Enabling Ultra Large-Scale Radio

Identification Systems
by
Kashif Ali
A thesis submitted to the
School of Computing
in conformity with the requirements for
the degree of Doctor of Philosophy
Queen’s University
Kingston, Ontario, Canada
August 2011
Copyright c _ Kashif Ali, 2011
Abstract
Radio Frequency IDentification (RFID) is growing prominence as an automated iden-
tification technology able to turn everyday objects into an ad-hoc network of mobile
nodes; which can track, trigger events and perform actions. Energy scavenging and
backscattering techniques are the foundation of low-cost identification solutions for
RFIDs. The performance of these two techniques, being wireless, significantly de-
pends on the underlying communication architecture and affect the overall operation
of RFID systems. Current RFID systems are based on a centralized master-slave
architecture hindering the overall performance, scalability and usability. Several pro-
posals have aimed at improving performance at the physical, medium access, and
application layers. Although such proposals achieve significant performance gains in
terms of reading range and reading rates, they require significant changes in both
software and hardware architectures while bounded by inherited performance bottle-
necks, i.e., master-slave architecture. Performance constraints need to be addressed in
order to further facilitate RFID adoption; especially for ultra large scale applications
such as Internet of Things.
A natural approach is re-thinking the distributed communication architecture of
RFID systems; wherein control and data tasks are decoupled from a central au-
thority and dispersed amongst spatially distributed low-power wireless devices. The
i
distributed architecture, by adjusting the tag’s reflectivity coefficient creates micro
interrogation zones which are interrogated in parallel. We investigate this promis-
ing direction in order to significantly increase the reading rates and reading range of
RFID tags, and also to enhance overall system scalability. We address the problems
of energy-efficient tag singulations, optimal power control schemes and load aware
reader placement algorithms for RFID systems. We modify the conventional set
cover approximation algorithm to determine the minimal number of RFID readers
with minimal overlapping and balanced number of tags amongst them. We show,
via extensive simulation analysis, that our approach has the potential to increase the
performance of RFID technology and hence, to enable RFID systems for ultra large
scale applications.
ii
Acknowledgments
First of all, I thank ALLAH (God) for his mercy, help and guidance.
It has been a great honour for me to be supervised by Dr. Hossam S. Hassanein. His
knowledge and guidance in research, and nice personality have been a great source
of support for me throughout my research. I would also like to thank Dr. Waleed
Al Salih, Dr. Abd-Elhamid Taha, Dr. Afzal Mawji, Sharief Oteafy, Ed Elkouche,
Abdallah Alma’aith and Louai Al-Awami, Hatem Abo Zied, Mervat Abu-Elkheir,
Xiangyang Zhang, Ashraf Alfagih, Abdulrahman Abahsain, Umair, Haroon Malik
and Pratik Dhrona for their support, collaboration, friendship and fruitful discussions.
I owe to my parents any success I make in my life. Their encouragement and support
is a key factor in any achievements I have ever made. I also want to thank my brother,
Kamran, sisters, Ambreen and Pinky, sister-in-law, Zara, and niece, Zainab, for their
love and support.
I would like to thank all members of the School of Computing, especially Telecommu-
nication Research Labs, at Queens University for their support, thoughtful discussions
and friendship.
iii
I would like to thank all anonymous conference and journal reviewers for their con-
tributions.
Special thanks to the Natural Science and Engineering Research Council and Ontario
Graduate Scholarship for their financial support.
iv
Statement of Originality
I hereby certify that this Ph.D. thesis is original and that all ideas and inventions
attributed to others have been properly referenced.
Kashif Ali
June 8
th
2011
v
List of Acronyms
AFSA Advanced Framed Slotted Aloha
ALE Application Level Events
ASAP Adaptive Slotted Aloha Protocol
ASC Approximate Set Cover
ASC-LB Approximate Set Cover with Load Balancing
ASK Amplitude Shift Keying
BST Binary Search Tree
CW Carrier Wave
CDMA Carrier Division Multiple Access
CMOS Complementary Metal Oxide Semiconductor
CSMA Carrier Sense Multiple Access
DCS Distributed Color Selection
DNS Domain Name Service
DFSA Dynamic Framed Slotted Aloha
DR-RFID Distributed Receiving Radio Frequency Identification
dB Decibel
dBm Power ratio in decibel
EBST Enhanced Binary Search Tree
EDFSA Enhanced Dynamic Framed Slotted Aloha
EPC Electronic Product Code
FDMA Frequency Division Multiple Access
GSM Global System for Mobile Communications
FSA Framed Slotted Aloha
HF High Frequency
HiQ Hierarchical Q-Learning
Hz Hertz
IC Integrated Circuit
vi
IFF Identify Friend or Foe
IoT Internet of Things
LF Low Frequency
LSB Least Significant Bit
MAC Medium Access Control
MACA Multiple Access with Collision Avoidance
MIMO Multiple Input Multiple Output
MHz Mega Hertz
MSB Most Significant Bit
NTE Neighbors and Tags Estimation
O-PDC Optimal Power-based Clustering
OD-PDC Optimal Discrete Power-based Clustering
ONS Object Naming Service
PIE Pulse Interval Encoding
PDC Power-based Distance Clustering
PML Physical Markup Language
PRQT Prefix Randomized Query Tree
QT Query Tree
RCS Radar Cross Section
RF Radio Frequency
RFID Radio Frequency Identification
RRE Redundant Reader Elimination
SDMA Space Division Multiple Access
SS Spread Spectrum
TDMA Time Division Multiple Access
UHF Ultra High Frequency
URL Universal Resource Locator
VDCS Variable-Maximum Distributed Color Selection
µW Micro-Watts
WSN Wireless Sensor Network
XML Extensible Markup Language
vii
Co-Authorship
1. K. Ali and H. S. Hassanein,“DR-RFID: Distributed Receiving RFID System
A novel paradigm”, submitted, IEEE Transaction on Mobile Computing, June
2011
2. W. Alsalih, K. Ali and H. S. Hassanein, “A Power Control Technique for Anti-
collision Schemes in RFID Systems”, submitted, Computer Networks, May 2011
3. K. Ali, W. Alsalih and H. S. Hassanein, “Set cover approximation algorithms
for load-aware readers placement in RFID networks”, accepted, to appear in
IEEE International Communication Conference (ICC), 6 pages, May 2011
4. K. Ali, W. Alsalih, and H. S. Hassanein, “Using neighbour and tag estimations
for redundant reader eliminations in RFID networks”, accepted, to appear in
IEEE Wireless Communication and Networking Conference (WCNC), 6 pages,
March 2011
5. K. Ali, S. Oteafy and H. S. Hassanein, “Energy-efficient parallel singulation in
RFID”, in proceedings of IEEE International Conference on Communications
(ICC), pp. 1-6, May 2010
6. K. Ali and H. S. Hassanein, “Parallel singulation in RFID systems”, in proceed-
ings of IEEE Global Communications Conference (Globecom), pp. 5594-5599,
Dec. 2009
7. K. Ali, A-E. Taha and H. S. Hassanein, “Medium access control in Radio Fre-
quency Identification”, Book chapter in “RFID and Sensor Networks: Archi-
tectures, Protocols, Security and Integrations”, edited by Y. Zhang, L. T. Yang
and J. Chen, CRC press, pp. 3-30, Nov. 2009
8. K. Ali and H. S. Hassanein, “Passive RFID tags and vehicle-assisted information
dissemination for ITS applications”, in proceedings of 34th IEEE Conference
on Local Computer Networks (LCN), pp. 688-694, Oct. 2009
viii
9. K. Ali and H. S. Hassanein, “Distributed receiving in RFID systems”, in pro-
ceedings of 34th IEEE Conference on Local Computer Networks (LCN), pp.
69-76, Oct. 2009
10. W. Alsalih, K. Ali and H. S. Hassanein, “Optimal distance-based clustering for
tag anti-collision in RFID systems”, in Proceedings of 33rdIEEE Conference on
Local Computer Networks (LCN), pp. 266-273, Oct. 2008
11. K. Ali, H. S. Hassanein and A-E. Taha, “RFID anti-collision protocol for dense
passive tag environments”, in proceedings of the 32nd IEEE Conference on
Local Computer Networks (LCN), pp. 819-824, Oct. 2007
ix
Contents
Abstract i
Acknowledgments iii
Statement of Originality v
List of Acronyms vi
Co-Authorship viii
Contents x
List of Tables xiii
List of Figures xiv
Chapter 1: Introduction 1
1.1 Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Thesis Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Document Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Chapter 2: Background and Literature Survey 7
2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 Enabling Technologies . . . . . . . . . . . . . . . . . . . . . . 10
2.1.3 System Architecture . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.4 Communication Protocols . . . . . . . . . . . . . . . . . . . . 18
2.2 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.1 System Design . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.2 Medium Access Control . . . . . . . . . . . . . . . . . . . . . 31
2.2.3 Tag Collisions . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2.4 Reader Collisions . . . . . . . . . . . . . . . . . . . . . . . . . 52
x
2.2.5 Coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Chapter 3: Power Control in RFID 64
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.2 The Power-based Distance Clustering (PDC) Anti-Collision Scheme . 66
3.3 Optimal Power-based Distance Clustering . . . . . . . . . . . . . . . 68
3.3.1 System model and problem definition . . . . . . . . . . . . . . 68
3.3.2 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.3.3 Delay analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.3.4 Optimizing the delay . . . . . . . . . . . . . . . . . . . . . . . 72
3.4 Optimal Discrete Power-based Distance Clustering (OD-PDC) . . . . 75
3.4.1 System model and problem definition . . . . . . . . . . . . . . 75
3.4.2 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.4.3 Delay analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.4.4 Optimal clustering algorithm . . . . . . . . . . . . . . . . . . 78
3.5 Simulation Setup and Results . . . . . . . . . . . . . . . . . . . . . . 79
3.5.1 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.5.2 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . 81
3.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Chapter 4: Distributed Receiving in RFID 91
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.3 DR-RFID: Distributed Receiving System for RFID . . . . . . . . . . 98
4.3.1 RFID Reader . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.3.2 RFID Fielder . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.3.3 RFID Tag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.4 Micro-Zoning in DR-RFID System . . . . . . . . . . . . . . . . . . . 104
4.4.1 Tag Reflectivity Co-efficient . . . . . . . . . . . . . . . . . . . 105
4.4.2 Dynamic Assignment of Reflectivity Index . . . . . . . . . . . 108
4.5 Parallel Singulation in DR-RFID . . . . . . . . . . . . . . . . . . . . 110
4.5.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.5.2 The Deterministic Algorithm . . . . . . . . . . . . . . . . . . 112
4.5.3 The Probabilistic Algorithm . . . . . . . . . . . . . . . . . . . 116
4.5.4 Delay Analysis of Deterministic Algorithm . . . . . . . . . . . 119
4.5.5 Delay Analysis of Probabilistic Algorithm . . . . . . . . . . . 121
4.6 Simulation Setup and Results . . . . . . . . . . . . . . . . . . . . . . 121
4.6.1 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.6.2 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . 124
xi
4.7 Work in Progress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Chapter 5: Coverage in RFID 140
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5.3 RFID Readers Deployment (Deterministic Deployment) . . . . . . . . 145
5.3.1 Assumptions and Definitions . . . . . . . . . . . . . . . . . . . 145
5.3.2 Providing Coverage . . . . . . . . . . . . . . . . . . . . . . . . 146
5.3.3 Reducing Overlap Among Readers . . . . . . . . . . . . . . . 148
5.3.4 Load Balancing . . . . . . . . . . . . . . . . . . . . . . . . . . 150
5.3.5 Displaced tags . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.4 RFID Redundant Readers Elimination (Random Deployment) . . . . 152
5.4.1 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . 152
5.4.2 Neighbors and Tags Estimation (NTE) based algorithm . . . . 152
5.4.3 Exemplary Scenario . . . . . . . . . . . . . . . . . . . . . . . . 155
5.4.4 Computational Complexity . . . . . . . . . . . . . . . . . . . . 157
5.5 Simulation Setup and Results . . . . . . . . . . . . . . . . . . . . . . 159
5.5.1 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.5.2 Simulation Results for Deterministic Deployment . . . . . . . 160
5.5.3 Simulation Results for Random Deployment . . . . . . . . . . 164
5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Chapter 6: Conclusion 170
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
6.2 Future Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Bibliography 174
xii
List of Tables
2.1 World-wide UHF RFID Frequency Allocation. . . . . . . . . . . . . . 14
2.2 Comparison of deterministic anti-collision algorithms. . . . . . . . . . 51
2.3 Comparison of probabilistic anti-collision algorithms. . . . . . . . . . 52
2.4 Comparison of reader collision algorithm. . . . . . . . . . . . . . . . . 57
4.1 Simulation parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.1 Tracing the NTE algorithm for the scenario of Fig. 5.2. . . . . . . . . 158
xiii
List of Figures
1.1 Tags and readers collision for a typical RFID system . . . . . . . . . 2
2.1 History timeline of advancements in RFID technology . . . . . . . . . 9
2.2 Schematic of an exemplary RFID tag. . . . . . . . . . . . . . . . . . . 11
2.3 Backscattering modulation concept of RFID tags . . . . . . . . . . . 12
2.4 Master-slave architecture in RFID systems. . . . . . . . . . . . . . . . 13
2.5 EPC Network Architecture . . . . . . . . . . . . . . . . . . . . . . . . 16
2.6 Multiple tags-to-reader collisions. . . . . . . . . . . . . . . . . . . . . 18
2.7 Multiple readers-to-tag collisions. . . . . . . . . . . . . . . . . . . . . 19
2.8 Execution trace of the binary search tree anti-collision . . . . . . . . . 22
2.9 Singulation tree for the example RFID system. . . . . . . . . . . . . . 23
2.10 Execution trace of the slotted Aloha anti-collision . . . . . . . . . . . 25
2.11 Query round for a single RFID tag . . . . . . . . . . . . . . . . . . . 28
2.12 Counter execution flow chart of the Q-algorithm . . . . . . . . . . . . 29
2.13 Power-based tags clustering . . . . . . . . . . . . . . . . . . . . . . . 33
2.14 Adaptive SDMA using directional antenna . . . . . . . . . . . . . . . 35
2.15 Frequency domain multiple access. . . . . . . . . . . . . . . . . . . . 36
2.16 Classification of the time division multiple access procedures. . . . . . 37
2.17 Classification of the deterministic TDMA schemes. . . . . . . . . . . 41
xiv
2.18 Taxonomy of the probabilistic TDMA-based schemes. . . . . . . . . . 46
2.19 Taxonomy of reader collisions schemes. . . . . . . . . . . . . . . . . . 53
3.1 Flow diagram of the PDC scheme. . . . . . . . . . . . . . . . . . . . . 68
3.2 An example of an interrogation zone of 3 clusters. . . . . . . . . . . . 69
3.3 An example of 3 transmission ranges and 2 clusters. . . . . . . . . . . 76
3.4 Total number of queries with the QT scheme. . . . . . . . . . . . . . 82
3.5 Total number of queries with the EBST scheme. . . . . . . . . . . . . 82
3.6 Total number of queries with the Q-Algorithm scheme. . . . . . . . . 83
3.7 Ratio of successful queries with the QT scheme. . . . . . . . . . . . . 84
3.8 Ratio of successful queries with the EBST scheme. . . . . . . . . . . . 85
3.9 Ratio of successful queries with the Q-Algorithm scheme. . . . . . . . 85
3.10 Ratio of collision queries with the QT scheme. . . . . . . . . . . . . . 86
3.11 Ratio of collision queries with the EBST scheme. . . . . . . . . . . . 87
3.12 Ratio of collision queries with the Q-Algorithm scheme. . . . . . . . . 87
3.13 Ratio of idle queries with the QT scheme. . . . . . . . . . . . . . . . 88
3.14 Ratio of idle queries with the EBST scheme. . . . . . . . . . . . . . . 89
3.15 Ratio of idle queries with the Q-Algorithm scheme. . . . . . . . . . . 89
4.1 Effect of tag power requirements on the operating range . . . . . . . . 97
4.2 Distributed receiving-based RFID architecture. . . . . . . . . . . . . . 99
4.3 Block diagram of fielder . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.4 Block diagram of passive tag . . . . . . . . . . . . . . . . . . . . . . . 103
4.5 Micro-zones in the DR-RFID system. . . . . . . . . . . . . . . . . . . 104
4.6 Tag schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.7 Tag reflectivity coefficient and assignment. . . . . . . . . . . . . . . . 109
xv
4.8 Timeout scenarios in parallel singulation. . . . . . . . . . . . . . . . . 115
4.9 Grid-based fielders’ deployment scheme. . . . . . . . . . . . . . . . . 123
4.10 Reading range improvement of DR-RFID . . . . . . . . . . . . . . . . 125
4.11 Overall query’s comparison between RFID and DR-RFID . . . . . . . 127
4.12 Collision comparison between RFID and DR-RFID . . . . . . . . . . 129
4.13 Reading reliability comparison. . . . . . . . . . . . . . . . . . . . . . 130
4.14 Effect of mobility on the tag reading rate. . . . . . . . . . . . . . . . 131
4.15 Reading cycles comparison for the multi-readers and DR-RFID. . . . 133
4.16 Fielder schematic for the DR-RFID system. . . . . . . . . . . . . . . 134
4.17 Fielder and power board prototype. . . . . . . . . . . . . . . . . . . . 136
4.18 Semi-passive tag schematic for the DR-RFID system. . . . . . . . . . 138
5.1 3D Coverage in RFID Networks. . . . . . . . . . . . . . . . . . . . . . 141
5.2 An RFID network with redundant readers. . . . . . . . . . . . . . . . 142
5.3 Snapshot of random RFID network from the simulator. . . . . . . . . 156
5.4 Number of readers required for various tag density. . . . . . . . . . . 161
5.5 Maximum load (tags) for various tag density. . . . . . . . . . . . . . . 162
5.6 Effect of tag density and interrogation area . . . . . . . . . . . . . . . 163
5.7 Redundant reader detected in sparse and dense network. . . . . . . . 165
5.8 Number of tag reads in sparse and dense environments. . . . . . . . . 166
5.9 Number of tags write in sparse and dense environments. . . . . . . . 168
xvi
1
Chapter 1
Introduction
Radio Frequency IDentification (RFID) is a prominently emerging automated identi-
fication technology. An RFID system is typically composed of an application, reader,
and a set of passive tags. A passive tag has neither a physical power source nor a
transceiver as it utilizes the energy-harvesting and reflection based communication,
for power and communication, respectively. RFID has an edge over other identifica-
tion systems such as barcode systems, optical character recognition systems, smart
cards, biometrics (voice, fingerprinting, retina scanning) . . . etc., in that it requires
no line-of-sight for communication, is robust to harsh physical environments, allows
simultaneous identification, in addition to being cost and power efficient. RFID seam-
lessly turns everyday objects into mobile network nodes that can be tracked, traced
and monitored. This emerging and rapidly evolving technology has the potential to
revolutionize the way we interact with physical objects. However, performance of an
RFID system, defined in terms of reading range, reading rate and reading coverage is
affected significantly by the magnitude of ultra large scale applications, e.g., Internet
of Things (IoT). Such performance constraints need to be addressed in order to fa-
cilitate efficient RFID continuing adoption. Coping with these challenges and hence
1.1. MOTIVATIONS 2
Figure 1.1: Tags and readers collision for a typical RFID system
enabling ultra large scale RFID systems is the main focus of this research.
In the rest of this chapter, we present the motivations behind our research, high-
light the thesis contributions and outline the thesis document.
1.1 Motivations
The nature of a tag’s reflection based communication dictates failure of the usage of
the conventional wireless Medium Access Control (MAC) techniques in RFID systems;
as tags cannot sense the medium, detect collisions, or sense the presence of other
channel traffic. This means that neither collision avoidance nor collision resolution
mechanism can be implemented in tags. Hence, MAC in RFID systems is confined to
resolving collisions only at the reader; anti-collision algorithms are needed for both
tag collisions and reader collisions.
1.1. MOTIVATIONS 3
Tag collisions occur when more than one tag is within the reader’s interrogation
zone and attempt to reply to the reader’s requests at the same time. This is illus-
trated in Fig. 1.1 as tags T4 and T5 are within reader R4’s interrogation zone. Tag
collisions are the most devastating, especially when passive tags, with their limited
communication functionalities, are involved. Reader collisions occur when one tag is
interrogated by more than one reader, as illustrated in Fig. 1.1 wherein the tag T1
is interrogated by readers R2, R3 and R5. Tag collisions and reader collisions result
in reduced reading rates, waste of reader’s available power; eventually affecting the
network lifetime.
Considering tag simplicity, the existing anti-collision algorithms usually employ
the Aloha-based contention avoidance schemes. However, these algorithms, because
of the centralized reader and tag communication architecture and the probabilistic
nature of their executions, are capped by theoretical upper bounds on reading rates.
Furthermore, mobility limit the number of tags that can be physically monitored
within the interrogation zone of an RFID reader. Therefore, existing anti-collision
algorithms do not scale to ultra large RFID applications having 10-100 thousands of
tags and thousands of readers.
In addition, reflection based communication also imposes limitations on the read-
ing range of RFID tags. The reflected signal strength faces significant attenuation as
it propagates twice the distance between the tag and reader, back and forth, hence
capping the upper bound on the operational range of passive tags to some tens of
meters. The operational power requirement of upcoming passive tag circuitry and
the maximal allowable Carrier Wave (CW) signal strength, in addition to attenu-
ation, also limit the operational range of a passive tag to a few meters. Existing
1.2. THESIS CONTRIBUTIONS 4
approaches fail to provide significant enhancements to reading rates of RFID tags;
therefore forcing more readers to maintain required coverage which further escalates
the collisions.
Deploying more readers to cover the required interrogation area will result in
redundant readers. For instance, readers R1 and R2, in Fig. 1.1, are both redundant
readers. R1 is redundant as it does not cover any tag. Whereas, R2 is redundant
as the only tag it covers (T1) is also covered by its neighboring readers R3 and R5,
both of which cover other tags as well, T2 and T3, respectively. The overlapping tags
need to be interrogated multiple times, individually by each reader. Such redundant
interrogations translate into unnecessary energy consumptions, hence lower reading
rates thereby affecting network lifetime.
1.2 Thesis Contributions
Our research aims primarily at enabling RFID for ultra large scale applications and
improving the performance of existing RFID systems; mainly in reading range, read-
