can be
following Figure. 10 Node-B sites each with three sectors achieved using a lower Eb/No which in turn provides
are simulated. 2000 UEs are randomly distributed over an flexibility for network operator to increase the number of
area of 6 × 6 km2. The UMTS pedestrian environment for users. This implementation obviously requires some
the micro cell is selected as a propagation model. The processing overhead on both the encoder and decoder sides,
mobiles are assumed to be moving at a speed of 3 km/h. but considering the increasing processing power of mobile
Downlink bit-rate of 64 kbps is selected and other link layer stations, this should present a major obstacle and the error
parameters and results are imported from simulations concealment methods described should provide considerable
discussed above. In the first run, an Eb/No value of 4.6 dB enhancements.
is used and the total throughput per cell is represented in
graphical form as shown in Error! Reference source not
References:
found..
[1] International Standard ISO/IEC 14496-2: Information
Technology - Coding of audio-visual objects-Part 2,
Visual, International Organization for Standardization.
2001.
[2] 3GPP, Technical Specification Group Radio Access
Network; Radio Link Control (RLC) protocol
specification; 3GPP TS 25.322, V4.12.0.
[3] Pereira, F., Ebrahimi, T., The MPEG-4 Book; Prentice
Hall PTR (July 20, 2002), ISBN-10: 0130616214.
Network;Radio interface protocol architecture; 3GPP
TS 25.301 (2002-09), Ver 5.2.0.
Figure 7. Throughput in downlink per cell at Eb/N0 – 4.6
dB
[5] 3GPP, Technical Specification Group Services and

System Aspects; Quality of Service (QoS) concept and
architecture, 3GPP TS 23.107 v4.0.0.
Network; MAC protocol Specification; 3GPP TS
25.321, V4.0.0.
Network; PDCP protocol specification; 3GPP TS
25.331 (2003-12), Ver 6.0.0.
[8] 3GPP, Technical Specification Group, Radio Access
Network; Broadcast/Multicast Control BMC; 3GPP TS
25.324, v4.0.0.
[9] Worrall, S., Sadka, A., Sweeney, P., Kondoz, A.,
Backward compatible user defined data insertion into
MPEG-4 bitstream. IEE Electronics letters.
[10] ATIS Technical Report T1.TR.74-2201: Objective
Video Quality Measurement using a Peak-Signal-to-
Noise Ratio (PSNR) Full Reference Technique. October
2001, Alliance for Telecommunications Industry
Solutions.
[11] Wolf, S., Pinson, M., Video Quality Measurement
Techniques. June 2002, ITS, NTIA Report 02-392.
[12] RFC-3016, RTP Payload Format for MPEG-4
Audio/Visual Streams, November 2000.
[13] 3GPP, Technical Specification Group Terminals;
Common test environment for UE conformance testing;
3GPP TS 34.108 (2003-12), Ver 4.9.0.
Network; UE radio transmission and reception (FDD);
3GPP TS 25.101 (2003-12), Ver 6.3.0.
[15] Laiho, J., Wacker, A., Novosad, T., Radio Network
Planning and Optimization for UMTS, Wiley, 2nd
edition (December 13, 2001).
[16] Holma, H., Toskala, A., WCDMA for UMTS: Radio
Access for Third Generation Mobile Communications,
Wiley, 3rd edition (June 21, 2000).
Dr. Bhumin Pathak received his M.Sc. and Ph.D. degree from
Oxford Brookes University, Oxford, UK. He has been working as
Systems Engineer at Airvana Inc., since March 2007.
Dr. Geoff Childs is Principal Lecturer at School of Technology at
Oxford Brookes University, Oxford, UK.
Dr. Maaruf Ali is Senior Lecturer at School of Technology at
Oxford Brookes University, Oxford, UK.
Comprehensive analysis of UMTS Authentication

and Key Agreement
Engr. Mujtaba Hassan1, Engr. Munaza Razzaq2 and Engr. Asim Shahzad3
1
Kohat University of Sciences and Technology, Kust Institute of Engineering Sciences,
Department of Computer Engineering, Kohat. Pakistan
engr.mujtabahassan@gmail.com
2
NWFP University of Engineering and Technology Peshawar, Abbottabad Campus
Near Post Graduate College No. 2 University Road Mandian, District Abbottabad, Pakistan.
engr.munaza_razzaq@yahoo.com
3
University of Engineering and Technology Taxila, Pakistan
asim.shahzad@uettaxila.edu.pk
and enhanced security features that are designed to stop

Abstract: This paper presents an analysis and evaluation threats [2], [3], [4], [5], [15]. These include: Mutual
of the security of UMTS. This paper provides information on the Authentication which allows the mobile user and serving
3rd generation mobile communication system, UMTS, its
network(SN) to authenticate each other [6], Network to
Authentication and Key Agreement (AKA) procedures and
security aspects. The AKA procedure is the essence of Network security that secure communication between
authenticating a user to the network and vice versa. AKA serving networks which suggested the use of IP security to
procedures in UMTS have increased security compared with do so, wider security scope, secure International Mobile
GSM. The new feature of two-way authentication eliminates the Subscriber identity (IMSI) usage, user to mobile station
problem with false base stations. This is a very important authentication where more flexibility in that security
security improvement. Even though the security has improved in features can be extended and enhanced as required by new
some areas, there are still security features that should be
improved. Some weaknesses are also pointed out in UMTS. One
threats and services plus GSM compatibility.
of the major weaknesses in UMTS is sending IMSI in plaintext.
We have simulated this weakness pointed out in the literature 2. UMTS Security Architecture
survey. In this paper we have shown simulation scenarios for an
attack on IMSI of MS when it sends a registration request to the The security architecture in UMTS is based on three
serving network. security principles: Authentication, Confidentiality and
Keywords: UMTS, AKA, IMSI, Security Integrity
1. Introduction 2.1 Authentication