ing rate and reading coverage. The main contributions of this thesis are:
1. Tags within the interrogation zone of a reader cause immense “in-the-air” data
collisions, and hence, jeopardize their readability. In general, a reader interro-
gates tags within its zone using maximal power; mainly to cover as many tags
as possible. This conflicts with the objective of minimizing tag collisions. We
overcome this problem by devising an optimal power stepping scheme that mini-
mizes tag collisions while covering the required interrogation area. We formulate
the problem as a closed-form optimization problem.
2. The centralized reader and tag communication architecture of an RFID system
1.2. THESIS CONTRIBUTIONS 5
places fundamental constraints on its performance; namely reading rate and
reading range. We overcome these limitations by assigning the existing cen-
tralized design of RFID system − the nominal task of receiving tags replies −
to spatially distributed devices within the interrogation zones of readers. The
goal of the proposed system, named Distributed Receiving RFID (DR-RFID), is
enhancing the reading reliability and increasing the reading range and reading
rate in a scalable manner.
3. We extend the DR-RFID system and introduce the concept of a micro-zone. In
it, the reader’s interrogation zone is divided into smaller zones, each of which
is managed by a set of distributed devices. We devise an algorithm which de-
termines the reader’s power level and alters the strength of the tag’s reflected
signal; with the objective of confining its readability within a smaller subset,
i.e., a micro-zone. They facilitate autonomous interrogation of tags. We pro-
pose parallel singulation algorithms for the DR-RFID system. In addition,
we also propose both deterministic and probabilistic variations of existing tag
anti-collision algorithms. Our parallel singulation alleviates the reading rate
bottlenecks.
4. We define and solve RFID coverage as a deterministic as well as random de-
ployment problem. For the former, we propose an approximation algorithm to
determine the minimal number of RFID readers required to maintain the re-
quested coverage while having minimal overlapping interrogation zones amongst
the readers. We also extend the algorithm to give a load-balancing deployment
strategy. We propose a heuristic algorithm to eliminate redundant readers in
1.3. DOCUMENT OUTLINE 6
random deployments. The goal is to achieve maximal coverage with the min-
imal possible readers by turning off redundant readers, i.e., readers without
additional coverage benefits. This is useful for large scale, dense and random
deployments of RFID systems.
1.3 Document Outline
The remainder of this document is organized as follows. Chapter 2 presents the
background and literature survey of RFID systems. Chapter 3 presents the power
control scheme for tag anti-collision scheme in RFID systems. This includes the basic
heuristic scheme, closed-form optimization scheme and discrete greedy optimization
scheme. Chapter 4 presents the distributed design for RFID system. The proposed
design uses distributed receiving entities to facilitate reading range and reading rates.
This also includes our micro-zone concept and parallel singulation algorithms, adopted
for both deterministic and probabilistic classes of anti-collision schemes. Chapter 5
presents RFID coverage both as a deterministic and a random deployment problem.
It includes the approximation algorithms for load-aware reader placement and the
greedy heuristic algorithm for redundant reader elimination, in deterministic and
random deployments, respectively. Finally, Chapter 6 concludes this document by
highlighting the main issues addressed in this thesis and outlining some of the future
research directions.
7
Chapter 2
Background and Literature Survey
2.1 Background
In this section, we briefly present the history of RFID technology, the enabling tech-
nologies, fundamental system architecture and prominent underlying communication
protocols.
2.1.1 History
The roots of RFID can be traced back to World War II when Germans enhanced
upon the Scottish physicist Sir Robert Alexander Watson-Watt radar to differentiate
between the enemy and their own planes. The approach was to enable pilots to
roll their planes as they returned to base, allowing the reflected signal to alter in
its amplitude. This method would allow the ground crew to differentiate between
the German planes and the Allied planes. This was conceptually the first single ID-
bit passive RFID system. During same era, the British, under Watson-Watt secret
project, developed the first active Identify Friend or Foe (IFF) system which involves
deploying an active transmitter on each of the British plane. The transmitter, when it
2.1. BACKGROUND 8
received signals from radar stations on the ground, begins broadcasting signal which
helps to identify it as a friendly plane. These two systems were conceptually the first
single bit passive and active RFID system.
Advances in radar and RF communications systems continued throughout the
1950s and 1960s. These advancements aided into commercialization of the 1-bit RF
based identification resulted in to electronic article surveillance tags, still use in goods
theft protection. The bit, depending on the successful payment, is either on or off
to detect any forgery or theft. The early RFID technology also found its way into
military applications. In 1970s, Los Alamos National Laboratory developed the first
transponder reader system to detect the trucks carrying nuclear materials in and out
of the secure facilities. The reader wakes up the transponder which then will respond
with its serial number and stored information. This system was commercialized in
mid 1980s, since then it is adopted for the automated toll payment system for roads,
bridges and tunnels across the globe. This identification system, advanced with a
passive Low-Frequency (LF) 125 KHz RFID tag, was then adopted in various other
applications, e.g., animal tracking and building access. Later on, due to unregulated
and unused spectrum, High-Frequency (HF) 13.56 MHz RFID tags were introduced.
These HF tags offered relatively faster data transfer rates. Today, the HF 13.56 MHz
RFID systems are used for access control, payment systems (e.g., speed-pass) and
contact-less smart cards for secure access control.
In the early 1990s, IBM engineers developed an Ultra-High Frequency (UHF)
RFID system which offered longer range and faster data transfers however, suffers
from wide adoption-scale adoption due its high associated cost. These high costs
were mainly due to low sales volume, uncertainty in return-of-investment and lack
2.1. BACKGROUND 9
Figure 2.1: History timeline of advancements in RFID technology, highlighting the
key applications and enabling technologies.
of international standardizations. However, this changed with the establishment of
Auto-ID centre, at the Massachusetts Institute of Technology, which introduced stor-
ing the serial number on a small microchip. This approach enabled low-cost RFID
tags which can then be placed on products to be tracked throughout the numerous
manufacturers supply chain.
Between 1999 and 2003, the Auto-ID Centre, with support of more than 100 large
end-user companies, key RFID vendors, and U.S. Department of Defence, developed
two important air interface protocols (Class 0 and Class 1), the Electronic Product
Code (EPC) numbering scheme and a network architecture to lookup tag data over
the network. The standardization of RFID, especially with evident of the second
2.1. BACKGROUND 10
generation standard, currently known as EPCGlobal Gen2, enable large-scale deploy-
ments and adoption of RFID solutions. Some of the world biggest retailers, e.g.,
Metro, Target, Tesco, Wal-Mart, and U.S. Department of Defence are evaluating the
EPC Gen2 technology for goods tracking in their supply chain. Other industries such
as pharmaceutical, automotive, defence, aviation, and many more are also deploy-
ing prototype systems to evaluate the technical feasibilities of RFID systems within
their business process. The major RFID related historical events are summarized in
Fig. 2.1.
2.1.2 Enabling Technologies
Advancements in energy harvesting, backscatter modulation and micro electro me-
chanical systems present the foundation of low-cost identification solutions provided
by RFID. The energy harvesting yields battery-less RFID tags which enables minimal
maintenance, longer lifetime and low cost RFID solutions. The backscattering yields
transceiver-less RFID tags which enable minimal power and low cost RFID solutions.
Energy scavenging (harvesting) technique harness incident radio waves and fa-
cilitate the operation of battery-less RFID tags. These incident radio waves, after
interacting with the antenna, produce high-frequency voltage which is rectified using
a diode. The rectified signal is smoothed out using a storage capacitor to provide con-
stant voltage. The voltage is then used to power the tag’s logic unit and nonvolatile
memory space. A similar circuitry with a smaller capacitance is used to de-modulate
the data transmitted by the reader. Using this process, known as envelope detection,
the tag processed the reader’s query request and acts accordingly. The schematic of
a simple energy harvesting scheme and envelope detection of a typical RFID tag is
2.1. BACKGROUND 11
Figure 2.2: Schematic of an exemplary RFID tag.
shown in Fig. 2.2. Radio frequency based energy harvesting enables a cost effective
solution as the tag circuitry does not require sophisticated signal processing blocks,
e.g., crystal frequency reference, synthesizer, power amplifier, low-noise amplifier and
batteries.
Backscattering reflects and modulates incident radio waves and facilitates transceiver-
less RFID tags. The tag modulates the signal scattered by the antenna to commu-
nicate with the reader. Conceptually, for the modulated backscattering, the RFID
tag iterates between three different states, namely normal operation, open circuit and
short circuit. These three states are depicted in Fig. 2.3. In the normal state, the
tag’s antenna is matched with the IC load to allow maximum possible power delivery
to the IC. This power, as is explained above, is used by the rectifier to charge up the
tag’s circuitry. The tag is guaranteed a continuous unmodulated carrier wave, to har-
vest upon, by the reader, as it stays in this state for a maximum allowable time. To
modulate the backscattering signal the tag transits between short and open state. In
2.1. BACKGROUND 12
the short state, the current flow steadily to the ground plane. The flow of the signal
to the ground, as is not used by the rectifying unit, scatters with minimal resistance,
resulting in detectable power signal at the reader. On the other hand, in the open
state, the signal path to the ground is blocked. This results in no current flow in the
antenna and hence, minimal radiation is reflected back. These two state, i.e., short
and open, allow a tag to modulate the bit as ’0’ or ’1’, respectively. Backscattering
modulation enables a cost effective solution as the tag does not require a transceiver
to communicate with the reader.
Figure 2.3: Backscattering modulation concept of RFID tags [36].
2.1.3 System Architecture
An RFID system is typically composed of an application host, a reader and a set of
tags. A tag is designed to store certain information, the size of which varies between
96 bits and 64K bytes. Tags can be either passive, semi-passive or active. A passive
tag has no physical power source and no transceiver as it uses the energy-harvesting
and backscattering modulation techniques for power source and communication, re-
spectively. While most common in the RFID market, the passive tag has limited
2.1. BACKGROUND 13
Tag
Reader Application Tag
Tag
Master Slave
Master Slave
Interrogation
and tag reply
Figure 2.4: Master-slave architecture in RFID systems.
functionality for processing and communication. It processes simple state machines
and has no medium sensing capabilities. A semi-passive tag has an on-board battery,
use to power its circuitry, but no transceiver hence, relies on backscattering modula-
tion for communication with the reader. An active tag, on the other hand, has both
on-board battery and transceiver and may also carry certain sensing capabilities for
example temperature or pressure. The RFID tags are listen-first devices, i.e., they
does not initiate communication unless are interrogated by the reader.
Tags targeted by a certain reader are said to be with in that reader’s interrogation
zone. Specifically, an interrogation zone is the physical distance within which the
strength of the electromagnetic waves, generated by the reader, is able to power the
tags, receive the tags’ signals, and successfully decode them — a process known as
singulation. At any arbitrary instant, a reader can only read one tag within its
interrogation zone and a tag can only be read by one reader. An RFID reader acts as
a master for the tag and a slave for the application host. Upon application request,
an RFID reader will interrogate the tags within its interrogation zone. The master-
slave architecture of the RIFD system is shown in Fig. 2.4. The read tag data is then
passed onto the middleware application where it is filtered accordingly.
RFID reader operates in wide UHF frequency ranges, i.e., between 860-960 MHz
2.1. BACKGROUND 14
Region Radio frequency band
Argentina, Chile, Costa Rica,
Dominican Republic, Uruguay 902-928 MHz
Australia 918-926 MHz
Brazil 902-907.5 MHz & 915-928 MHz
Canada and U.S.A 902-928 MHz
China 917-922 MHz
EU and Russia 865-868 MHz
Hong Kong 920-925 MHz & 865-868 MHz
Japan 952-954 MHz
Korea 908.5-914 MHz
Malaysia 866-869 MHz & 919-923 MHz
Mexico 915 MHz
New Zealand 864-868
Singapore 923-925 MHz & 866-869 MHz
South Africa 865.6-867.6 MHz & 917-921 MHz
Taiwan 922-928 MHz
Thailand 920-925 MHz
Table 2.1: World-wide UHF RFID Frequency Allocation.
and 2.4-2.45 GHz. The worldwide spectrum regulates within the 860-960 MHz is
complex, as it competes with cellular hence, the different RFID spectrum varies for
different region. For instance, in North America and European Union the allowed
RFID frequency is between 902-928 MHz and 865-868 MHz, respectively. The world-
wide UHF band allocation for RFID systems is summarized in Table 2.1. The 2.4-2.45
GHz band is an attractive alternate because of its unlicensed spectrum. However, it
receives high interference from other devices, e.g., Zibgee and WiFi, and has reduce
reading range (1-3m at 2.4 GHz vs. 2-10m at 900 MHz).
RFID readers collect large amounts of data, most of which are redundant or
irrelevant for the time instance. An RFID middleware filter out the redundant and
irrelevant data to be allowed as input for localization and tracking, managing the
2.1. BACKGROUND 15
tagged-object historical data, trigger associated events and so forth. The middleware
also encodes the data, to pass on to the reader, so it can be stored in proper format
within the tag memory. According to the EPCglobal standardization efforts, the
following components are the required modules to properly associate the middleware
applications with the tagged object. The EPC network architecture, along with
components and layers, is shown in Fig. 2.5.
1. Electronic Product Code: Electronic Product Code (EPC) is the tag data which
frequently is polled by the reader(s) on a regular basis. Due to frequent nature
of these polling, the EPC data needs to be filtered by the middleware to remove
duplicate codes.
2. EPC Middleware: The middleware facilitate the exchange of data between the
RFID reader, or a network of readers, and an RFID application. The middle-
ware instructs reader(s) for singulation, accumulates and filters the raw EPC
data. For instance, after initial interrogation, the middleware needs only to
maintain the database for those existing or new tags which are either leav-
ing or entering, respectively, within the interrogation zone of the reader. The
processed and useful EPC data is then passed onto the RFID application.
3. Application Level Events: The Application Level Events (ALE) is an Interface
definition that defines the interface between the EPC-enabled RFID application
and the EPC-enabled RFID reader or a network of readers.
4. Object Naming Service: The Object Naming Service (ONS) translates the bytes
in the EPC tag data into the web address (URL) of the object data, i.e., returns
the network-addressable endpoints mapping to a particular EPC code data. The
2.1. BACKGROUND 16
Figure 2.5: EPC Network Architecture (reproduced from [40]).
2.1. BACKGROUND 17
ONS is an automated network service, allowing the end-user to execute Domain
Name Service (DNS) like queries. In order to use DNS to find information
about an item, the items EPC must be converted into a format that DNS can
understand, which is the typical, dot delimited, left to right form of all domain-
names. The ONS resolution process requires that the EPC being asked about is
in its pure identity URI form as defined by the EPCglobal Tag Data Standard
[EPC] (e.g., urn:epc:id:sgtin:0614141.100734.1245).
5. The Physical Markup Language: In a supply chain application, the data needs
to be available to multiple participating organizations, and it makes sense to
store it in an XML-based format that can be read consistently by all parties,
regardless of their information systems. The Physical Markup Language (PML)
provide this service to the participating application. The PML, based on XML,
provides a format for storing and retrieving data about objects that can be
parsed by multiple RFID applications.
6. EPC Information Services: The EPC Information Service is an inter-application
data exchange standard that defines a secure, web services based data exchange
mechanism. The information service is one of the core service and allow the
business partners to access and search for EPC data. The EPC Discovery
Service is the search engine for the EPC related data which returns the URL
locations that have the requested data. Multiple discovery service may be run,
each with limited scope (regional, facility wide, local, etc.)
2.1. BACKGROUND 18
Reader
Tag
Tag
Tag
Tag
Figure 2.6: Multiple tags-to-reader collisions.
2.1.4 Communication Protocols
The simultaneous wireless medium access results in collisions that undermine a RFID
system’s overall performance. To sustain system operability, efficient mechanisms for
Medium Access Control (MAC) are required. As with other radio-based systems, the
main objective of mechanisms aimed at regulating medium access, is to reduce colli-
sions, be it in a proactive or a reactive manner. Proactively, collisions are avoided by
distributing sufficient information about the access requirements of elements sharing
the medium. Reactive mechanisms respond to collisions and attempt to speed the
system’s recovery from a collision stall. Conventional collision avoidance methods,
such as Carrier Sense Multiple Access (CSMA) cannot be adopted for RFID sys-
tems, especially when passive tags are used due to power limitations and its basis
of reflection-based communication, i.e., use of backscattering modulation. Avoidance
mechanisms also carry the risk of increasing the overall cost of tags and shrinking the
potential interrogation zone for a reader. Proposals for RFID systems have there-
fore favoured reactive approaches to alleviate collisions. Specific to RFID systems,
collisions can be classified based on the type of entities involved:
2.1. BACKGROUND 19
Reader1
Reader2 Tag
Figure 2.7: Multiple readers-to-tag collisions.
Tags-to-reader Collisions: Occur when more than one tag within a reader’s inter-
rogation zone attempts to reply to the reader’s requests at the same time, depicted
in Fig.2.6. Tags-to-reader collisions are the most devastating, especially when pas-
sive tags are involved. They result in reduced reading rates, wasted resources, and
increased delay.
Readers-to-tag Collisions: Occur when one tag is interrogated by more than one
reader, as shown in Fig. 2.7. In such a scenario, multiple readers try to singulate a
single tag which results in corruption of the tag’s internal state. As a result, the tag
may not be detectable.
Reader-to-reader Collisions: Are result of the conventional frequency interfer-
ences, i.e., multiple readers within each other’s interference zones are locked on the
same frequencies. Existing mechanisms such as frequency-hopping, dynamic fre-
quency allocation and dynamic power adjustment are utilized to hinder these col-
lisions.
Tag collisions, both Tags-to-reader and Readers-to-tag, are the most common
source of collisions in RFID systems and are resolved by the reader utilizing tech-
niques collectively known as anti-collisions schemes. These anti-collisions schemes
are classified in literature to be either deterministic or probabilistic [42, 96].
2.1. BACKGROUND 20
Deterministic anti-collision scheme
In deterministic mechanisms the reader splits and identifies a set of tags to respond
in a given time. Splitting is based on contention information obtained from the
previous interrogation cycle and attempts to reduce contention for the next cycle.
Deterministic anti-collision mechanisms fall under the general algorithmic classifica-
tion of tree-based algorithms because of their splitting approach. The deterministic
mechanisms utilize either tags’ serial numbers (identification codes) or randomly gen-
erated numbers to be used for splitting of the tree branches. The tree-based algorithm
is a 2-way handshake algorithm involving sequences of interaction between the reader
and the tags, known as the interrogation cycle. The objective of these interrogation
cycles is to split the tags, using their serial numbers (IDs) or randomly generated
numbers, into a more manageable set of tags. The splitting of the tree, mostly binary
trees, into two branches (leaves) is based on the bit collisions and their respective
positions; the location of which is obtained from the previous interrogation cycle.
Obtaining collision information at the bit level requires that the precise bit position
of collision is identifiable by the reader. For this particular reason, normally, either
the Manchester coding or the NRZ (non-return-to-zero) bit coding is used by the
RFID reader.
In order to understand the deterministic tree-based algorithm, we trace the exe-
cution of the conventional binary search tree algorithm [102]. The objective of using
the binary search algorithm with multiple interrogation cycles is to singulate a tag
from a larger set. The 2-way handshakes during each interrogation cycle (iteration)
are the commands that the reader broadcasts based on its previous iteration for the
following iteration. The set of commands constitutes four main commands, which are
2.1. BACKGROUND 21
as follows.
1. REQUEST: Carries a serial number to the tag as a parameter. If a tag’s own
serial number is less than or equal to the received serial number, the tag sends
its own serial number back to the reader. Otherwise, the tag will not respond.
2. SELECT: Carries a serial number to the tag as a parameter. Only a tag with
an identical serial number is selected for the processing of other commands, e.g.
reading and writing data. Only the selected tag will continue to respond to the
reader commands.
3. READ DATA: The selected tag sends stored data to the reader.
4. UNSELECT: Cancels the selection of an up-until-now active tag, i.e. mutes the
tag. Beyond this, the unselected tag becomes completely inactive and does not
respond to further REQUEST commands until is reset by the reader.
Let us assume that there are three RFID tags within the interrogation zone of
the reader with 4-bit IDs of 1010
b
, 1011
b
, and 1110
b
. The singulation process starts
with the RFID reader sending the highest possible serial number. The purpose is to
make all tags respond so that the exact bit collisions among all the tags’ ID’s could
be determined. In our example, shown in Fig. 2.8, the first iteration of the algorithm
begins with the reader transmitting the REQUEST command with the serial number
1111
b
as an argument. The serial number 1111
b
is the highest possible in this case.
As the serial number of the tags in the interrogation zone is less than the requested
serial number, all three tags reply back with their IDs. This first iteration results in
collisions (C) at position 0 and 2, starting from the Least Significant Bit (LSB), i.e.,
1C1C.
2.1. BACKGROUND 22
Figure 2.8: Execution trace of the binary search tree anti-collision algorithm using
three tags and a single reader.
The third bit is the highest valued bit at which the collision has occurred during
the first iteration. This implies that there is at least one tag between 1100
b
and
1011
b
. The binary search algorithm at this point splits the search into two subsets in
an attempt to limit the search zone for subsequent interaction. The algorithm sets
the third bit of the request to 0, i.e., 1011
b
. The least significant bits after the third
bit are all set to 1 in an attempt to capture all the tags whose two most significant
bits are 10. The reader now broadcasts the command REQUEST with argument
1011
b
. Two tags fulfill the command’s criteria, i.e., tag1 and tag2. The response
back from these two tags causes collision at the bit position 0, i.e., 101C. The reader,
repeating the same splitting procedure, now chooses 1010
b
as the requesting string
for the subsequent iteration, i.e., the second iteration. This request is fulfilled by only
2.1. BACKGROUND 23
1111
1011
1010 1011
1111
1C1C
101C
1010
1 0
1 0
Figure 2.9: Singulation tree for the example RFID system.
one tag tag1 which is now singulated. For now, no further iterations are required, as
the reader has successfully detected a single tag without collision. Using subsequent
SELECT commands, tag1 is selected using the detected tag ID address and can now
be read or written by the reader without interference from other tags. At this point,
other tags are silent as the READ DATA command is selective. The tree, from the
request point of view, is shown in Fig. 2.9. The string shown in the tree nodes are
requested and sent by the reader, whereas the string next to each circle is the received
response. The node splits into two children nodes, appending 0 and 1 at the Most
Significant Bit (MSB) collision, to the left and right child node, respectively. This
process continues until tag 1010
b
is singulated, which is distinguished in the figure by
a double concentric circle.
After the completion of the required read/write operation, the selected tag (tag1)
is muted by the use of the UNSELECT command. The muted tag is completely
inactive and does not reply to any further REQUEST commands. Muting the tags
reduces the number of responding tags in the subsequent iteration of the singulation
process, therefore resulting in lower collisions and less iterations. Referring back to
2.1. BACKGROUND 24
Fig. 2.8, the singulation of the tag (tag2) requires one less iteration, compared with
tag1’s singulation. The muting of tag1 has created a positive impact by yielding low
collision as its serial number; otherwise, it would have collided with the tag2 serial
number.
Probabilistic anti-collision scheme
In the probabilistic mechanisms the reader communicates the frame length and the
tag picks a particular slot in the frame for transmission. The reader repeats this
process until all tags have been successfully transmitted at least once. Reader con-
trolled synchronization is necessary as the tag has to transmit within its slotted time
frame. The frame size may be adjusted based on the collision, idle and successful
frame information from a previous interrogation cycle for the subsequent cycle. This
encourages frame adaptability according to tag density and distribution, thereby re-
ducing idle and collision frames. The probabilistic approach is faster in comparison
to the deterministic approach because of its low overhead. However, it may suffers
from tag starvation syndrome.
We will now illustrate the details of the conventional slotted Aloha anti-collision
procedure [42] using the last example. Consider again three tags with IDs of 1010
b
,
1011
b
and 1110
b
. Similar to the deterministic approach, the probabilistic approach
is a chit-chat protocol that uses a sequence of interactive commands. The set of
commands are as follows:
1. REQUEST: Synchronizes and prompts all tags to transmit their serial number
to the reader using one of the time slots that follow. For our sampler RFID
system, there are three available time slots.
2.1. BACKGROUND 25
2. SELECT: Sends a specific serial number to the tag as a parameter. Only the
matching tag, i.e., the tag whose serial number matches with the passed pa-
rameter, is flagged for further operation, allowing it to read and write. Tags
with conflicting serial numbers however, are still responsive to the REQUEST
command.
3. READ DATA: The selected tag sends stored data to the reader. In real RFID
systems, other commands will also be available but they are omitted here for
the purpose of simplicity.
Downlink
(R => T)
REQUEST
SELECT
1110
Uplink
(T => R)
Tag1
Tag2
Tag3
1110
1011
READ/
WRITE
1010
1 2
Collision
1110
1110
3
Figure 2.10: Execution trace of the conventional slotted Aloha protocol, using exam-
ple RFID system of three tags and a single reader.
The execution trace of the slotted Aloha protocol is shown in Fig. 2.10. After
a certain time interval, the reader periodically transmits the REQUEST command.
After receiving the REQUEST command, the tag randomly selects one of the three
available slots. The tag uses the slot to transmit its serial number back to the reader.
Due to a random slot selection method, there is a collision between tag1 and tag2 in
slot1. Only tag3 is successful in transmitting its serial number back to the reader. The
successful tag is selected for subsequent reading and writing after it is been selected
2.1. BACKGROUND 26
using the SELECT command. If no transmission was successful, the REQUEST
command is iteratively repeated until the serial number can be received successfully
and no collision is observed in the frames. As we’ve witnessed, in the probabilistic
anti-collision algorithms, the tag randomly selects a slot number in the frame and
responds to the reader using the slot it selected. When the number of tags is small,
the probability of tag collision is low, and the time used to identify all tags is rela-
tively short. As the number of tags increases, however, the probability of collision
becomes higher and the time used to identify the tags increases rapidly. Therefore,
the probabilistic algorithm performance depends significantly on the number of tags
within the interrogation zone of the reader.
EPC Class-1 Gen-2 RFID standard
The EPCglobal UHF Class-1 Gen-2 protocol defines the standard physical and medium
access requirements for the passive RFID tags. According to the defined standard, the
reader communicate with tag using RF modulation. The tags receives its operating
energy, using the unmodulated Continue RF Wave (CW), and communication link,
both down-link (forward) and up-link (backward), from the reader’s RF signal. We
will now explain the physical and medium access layer as is described the EPCglobal.
The specification defines a number of physical layer options for both forward-link
and backward-link. All forward-link communication uses Amplitude Shift Keying
(ASK), where bits are indicated by brief periods of low amplitude, and Pulse Interval
Encoding (PIE). The time between low amplitude periods differentiates a zero or a
one, and the reader can choose pulse durations that result in down-link rates ranging
from 26.7 kbps to 128 kbps. The backward-link rate is determined partially by the
2.1. BACKGROUND 27
forward-link preamble and partially by a bit field set in the Query command which
starts each query round, allowing for the backward-link frequency ranging from 40
kHz to 640 kHz. Along with setting the backward-link frequency, the reader also sets
one of four up-link encodings, namely FM0, Miller-2, Miller-4, or Miller-8. When
using FM0, one bit is transmitted during each cycle and the data rate is equivalent
to the link frequency. However, FM0 is highly susceptible to noise and interference
which motivated the addition of the Miller encodings. While these are more robust
to errors with a higher number being more robust, their link rates are reduced by a
factor of 2, 4, or 8, depending on the encoding.
The specification also defines the medium access control protocol, based on Framed
Slotted Aloha (FSA). In FSA scheme, each frame has a fixed number of slot and the
tags response by randomly picking the slot from each frame. The tag interrogation
process starts off by reader transmitting the CW to power-up the tags within its
interrogation range. The reader, afterwards, initialize the number of frames and slots
within each each frame. The number of frames and slots are dynamically selected
according to tags population. In literature, the individual frame is known as Query
Round. The sequence of interrogation commands, i.e., Query, Ack, Query Rep, etc.
that make up the Query Round, for a single tag, is shown in Fig. 2.11.
At the beginning of a Query Round, the reader can optionally transmit a Select
command which limits the number of active tags by providing a bit mask, as only
tags with IDs (or memory locations) that match this mask will respond in the subse-
quent round. A Query command is then transmitted which contains the fields that
determine the backward-link frequency and data encoding, the Q parameter which
determines the number of slots in the Query Round, and a target parameter. When a
2.1. BACKGROUND 28
Figure 1: C1G2 Protocol (Courtesy of EPCglobal)
can optionally transmit a Select command which limits the
number of active tags by providing a bit mask, as only tags
with ID’s (or memory locations) that match this mask will re-
spond in the subsequent round. A Query command is then
transmitted which contains the fields that determine the up-
link frequency and data encoding, the Q parameter which de-
termines the number of slots in the Query Round, and a Target
parameter. When a tag receives a Query command, it chooses
a random number in the range (0, 2
Q
- 1), where Q is in the
range (0, 15), and the value is stored in the slot counter of the
tag.
If a tag stores a 0 in its slot counter, it will immediately
transmit a 16 bit random number (RN16). Upon receiving
the RN16, the reader will echo the RN16 in an ACK packet
at which point, if the tag successful receives the ACK with
the correct random number, the tag will backscatter its ID (re-
ferred to as EPC in Figure 1). This message sequence is shown
in the upper half of Figure 1.
While powered up, tags maintain an Inventoried flag which
can be in one of two states, A or B. In the Query command the
Target parameter is set to either A or B, and only tags with a
matching Inventoried flag will respond during the round. After
a tag transmits its ID, a subsequent QueryRepeat command
will cause the tag to toggle its Inventoried flag. If the ID was
not successfully received by the reader, a NAK command is
sent which resets the tag so that a subsequent QueryRepeat
will not result in the flag being changed. In this case, the tag
will be active in the next round.
1
Along with toggling the Inventoried flag of a tag, a
QueryRepeat signals the end of the slot. On receiving the
command the remaining tags will decrement their slot counter
and respond with a RN16 if their slot counter is set to 0. The
process then repeats, with the number of QueryRepeats being
equal to the number of slots set using the Q parameter.
1
The C1G2 standard specifies a timeout period between tag
responses and subsequent reader commands. Consequently,
even if a NAK is lost the tag will be reset by the timeout and
will not toggle the Inventoried flag at the next QueryRepeat.
Of course, tags may choose the same random number ini-
tially with the result that multiple tags reply during a slot. In
this case, a collision occurs and the tags will not be ACKed
during the round. However, these tags will be active in the
next round, where they will choose a new random number. In
the lower half of Figure 1, a scenario is shown with three tags,
two of which choose an initial random value of 0 resulting in
a collision in the first slot, and one which chooses an initial
value of 2 and so responds after the second QueryRepeat.
As well as collisions, if there are no tags with a slot counter
of 0, a slot may be empty. In order to reduce empty slots and
collisions, the reader changes the Q parameter for every Query
Round in order to best accommodate the number of tags re-
maining to be read. This is done by determing the number of
empty slots, slots with collisions, and slots with only a single
tag reply during the course of the Query Round. Using this in-
formation, the appropriate Q parameter can be set for the next
round. In [16], it is shown that the optimum is seen when the
number of slots is set equal to the number of tags, and approx-
imately 35% of the slots would see only a single tag response.
While there are a number of strategies for reading a tag set,
a Query Cycle generally consists of a series of Query Rounds
with the target set to A. As tags are read, and their Invento-
ried flags are set to B, they become inactive during subsequent
Query Rounds. When a Query elicits no tag responses, as all
tags have set their Inventoried flags to B, the Query Cycle ends
and the reader powers down. When powered down, the Inven-
toried flag of each tag is reset to A, and all tags will conse-
quently be active during the next Query Cycle.
2
2
Note that a reader can be configured such that the state of a
tag’s Inventoried flag will persist across cycles. As this per-
sistence time is defined to be anywhere between 500 ms to
5 seconds, this configuration option is not considered in this
study.
Figure 2.11: Query round for a single RFID tag (reproduced from [40]).
tag receives a Query command, it chooses a random number in the range (0, 2
Q
−1),
where Q is in the range (0, 15), and the value is stored in the slot counter of the tag.
If a tag stores a 0 in its slot counter, it will immediately transmit a 16 bit random
number (RN16). Upon receiving the RN16, the reader will echo the RN16 in an ACK
packet at which point, if the tag successful receives the ACK with the correct random
number, the tag will backscatter its ID (referred to as EPC in Fig. 2.11). While
powered up, tags maintain an Inventoried flag which can be in one of two states, A
or B. In the Query command the Target parameter is set to either A or B, and only
tags with a matching Inventoried flag will respond during the round. After a tag
transmits its ID, a subsequent QueryRepeat command will cause the tag to toggle
its Inventoried flag. If the ID was not successfully received by the reader, a NACK
command is sent which resets the tag so that a subsequent QueryRepeat will cause
the flag to change. In this case, the tag will be active in the next round.
Along with toggling the Inventoried flag of a tag, a QueryRepeat signals the end
of the slot. On receiving the command the remaining tags will decrement their slot
2.1. BACKGROUND 29
counter and respond with a RN16 if their slot counter is set to 0. The process then
repeats, with the number of QueryRepeats being equal to the number of slots set using
the Q parameter. Of course, tags may choose the same random number initially with
the result that multiple tags reply during a slot. In this case, a collision occurs and
the tags will not be ACK’ed during the round. However, these tags will be active in
the next round, where they will choose a new random number. In the lower half of
Fig. 2.11, a scenario is shown with three tags, two of which choose an initial random
value of 0 resulting in a collision in the first slot, and one which chooses an initial
value of 2 and so responds after the second QueryRepeat. If there are no tags with a
slot counter of 0, a slot will go empty.
Figure 2.12: Counter execution flow chart of the Q-algorithm (reproduced from [40]).
To reduce empty slots and collisions, the reader changes the Q parameter for every
Query Round in order to best accommodate the number of tags remaining to be read.
The algorithm, known as Q-algorithm, achieves this by determining the number of
2.2. LITERATURE SURVEY 30
empty slots, slots with collisions, and slots with only a single tag reply during the
course of the Query Round. The execution flow chart of the Q-algorithm is shown
in Fig. 2.12. The Q-algorithm evaluates the tags’ replies, for the current frame, to
determine the next frame size. The Q-algorithm classifies the replies as idle, collision,
and successful slots and accordingly update the Q value for the subsequent query
cycles. The value of Q parameter is added and subtracted by a fixed constant C if
current slot resulted in collision and idle, respectively. In case of successful slot, the Q
parameters remains unchanged. The slot are updated, incremented or decremented,
to avoid mis-match between the number of slots and number of tags. For instance,
in case of idle slot, the number of slots are decremented as it imply that the number
of available slots exceed the number of tags. Similarly, in case of collision slot, the
number of tags exceeds the number of slots. In [72], it is shown that the optimum is
seen when the number of slots is set equal to the number of tags, and approximately
35% of the slots would see only a single tag response.
2.2 Literature Survey
In this section, we will survey RFID literature for system designs, tags anti-collision
and readers anti-collision protocols.
2.2.1 System Design
RFID is a dynamic and active research area. System level development includes
antenna designs, RFID IC with enhanced RF sensitivity, ultra low power tag IC, low-
power backscattering modulation technique and so forth. The antenna design involves
tradeoff between antenna gains, impedance and bandwidth while satisfying numerous
2.2. LITERATURE SURVEY 31
requirements such as size, form, reading range, support for mobility, cost and reading
reliability [91]. Various antenna configurations such as patch antenna [42], folded
dipole antenna [108], asymmetric antenna [27] and use of multiple tag antenna [49]
have been proposed in literature. A detailed comparative performance evaluation of
various antenna configurations can be found in [45, 91, 101]. Low-power IC design
for passive tags [34, 36, 42, 57], as low as 2.7µW, has been proposed in the literature
to enhance the operating range of the passive RFID tags. Improvements in the
backscattering and modulation techniques are also proposed in the literature. Various
analytical studies on the Radar Cross Section (RCS), an important parameter of in
determining the quality of the backscattering communication, for the passive tags [81],
its differential scalar scattering [82] and to enhance the bit error rate [46], have been
performed. Energy-aware modulation techniques [13, 67, 98] also been proposed to
reduce tag’s energy consumption while using backscattering communication. Wireless
Sensor Network (WSN) and RFID integration provides multi-hop communication to
the network of RFID readers [70]. In such a setting, the WSN node and RDID reader
individually sense, read tags and perform multi-hop data communication with other
interrogating readers.
2.2.2 Medium Access Control
In a radio based system, MAC protocols provide mechanisms for channel access con-
trol, allowing multiple devices to share the same physical medium. The most promi-
nent MAC techniques are based on the Carrier Sense Multiple Access (CSMA), Mul-
tiple Access with Collision Avoidance (MACA), or the Aloha protocol. RFID systems
use half-duplex, point-to-point communication links in which communication between
2.2. LITERATURE SURVEY 32
the reader and the tag is based on backscatter modulation. In backscattering, the
tag sends its serial number by adjusting the antenna reflectivity to modulate the
reflected signal. The nature of reflection based communication results in failure of
conventional MAC techniques; as tags cannot sense the medium, detect collisions, or
sense the presence of other channel traffic. This means that no collision avoidance or
resolution mechanisms could be implemented at the tags. Hence, MAC in RFID sys-
tems is confined to resolving of collisions only at the reader, resulting in anti-collision
algorithms for both tag collisions and reader collisions, mainly employing Aloha-based
contention avoidance schemes.
RFID systems exhibit very short durations of very high bursty activity followed by
relatively long durations of inactivity. Generally speaking, four different approaches
exist in the literature in handling multi-access issues in radio technologies. The ap-
proaches regulate access based on time, space (location), frequency, or code. Code
Division Multiple Access (CDMA) supports high-rate data multiplexing using the
spread-spectrum (SS) technique. In conventional SS, each user encodes data pack-
ets using an orthogonal spreading code, allowing successful transmission by multi-
ple users. However, complex receiver design and higher computational and energy
requirements restrained adaptation of CDMA technology in RFID systems. A com-
bination of TDMA and CDMA based schemes has been investigated recently [75].
However, the CDMA scheme is a largely unexplored area in RFID systems.
Space Division Multiple Access (SDMA) [42, 99] spatially reuses the channel. The
philosophy behind the SDMA procedure is that practically at any given time there
will be a limited number of RFID tags spatially co-located at the same position, inde-
pendent of the total number of tags that may be present in the overall interrogation
2.2. LITERATURE SURVEY 33
Reader
d’
T’
T’
T’
T’
d
d’’
T"
T"
T"
T"
T"
Figure 2.13: Clustering based on the distance between the reader and the tags, using
reader’s power levels to adjust its interrogation zone.
zone. Therefore, by spatially isolating the tags, the interference caused by other tags
can be minimized. Numerous spatial isolation techniques exist with the most notable
ones being: adjusting the reader power levels (power control) [10, 12, 60], using adap-
tive arrays and Multiple Input Multiple Output (MIMO) antennae [65], as well as
the use of electronically controlled directional antenna [42].
The power control cluster-based algorithm [10, 12] divides the interrogation zone
into smaller clusters based on the distance to the reader by adjusting the reader’s
power levels. Tags in each cluster are read separately. Since the number of tags in
a cluster is less than that in the whole interrogation zone, collisions are minimized.
An example of such a partitioning is shown in Fig. 2.13, where the interrogation zone
is divided into three clusters: d, d