Authentication is provided to assure the claimed identity of
The Universal Mobile Telecommunications System (UMTS) an entity. A node that wants to authenticate itself to
is one of the new ‘third generation’ (3G) mobile cellular someone has to show its own identity. This can be done
communication systems being developed within the either by showing knowledge of a secret only the nodes
framework defined by the International involved knows; or by letting a third party that both nodes
Telecommunications Union (ITU) known as IMT-20001. trusts, vouch for their identities.
UMTS security builds on the success of Global System for Authentication in UMTS is divided into two parts:
Mobile communications (GSM) by providing new and
• Authentication of the user towards the network
enhanced security features. UMTS aims to provide a
• Authentication of the network towards the user
broadband, packet-based service for transmitting video, text,
digitized voice, and multimedia at data rates of up to 2 2.2 Confidentiality
Mbps while remaining cost effective. UMTS utilizes Code Confidentiality is to keep information secured from
Division Multiple Access (CDMA) as it is far better suited unwanted parties. With more and more people using the
for fast data stream transfer. Although GSM security has terminals for both personal and business calls (e.g. online
been very successful but GSM suffers from security services like banking) the need for keeping the
problems such as weak authentication and encryption communication secure grows rapidly. Confidentiality in
algorithms, short secret key length (only 32 bits) with no UMTS is achieved by ciphering communications between
network authentication. This has lead to false base station the subscriber and the network and by referring to the
attack and lack of data integrity, allowing denial of service subscriber by temporary (local) identities instead of using
attacks, limited encryption scope and insecure key the global identity, IMSI. The properties that should be
transmission. An objective of the UMTS security design was confidential are:
to address weaknesses [1] in GSM. UMTS introduces new • The identity of the subscriber
• The current location of the subscriber • Attach Request

• User data (both voice and data communications • Detach request
should be kept confidential). • Connection re-establishment request
Registration of a subscriber in a SN typically occurs when
2.3 Integrity
the user goes to another country. The first time the
Sometimes a message’s origin or contents have to be subscriber then connects to the SN, he gets registered in the
verified. Even though it might come from a previously SN. Service Request is the possibility for higher-level
authenticated party, the message may have been tampered protocols/applications to ask for AKA to be performed. E.g.
with. To avoid this, integrity protection is necessary. The performing AKA to increase security before an online
message itself might not even have to be confidential; the banking transaction. The terminal updates the Home
important thing is that it’s genuine. The method for Location Register (HLR) regularly with its position in
integrity protection in UMTS is to generate stamps to be Location Update Requests. Attach request and detach
added to messages. The stamps can only be generated at the request are procedures to connect and disconnect the
nodes that know the keys derivate of the pre-shared secret subscriber to the network. Connection re-establishment
key, K. They are stored in the Universal Subscriber Identity request is performed when the maximum number of local
module (USIM) and the Authentication Centre (AuC). It is authentications has been conducted. In the following an
very important to offer integrity protection, especially since overview of how the UMTS AKA protocol works is given:
the SN often is operated by another operator than the
subscriber’s own operator. The property that should be 3.2 Procedures
integrity protected is: Signaling messages and signaling Authentication and key agreement (Fig. 2) [16] consists of
data. two procedures: First, the Home Environment (HE)
distributes authentication information to the SN. Second, an
3. Authentication and Key Agreement authentication exchange is run between the user and the SN.
The authentication is performed by the Authentication and
Key Agreement (AKA) procedure [7]. The AKA procedure
is built on the RIJNDAEL block cipher [8]. In addition to
authentication, AKA procedure also results in the Cipher
Key (CK) and the Integrity Key (IK). In UMTS, only the
encryption mode of the RIJNDAEL block cipher is used [9]
as an iterated hash function [10]. The block and key length
have been set to 128-bit. The USIM AKA (Fig. 1) is chosen
in such a way as to achieve maximum compatibility with the
current GSM/GPRS security architecture. [11], [12]. USIM
AKA is a one-pass challenge response protocol [13], [14].
Figure 2. Overview of Authentication and Key Agreement

[16]
Figure 2 shows that, after receiving an authentication

information request, the HE generate an ordered array of n
authentication vectors. Each authentication vector (AV)
consists of five components (and hence may be called a
UMTS ‘quintet’ in analogy to GSM ‘triplets’): A random
number RAND, an expected response XRES, a cipher key
Figure 1. UMTS Authentication and Key Agreement [8]
CK, an integrity key IK and an authentication token AUTN.
This array of n authentication vectors is then sent from the
3.1 When to use AKA HE to the SN. It is good for n authentication exchanges
• Registration of a user in a SN between the SN and the USIM. In an authentication
• After a service request exchange the SN first selects the next (the i-th) AVfrom the
• Location Update Request array and sends the parameters RAND(i) and AUTN(i) to
the user. The USIM checks whether AUTN(i) can be

accepted and, if so, produces a response RES(i) which is
sent back to the SN. AUTN(i) can only be accepted if the
sequence number contained in this token is fresh. [16] The
USIM also computes CK(i) and IK(i). The SN compares the
received RES(i) with XRES(i). If they match, the SN
considers the authentication exchange to be successfully
completed. The established keys CK(i) and IK(i) will then
be transferred by the USIM to the mobile equipment(ME)
and by the Visitor Location Register (VLR) or Serving
General packet Radio Service Support Node(SGSN) to the
Radio Network Controller (RNC); the keys are then used by
the ciphering and integrity functions in the Mobile Station Figure 3. AV and Key Generation in AuC [17]
(MS) and in the RNC.
.
4. AKA Algorithms 5.2 Functions in USIM
The security features of UMTS are fulfilled with a set of To generate the output keys in the USIM it has only one of
cryptographic functions and algorithms.[17] A total of 10 the four parameters that the AuC has the pre-shared secret
functions are needed to perform all the necessary features, key (K). The rest of the parameters it has to receive from the
f0-f5, f1*, f5*, f8 and f9. AuC.
Table 1: Authentication Functions [17]
Functio Description Output