, and d

. When the reader sends a request, only
2.2. LITERATURE SURVEY 34
those tags in the current cluster respond. For instance, assume the reader has read
the tags from cluster d and has just sent a request to cluster d

. In that instance only
tags from cluster d

, i.e., tags marked as T

, will respond to that request. After all
tags marked as T

have been read, they are put into a sleep mode. Then, the reader’s
zone is increased to d

and a new request is sent, and so on. Similar transmission
control schemes have been used for reader collision avoidance [60], where reader’s
power levels are adjusted to reduce the interrogation overlaps.
Smart antenna techniques such as adaptive array antenna and MIMO antenna [65]
are used to maximize system throughput by reducing the collisions. The antenna
array adaptively nullifies interferences from other array elements, i.e., one element
receives the desired signal whereas other array elements are used to remove the inter-
ference in order to maximize the received signal strength. With MIMO, however, each
antenna element receives a superposition of the multiple transmitted streams with dif-
ferent spatial signatures. These differences are used to separate multiple streams with
signal processing at the receiver. These two techniques lower tag collisions, but with
significant increase in the reader complexity and monetary cost.
Electronically controlled directional antennas [42] are used to adaptively adjust the
directional beam of the RFID reader at subset of tags, one at a time. Such techniques
are commonly known as adaptive space division multiple access and are depicted in
Fig. 2.14. To read a specific tag or all the tags within the interrogation zone, the reader
scans the area using the directional beam until the desired tag is detected or, in the
latter case, when all the tags have been read by the reader. The SDMA technique is
effective in reducing collisions but with a relatively high implementation cost because
of sophisticated antenna systems, and is therefore restricted to a few specialized
2.2. LITERATURE SURVEY 35
Reader
Tag
Tag
Tag
Tag
Tag
Tag
Tag
Interrogation Zone of reader
Figure 2.14: Adaptive SDMA with an electronically controlled directional antenna.
applications. One exceptional approach is the aforementioned power control scheme
[10], which is low in cost and is effective in reducing both reader and tag collisions.
Frequency Domain Multiple Access (FDMA) proposes the use of multiple chan-
nels, each with distinct carrier frequencies for communication. In the context of
RFID systems, this implies the availability of broadcasting frequency for the reader
and several frequencies for the tags that they can lock on to. The reader, using the
broadcasting frequency, synchronizes and issues the interrogation commands. The
tags reply to the reader using one of the many available frequencies, shown as f
1
, f
2
,
f
3
and f
4
, in Fig. 2.15. The advantage of FDMA is the availability of non-interfering
frequencies for concurrent communication by multiple tags and readers. The FDMA
2.2. LITERATURE SURVEY 36
Tag
Tag
Tag
Tag
Tag
Interrogation Zone
of the reader
Reader
Broadcast/synchronisation
f
f
f
f
f
B
1
2
4
3
Figure 2.15: Frequency domain multiple access.
technique has not been significantly utilized in RFID systems. The reason is the
impracticality for tags and relatively high cost of the readers as a dedicated receiver
must be provided for every reception channel, restricting the use of FDMA for limited
and specialized applications.
Time Domain Multiple Access (TDMA) relates to techniques where the available
channel is divided, along the time dimension, between potential participants. TDMA,
in a modified and mostly hybrid manner, is used in other prominent networks, e.g.,
GSM, Bluetooth and IEEE 802.16 (WiMax). The TDMA technique is by far the most
dominant medium access protocols in RFID systems. This is because of TDMA sim-
plicity, low processing overhead for passive tags and low complexity (computational,
processing, and monetary cost) compared with other available procedures such as
2.2. LITERATURE SURVEY 37
Figure 2.16: Classification of the time division multiple access procedures.
FDMA, CDMA, SDMA, etc. In the context of the RFID system, the TDMA proce-
dure is further classified into tag-driven and reader-driven, as is shown in Fig. 2.16.
The tag-driven procedure operates in an asynchronous fashion as the reader does not
control the data transfer. Tag-driven procedures are naturally very slow and inflexible
with limited applicability. Most of the TDMA procedures are therefore based on the
reader-driven approach. The reader-driven approach is a synchronous mechanism,
with all the tags’ data transfer handled by the reader, i.e., control when the tags
transfer data and selects which ones, amongst the tag population, will transmit its
data at any given time. This selection process, i.e., isolating an individual tag from a
large group of tags, is known as the singulation process. The reader-driven approach
is used to tackle the collisions in RFID systems.
2.2. LITERATURE SURVEY 38
Tag collision schemes usually involve a logical partitioning of the tag popula-
tion using tree-based algorithms or probabilistic-framed Aloha schemes, into a more
manageable set of tags. Reader collisions employ conventional contention resolution
approaches such as scheduling, interference learning, and coloring schemes to tackle
reader collisions, i.e., reader-to-reader and readers-to-tags collisions. The various col-
lision resolution schemes for tags and readers’ collisions are classified based on their
underlying techniques, and are discussed next, in a sequential manner, in the following
sections.
2.2.3 Tag Collisions
Tag collisions are the most common source of collisions in RFID systems. They take
place when multiple tags are within the interrogation zone of a reader and simulta-
neously reply to the reader’s commands. The passive RFID tag is by design reader-
driven; as it does not have carrier sensing nor inter-tag communication capabilities.
Therefore, tag collisions are resolved by the reader utilizing techniques collectively
known as tags anti-collision schemes. These anti-collisions schemes are classified in
literature, as depicted in Fig. 2.16, to be either deterministic or probabilistic [42, 96].
In deterministic mechanisms the reader splits and identifies a set of tags to re-
spond in a given time. Splitting is based on contention information obtained from the
previous interrogation cycle and attempts to reduce contention for the next cycle. De-
terministic anti-collision mechanisms fall under the general algorithmic classification
of tree-based algorithms because of their splitting approach. The deterministic mech-
anisms utilize either tags’ serial numbers (identification codes) or randomly generated
numbers to be used for splitting of the tree branches. Under certain circumstances,
2.2. LITERATURE SURVEY 39
the deterministic approach may take a considerably longer time. However, it does
not suffer from the tag starvation problem. In tag starvation, a tag may not be
identifiable for a long time, and in the worst case, might not be able to be read at all.
In probabilistic mechanisms, the reader communicates the frame length, and the
tag, randomly transmits a particular slot in the frame. The frame size may be ad-
justed, based on the information from the previous interrogation cycle, encouraging
adaptability according to tag density and distribution. The frame process is repeated
until all the tags have been identified. The probabilistic approach is fast, due to its
low overhead, but suffers from tag starvation syndrome.
Deterministic Schemes
Deterministic anti-collision mechanisms are essentially tree-based anti-collision algo-
rithms. The tree-based algorithm is a 2-way handshake algorithm involving sequences
of interaction between the reader and the tags, known as the interrogation cycle. The
objective of these interrogation cycles is to split the tags, using their serial numbers
(IDs) or randomly generated numbers, into a more manageable set of tags. The split-
ting of the tree, mostly binary trees, into two branches (leaves) is based on the bit
collisions and their respective positions; the location of which is obtained from the
previous interrogation cycle. Obtaining collision information at the bit level requires
that the precise bit position of collision is identifiable by the reader. For this partic-
ular reason, normally, either the Manchester coding or the NRZ (non-return-to-zero)
bit coding is used by the RFID reader.
We further classify the deterministic anti-collision mechanisms, as shown in Fig. 2.17,
based on whether they use a collision tracking or a collision detection approach.
2.2. LITERATURE SURVEY 40
Collision Tracking: In the collision tracking method, both the reader and the tag
maintain a certain amount of collision information from the previous interrogation
cycle. The information, mostly as pointer to the most recent query, collision bit,
or node in the tree, is used to send a subsequent query for the next interrogation
cycle. The binary search tree algorithm (explained above using an example RFID
system) and its variants [17, 18, 32, 41, 43, 54, 71, 80, 105, 106, 113, 116]), fall under the
collision tracking class of the deterministic anti-collision algorithms. The variants
of conventional binary tree algorithms differ in their enhancement objectives, i.e.,
shortening execution time, reducing memory usage, reducing processing overhead or
eliminating the unnecessary iterations.
In the conventional binary tree algorithm, when a tag is successfully singulated,
read, and muted, the singulation process initiates from the root for the subsequent
tag. This was seen, for instance, in our example above where the reader sends out the
initial query of 1111
b
, i.e., the highest possible serial number, after the first tag, i.e.,
tag1, is singulated and unselected. This results in unnecessary overhead, both from
communication and processing perspectives. The overhead can be eliminated, after
successful singulation, by setting the starting query point to the queued query from
the last interrogation cycle. This translates to selecting either the parent node of the
current query node or its sibling node [54, 105]. This simple modification speeds up
the interrogation process and reduces overhead by utilizing the saved (queued) infor-
mation. The conventional binary tree-based algorithm recognizes only a single tag at
a time through setting the most significant collision bit to either 1 or 0 during the
interrogation process. The work in [32, 71] notes that this results in low performance
and claims that without any fundamental change in the tree navigation process two
2.2. LITERATURE SURVEY 41
Collision detecting
methods
Collision tracking
methods
Memoryless
Prefix overhead
reduction
Hybrid Statistical
Energy
aware
Binary Tree
Deterministic
Quary Tree
Figure 2.17: Classification of the deterministic TDMA schemes.
tags are identifiable at one time. This takes place during the last bit of the colli-
sion sequence, i.e., when the least significant bit collision has been resolved. This is
evident, as the least significant bit collision indicates the presence of two tags, i.e.,
with collision bits of 0 and 1, therefore speeding up interrogation process. The use
of other mechanisms such as interference cancellations [112], utilizing the similarities
among the tags’ IDs [58], using height-oriented, depth-first search algorithms with
tag state support [60] and secure variation of the tree algorithms [17, 106] have also
been investigated. All these approaches enhance the binary tree algorithm without
deterring significantly from the conventional algorithm.
We further classify the collision tracking approach to be either statistical, hybrid,
or energy aware, as is shown in Fig. 2.17. The statistical approach exploits informa-
tion stored in previous interrogation cycles (iterations) to assist in the forthcoming
cycle. Adaptive binary splitting algorithms [26, 76, 77] initiate the conventional tree
2.2. LITERATURE SURVEY 42
splitting procedure by using the information obtained from the last reading cycle.
The fixed tags, i.e., static tags, which have been identified in the previous cycles will
be re-identified again by the reader in the next reading cycle. As the reader knows
about the fixed tags it can avoid collisions, whereas conventional methods are used
to identify the mobile tags. The mobile tag is defined as the tag just coming in, and
therefore was not present during the last reading cycle, or has moved out of the inter-
rogation zone of the reader. These adaptive splitting algorithms significantly reduce
the number of collisions while supporting mobility but at the cost of substantial stor-
age requirements. In such scenarios, where a significant portion of tags are mobile,
the adaptive splitting algorithms will perform no better than the conventional binary
algorithm.
The hybrid tree approach [94] is a combination of the deterministic and probabilis-
tic approach. It combines the tree-based protocol with a slotted back-off mechanism.
The algorithm uses a 4-ary tree instead of the conventional binary tree, with slotted
back-off aimed at reducing the number of unwanted and unnecessary query com-
mands. Upon responding to the reader, a tag sets its back-off timer using a part of
its ID. If there is a collision the reader can partially deduce how the IDs of the tags
are distributed and potentially reduce the unwanted idle cycles. Like the statistical
approach of [76], this approach utilizes the information from previous reading cycles
to avoid collision for the static tags and starts the querying process from the root in
order to accommodate the tag mobility. In addition to inheriting the drawbacks of
the statistical approaches, the hybrid approach increases the tags’ cost as it requires
additional functionality, e.g., back-off timer.
The MAC scheme’s objective of reducing collisions usually comes at the expense
2.2. LITERATURE SURVEY 43
of increasing energy consumption. This energy consumption occurs because of addi-
tional processing and communication overhead imposed by the MAC protocols. The
energy-aware approaches [78, 87, 115, 116] address the energy issues, mainly the tag
energy consumption, while maintaining low collisions. The approach in [116], while
utilizing the conventional binary tree algorithm, involves the reader taking action if
bit collision is detected. This translates to the reader sending a special symbol to
the tags to stop sending data back. Fewer bits are thus sent by the tags, indirectly
reducing energy consumption, by reducing communication overhead. More efficient
schemes [90] use multiple time slots per tree node and three different anti-collision
protocols. The motive is that in order to detect the tags, the binary tree algorithm
relies on the collision as it transmits redundant queries to the tags. This is important
in the conventional binary tree algorithm as it needs to determine which sub-tree to
query in the subsequent interrogation cycle. The tags are allowed to transmit re-
sponses within a slotted time frame, thereby avoiding collisions. The queries are used
to find colliding sub-trees as well as to read the tags. These two mechanisms help to
read more tags with fewer queries, thereby reducing the energy consumption at the
tags.
Collision Detection: Collision detection methods are stateless algorithms in
that they do not exercise tracking. Relative to collision tracking, detection algorithms
may take longer to read all tags within an interrogation zone. Their main advantage is
in their non-storage requirements, especially for the tags and their comparatively low
communication overhead. We classify the collision detection methods into memoryless
and prefix reduction, as is shown in Fig. 2.17.
Query tree protocol (QT) [63] and variants [31, 77, 116] are prominent examples of
2.2. LITERATURE SURVEY 44
the memoryless approach in which each tag does not need additional memory beyond
storing its serial number. This means that no storage is required for random serial
numbers, pointers or states, as it was required by the binary tree of the collision
tracking algorithms. The QT uses the tag’s serial number for the tree splitting mech-
anism. The reader broadcasts the query, using a string of bits as an argument. The
tag receiving the query matches the most significant bit of its own serial number with
the broadcasted serial number. If there is a match the tag transmits the remaining
least significant portion of its serial number, otherwise, it remains silent. The reader
maintains the queue for the transmitted strings of bits. At the beginning of the
frame, the queue is initialized with two 1-bit strings, 0 and 1. The reader pops a bit
string from the queue and broadcasts to all tags in its interrogation zone. If the tag
responses collide, the reader pushes two 1-bit longer bit strings, compared to the last
bit string transmitted, into the queue. By expanding the query until there is either
a response or no response — all the tags will respond eventually — the query bit
string traverses from the most significant to the least significant bits for the possible
serial numbers. Contrary to binary trees, a QT imposes simple functions and requires
no state or pointer maintenance by the tag. However, the tag reading delay for the
QT protocol is significantly affected by the numerical distribution of the tag serial
number, in that the reading delay is increased when tags have minimal difference in
their serial numbers.
The QT and its variants has noticeable overhead in terms of the number of bits
that are sent as query argument, i.e., the string bits. The reason is the string bits have
to traverse, from most significant to least significant, equal to the length of the serial
number. This results in communication overhead as redundant bits are transmitted,
2.2. LITERATURE SURVEY 45
both by tags and readers, during the interrogation cycles.
Prefix overhead reduction algorithms [28, 29, 80, 113] maintain prefix and iteration
overhead reduction approaches which enhance the performance of the memoryless
mechanisms. Prefix Randomized Query Tree (PRQT) [113], falls under the prefix
reduction mechanism and overcomes the limitation of the QT-based algorithm, i.e.,
create longer reading cycles due to tag length and its distributions. PRQT similar
to QT is a stateless approach however, and it differs from QT in that it uses prefixes
chosen randomly by tags (rather than using their ID-based prefixes), thereby pro-
viding additional memory space. Each tag generates a random length prefix which
is then used during the interrogation process. The optimal length prefix depends
on an estimate of the number of tags within the interrogation zone. Tag estimation
is proposed in the adaptive optimal incremental prefix length algorithm [29]. The
algorithm begins by setting a small initial prefix length followed by the polling of
all possible prefixes. The initial prefix length is then increased repeatedly until the
collision ratio satisfies a prescribed condition. By reducing the prefix overhead, the
collision detection mechanisms enjoy a faster reading rate and lower overheads with
marginal need of additional memory at the tags.
Probabilistic Schemes
In the probabilistic mechanisms the reader communicates the frame length and the
tag picks a particular slot in the frame for transmission. The reader repeats this
process until all tags have been successfully transmitted at least once. Reader con-
trolled synchronization is necessary as the tag has to transmit within its slotted time
frame. The frame size may be adjusted based on the collision, idle and successful
2.2. LITERATURE SURVEY 46
Statistical Estimation Reservation
Dynamic
Static
Adaptive
Non-adaptive
Probabilistic
Figure 2.18: Taxonomy of the probabilistic TDMA-based schemes.
frame information from a previous interrogation cycle for the subsequent cycle. This
encourages frame adaptability according to tag density and distribution, thereby re-
ducing idle and collision frames. The probabilistic approach is faster in comparison
to the deterministic approach because of its low overhead. However, it may suffer
from tag starvation syndrome.
Generally speaking, and as shown in Fig. 2.18, we classify the probabilistic mech-
anisms into two main groups: static and dynamic. The number of slots in the static
approach is fixed, making it mostly used in instances of low tag density. An example
of static algorithms includes those based on Aloha and framed slotted Aloha. While
a fixed frame size leads to a simple implementation, static algorithms risk certain
inefficiencies. For example, long frames used with a small number of tags, or vice
2.2. LITERATURE SURVEY 47
versa, results in delays and resource underutilization. Such issues are addressed in
dynamic approaches in which the frame size is adjusted based on the collision rate
from tags within an interrogation zone. For example, the Dynamic Framed Slotted
Aloha (DFSA) algorithm [42] determines the frame size based on information such
as the number of slots used to identify the tag and the number of the slots with
collisions.
Variations of the dynamic framed slotted Aloha algorithm differ in frame size by
changing approach. In one such variation, the frame size is adjusted based on some
predefined threshold. For instance, the frame size is increased when the number of
collision slots exceeds a preset threshold. As the collisions decrease, the frame size is
reduced to its previous value. This dynamic adjustment of frame size allows readers
to change its frame size depending on the tag density. That is to say, the frame size
is increased as with an increase in number of tags increases the collisions, and vice
versa.
In another variation, the interrogation cycle begins by setting the initial frame
size to two or four. The frame size is incrementally increased with unsuccessful
transmissions until at least one tag transmits successfully. If at least one tag is
successfully identified, the current reading process is aborted and reinitiated from the
beginning at its initial frame size. Despite the general performance of the dynamic
framed slotted Aloha and its variants, changing the frame size only may not be
sufficient given the number of tags, as the frame size cannot be infinitely increased.
For example, in the second variation above, when the number of tags is small the
reader can identify all tags without much collision. If the number of tags is large,
however, the number of slots needs to exponentially increase since the reader always
2.2. LITERATURE SURVEY 48
starts with the initial minimum frame size.
We further classify the dynamic algorithms as adaptive or non-adaptive (see
Fig. 2.18). Adaptive algorithms use statistical information for frame size adjustment.
Adaptive algorithms accommodate tag mobility more than non-adaptive algorithms
because they reduce the probability of tag collisions while simultaneously expediting
the identification of the RFID tags. Dynamic algorithms (both adaptive and non-
adaptive) can be further classified based on their use of reservation, estimation, or
statistical techniques. Estimation based approaches are the most important Aloha
based schemes [21,23,59,66,85,86,97,102] and utilize tag count estimation techniques
to adjust the frame accordingly. The Advanced Framed Slotted Aloha (AFSA) algo-
rithm [102] estimates the number of tags, prior to initiating the reading process, and
adjusts the frame size based on the estimation. However, AFSA has to increase the
frame size indefinitely, which is impractical with high tag density.
The Enhanced Dynamic Framed Slotted Aloha (EDFSA) [66] overcomes the lim-
itation of AFSA by bounding the estimation based on maximum frame size, i.e.,
setting an upper bound on the frame size. EDFSA initially estimates the number of
unread tags using the AFSA estimation function and then partitions the unread tags
into a number of groups. Only one group of tags is allowed to respond at one time.
The number of groups that give output to the maximum throughput for the subse-
quent reading cycle is calculated and adjusted after every interrogation cycle. The
reader transmits the number of tag groups and a random number to the tags when it
broadcasts a request. A tag receiving the request generates a new number from the
received random number and the tag’s own serial number and divides the new number
by the number of tag groups. Only the tags having the remainder of zero respond
2.2. LITERATURE SURVEY 49
to the request. The estimation and grouping of the unread tags are performed after
each read cycle until all the tags have been read. Although EDFSA overcomes the
limitation of AFSA protocol, it has longer reading cycles and additional functionality
at the tag.
Statistical algorithms [30, 43, 44, 59] exploit statistical information to improve the
read time of the RFID systems. Adaptive Slotted Aloha Protocol (ASAP) [59] utilizes
information relative to the tag population, from previous interrogation cycles and
reading processes, to estimate the number of tags presently within the interrogation
zone of the reader. The ML-based estimation algorithm is used for this purpose.
The frame size is adjusted optimally to reflect the tag estimation. The mobility
is supported by accounting for tag arrival and departure rate, while initiating the
estimation and frame adjustment at the beginning of every interrogation cycle. The
statistical algorithm resembles the deterministic anti-collision category and shares the
same pros and cons.
Discussion
Most of the anti-collision protocols discussed so far in this chapter attempt to max-
imize specific gains at the expense of other performance merits. In minimizing col-
lision, for example, an algorithm increases the reading rate. This increase, however,
comes at the expense of additional memory, overhead communication and processing
requirements that may increase the cost of a tag. In what follows, we discuss some
of the performance merits traded off in the design of anti-collision algorithms.
1. Speed: the rate at which tags can be read. This is a general objective that is
sought by most proposals.
2.2. LITERATURE SURVEY 50
2. Overhead: the communication and processing overhead for the tags and the
readers. The communication overhead includes sending additional bits which
otherwise would not be sent. The processing overhead includes additional in-