n
f0 Random challenge generating RAND
function
f1 Network authentication function MAC-
A/XMAC-A
f1* Re-synchronization message MAC-
authentication function S/XMAC-S
f2 User authentication function RES/XRES

f3 Cipher key derivation function CK
Figure 4. RES Generation in USIM [17]
f4 Integrity key derivation function IK
f5 Anonymity key derivation function AK When the USIM receives the (RAND||AUTN) pair it starts
by generating the Anonymity Key (AK) by applying
f5* Anonymity key derivation function AK function f5 on the received RAND. By XOR-in the AK with
for the resynchronization message
function the (SQN XOR AK) from the Authentication Token, the
f8 Confidentiality key stream <Key stream sequence number of the AuC is revealed (SQNHE). [17] The
generating function block> secret key K is then used with the received AMF, SQN and
f9 Integrity stamp generating function MAC- RAND to generate the Expected Message Authentication
I/XMAC-I Code (XMAC-A). This is then compared with the MAC-A.
If the X-MAC and MAC matches, the USIM have
5. Key Generation functions authenticated that the message (RAND||AUTN pair) is
originated in its HE.
The functions f1-f5* are called key generating functions and
are used in the initial Authentication and Key Agreement
6. Authentication parameters
procedures.
The parameters used in the Authentication and Key
5.1 Functions in AuC
Agreement procedure are: AV, AUTN, RES and XRES,
When generating a new AV the AuC reads the stored value MAC-A and XMAC-A, AUTS, MAC-S and XMAC-S [17].
of the sequence number, SQNHE and then generates a new
SQN and a random challenge RAND [17].
Table2: Size of Authentication Parameters [17]
Figure 6. Confidentiality function
Table 4: Input parameters to confidentiality function [17]

7. Integrity function
The integrity protection of signaling messages between
Mobile Equipment (ME) and RNC starts during the security
mode set-up as soon as the integrity key and integrity
protection algorithm is known. A MAC function is applied
to each individual signaling message at the RRC layer of
UMTS Terrestrial Radio Access Network (UTRAN)
protocol stack [14] Both f8 and f9 algorithms are based on KASUMI algorithm.
The block cipher KASUMI is a modification of MISTY1
[14], [18]. KASUMI has been tested by the design team and
independent evaluation teams using crypt analytical
methods [19]. KASUMI constructions have also been
proven to provide pseudo randomness.
9. Weaknesses in UMTS security

mechanisms
Figure 5. Integrity Function [14] To sum up, the main weaknesses in UMTS security
mechanism are:
Figure 5 illustrates the use of integrity algorithm f9 to
• Integrity keys used between UE and RNC
authenticate the data integrity of an RRC signaling
generated in VLR/SGSN are transmitted
message. Input Parameters to the integrity function are
unencrypted to the RNC (and sometimes between
COUNT, IK, FRESH and Message [17].
RNCs).
• IMSI is transmitted in unencrypted form.
8. Confidentiality function • For a short time during signaling procedures,
In the 3G Security, user data and some signaling signaling data are unprotected and hence exposed
information elements are considered sensitive and may be to tampering.
confidentiality protected [14]. The need for a protected
mode of transmission is fulfilled by a confidentiality 10. Simulation Scenarios
function f8 as shown in Fig.6 [14]. The encryption function
Our simulation which identifies the problem is developed
is applied on dedicated channels between the ME and the
and tested on the version designed for the Microsoft
RNC
Windows environment, MATLAB R2008a. Figure 7 is the
[14].
screenshot for the normal AKA procedure and figure 8
shows that an intruder has captured the IMSI of MS, as it
was transmitted in plain text.
Figure 7. AKA normal procedure
It can be seen from Figure 8 and 9 that after the intruder has
captured IMSI, it will send this IMSI to VLR/SGSN and has
authenticated itself as an original MS although it is an
attacker that is acting as an authenticated user. Figure 9. An intruder launches the attack
Rest of the five USIM functions can be generated through

SIM cloning process by the attacker in order to prove that it
is the real user which has requested the service.
This problem can be avoided by proposing such algorithm
which also encrypts the IMSI during authentication process.
This problem has also been identified in [22], [23].
11. Further developments in UMTS security