terrogation cycles or data crunching beside the nominal.
3. State: the amount of state that can be reliably stored on the tag. Additional
storage means additional memory. A reader may maintain statistical informa-
tion but the storage cost at the reader is significantly less than at the tag.
4. Mobility: the ability to accommodate tags which enter and leave an interroga-
tion zone during the interrogation process.
5. Scalability: the ability to accommodate high tag deployment densities.
6. Cost: in terms of additional functionality and memory requirements at both
tags and reader. Increase in the cost also reflects an increase in the device
monetary value.
Table 2.2 and Table 2.3 show the scheme improvements, depicted by a check-
mark (), and the tradeoffs, depicted by a down arrow (↓), for the deterministic
and probabilistic anti-collision algorithms, respectively. A scheme may not affect cer-
tain metric but may do so indirectly. This shown by −. For instance, the prefix
reduction scheme’s main objective is to reduce the communication overhead which
indirectly reduces the time it requires to interrogate the tags, thereby accelerating
the tag reading rate. It is evident from the tables that each of the schemes, either
deterministic or probabilistic, come with a tradeoff and the suitability of each varies
under different circumstances. Therefore, with diverse sets of RIFD applications, no
2.2. LITERATURE SURVEY 51
Approach Speed Overhead Stateful Mobility Scalability Cost
Energy-aware - ↓ ↓
[78, 87, 116]
Hybrid ↓ - ↓
[94]
Memoryless ↓ - -
[31, 63, 77, 116]
Prefix - ↓ - - -
[28, 29, 61, 80, 113]
Statistical - ↓ ↓
[26, 76, 77]
Table 2.2: Comparison of deterministic anti-collision algorithms.
single existing scheme fulfills all performance metrics. For example, in a typical ware-
house scenario, the reader deployed at the docking doors or the conveyor belts is to
interrogate hundreds of mobile tags, or possibly thousands with item-level tagging,
as they move from one section (e.g., docking doors) to another (e.g., sorting) section
of the warehouse. Therefore, it is of utmost importance to be able to read all the
tags encompassing a certain performance metric, as speed, mobility and scalability
matter more than overhead reduction. For such scenarios, the statistical approach
is the most promising, compared to other high-speed reading schemes such as hy-
brid, with its limited scalability or memoryless, with its low support for mobility. An
RFID based access control system does not require scalability and high reading rates,
therefore any generic tree-based algorithm is sufficient.
2.2. LITERATURE SURVEY 52
Approach Speed Overhead Stateful Scalability Cost
Static ↓ - ↓ -
[42]
Reservation ↓ ↓ - ↓
[107]
Estimation ↓ ↓ - -
[21, 23, 58, 66, 85, 86, 102]
Statistical - ↓ ↓
[30, 43, 44, 59]
Table 2.3: Comparison of probabilistic anti-collision algorithms.
2.2.4 Reader Collisions
Readers with intersecting interrogation zones may interfere with each other’s oper-
ation to the extent of barring a reader from identifying tags within its reach. As
with other radio systems, a reader’s operation may also be affected by other read-
ers even if their interrogation zone does not overlap. In either case, a reader colli-
sion results [39, 68]. While the use of multiple frequencies may effectively minimize
reader collisions, RFID tags are limited functionality devices that are incapable of
differentiating between multiple readers and they cannot be mass produced to com-
municate using multiple frequencies. We classify the reader collision schemes, shown
in Fig. 2.19. Multiple mechanisms based on scheduling, coverage and learning are
adopted to resolve collisions.
In the scheduling based approach, frequency and an associated time slot is sched-
uled for the reader. Scheduling can be centralized or distributed, and can support
2.2. LITERATURE SURVEY 53
Coverage
Scheduling Learning
With Tag
support
Without Tag
support
Neural
Network
Online
Learning
Genetic
Algorithms
Reader
Collision
Figure 2.19: Taxonomy of reader collisions schemes.
both static and dynamic frequency assignments. Occasionally, the scheduling is en-
hanced with tag support, i.e., additional data is stored within tags at the expense
of additional memory requirement. However, scheduling without tag support is the
most common approach. Algorithms such as colorwave [103] fall under this cate-
gory. Learning based approaches, on the other hand, are based on hierarchical online
learning, genetic algorithms, and neural network methods. The learning approach
attempts to minimize collision by learning the collision patterns and assigning fre-
quencies based on the learned pattern. Algorithm such as HiQ [51] falls under this
category. Alternate approaches such as the use of beacon channels [15], central co-
operators [105], and handling reader collisions as coverage problems [19] also exist in
literature.
To achieve optimal frequency channel assignment, a pair of distributed algorithms
called Distributed Color Selection (DCS) and Variable-Maximum Distributed Color
Selection (VDCS) are introduced [103]. Both DCS and VDCS fall under the schedul-
ing category. An objective of both algorithms, simply known as colorwave, is to color
each node of the multiple reader networks using multiple colors in such a manner that
2.2. LITERATURE SURVEY 54
the possibility of having two adjacent nodes with the same color is minimized. With
color representing different frequencies, the difference between an adjacent node’s
color could indicate the availability of non-interfering frequencies. Colorwave op-
timizes the minimum number of colors and therefore frequencies required, up to a
predefined maximum value while maintaining a configured successful transmission
rate by the readers. Colorwave is a TDMA based approach, wherein each reader
randomly selects a time slot to transmit using its assigned frequency. Upon collision,
the reader selects a new time slot and informs its adjacent nodes of the selection. If
any neighbor is scheduled to use the same frequency (color), it reinitiates itself, se-
lects a new frequency and informs its neighbor of its actions. The process of selecting
frequency and scheduling time slot is repeated, as required. The maximum allowable
color is increased if the successful transmission rate drops below the configured value.
The colorwave reduces the reader collision up to certain network scales, as in practice
only a limited numbers of frequencies are available that can be utilized.
The Hierarchical Q-Learning (HiQ) algorithm [51] is a hierarchical, online learn-
ing algorithm that finds the dynamic solution to reader collision problems. HiQ,
similar to Colorwave, tries to minimize the reader-to-reader collisions by assigning
frequencies to each reader over time. The frequency assignment however, is based on
learning the collision patterns among neighbors and then assigning the optimal fre-
quency. The optimal or near-optimal frequency assignment is achieved by repetitive
environment interaction, i.e., collision patterns and estimation. HiQ utilizes three
basic hierarchical tiers, in some form already present in the existing RFID systems,
namely: readers, R-servers, and Q-servers. At the lowest tier are the RFID readers.
The RFID reader communicates using prescheduled frequency in the reserved time
2.2. LITERATURE SURVEY 55
slots. The reader detects the collisions amongst the neighboring readers, i.e., the read-
ers with overlapping interrogation zones. This statistical information, as explained
earlier, is utilized for optimal frequency assignments; the decision being made by up-
per tiers, i.e., the reader-level server (R-server). A one-on-one relation exists between
the low-level reader and the R-server. The Q-learning server (Q-server) is the high-
est tier and is responsible for resource allocation, by finding an optimum scheduling
policy based on the underlying readers’ learning experience or collision information.
The HiQ learning algorithm is distributed at lower-level (readers) with centralized
scheduling at the top-most tier. HiQ provides optimal scheduling, resulting in higher
throughput, but at the cost of a centralized solution.
Other learning based approaches include neural networks and genetic algorithms.
A neural network approach is a fixed channel solution where each neuron represents
an RFID reader. The constraint and optimization criteria of the system are translated
into an energy function, which the neural network tries to minimize. The outcome
of these neurons determines if the channel may be usable for the represented RFID
reader. Genetic algorithms are a form of blind local search. In the genetic algorithm
approach to the channel assignment, a single solution is an assignment scheme for all
RFID readers. The genetic algorithm takes a set of solutions as the population and
begins the evolutionary process, which is repeated until a solution is found with no
interference.
The learning and scheduling approaches are mainly targeted towards reader-to-
reader collision. Alternate procedures such as beacon channels [15] and central co-
operators [105] are used to tackle multiple readers-to-tag collisions. The distributed
2.2. LITERATURE SURVEY 56
protocol, Pulse [15], is based on the beaconing mechanism. The reader, while read-
ing the tags within its interrogation zone, periodically broadcasts a beacon using the
control channel. Any other reader that wants to read the tags, i.e., tags within the
overlapping interrogation zone, has to ensure that no beacon has been transmitted on
the control channel before it can initiate its interrogation process. The latter reader
has to wait until the first reader has executed its interrogation process because the
control channel is idle for a specific time period before triggering its own beacon.
The pulse protocol however, is inclined toward eliminating the collision caused by
mobile RFID readers. In the central cooperative approach [105], a centralized device
is responsible for the multiplexing of tags’ data to multiple readers. Using a control
cooperator, the present ’multiple points to multiple points’ collision problem (caused
when multiple readers are trying to access multiple tags) translates into the classical
problem of ’multiple point to one point’ collision. The central cooperator multiplexes
multiple reader requests into a single request for the tags. The responses from the tags
are de-multiplexed to the individual readers separately. The central cooperator pro-
vides an affective approach in handling multiple readers-to-tag collisions. However,
it requires a device with all the functionality of a conventional RFID reader.
The reader collision avoidance schemes include frequency assignment scheduling
and online learning approaches to avoid reader collisions. Table 2.4 shows a contrast
between various algorithms for resolving the collision problem. The algorithms vary
and have fundamental differences. For instance, some are centrally controlled, i.e.,
using central authority for channel assignment and multiplexing the reader queries.
2.2. LITERATURE SURVEY 57
Centralized Distributed Fixed Dynamic
Method Approach control control Channel Channel
DCS [103] scheduling - -
VDCS [103] scheduling - -
HiQ [51] learning -
Neural learning - -
Annealing [69] learning - -
Genetic learning - -
Redundancy [19] coverage - - -
CC [105] coverage - -
Table 2.4: Comparison of reader collision algorithm.
Reader collision, especially reader-to-reader interference, is similar to a frequency as-
signment problem in conventional wireless communication. However, the fundamen-
tal difference is that RFID tags, especially passive tags, are unable to differentiate
between various readers. Therefore, two readers communicating with the tag must
communicate at different time slots or make use of sessions. According to the EPC
Gen2 standard, the tag shall provide support for up to four sessions, allowing two
or more readers to independently interrogate the tag. The readers, step-by-step, se-
lectively singulate the tags into their respective sessions and move them into other
respective sessions upon completion. This allows multiple readers to interrogate com-
mon populations of tags. Although effective, this requires extra memory at the tag,
increasing the cost.
2.2.5 Coverage
Existing schemes handle coverage and deployment issues for RFID networks in two
approaches: planned or random.
2.2. LITERATURE SURVEY 58
In the planned approach, algorithms are developed to find optimal placement of
readers in terms of maximal coverage and minimum number of readers, and under
some constraints on where readers can be located [25, 50, 83, 92, 93, 104, 110]. In the
optimal grid coverage approach [83], readers are deployed in a grid where the distance
between two neighboring readers is determined by the interrogation range of readers.
After deployment, the reader(s) not covering any tag can be turned off. Another
similar approach is the honey grid [110] in which the readers are deployed in rings
of different sizes and centred at the centre of the interrogation zone. When rings are
numbered based on their radii, the i
th
ring contains 6i readers. Similar to the regular
grid approach, after deployment, readers not covering any tags can be turned off. It
has been proven that both schemes provide maximal possible coverage. However, that
comes at the cost of large number of readers, significant readers collisions and unfair
load distribution. Some reader deployment schemes that take into account tags and
readers orientation have been proposed [104]. The scheme proposed in [104] finds an
optimal number of readers, their locations out of a set of pre-fixed locations and their
antennas orientation to maximize readability. The scheme, however, is not generic;
it is only being proposed for portal accuracy in supply chain management scenarios.
Furthermore, like other schemes in the literature, it considers only placement without
load balancing.
In the random deployments, algorithms are devised to find redundant readers in a
given deployment strategy [16,52,53]. In the RRE scheme [16], each reader broadcasts
a message carrying its own identity and its tag count, which is the number of tags
within its region. Each tag stores the identity of the reader having the maximum tag
count it received. Access to a particular tag is granted to the reader covering that
2.2. LITERATURE SURVEY 59
tag with the maximum tag count (i.e., the reader of the identity stored in the tag).
Readers that are not granted access to any tag are marked redundant and are turned
off. The RRE scheme has a lot of communication overhead as it requires frequent
write operations to the tag. To mitigate this overhead, the LEO and the LEO+RRE
algorithms [52] introduce the concept of ”first-read first-own” principle: an RFID
reader only write its own identity into a tag memory. The reader that manages
to singulate the tag and writes the identity first is granted access. In case of the
LEO+RRE algorithm [52], upon completion of LEO execution, the RRE algorithm
is then used to further eliminate any redundant readers. Various other redundant
reader elimination schemes have been proposed in the literature [53, 111], and they
try to make enhancements upon the RRE and LEO schemes. A load balancing
scheme utilizing tag storage space has also been proposed [37]. The basic idea is
that each reader writes its tags count into tag memory space, and the tag picks and
responses back to the reader with the lowest load. Various other middleware-based
load balancing schemes have also been proposed in the literature [24].
Coverage, in context of node placement, has also been formulated as art gallery
problem, Oceans coverage problem, coverage for robots and coverage in wireless sen-
sor networks. In the art gallery problem [84], the problem translate into finding the
minimum set of locations for security guard, inside a polygonal art gallery, such that
the gallery is visible by at least one of the guard. Various randomized algorithms
have been proposed to solve the art gallery problem. The art gallery problem for
the 2D and 3D case is linear and NP-hard, respectively, wherein an approximation
algorithm is presented for the later case [73]. The art gallery based coverage problem
has many real world applications, such as placement of antennas for cellular telephone
2.2. LITERATURE SURVEY 60
companies, and placement of cameras for security purposes in banks and supermar-
kets. However, the analogy of art gallery problem has limited applications for RFID
systems as they have limited coverage, i.e., reading range.
In ocean coverage problem [48], the authors are interested in monitoring the
Oceans pigments concentrations, using satellites based remote sensing. Using nu-
merical analysis, the results show that merging data from three satellites can increase
ocean coverage by 58% for one day and 62% for four days. The exceptional coverage,
using limited satellites was made possible as the satellite nodes had different orbits
and therefore was able to sense the same ocean location at different times of day.
Such a multi-sensing solutions, due to large associated cost, is justifiable only for
large coverage (e.g., Oceans) and hence, can not applied directly to RFID systems.
The coverage concept with regard to the many-robot systems was introduced
in [47]. The author herein define three types of coverage: blanket coverage, barrier
coverage, and sweep coverage. In the blanket coverage, the goal is to achieve a static
arrangement of sensors that maximizes the total detection area. In barrier coverage
the goal is to achieve a static arrangement of nodes that minimizes the probability of
undetected penetration through the barrier, whereas the sweep coverage is more or
less equivalent to a moving barrier. New applications arise in the context of mobile
WSNs, where sensors have locomotion capabilities. Thus, the nodes can spread out
such that the area covered by the network is maximized and can relocate to handle
sensor failures. These solution can be adopted to mobile RFID readers however, is
an open research direction.
In wireless sensor networks, deployment of nodes are subject to multiple objectives.
The main objectives are to minimize the number of nodes, extending the network
2.2. LITERATURE SURVEY 61
lifetime [35] and increasing the network fidelity [114]. These objectives are addressed
in the context of area coverage, point coverage and barrier coverage. The most studied
coverage problem is the area coverage problem, where the main objective of the
sensor network is to cover (monitor) an area, using both centralized and distributed
algorithms. Various energy efficient variations of the area coverage also have been
proposed in the literature. These variation utilizes diverse techniques such as node
duty cycling and node scheduling, connected random coverage, deterministic coverage
and k-coverage. The main objective is to minimize the number of nodes required to
maintain certain coverage while maintaining network connectivity.
The point coverage scenario considers a limited number of points (targets) with
known location that need to be monitored. A large number of sensors are dispersed
randomly in close proximity to the targets; mainly is used for surveillance applications.
In such applications, requirement is that every target must be monitored at all times
by at least one sensor. In [20] authors proposed a solution wherein the disjoint
sets are modelled as disjoint set covers, such that every cover completely monitors
all the target points. The authors prove that the disjoint set coverage problem is
NP-complete and and propose an efficient heuristic for set covers computation using
a mixed integer programming formulation. In [56], authors consider the scenario of
deterministic point coverage, i.e., the possibility for explicit placement of set of sensor
nodes at certain points. Given these set of points, the objective is to determine a
minimum number of sensor nodes and their location such that the given points are
covered and the sensors deployed are connected. In the barrier coverage [74, 89], the
goal is to minimize the probability of undetected penetration through the barrier, i.e.,
the sensor network. Such coverage involve three tiers mobility, i.e., target mobility,
2.3. SUMMARY 62
sensor mobility, and data collector mobility.
Unlike RFID, the WSNs node deployment and hence, coverage, is very application
dependant. However, similar to RFID, the most common goal is to reduce the num-
ber of deployed sensor node while maintaining an acceptable coverage of the event
sensing area. Therefore the solutions from the wireless sensor network coverage can
be adopted for RFID systems however, is a largely unexplored research area.
2.3 Summary
In this chapter we present the background and literature survey on medium access
control and coverage in RFID systems. Energy harvesting and backscattering modula-
tion are the main foundation technologies of low-cost identification solutions provided
by RFID. The energy harvesting yields battery-less RFID tags which enables mini-
mal maintenance, longer lifetime and low cost RFID solutions. The backscattering
yields transceiver-less RFID tags which enable minimal power and low cost RFID
solutions. RFID system employs a centralized communication architecture wherein
the reader simultaneously interrogates and process the tags, within its reading range.
This centralized and simultaneous interrogations result into collisions.
Due the simplicity of the RFID tags, it is not possible to use the conventional
wireless medium access and collision avoidance protocols. To this extend, RFID em-
ployes deterministic slotted aloha and probabilistic framed slotted aloha anti-collision
algorithms. In the deterministic algorithms, the reader splits and identifies a set of
tags to respond in a given time. Splitting is based on contention information obtained
from the previous interrogation cycle and attempts to reduce contention for the next
cycle. In the probabilistic algorithms, the reader communicates the frame length, and
2.3. SUMMARY 63
the tag, randomly transmits a particular slot in the frame. The frame size may be
adjusted, based on the information from the previous interrogation cycle, encouraging
adaptability according to tag density and distribution. Under certain circumstances,
the deterministic algorithms may take a considerably longer time. On the other hand,
the probabilistic algorithms are fast, due to its low overhead, but may suffers from
tag starvation syndrome.
64
Chapter 3
Power Control in RFID
3.1 Introduction
The focus of our work in this chapter is to overcome tags-to-reader collisions. While
several schemes have been proposed to deal with tag collisions in RFID systems,
the interrogation delay is still a problem for some applications that involve dense
and/or fast moving passive tags. This causes immense data collisions at the reader.
A promising opportunity to avoid a significant amount of collisions is to partition
the interrogation zone spatially into smaller clusters, and to have tags in each cluster
being read separately (i.e., one cluster at a time). Any existing anti-collision scheme
can be used to resolve collisions in a single cluster. Such a scheme should reduce the
number of collisions as it reduces the number of tags that may respond at the same
time. To the best of our knowledge, the proposed Power-based Distance Clustering
(PDC) scheme [10] is the first to exploit such an avoidance approach. In PDC,
the reader tunes transmission power so that tags within the interrogation zone are
clustered based on their distance from the reader. Tags which are being interrogated
within the current cluster will not respond to the reader’s queries for the subsequent
3.1. INTRODUCTION 65
clusters; as once read, these tags are forced into a sleep mode.
We studied the viability and efficiency of the PDC scheme in [10]. However, it was
not clear how to find the best partitioning scheme. Indeed, having too many clusters
may result in many empty clusters, which is an extra overhead, and few clusters
may result in having crowded clusters; both situations affect the performance of that
approach significantly.
The contributions of this chapter are as follows. We formulate two optimization
problems for finding the clustering scheme that minimizes the number of collisions
and, hence, reduces the interrogation delay. The first problem targets an ideal set-
ting in which the transmission power and, hence, the transmission range of the RFID
reader can be tuned with high precision. In the second problem, we relax this as-
sumption and consider an RFID reader with finite, discrete transmission ranges. For
each optimization problem, we present a mathematical delay analysis and devise an
efficient method to find the optimal clustering scheme. Our proposed methods have
been designed to adapt to different system environments such as the number of tags
and their distribution. We also show the results of several experiments, using the
ns-2 simulator [7], which verify the effectiveness and performance improvements of
the PDC scheme in terms of number of collisions, reading rates, and delay.
The rest of this chapter is organized as follows. Section 3.2 introduces our PDC
scheme. Section 3.3 presents a mathematical analysis and an optimal algorithm for
clustering the interrogation zone of an RFID reader whose transmission range can
be tuned with high precision. Section 3.4 presents a mathematical analysis and an
optimal algorithm for clustering the interrogation zone of an RFID reader with a finite
number of transmission ranges. Section 3.5 presents the results of our experiments.
3.2. THE POWER-BASED DISTANCE CLUSTERING (PDC)
ANTI-COLLISION SCHEME 66
Finally, Section 3.6 summarizes our work.
3.2 The Power-based Distance Clustering (PDC) Anti-Collision Scheme
The PDC is a divide-and-conquer anti-collision scheme in which tags are divided into
clusters based on their distance to the reader. Tags in different clusters are then read
separately (one cluster at a time). This has the potential to reduce the number of tags
which can be concurrently active, lower the collision probability, and, hence, expedite
the interrogation process. Partitioning the interrogation zone can be achieved by
controlling the readers antenna power level. The reflected power density and the
reader range can be computed using the following formulas [42] [115]:
S =
λ
2
.P
reader
.G
Reader
.G
Tag
(4π)
2
R
4
(3.1)
and
R =
λ

.
4
¸
k.P
reader
.G
2
Reader
.G
2
Tag
P
back
(3.2)
where S is the reflected power density, λ is the wavelength of the emitted electromag-
netic wave, P
reader
is the power supplied to the reader’s antenna, R is the distance
between the reader and the tag, G
Reader
and G
Tag
are respectively the antenna gain
for the reader and the tag, and P
back
is the power received by the reader from the tag.
It follows from (3.1) that the power density reflected back by the antenna is propor-
tional to the fourth root of the power transmitted by the reader. Also it follows from
(3.2) that the reading distance between the tag and the reader can be changed by
changing the power supplied to the reader’s antenna while maintaining P
back
. Hence,
the interrogation range of the reader can be reduced by lowering P
reader
.
3.2. THE POWER-BASED DISTANCE CLUSTERING (PDC)
ANTI-COLLISION SCHEME 67
An example of the partitioning is shown in Fig. 3.2, where the interrogation zone
is divided into three clusters: D
1
, D
2
, and D
3
. When the reader sends a request
to a particular cluster, only those tags in that particular cluster may respond. For
example, assume that the reader just sent a request (query) to cluster D
2
after it
had finished reading tags from cluster D
1
. Then only tags from cluster D
2
(i.e., tags
marked as t
2
) will respond to that request. After all tags marked as t
2
have been read
successfully, they are put into a sleep mode. Then, the reader transmission range is
increased to cover cluster D
3
and a new request is sent. Now only tags from cluster
D
3
(i.e., tags marked as t
3
) will respond to the reader requests. This identification
process continues until the reader reaches its maximum interrogation range. The flow
diagram of the PDC scheme is illustrated in Fig. 3.1. Any anti-collision scheme can
be used to resolve collisions within a single cluster.
The scheme is sensitive to the size and the number of clusters. A large number of
clusters in a sparse tag environments yields longer delays as a result of many empty
clusters and idle cycles. On the other hand, a small number of clusters may result in
having too many tags in one cluster which renders the scheme ineffective, especially
in dense tag environments.
The Static PDC, which we proposed in [10], divides the interrogation zone into
clusters based on a fixed stepping value d. This means that the transmission range
of a cluster is more than that of the previous cluster by a fixed value d. While very
simple, this scheme does not divide the interrogation zone into equal-size clusters and
does not adapt to different tags densities. However, it has the ability to be integrated
with any anti-collision scheme and it has shown acceptable improvements as shown
in Section 3.5.
3.3. OPTIMAL POWER-BASED DISTANCE CLUSTERING 68
Figure 3.1: Flow diagram of the PDC scheme.
3.3 Optimal Power-based Distance Clustering
In this section we present our theoretical analysis and methodology for finding the
Optimal Power-based Distance Clustering (O-PDC) scheme. This scheme targets
RFID readers whose transmission power can be tuned with a high precision so that
the transmission range can be controlled with high accuracy.
3.3.1 System model and problem definition
We consider an RFID system consisting of an RFID reader and n passive tags. The
interrogation zone is modelled as a circle centred at the RFID reader with a radius
of R units; the RFID reader can communicate with, and read, all tags located within
3.3. OPTIMAL POWER-BASED DISTANCE CLUSTERING 69
Reader
T
T
T
T
T’
T’
T’
T’’
T’’
T’’
T’’
T’
T’’
T’’
d
d’
d’’
Figure 3.2: An example of an interrogation zone of 3 clusters.
its interrogation zone.
The interrogation zone is divided into k equal size clusters, and tags in each cluster
will be read separately (existing protocols use k = 1). In general, any anti-collision
protocol can be used to resolve collisions in a particular cluster. The RFID reader
and tags will go through several cycles during the reading process; in each cycle the
RFID reader sends a request and zero, one, or more tags respond to the request by
sending their IDs. An idle cycle is one in which no tag responds, a successful cycle
is one in which exactly one tag responds, and a collision cycle is one in which two
or more tags respond. Upon detecting a collision, the RFID reader will send a more
restricted request which excludes some of the colliding tags. Eventually, a tag will
be read successfully. This process will be repeated until all tags are read successfully.
Now, the problem can be defined as follows:
3.3. OPTIMAL POWER-BASED DISTANCE CLUSTERING 70
Find the optimal number of clusters that minimizes the total number of
cycles required to read all tags.
3.3.2 Assumptions
To resolve collisions within a single cluster, we use the Query Tree (QT) protocol [64].
Nevertheless, any existing tree-based protocol can be equally used for that purpose.
We also assume that tags are uniformly distributed over the interrogation zone. We
also assume that the total number of tags is known to the reader; several schemes in
the literature (e.g., [62]) are able to find a precise estimation to the number of tags
in a negligible time.
3.3.3 Delay analysis
Let f(n) denote the number of cycles required to read a set of n tags, and let E[f(n)]
denote the expected value for f(n). For the QT protocol, it has been shown in [64]
that for n ≥ 4,
2.881n −1 ≤ E[f(n)] ≤ 2.887n −1 (3.3)
To generalize our analysis to other tree-based anti-collision schemes (e.g., the recent
work in [95]
1
), we will have:
E[f(n)] = c n −1, (3.4)
where c is a constant.
Now, assume that the interrogation zone is divided into k equal size clusters, and tags
are uniformly distributed over the interrogation zone. Let g(n, k) denote the number
of cycles required to read a set of n tags uniformly distributed over k clusters and
1
The scheme proposed in [95] has f(n) = 2 n −1.
3.3. OPTIMAL POWER-BASED DISTANCE CLUSTERING 71
using distance-based clustering. Thereby, we deduce the following lemma.
Lemma 1. If we have n tags, and the interrogation zone is divided into k equal size
clusters, then
E[g(n, k)] = c n −k + 2E[s], (3.5)
where s is the number of empty clusters (i.e., clusters which do not contain any tag).
Proof. We have s empty clusters and t nonempty clusters. Let n
i
denote the number
of tags in the i
th
nonempty cluster. For each empty cluster, we will have a single idle
cycle; this results in a total of s cycles. The expected number of cycles required to
read tags in the i
th
nonempty cluster is c n
i
− 1. Therefore, the expected number
of cycles required to read tags in all nonempty clusters is c n − t. Therefore, the
expected total number of cycles is c n−t +s, but t = k−s, so we get c n−k+2s.
Now, we deduce the following lemma for the expectation of the number of empty
clusters.
Lemma 2. If tags are uniformly distributed over the interrogation zone, then
E[s] = k
_
k −1
k
_
n
(3.6)
Proof. Let P
i
denote the probability that the i
th
disk is empty (i.e., no tag is located
3.3. OPTIMAL POWER-BASED DISTANCE CLUSTERING 72
in that cluster). Then,
E[s] =
k