Work on the next UMTS release has started. This will
introduce new security features. Many of these features will
be introduced to secure the new services which will be
introduced, e.g. presence services, push services and
multicast/broadcast services. Looking more into the future,
mobile cellular systems will have to accommodate a variety
of different radio access networks including short-range
wireless technologies, connected to a common core network.
On the user side the concept of a monolithic terminal, as we
know it, is dissolving. Distributed terminal architectures are
appearing whose components are interconnected by short-
Figure 8. An intruder captures the IMSI range radio links. These new developments represent a major
challenge to the UMTS security architecture. A collaborative
research project funded by the European Union and called
SHAMAN (Security for Heterogeneous Access in Mobile
Applications and Networks) have tackled these issues. A
separate project is also underway to identify research topics
in the area of mobile communications; this project is called
PAMPAS (Pioneering Advanced Mobile Privacy and
Security).
11. Conclusion key generation functions f1, f1*, f2, f3, f4, f5 and f5*,
Document 2: Algorithm Specification, April 2001.
AKA procedures in UMTS have increased security compared [10] 3GPP TR 35.909 V4.0.0, Technical Specification
with GSM. The new feature of two-way authentication Group Services and System Aspects, 3G Security,
eliminates the problem with false base stations. This is a very Specification of the MILENAGE. Algorithm Set: An
important security improvement. example algorithm set for the 3GPP authentication
Even though the security has improved in some areas, there and key generation functions f1, f1*, f2, f3, f4, f5 and
are still security features that should be improved. It is not f5*, design and evaluation, April 2001.
sufficient to just require integrity protection on signaling [11] C. J. Mitchell, “Security for Mobility”, Institute of
messages. All messages should be integrity checked, but Electrical Engineers, December, 2004.
indirectly by requiring confidentiality protection together [12] 3GPP TS 33.102 (5.2.0), “3G Security; Security
with integrity. AKA concept is used to perform Architecture”, Release 5, June, 2003.
authentication of the user and network, as opposed to 2G [13] ISO/IEC 9798-4: "Information technology – Security
systems, which only authenticated users in a system. The techniques - Entity authentication - Part 4: Mechanisms
confidentiality algorithm is stronger than its GSM using a cryptographic check function"
predecessor. The integrity mechanism works independent of [14] Evaluation of UMTS security architecture and services
confidentiality protection and provides protection against 1-4244-9701-0/06/$20.00 ©2006 IEEE
active attacks. The design of cryptographic algorithms is open [15] “Extension of Authentication and Key Agreement
and they are extensively crypto analyzed. Moreover, the Protocol (AKA) for Universal Mobile
architecture is flexible and more algorithms can be added Telecommunication System (UMTS)International
easily. Journal of Theoretical and Applied Computer
Although 3G Security marks a large step forward however Sciences” Volume 1 Number 1 (2006) pp. 109–118 (c)
there are some short comings. IMSI is sent in clear text when GBS Publishers and Distributors (India)
allocating TMSI to the user. In this paper this problem is http://www.gbspublisher.com/ijtacs.htm
discussed in detail with the help of simulation scenarios. [16] UMTS Security by K. Boman, G. Horn, P. Howard,
Some future work on UMTS security architecture is also and V. Niemi October 2002 issue of IEE Electronics &
elaborated in this paper. Communication Engineering Journal.
[17] UMTS Authentication and Key Agreement - A
comprehensive illustration of AKA procedures within
the UMTS system By Jon Robert Dohmen , Lars Sømo
References
Olaussen, , Grimstad - Norway, May 2001
[1] “3G Security; Security principles and objectives”, [18] M. Matsui, “Block encryption algorithm MISTY” in
Release 4, March, 2001. Proceedings of Fast Software Encryption (FSE’97),
[2] Johnson, M. (2002). Revenue Assurance, “Fraud and Volume 1267, Springer-Verlag, 1997.
Security in 3G Telecom Services. VP Business [19] 3GPP TR 33.909 V1.0.0 (2000-12) Technical Report;
Development Visual Wireless AB”, Journal of 3rd Generation Partnership Project; Technical
Economic Management, 2002, Volume 1, Issue 2. Specification Group Services and System Aspects;
[3] Stalling, W. Cryptography and Network Security, Report on the Evaluation of 3GPP Standard
Principles and Practice. 3rd edition. USA, Prentice Confidentiality and Integrity Algorithms (Release
Hall. 2003 1999)
[4] Stefan, P, and Fridrich R. (1998). “Authentication [20] S. Babbage and L. Frisch, “On MISTY1 higher order
Schemes for 3G mobile radio Systems”. The Ninth differential cryptanalysis”, in Proceeding of
IEEE International Symposium on, 1998. International Conference on Information Security and
[5] Zhang, M. and Fang, Y. (2005).” Security Analysis Cryptology (ICISC 2000), Lecture Notes in Computer
and Enhancements of 3GPP Authentication and Key Science Volume. 2015, Springer-Verlag, 2001.
Agreement Protocol”. IEEE Transactions on wireless [21] U. Kühn, “Cryptanalysis of reduced-round MISTY”,
Communications, Vol. 4, No. 2. 2005 in Proceedings of Eurocrypt’01, Lecture Notes in
[6] 3GPP TS 21.133. “3GPP Security; Security Computer Science, Volume 2045, Springer-Verlag,
Architecture”. 2001.
[7] 3GPP TS 33.102, Technical Specification Group [22] G. M. Koien, “Privacy enhanced cellular access
Services and System Aspects,” 3G Security. Security security” ACM - 2005
Architecture”, V 4.2.0 September 2001. [23] ” S.Y. A.-R. A. Mustafa Al-Fayoumi, Shadi Nashwan,
[8] Daemen J, Rijmen V. AES Proposal: Rijndael. “A new hybrid approach of symmetric/asymmetric
Available: authentication protocol for future mobile networks
IEEE - 2007
http://csrc.nist.gov/encryption/aes/round2/AESAlgs/Rij
ndael
[9] 3GPP TS 35.206 V4.0.0, Technical Specification Group
Services and System Aspects, 3G Security,
Specification the MILENAGE Algorithm Set: An
example algorithm set for the 3GPP authentication and
Automation of Data Warehouse, Extraction,

Transformation and Loading Update Cycle
Atif Amin1 and Abdul Aziz2
1
atif.amin @ucp.edu.pk
2
aziz@ucp.edu.pk
Abstract: Business enterprises invest lots of money to develop business advantage. In many organizations the way for users
data warehouse that gives them real, constant and up to date is to obtain timely information for correct decision. The
data for decision making. To keep data warehouse update, fundamental role of data warehouse is to provide correct
traditionally, data warehouses are updated periodically. decision making system. To achieve these kinds of
Periodic updates make a delay between operational data and information an application / tools are required for its
warehouse data. These updates are triggered on time set; some implementation. Their requirements include easy to access
may set it to evening time when there is no load of work on
data, scope and accuracy. They have also on-line analytical
systems. This fixing of time does not work in every case. Many
companies run day and night without any break, then in these
processing (OLAP) based on relation databases.
situations periodic updates stale warehouse. This delay depends Decision support systems (DSS), was named given in
upon the periodic interval, as interval time increase the 1970s to information systems designed to help managerial
difference between operational and warehouse data also staff for making decisions. Managerial problems are
increase. The most recent data is unavailable for the analysis ranging from assigning budget and choosing correct site
because it resides in operational data sources. For timely and locations for business etc. The basic idea behind developing
effective decision making warehouse should be updated as soon these systems was that mangers could create and operate
as possible. Extraction, Transformation and Loading (ETL) are these systems at their own. Therefore, in 1980s number of
designed tools for the updating of warehouse. When warehouse organizations called it executive information system (EIS)
is refreshed for the update purpose, it often gets stuck due to [3].
overloading on resources. Perfect time should be chosen for the
The basic idea behind EIS was the mangers needs
updating of warehouse, so that we can utilize our resources
standard information about their firms and the external
efficiently. Warehouse is not updated once, this is cyclic process.
We are introducing automation for ETL, Our proposed environment related to business. This information includes
framework will select best time to complete the process, so that the time history of problems and their output for predicting
warehouse gets updated automatically as soon as resources are their future state, so that manger could instantly know what
available without compromising on data warehouse usage. is going on. The EIS system does not have analytical
advantages of DSS. Some writers say that EIS is used by
Keywords: ETL, Updating, Loading, Data Warehouse. senior managers and DSS is used by junior staff.
Although very useful, the EIS and DSS often lacked a
1. Introduction strong database. Generally, information gathered for one
database cannot be used for another database. Managerial
Computers were used for data transactions and to provide
decision making, required consideration of the past and the
information to support decision making. As early as the
future, but not just the present. As a result, DSS and EIS
merit of placing information in different platform for
have to create their own databases, an area which was not
decision making were used. This approach is for easy to
their prime expertise. This activity is demanding and time
access needed data, improves system response time and
consuming.
assures security and data integrity. These systems were
pioneer to use this approach. Its end user saw many
applications for example executive summary and etc for
having specially prepared data.
Before two decades, organizations developed data
warehouse to provide users decision support system. There
are different approaches from earlier systems. One is the use
of special purpose system which task was to clean data,
extract useful data and loading all data into data warehouse.
Depending on the application needs many software can be
used to store data. Enhanced data access tools make it easy
for end user to access, transform, analyze and display
computed information without writing queries.
Many organizations are becoming customer focused.
They are using data mining to provide information for Figure 1. ETL follow in Data Warehouse
Data of the interest is across the different sources, for the time near to no use. We have to utilize the time of below
centralize warehouse it is collected over the heterogeneous average of resources for the updating of warehouse; it will
sources. Data is firstly extracted from the desired source, be selected on historical records of resources.
then transformed and cleansed respectively according to
globalize rules. In figure 1 basic cycle of data warehouse 3. Related Work
updating is shown. ETL is the process of managing Many researchers [2], [3], [7] have done a lots of work on
operational source data uniform so that it can be loaded into the ETL life cycle. They have purposed different techniques
the warehouse. In order to keep data warehouse up to date, regarding updating time. This process stuck out the data
operational source data is loaded into the warehouse timely. warehouse when ETL is being performed, because in
Loading into the warehouse is done by ETL tool. Warehouse Extract-Transform-Load makes warehouse unstable. When
should be intelligent enough to make decision for updating. this process finishes the warehouse is updated not only with
data but also with Meta data repository as shown in figure 2.
2. Common Problem It becomes stable when complete updating has been
Data warehouse is repository of business data that is drawn performed over it. We review different techniques that have
from multiple sources across the heterogeneous data stores. been used and proposed earlier. Many proposals consider
Since warehouse implement materialized view of data. Data that the development phase of data warehouse is totally
in the warehouse is stored and information is extracted from different from the development of RDBMS. Either they
the warehouse. Whenever data arrives in the warehouse, the include extra stages, like workload refinement [10] or the
extracted metalized information should also be implemented ETL process [11, 12]. Moreover, they provide different
according to current data. methods for the requirements analysis phase. Several
As we know that warehouse is the rising technique for the authors argue the importance of metadata as an integral part
retrieval of data from the distributed and heterogeneous data of the data warehouse design process [13, 14, 15].
sources. A data warehouse is a repository of integrated
3.1. Offline ETL Process
information, available for queries and analysis (e.g. decision
support and data mining) [1]. When relevant information This technique has become obsolete; it is not being used by
becomes the part of warehouse, the information extracted any modern data warehouse repository. We would like to
from source is translated into a common model called discuss it. Data in the warehouse was loaded offline. During
Relational Model, and integrated with the existing data of updating process, data warehouse was shut down from
warehouse. Data mining and querying to data warehouse functionality. In organizations every concern authorities was
can be done quickly and efficiently because formation of well informed that updating of data warehouse is in
data already exists due to warehouse property, and all progress and warehouse will not be working for some time
differences found resolved in warehouse. interval. When data in the warehouse was updated,
intimation to the all concern authorities of organization sent
that they can use it.
This technique was also called manually updating the
data warehouse. If any problem to the warehouse comes, all
others have to suffer from using it. During maintaining
phase warehouse was not capable of performing its
functionality. Warehouse is not one time activity, every time
it needs to be updating to capture new requirements and
more data. Most of the time warehouse was not functioning
due to its maintain process. This technique was very
resource consuming because lots of resource were left
unused most of the time due to unavailability of warehouse
functionality.
3.2. Periodic ETL Process
Figure 2. The ETL process When offline ETL process was not functioning well for the
We can think of data warehouse as a defining and storing goodwill of the organization, a periodic ETL process was
integrated materialized view over the multiple data sources. introduced. This technique does not automate the process of
Figure 2 shows a cyclic process; it will be executed each updating but it somehow has semi automated this task. In
time when warehouse would be refreshed. Warehouse this technique off-hours were used for the updating process.
should be refreshed with the new operational source data When there was no need of warehouse. It also stops the
timely, so that analysis can be performed over it. Refresh of functionality of the warehouse while updating. It does not
warehouse requires the maximum utilization of matter because in off-peak hours there is no need of data
computerized resources. In business environment computing warehouse. Warehouse is mostly needed because when there
resources are being used for gaining the efficiency in the is management staff and they have to make decision. Now
business. If in peak hours this updating starts it will acquire problem arises how to select off-peak hours.
the use of these resources and will slow down the business 3.2.1 Daily Periodic Activity
system. These resources are not used all the time.
Sometimes, these recourses are used at average and some Daily periodic activity is best for those organizations which
need up to date data for correct decision making and these
organizations works in one shift only. If they work in the extra business needs. During maintenance time no user can
morning time and closed their work in the evening. We can communicate with the warehouse for getting results out of
see that their all resources remain unused when their office it.
remains closed from evening to morning up till new day
3.2.5 Yearly Periodic Activity
starts. Office close time will be the peak hours for that
organization which works for one shift only. Their There are some organizations which update their data
warehouse administrator will chose evening time for the warehouse on annual basis. Mostly these companies make
updating of warehouse. It will utilize this resource when their decision and policies at the beginning of every year.
these were at not in use. Our primary goal is to maximize Once they have decided their work plan then it will remain
the use of resource to save our cost. This is online updation in effect for rest of the year. In market different
because in this technique there is no need to shut down the organizations have different plans and policies. They use
activities of warehouse. Administrator is also not necessary the strategy which suits them best. Moreover, it also
to be there when warehouse is being updating. When depends on the business nature and market trends.
warehouse is updated, it automatically sends confirmation 3.3. Online periodic Queue Activity
report to the concerned person that the warehouse has been
An extensive study on the modeling of ETL jobs is
updated.
published by Simitsis and Vassiliadisl [5][9]. Online
3.2.2 Weekly Periodic Activity periodic queue is best suited for that environment where
Weekly periodic activity is best for those organizations different heterogeneous sources update the warehouse.
which need up to date data for correct decision making but Business situation does not remain the same for a long time,
these organizations works in all shifts in working days i.e. situation change quite frequently. One operation source may
they work round the clock and close their work in the be in one time zone and other may be in different time zone.
weekend holidays. We can see that their all resources That is when one is using warehouse and other needs to
remain unused when their office remains closed. Warehouse update warehouse online because this is peak off time of
administrator will chose week holidays for the updating of that organization. In this scenario which is the best time to
warehouse. It will utilize resource when these were in no update warehouse.
use. Our always goal is to maximize the use of resource to We see that none of the given solution works for that kind
save our cost. This same situation can be reciprocal in an of problem where operational data stores are at different
organization if it works at week ends and holidays, and places and at different time zones. Then researcher
remains closed during week days. It depends on introduced online periodic queue activity. This activity
organization to organization according to their structural allows warehouse to have queue attached with it.
needs; their updating will be performed in those weeks
holidays respectively. This is online updating because in Operational Source
Operational Source
this technique there is no need to shut down the activities of
warehouse. Administrator is also not necessary to be there
when warehouse is being updating. When warehouse is
updated, it automatically sends confirmation report to the
Operational Source
concerned person that warehouse has been updated. Warehouse
Operational Source
3.2.3 Monthly Periodic Activity
Monthly periodic activity is best for those organizations
which need up to date data for correct decision making but
these organizations works in all shift in working days as
well as on holidays also. If they work in the all the shifts
and does not close their work even in the weekends,
holidays. Their warehouse administrator will chose closing
date of the month for the updating of warehouse. It depends Queue
on organization to organization according to their structural