i=1
P
i
(3.7)
=
k

i=1
_
k −1
k
_
n
(3.8)
= k
_
k −1
k
_
n
(3.9)

The following theorem is a direct result from Lemma 1 and Lemma 2.
Theorem 1. If we have n tags uniformly distributed over k equal size clusters, then
E[g(n, k)] = c n −k + 2k
_
k −1
k
_
n
(3.10)
3.3.4 Optimizing the delay
The objective of this sub-section is to find the optimal number of clusters (i.e., the
value of k that minimizes E[g(n, k)]). We start with the following lemma.
Lemma 3. E[g(n, k)] is a convex function over the interval [1, ∞).
Proof. E[g(n, k)] is twice differentiable with respect to k over the interval (0, ∞), and
d
2
dk
2
E[g(n, k)] =
2n
k
3
(n −1)
_
k −1
k
_
n−2
(3.11)
Since n ≥ 1 and k ≥ 1,
d
2
dk
2
E[g(n, k)] is non-negative. Therefore, E[g(n, k)] is convex
over the interval [1, ∞).
3.3. OPTIMAL POWER-BASED DISTANCE CLUSTERING 73
From lemma 3, we know that the optimal value for k is the one at which
d
dk
E[g(n, k)] =
0. Therefore, we need to solve the following equation.
−1 + 2
_
k −1
k
_
n
+
2n
k
_
k −1
k
_
n−1
= 0 (3.12)
Let x
opt
denote the solution to equation 3.12. In general, there is no closed-form
solution to equation 3.12. However, it can be solved by numerical methods. Moreover,
since the number of clusters is integer, we just need to find an integer x
int
, such that
x
int
≤ x
opt
≤ x
int
+ 1; the optimal number of clusters is either x
int
or x
int
+ 1. The
following lemma shows that 1 ≤ x
int
≤ n −1.
Lemma 4. When there are n tags, the optimal number of clusters is at most n.
Proof. To prove lemma 4, it suffices to show that the expected number of cycles with
n clusters is less than that with n+i clusters, where i ≥ 1, (i.e., we need to show that
E[g(n, n)] < E[g(n, n + i)]). With n + i clusters, we are certain (with probability 1)
that at least i clusters are empty, which results in i idle cycles. Thus,
E[g(n, n + i)] = E[g(n, n + i)[at least i clusters are empty]
= i + E[g(n, n)]
> E[g(n, n)] (3.13)

Since E[g(n, k)] is convex and the optimal number of clusters is at most n,
d
dk
E[g(n, k)]
is a non-decreasing function. Therefore, one can find the optimal solution through
binary search over the set ¦1, 2, , n¦. Algorithm 1 finds the optimal number of
3.3. OPTIMAL POWER-BASED DISTANCE CLUSTERING 74
clusters in O(log n).
Algorithm 1: Optimal Clustering.
Function Derivative(n,i) 1
Input: n: the number of tags.
i: an integer between 1 and n.
Output:
d
dk
E[g(n, k)] at k = i.
begin 2
return −1 + 2
_
i−1
i
_
n
+
2n
i
_
i−1
i
_
n−1
; 3
end 4
5
Function FindOptimal(n)
Input: n: the number of tags.
Output: the optimal number of clusters.
begin 6
left = 1; 7
right = n; 8
while right-left>1 do 9
if Derivative(n, ¸
left+right
2
|) < 0 then 10
11
left = ¸
left+right
2
|;
else 12
right = ¸
left+right
2
|; 13
end 14
end 15
if [Derivative(n, left)[ < [Derivative(n, right)[ then 16
optimal = left; 17
else 18
optimal = right; 19
end 20
return optimal ; 21
end 22
3.4. OPTIMAL DISCRETE POWER-BASED DISTANCE
CLUSTERING (OD-PDC) 75
3.4 Optimal Discrete Power-based Distance Clustering (OD-PDC)
In the previous section, we presented an exact optimal scheme that is suitable for
an ideal setting in which RFID tags are uniformly distributed and the RFID reader
has a precisely tuneable transmission range. In this section we present a near-optimal
scheme that suits RFID readers with discrete transmission ranges (i.e., a finite number
of transmission ranges). To the best of our knowledge, this applies to all today’s
commercial RFID readers (e.g., SkyeTek’s M10 RFID reader [8]). Furthermore, the
scheme we present here is not limited to any particular distribution for the tags nor
to any particular anti-collision protocol to resolve collisions in single clusters. We
present a dynamic-programming algorithm that finds a near-optimal set of clusters
for this setting.
3.4.1 System model and problem definition
We consider an RFID system consisting of an RFID reader and n passive tags. The
interrogation zone is modelled as a circle centred at the RFID reader with a radius
of R units; the RFID reader can communicate with, and read, all tags located within
its interrogation zone.
The RFID reader has k discrete transmission ranges Tr
1
, Tr
2
, . . . , Tr
k
; where
Tr
1
< Tr
2
< < Tr
k
= R. We also use Tr
0
= 0 to denote a virtual transmission
range. Each transmission range may and may not be used in interrogating RFID
tags; this makes it possible to divide the interrogation zone into up to k clusters. A
cluster is the area between two consecutive active (i.e., used) transmission ranges as
shown in Fig. 3.3. For example, when k = 3, we can have 3 clusters if all transmission
ranges are active. We may also have two clusters if only Tr
1
and Tr
3
are active, and
3.4. OPTIMAL DISCRETE POWER-BASED DISTANCE
CLUSTERING (OD-PDC) 76
Inactive transmission range
Active transmission range A cluster
RFID reader RFID tag
Figure 3.3: An example of 3 transmission ranges and 2 clusters.
we may have only one cluster if only Tr
3
is active. Note that Tr
k
must be active
in order to cover the whole interrogation zone. A clustering scheme is defined by
a subset of transmission ranges to be active. A clustering scheme defines the set of
clusters used in reading all tags in the interrogation zone. Now, the problem can be
defined as follows:
Find a clustering scheme that covers the whole interrogation zone, such
that the total number of cycles required to read all tags is minimized.
3.4. OPTIMAL DISCRETE POWER-BASED DISTANCE
CLUSTERING (OD-PDC) 77
3.4.2 Assumptions
To resolve collisions within a single cluster, any anti-collision scheme can be used.
We assume that the distribution of tags is known; yet we do not assume a particular
distribution. In fact all what is required by this scheme is the probability that a par-
ticular cluster is empty of tags; a detailed distribution does not have to be available.
We also assume that the total number of tags is known to the reader.
3.4.3 Delay analysis
This subsection explains how each cluster decreases or increases the total number of
cycles. A cluster is defined by a pair of transmission ranges that bound it (e.g, the
cluster (i, j) is the area between Tr
i
and Tr
j
). It is obvious that more clusters results
in less collisions; we can for example add more clusters until each cluster has at most
one tag and, hence, there will be no collisions. However, increasing the number of
clusters arbitrarily results in having many empty clusters and, hence, additional idle
cycles. Based on this observation, we should maximize the number of non-empty
clusters and minimize the number of empty clusters. While the negative effect of an
empty cluster is straight forward (which is one additional idle cycle), quantifying the
positive effect of a non-empty cluster depends heavily on the anti-collision scheme used
within single clusters. To make a general optimization scheme, we assume that an
empty cluster adds one extra idle and a non-empty cluster saves d cycles. Therefore,
the objective is:
MAX E[α]d −E[β], (3.14)
where α is the number of non-empty clusters and β is the number of empty clusters.
For some anti-collision schemes, finding the exact value of d is trivial. For example,
3.4. OPTIMAL DISCRETE POWER-BASED DISTANCE
CLUSTERING (OD-PDC) 78
the binary anti-collision algorithm with backtracking has d = 1 [95]. On the other
hand, it is not easy to find the exact value of d for some probabilistic anti-collision
schemes. Nevertheless, we can use d = 1 to reflect the objective of maximizing the
number of non-empty clusters and minimizing the number of empty clusters regardless
of the anti-collision scheme being used to resolve collisions within single clusters.
Each cluster will contribute to the objective function in (3.14). Let’s give each
cluster a rank based on its expected contribution to the objective function, and let
h(i, j) denote the rank of a cluster (i, j). h(i, j) can be computed as follows.
h(i, j) = P(n
(i,j)
> 0) ∗ d −P(n
(i,j)
= 0), (3.15)
where n
(i,j)
is the number of tags located in the cluster (i, j) and P(e) is the probability
of an event e, which can be computed based on the distribution of tags. This is based
on the fact that a non-empty cluster saves d cycles and an empty one adds an extra
idle cycle. When the rank of a cluster is negative, it means that the cluster is expected
to add an extra cycle rather than to save cycles. The rank of a clustering scheme
cl, which is denoted by H(cl), is the sum of the ranks of clusters composing cl. The
optimal clustering scheme is one with the maximum rank.
3.4.4 Optimal clustering algorithm
In this sub-section we present an algorithm that finds the optimal clustering scheme
(a clustering scheme is defined by a subset of transmission ranges to be active). In
general when there are k transmission ranges, there will be 2
k−1
possible clustering
schemes that cover the whole interrogation zone; it is actually equivalent to the
number of all subsets of a set of k −1 elements. We present a dynamic programming
3.5. SIMULATION SETUP AND RESULTS 79
algorithm to solve this problem. This algorithm finds the clustering scheme whose
rank is maximum.
We start with some definitions. Let CL[i] denote the set of all clustering schemes
in which Tr
i
is active and Tr
j
is inactive for all j > i. Let M[i] denote the maximum
rank of a clustering scheme in CL[i] (i.e., M[i] = MAX
cl∈CL[i]
H(cl)). It is obvious
that when we have k transmission ranges, CL[k] will belong to the optimal clustering
scheme. The algorithm is described in Algorithm 2.
Algorithm 2: Optimal Discrete Clustering.
Function Find Optimal (n, k) 1
Input: n: the number of tags.
k: the number of transmission ranges.
Output: The optimal clustering.
M[0] = 0; 2
for i = 1 to k do 3
Next[i] = 0; 4
M[i] = h(i, 0); 5
for j = 1 to i −1 do 6
if M[j] + h(i, j) > M[i] then 7
M[i] = M[j] + h(i, j); 8
Next[i] = j; 9
end 10
end 11
end 12
return (M[k], Next[ ]); 13
3.5 Simulation Setup and Results
In this section, we present the results of a simulation based study we conducted
to evaluate the performance of the PDC schemes. We investigate the performance
improvements that can be made by the three PDC schemes presented: the static PDC
3.5. SIMULATION SETUP AND RESULTS 80
which is presented in Section 3.2, the Optimal PDC (O-PDC) which is presented in
Section 3.3, and the Optimal Discrete PDC (OD-PDC) which is presented in Section
3.4. We study the performance of these schemes by generating random RFID networks
(topologies) and comparing the performance of a particular pure classical anti-collision
scheme (i.e., without PDC) with that of the PDC scheme integrated with the same
classical scheme. We may, for example, compare the performance of the pure QT
scheme with that of the O-PDC integrated with the QT scheme (i.e., the O-PDC
scheme withe the QT scheme being used to resolve collisions within single clusters).
3.5.1 Simulation Setup
We extended the ns-2 simulator [7] to implement RFID. We generate random RFID
topologies in which tags are uniformly distributed in a grid of 20 20 m
2
. A single
reader is located at the centre of the grid. The reader has a maximum transmission
range of 10 m. For the PDC scheme, we have a stepping value of 0.5 m (i.e., d =
0.5 m). For the OD-PDC, we have 30 transmission ranges. Tags have randomly
generated IDs. In each RFID network, an anti-collision scheme is applied and its
operation continues until all tags are successfully identified by the reader. We use
several performance metrics to evaluate and compare different schemes. The main
metric is the total number of queries (cycles) needed to identify all tags, which is a
direct indicator to the performance of different schemes. We also consider the ratio
of successful queries, the ratio of collision queries, and the ratio of idle queries. The
results are averaged over 20 randomly generated topologies.
Our PDC schemes are compared with three existing schemes, namely the QT
scheme, the EBST scheme, and the EPC Q-algorithm. To compare the performance
3.5. SIMULATION SETUP AND RESULTS 81
of our PDC schemes with these existing schemes, we show, for each performance
metric, the result of the existing scheme without any clustering and the results of
the PDC schemes when integrated with that particular existing scheme. This clearly
shows the improvements that can be achieved using different PDC schemes. The
O-PDC scheme is compared with the QT and EBST schemes only because they meet
the assumptions made for the O-PDC scheme. On the other hand, the PDC scheme
and the OD-PDC scheme do not have any assumptions on the classical anti-collision
scheme being used to resolve collisions in single clusters and, hence, they are compared
with the QT scheme, the EBST scheme, and the Q-Algorithm scheme.
3.5.2 Simulation Results
Total number of queries
The main indicator for the efficiency of an anti-collision scheme is the total number
of queries needed to identify all tags. Fig. 3.4 shows the improvements made by the
PDC schemes when integrated with the QT scheme as compared to the pure QT
scheme. The O-PDC scheme saves around 35% of the queries. The OD-PDC scheme
saves around 20% of th queries. Fig. 3.5 shows the results of the PDC schemes when
integrated with the EBST scheme. The O-PDC scheme saves around 25% of the
queries. The OD-PDC scheme saves around 10% of the queries. Fig. 3.6 shows the
results of the PDC schemes when integrated with the Q-Algorithm scheme. The OD-
PDC scheme saves around 25% of the queries. The static PDC scheme saves around
10% of the queries.
3.5. SIMULATION SETUP AND RESULTS 82
500
1000
1500
2000
2500
3000
3500
4000
100 200 300 400 500 600 700 800 900
N
u
m
b
e
r