Figure 3. Online Periodic queue activity
needs; their updating will be performed in those closing
month’s days when there is less use of warehouse resources, In figure 3 it is shown that queue is attached with the
because normally these organizations does not make warehouse. Each operation source sends its ETL output to
decisions at the closing of the months. the warehouse for the updating. It stores that into queue and
update one by one. This technique allows writing as much
3.2.4 Quarterly Periodic Activity
time as operational data sources needs to do so. Warehouse
Quarterly periodic activity is best for those organizations keeps updating itself whenever it gets off-hours. There are
which need up to date data for correct decision making but some problems with these techniques that are when
these organizations works in all shift in working days as operational data source keeps on sending their updating
well as on holidays. If they work in the all the shifts and request in the warehouse queue for the updating of
does not close their work in the weekends, holidays. warehouse, and warehouse does not get time for the
However, we require some extra time for the updating of updating. This creates problem of buffer overflow.
data warehouse. In this time updating is performed offline Warehouse would not be updating as it was desired. Lastly
and data warehouse is also maintained to accommodate it has another problem that is if more and more updating
request will be coming from the different data source for limits. We called the threshold that is average of maximum
the updating of warehouse, it will keep busy warehouse and minimum.
mostly in the updating and it will reduces the productive
work out of warehouse. This will decrease the efficiency of
the warehouse because we know that updating stuck the
warehouse system.
4. Proposed Solution
All above techniques waits for the off-hours so that
resources gets free and ETL can be performed easily. In
reality this is not the case with multinational organizations
because these organizations run round the clock. There are
no holidays even on Christmas, Eids and Easters. Although
employees get off individually but these organizations has
maintained a cycle that if one employee will be on holidays
other will be working for it alternatively, on top of that, Figure 5. Threshold of Resource
customers satisfaction and quality matters to them. In this
type of working environment where organizations remain Figure 4 and figure 5 explain the overall load on
busy round the clock; it is very difficult to find any off-hours resource. When it is near or under the threshold more load
and warehouse shut down also creates a problem for them. can be applied as it is applied in the off-hours in the
Our proposed framework will identify the times for the periodic updating activity.
updating of the warehouse using prediction technique Now we see resources utilization that participates in the
applying historical data. working of organization and updating of warehouse. This
It identifies all the resources that are used by the resource can be any server that is designed for both work.
warehouse while it is being maintained and updated. It We anticipate that resource that participates in duel actions
keeps the record each resource that is utilized with and remains overburden and causes the system malfunction from
without updating warehouse. It also identifies those the warehouse. In figure 6 we maintain a graph that checks
machines that get maximum use of warehouse. We will the load on that resource during the office hours and after
attach histogram with each resource and machines that the office hours. It maintains a history record set of these
needs data. They will observe the utilization of the activities on daily basis. We have shown a history record set
resources. It will calculate the threshold limit for the each of day Monday in table 1. We have picked those record set
resource and machines. This observation keeps on checking time and their duration that stay below the average
at what level of load this resource perform in timely manner threshold time.
and what stage it would start sticking or malfunctioning.
Once we identify these things we will notice that there Table 1: Resources duration with their status
will be some times where updating can be performed easily. Resources Day Time Duration Status
This framework identifies that times when office is working Res 1 Mon 6:00 Am 4 hours Below
but concerned resources are not being used. It will apply Res 2 Mon 5:00 Am 4 hours Below
updating process at that time and will starts observing load Res 3 Mon 12:00 10 hours Below
of the overall working environments up till functionality is Am
being performed well. When loads gets back its normal Res 5 Mon 7:00 Am 3 hours Below
stage of threshold we stop updating the warehouse at that
stage and will resume from that stage once updating will be We show some calculations that will get time for the
completed. warehouse updating during the office hours. We gets the
overlapped time that is free for all the activities. Table 1 is
the history of one day, we gets two hours free for the
updating of the warehouse. This overlap time is from 7:00
Am to the 9:00 Am where all the resources are free and
warehouse updating activity will not makes them in the
state of the stuck system. This recording of resources
duration and their status will be a continuous process. As
long as this process goes old, it will start giving the true
prediction of time for the warehouse updating. This
philosophy is improved with the passage of time, as
statistical population/data size grows prediction becomes
more and more accurate.
There will be a situation where ETL update cycle will
predict to update the warehouse but current status would not
Figure 4. Minimum and maximum of resource utilization
allow doing so because updating in this situation will cause
These are the maximum and minimum load limit where warehouse failure. We will skip that time and wait for
resource can work properly as shown in figure 4 by the another time.
warehouse buffer overflow due to the limited request of

update. Each data source will have a request of maximum of
one job in warehouse queue. Operational data sources will
contain there updating request in their queues until it does
not ask.
Figure 6. Resource utilization graph below and above

threshold point
Lastly, there may be a situation where current utilization
of resources and predicted time for the updating of
warehouse will allows to update warehouse smoothly but
when we will start updating the warehouse, the utilization
of the resource increase due to external use of resources. At
that time we will stop the updating process and give the
Figure 8. ETL update cycle architecture
utilization to the system environment and we will wait for
the next suitable time, and this is completely automatic, self These queues will give equal opportunity to each
activation and stopage. operational data source. It avoids them from starvation. In
Figure 7 gives the utilization picture of the resource that previous architectures some operation data sources may be
is below to its threshold time. We consider these as a free of starved due to the rapid updating of another warehouse.
utilization. We record the history of those resources with
their reason being free, which remains below average
mostly. If this case continues to be happened, we removed 5. Conclusion
them from the list of resources that we inspect because these
resources lie in the dual usage of system and warehouse. In this paper we discussed the issue regarding ETL update
cycle. We argued that update cycle method is very efficient
compared to complete loading of data from operational data
sources because in multinational organizations it is very
difficult to find off-hours easily. However, ETL update cycle
method is preferable in general because it gives maximum
utilization of resources and gets maximum throughput out
of it and availability of warehouse so that different users
gets maximum queries from it. Since ETL update cycle
requires extra checking cost for resources utilization every
time that is why the cost is increased, but this becomes
minimal when organizations gets more business due to
availability of warehouse.
Our main contribution is to provide certain rules for the
ETL update cycle that makes it fully automated process and
Figure 7. Resource Utilization below threshold point efficient without the human interaction with the data
warehouse. It selects its best time automatically for loading
There may comes a situation when there does not exist
data into the warehouse without compromising of its
any time frame when resources are not below threshold time
functionality and response time to the users. If warehouse
but are not at the maximum level of resource utilization.
gets some data from their operational connected data
Then we calculate the approximate load of ETL process
sources, then it give first priority to users and second
with its duration. We predict the load on the resource after
priority to updating otherwise our algorithms will not
putting the ETL update cost. If this remains at normal level
occupy the resources un-necessary. Warehouse does not
that is below the maximum level, we start updating the
need to update every time it gets updated call by some
warehouse until any of the resource goes beyond the
predefined time schedule. Our proposed mechanism saves
maximum load limit.
the time and cost as compared to periodic update and other
Our architecture of warehouse would also be modified
existing techniques for ETL.
with little changes. It has been shown in figure 8. It adds the
queue with each operational data source. This queue will
send the updating request when they will get response from
the warehouse that their pervious request has been updated.
This approach wills benefits in many ways. It will avoid
References Authors Profile

[1] Gupta, A., Jagadish, H. V., Mumick, "Data Integration
using Self-Maintainable Views," EDBT, pp. 140-144, Atif Amin received his B.S. in Computer Science degree from
University of Central Punjab in 2008. He has been the winner of
1996
All Pakistan Software Competition, Softcom ‘08 in 2008 where he
[2] JÄorg, T., Dessloch, "Towards generating ETL won first position. He has been chairman IEEE from 2007-2008.
processes for incremental loading," IDEAS, pp. 101- He is now doing M. S. in Computer Science from University of
110, 2008 Central Punjab.
[3] JÄorg, T., Dessloch, "Formalizing ETL Jobs for
Incremental Loading of DataWarehouses," BTW pp. Prof. Dr. Abdul Aziz did his M.Sc. from University of the
327-346, 2009 Punjab, Pakistan in 1989; M.Phil and Ph.D in Computer Science
from University of East Anglia, UK. He secured many honors and
[4] Kimball, R., Caserta, The Data Warehouse ETL
awards during his academic career from various institutions. He is
Toolkit: Practical Techniques for Extracting, leaning, currently working as full Professor at the University of Central
Conforming, and Delivering Data, John Wiley & Punjab, Lahore, Pakistan. He is the founder and Chair of Data
Sons, 2004 Mining Research Group at UCP. Dr. Aziz has delivered lectures at
[5] Panos Vassiliadis, Alkis Simitsis, and Spiros many universities as guest speaker. He has published text books
Skiadopoulos, "Conceptual modeling for ETL and large number of research papers in different refereed
processes," In DOLAP, pp. 14–21, 2002 international journals and conferences. His research interests
[6] Alkis Simitsis, "Mapping conceptual to logical models include Knowledge Discovery in Databases (KDD) - Data Mining,
for ETL processes," In DOLAP, pp. 67–76, 2005 Pattern Recognition, Data Warehousing and Machine Learning.
[7] Alkis Simitsis, Panos Vassiliadis, and Timos K. Sellis, He is member of editorial board and referee for various well
"Optimizing ETL Processes in Data Warehouses," In known international journals / conferences including IEEE
ICDE, pp. 564–575, 2005 publications. (e-mail: aziz@ucp.edu.pk).
[8] Alkis Simitsis, Panos Vassiliadis, Manolis Terrovitis,
and Spiros Skiadopoulos, "Graph-Based Modeling of
ETL Activities with Multi-level Transformations and
Updates," In DaWaK, pp. 43–52, 2005
[9] Labio, W., Garcia-Molina,"Efficient Snapshot
Differential Algorithms for Data Warehousing,"
VLDB, pp. 63-74, 1996
[10] M. Golfarelli and S. Rizzi. A methodological
framework for data warehouse design. In I.-Y. Song
and T.J. Teorey, editors, Proceedings of the 1st ACM
International Workshop on Data Warehousing and
OLAP, DOLAP’98, pp. 3–9. ACM Press, 1998.
[11] R. Kimball, L. Reeves, M. Ross, and W.
Thornthwaite. The Data Warehouse Lifecycle
Toolkit: Expert Methods for Designing, Developing,
and Deploying Data Warehouses. Wiley, 1998.
[12] S. Luján-Mora and J. Trujillo. A comprehensive
method for data warehouse design. Proceedings of the
5th International Workshop on Design and
Management of Data Warehouses, DMDW’03.
CEUR Workshop Proceedings, 2003.
[13] C. Ballard, D. Herreman, D. Schau, R. Bell, E. Kim,
and A. Valencic. Data Modeling Techniques for Data
Warehousing. IBM Redbooks SG24-2238-00, 1998.
[14] L. Carneiro and A. Brayner. X-META: A
methodology for data warehouse design with
metadata management. In [157], pp. 13–22.
[15] F. Paim, A. Carvalho, and J. Castro. Towards a
methodology for requirements analysis of data
warehouse systems. In Proceedings of the 16th
Brazilian Symposium on Software Engineering,
SBES’02, pp. 1–16, 2002.

Vol2 No 2

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Vol2 No 2

Uploaded by

Copyright:

Available Formats

(IJCNS) International Journal of Computer and Network Security, 1

Vol. 2, No. 2, February 2010

Multiple region Multiple Quality ROI Encoding

For some applications, only a subsection of the image

Figure. 1(b) A metaphase cell’s karyotype[5] 4.1 Classification of wavelets

5.1 ROI Coding

(LL subband) for a number of decomposition levels,

Figure 3. Original medical Image of

Space –Frequency Quantization (SFQ) technique exploits

In Fig. 3 an Image is taken showing the abdominal

Fig. 6 an image of a newborn baby affected by

Electronics, Vol. 46, No. 4, pp. 1103-1127, November

[7] K.P.Soman and K.I.Ramachandran -Insight Into

7. Conclusions Authors Profile

[1] Shaou-Gang Miaou, Shih-Tse Chen, Shu-Nien Chao ,

2- Dimension Automatic Microscope Moving Stage

2. Materials and Methods

2.1 To prepare the blood smear sample and set up the

Figure 1. Microscope moving stage

The basic principal for diagnostic in red blood cell is using

stage is made as large as possible and exposes the full

Figure 3. Scanning area

Figure 6. The characteristic of encoder 2 signal

Figure 4. Moving stage design

2.3.2 Digital Electronics Design:

The digital electronics architecture has two main

2.3.3.2 Stepping motor: control moving stage

Home 1:Home 2:Manual

Start. 1:Posit. #:Exit

Figure 10. Main menu of control program

3. RESULTS AND DISCUSSION

To design main menu as shown in Figure 10 to control the

4. Conclusion microscopes, accepted for publication on Nucl. Instr.

A Large Block Cipher Involving a Key

P = Mix (P), (2.5) Algorithm for Decryption

Obviously, here the length of the plain text block is 16 x

whose size is 16 x 16.

On comparing (3.9) and (3.11), we find that the cipher texts

Dr. S. Durga Bhavani is presently working as Professor in School

Mr. D. S. R. Murthy obtained B. E. (Electronics) from Bangalore

Electrocardiogram Prediction Using

Figure 1. Concept of processing for error convergence-type

2.2 Single Output Neuron Network System

ε i = yAi − zAi ( i = 1, 2,L , n ) (20)

where x is the input signal vector, x is the input signal, xAi is (

shown in expression (39). An output signal of ECP is shown

Aij ≠ 0 ( i = 1, 2,L, m; j = 1, 2,L, ni ) (33)

where the first figure of suffix of each sign is the parallel

nonlinear predictor using NNs is easier by strengthening of s (τ ) = ∑ σ 1 ( p )(τ ) u (τ − p )

in ECNNS is comparatively easy. However, learning for NN p=0 q= p

at the high step is difficult because it is guessed that the

Figure 5. Error convergence parallel-type recurrent neuron

x(jτ ) z1(τ −1) z1(τ ) -

Figure 6. Predictor using error convergence-type neuron network system

Steps The 1st step The 2nd step

Segmentation Method Based On Circular Edge

segmentation and the results are shown in Fig.5 and Fig.6

5. Conclusions indicate the domination of linear, circular and horizontal

Design of a 3.3GHz Band Application LC VCO for

Voltage Controlled Oscillators (VCOs’) are used in wireless

In order to reduce the power consumption, phase noise, tuning Where,

Tuning range can be determined as follows [5]:

Where, W0max is the maximum frequency of operation, W0min is the