o
f

Q
u
e
r
i
e
s
Number of Tags
QT
PDC(QT)
O-PDC(QT)
OD-PDC(QT)
Figure 3.4: Total number of queries with the QT scheme.
200
400
600
800
1000
1200
1400
1600
1800
100 200 300 400 500 600 700 800 900
N
u
m
b
e
r

o
f

Q
u
e
r
i
e
s
Number of Tags
EBST
PDC(EBST)
O-PDC(EBST)
OD-PDC(EBST)
Figure 3.5: Total number of queries with the EBST scheme.
3.5. SIMULATION SETUP AND RESULTS 83
0
500
1000
1500
2000
2500
3000
100 200 300 400 500 600 700 800 900
N
u
m
b
e
r

o
f

Q
u
e
r
i
e
s
Number of Tags
Q-Algo
PDC(Q-Algo)
OD-PDC(Q-Algo)
Figure 3.6: Total number of queries with the Q-Algorithm scheme.
Ratio of successful queries
To understand the behaviour of different schemes, we look into the ratio of successful
queries to the total number of queries in different schemes. In fact, this metric is
inversely proportional to the total number of queries. This is because the total number
of successful queries should be the same for all schemes; it is actually the same as
the number of tags. However, this metric helps to see how the behaviour of different
schemes changes with different tag densities. Fig. 3.7 shows the ratio of successful
queries achieved by the PDC schemes when integrated with the QT scheme and that
of the pure QT scheme. Fig. 3.8 shows the ratio of successful queries achieved by the
PDC schemes when integrated with the EBST scheme and that of the pure EBST
scheme. It is not a surprise to see constant ratios for the pure QT scheme and for
the pure EBST scheme. The reason is that, for these two schemes, the total number
of queries is a linear function of the number of tags. The nice observation here is the
3.5. SIMULATION SETUP AND RESULTS 84
0.22
0.23
0.24
0.25
0.26
0.27
0.28
0.29
0.3
0.31
100 200 300 400 500 600 700 800 900
R
a
t
i
o

o
f

S
u
c
c
e
s
s
f
u
l

Q
u
e
r
i
e
s
Number of Tags
QT
PDC(QT)
O-PDC(QT)
OD-PDC(QT)
Figure 3.7: Ratio of successful queries with the QT scheme.
stable behaviour of the PDC schemes, and specially the O-PDC scheme. Indeed, this
means that the improvements made by the O-PDC scheme are the same regardless
of the number of tags. Fig. 3.9 shows the ratio of successful queries achieved by the
PDC schemes when integrated with the Q-Algorithm scheme and that of the pure
Q-Algorithm scheme. The probabilistic nature of the Q-Algorithm makes the ratio
of successful queries less stable than those of the QT and the EBST schemes. This
directly affects the stability of the same ratio for the PDC schemes when integrated
with the Q-Algorithm. However, as shown in Fig. 3.9, the improvements made by
the PDC schemes stay significant regardless of the tags densities.
Ratio of collision queries
Fig. 3.10 shows the ratio of collision queries observed using the PDC schemes when
integrated with the QT scheme and that of the pure QT scheme. Fig. 3.11 shows
3.5. SIMULATION SETUP AND RESULTS 85
0.5
0.52
0.54
0.56
0.58
0.6
0.62
100 200 300 400 500 600 700 800 900
R
a
t
i
o

o
f

S
u
c
c
e
s
s
f
u
l

Q
u
e
r
i
e
s
Number of Tags
EBST
PDC(EBST)
O-PDC(EBST)
OD-PDC(EBST)
Figure 3.8: Ratio of successful queries with the EBST scheme.
0.33
0.34
0.35
0.36
0.37
0.38
0.39
0.4
0.41
0.42
100 200 300 400 500 600 700 800 900
R
a
t
i
o

o
f

S
u
c
c
e
s
s
f
u
l

Q
u
e
r
i
e
s
Number of Tags
Q-Algo
PDC(Q-Algo)
OD-PDC(Q-Algo)
Figure 3.9: Ratio of successful queries with the Q-Algorithm scheme.
3.5. SIMULATION SETUP AND RESULTS 86
0.27
0.28
0.29
0.3
0.31
0.32
0.33
0.34
100 200 300 400 500 600 700 800 900
R
a
t
i
o

o
f

C
o
l
l
i
s
i
o
n

Q
u
e
r
i
e
s
Number of Tags
QT
PDC(QT)
O-PDC(QT)
OD-PDC(QT)
Figure 3.10: Ratio of collision queries with the QT scheme.
the ratio of collision queries achieved by the PDC schemes when integrated with the
EBST scheme and that of the pure EBST scheme. These ratios for the pure QT
and EBST schemes are almost constant for the same reason mentioned earlier for the
constant ratio of successful queries. The PDC schemes also show stable ratios over
varying tag densities. Fig. 3.12 shows the ratio of collision queries achieved by the
PDC schemes when integrated with the Q-Algorithm scheme and that of the pure
Q-Algorithm scheme. Values for this metric has nothing to do with the performance
because they are relative to the total number of queries achieved by different schemes.
Ratio of idle queries
Fig. 3.13 shows the ratio of idle queries achieved by the PDC schemes when integrated
with the QT scheme and that of the pure QT scheme. Fig. 3.14 shows the ratio of
idle queries achieved by the PDC schemes when integrated with the EBST scheme
3.5. SIMULATION SETUP AND RESULTS 87
0.3
0.32
0.34
0.36
0.38
0.4
0.42
0.44
0.46
0.48
0.5
100 200 300 400 500 600 700 800 900
R
a
t
i
o

o
f

C
o
l
l
i
s
i
o
n

Q
u
e
r
i
e
s
Number of Tags
EBST
PDC(EBST)
O-PDC(EBST)
OD-PDC(EBST)
Figure 3.11: Ratio of collision queries with the EBST scheme.
0.36
0.37
0.38
0.39
0.4
0.41
0.42
0.43
0.44
0.45
100 200 300 400 500 600 700 800 900
R
a
t
i
o

o
f

C
o
l
l
i
s
i
o
n

Q
u
e
r
i
e
s
Number of Tags
Q-Algo
PDC(Q-Algo)
OD-PDC(Q-Algo)
Figure 3.12: Ratio of collision queries with the Q-Algorithm scheme.
3.6. SUMMARY 88
0.42
0.425
0.43
0.435
0.44
0.445
100 200 300 400 500 600 700 800 900
R
a
t
i
o

o
f

I
d
l
e

Q
u
e
r
i
e
s
Number of Tags
QT
PDC(QT)
O-PDC(QT)
OD-PDC(QT)
Figure 3.13: Ratio of idle queries with the QT scheme.
and that of the pure EBST scheme. As explained in Chapter 2, the EBST scheme
does not have any idle queries. This makes these ratios for the PDC schemes, when
integrated with the EBST scheme, very low; idle collisions come only from empty
clusters. When the PDC schemes are compared with the QT and the Q-Algorithm
schemes, no significant changes are observed between the PDC schemes and the pure
classical ones. As is the case of the ratio of collision queries, values for this metric has
nothing to do with the performance because they are relative to the total number of
queries achieved by different schemes.
3.6 Summary
In this chapter we introduce the power-based distance clustering scheme for tag col-
lision resolution in RFID systems. The main idea in our approach is to divide tags
3.6. SUMMARY 89
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
100 200 300 400 500 600 700 800 900
R
a
t
i
o

o
f

I
d
l
e

Q
u
e
r
i
e
s
Number of Tags
EBST
PDC(EBST)
O-PDC(EBST)
OD-PDC(EBST)
Figure 3.14: Ratio of idle queries with the EBST scheme.
0.14
0.16
0.18
0.2
0.22
0.24
0.26
0.28
100 200 300 400 500 600 700 800 900
R
a
t
i
o

o
f

I
d
l
e

Q
u
e
r
i
e
s
Number of Tags
Q-Algo
PDC(Q-Algo)
OD-PDC(Q-Algo)
Figure 3.15: Ratio of idle queries with the Q-Algorithm scheme.
3.6. SUMMARY 90
into clusters based on their distance to the reader, and read tags in each cluster sep-
arately. Since the number of tags in a single cluster is less than that in the whole
interrogation zone, the likelihood of a collision is reduced. However, the number of
clusters has to be selected carefully, as having too many clusters results in having
many empty clusters, which causes a significant number of idle cycles. Our approach
finds a balance between reducing the likelihood of collisions and reducing the number
of empty clusters by finding an optimal number of clusters. Theoretical analysis and
simulations have been presented to verify performance improvements of our approach.
Moreover, our approach can be integrated with any existing anti-collision scheme and
improve its performance.
91
Chapter 4
Distributed Receiving in RFID
4.1 Introduction
In RFID systems, a single reader broadcasts the un-modulated signal (on forward link)
and receives the backscattered modulated signal from the RFID tags (on backward
link). In the forward-link, the RFID reader broadcasts un-modulated Carrier Wave
(CW) signals and interrogation queries. Each passive tag, in response to the queries,
would harvest its circuitry power from the CW signal incident on its antenna. The
operational power requirement of a passive tag circuitry, anywhere between 2.5µW −
100µW [34, 36, 42, 57] and the maximal allowable CW signal power, in addition to
attenuation, limit the operational range of a passive tag to a few meters. In the
backward-link tags communicate with the reader by reflecting and modulating the
incident CW signal on their respective antennas. The reflected signal strength faces
significant attenuation as it propagates twice the distance between the tag and reader,
hence caps the upper bound on the operational range of the semi-passive tags at some
tens of meters.
4.1. INTRODUCTION 92
Furthermore, the inherited centralized architecture coupled with RFID tags’ in-
ability to sense carrier channels create collisions at the reader. Collisions are the most
prominent hindrance to performance in RFID systems; they lower reading rates, data
rates and jeopardize system operational capacity. Currently, numerous solutions have
been proposed, both at the system-level and protocol-level, to attain maximal oper-
ational performance for RFID systems. For instance, at the system-level, design of
low-power circuits [34], higher gain antennas [36], directional antennas [42], etc. have
been proposed to enhance the operational range of RFID tags. Similarly, at the
protocol-level, energy-aware anti-collision protocols [79, 88, 116], energy-aware mod-
ulation schemes [13, 100], sophisticated anti-collision schemes [12, 42, 96], etc. have
been proposed to overcome collisions.
However, existing and related proposed solutions are based on centralized com-
munication architectures, therefore inherit their limitations. These limitations place
upper bounds on performance, such as reading rate [91], operational range [42], band-
width [36] and reading reliability [104]. Imminent RFID applications (e.g., intelligent
transportation [9]), item-level tagging [14] and mobility demand performance beyond
these bounds. It is anticipated that promising trends in CMOS technologies may
help in mitigation of some root causes. For instance, reducing the tag circuit power
requirements by at least a factor of ten [36], and hence, decoupling passive tags from
the performance of the forward-link channel. However, it will also open the door
to challenges that require new initiatives; passive tags will be constrained by the
backward-link channel performance. In addition, item-level tagging and mobile tags
(potentially numbered in thousands [14]) will also prove impossible to interrogate us-
ing the existing centralized systems, provoking even stronger advocates to tackle the
4.1. INTRODUCTION 93
aforementioned limitations. Indeed, there is a need for new communication models
and singulation approaches for RFID systems.
The focus of our work in this chapter is to lessen the limitations and alleviate the
bottlenecks imposed by centralized architectures; with the goal of boosting the oper-
ating performance of RFID systems to meet both existing and future requirements of
ultra large scale RFID applications. To this end, we propose the distributed receiving-
based communication architecture, namely Distributed Receiving RFID (DR-RFID)
system.
Our main contributions in this chapter are as follows. We introduce our dis-
tributed receiving paradigm in RFID systems. In the proposed paradigm, the task
of collecting the modulated backscattered signals is dispersed to spatially distributed
entities within the reader interrogation zone. This model supports multipoint com-
munication and hence eliminates the redundant interrogation of same set of tags, by
overlapping RFID readers. We also present the novel notion of micro-interrogation
zones. The micro-zones are formed by dynamically adjusting reflectivity coefficients
of RFID tags. Micro-zones spatially isolate tags within the reader’s interrogation
zone; hence facilitating high bandwidth and data rates. Furthermore, we also devise
a parallel singulation algorithm which exploits distributed receiving and micro-zones
to mitigate tag collisions. The algorithm singulate tags, from each micro-zone, in a
parallel and autonomous manner. Parallel singulation assists in achieving higher read-
ing rate under both dense and mobile tag environments. We evaluate the proposed
system, via comprehensive simulations, using the proposed DR-RFID and existing
RFID systems, implementation on the ns-2 simulator [7]. The results validate the
proposed communication architecture and singulation algorithm, and demonstrate
4.2. MOTIVATION 94
substantial improvement in system performance, usability and enhanced scalability
under diverse distribution of tags.
The rest of the chapter is organized as follows. Section 4.2 elaborates on the mo-
tivation behind the proposed system. Section 4.3 explains the distributed receiving-
based architecture and details its components. Section 4.4 explains the concept of
micro-zones and the algorithm formalizing its formation. Section 4.5 explains the par-
allel singulation algorithm and analyzes its running complexity. Section 4.6 provides
detailed information about the simulation methodology and performance analysis.
Section 4.7 outlines the work-in-progress on the proposed system. Section 4.8 sum-
marizes our work.
4.2 Motivation
Communication in RFID follows a centralized-control model, usually termed as a
master-slave, as is depicted in Fig. 2.4. Although the centralized approach has been
widely employed in RFID systems, it has significant drawbacks and limitations, which
hinder overall system performance. In this section, we elaborate on the limitations
inherently imposed by existing RFID communication architectures.
Reading Range: The least available power for tag circuitry and the strength of
backscattered signal are two key requirements in determining the reading range for
passive and semi-passive tags; respectively. In addition, in the case of passive tags,
other major parameters include reader operating power P
ERIP
and tag mismatch
factor τ. Analytically, the operating distance between a reader and a passive tag is
given by
d
(r,t)
=
λ

_
P
ERIP
.G
r
.G
t

P
min
(4.1)
4.2. MOTIVATION 95
As such, under the assumption of a dipole antenna (G
r
= G
t
= 1.6) operating at
900MHz, a tag with a modest power requirement P
min
of 16µW [57] and a mis-
match factor τ of 0.5, the maximum theoretical operating distance of the passive tag,
following North American regulations, is 15 meters.
The reading range of the semi-passive tag is a function of the strength of the
modulated reflected signal
P
rcvd
=
λ
2
.P
ERIP
.G
r
2
.G
t
(4π)
2
(d
(r,t)
)
4
(4.2)
From (4.1) and (4.2) it is evident that the tag reading range can be increased by
adjusting following variables. First, by decreasing the power requirements for the tag
circuitry P
min
. In doing so the tag will be operative at a longer distance however, it is
then constrained by the integrity of the received signal (4.2) on by the backward-link.
Second, by improving the reader’s antenna sensitivity, i.e., (
P
rcvd
P
ERIP
) by using arrays
of antennas, higher gain antennas [36] or directional antennas [42]. However, such
solutions, due to their high associated monetary costs, are restricted only to limited
applications.
Reading Rate: RFID applications, such as item-level tagging for the supply chain
and retail industry, generally involve dense and mobile passive tag distributions. Such
environment constitutes potentially thousands of mobile tag within the interrogation
zone of a single reader, and as a consequence, causes immense data collisions at the
reader [42, 96]. Reading rate is defined as the number of tags that can be singu-
lated per time unit. Lower tags’ collisions and the ability of the RFID reader to
handle multiple tags at one time imply high reading rates. However, the broadcast
nature of the backscattered signal from the tags and reader inability to process them
4.2. MOTIVATION 96
in parallel affect the reading rates of current RFID systems. Furthermore, limited
available bandwidth and low data rates further reduce the reading rates. For in-
stance, theoretically, existing RFID reader can only support reading rates of up to
700 tags/second [91]. These reading rates (which may be lower in practice) are not
adequate for typical item-level applications involving thousands of moving tags.
Reading Reliability: Reading reliability is the probability a tag is readable regard-
less of the reader and tag antennas’ orientations. Having high reading reliability is of
utmost importance for certain applications. For instance, self-checkout counter at a
department store – a miss-read tag is lost revenue. The power received by the RFID
tag, considering its antenna orientation, i.e., azimuth and latitude angles of the reader

r
, φ
r
) and the tag (θ
t
, φ
t
) [104], is
P
tag
rcvd
=
P
ERIP
.G
t

t
, φ
t
).G
r

r
, φ
r
).λ
2
(4λd
(r,t)
)
2
(4.3)
The tag orientation proportionally affects the readability and in some cases, e.g.,
perpendicular orientation between the tag and the reader antenna, can result in zero
readability. Due to the current centralized communication architecture, the CW sig-
nal is collected and processed by the emitting reader, which thins out the probability
of having decent strength signal for different antenna’s orientations and hence throt-
tles readability.
Forward-link: Emerging CMOS fabrication technology promises to significantly re-
duce the tag’s IC power requirements, at least, by a factor of ten. The effect of IC
power requirements on the tag reading range is illustrated in Fig. 4.1. The three
horizontal lines, from top to bottom, are the minimum power (in dBm) requirement
of existing tags; −10dBm (100µW) [36] and −26dBm (2.5µW) [34] and anticipated
4.2. MOTIVATION 97
-120
-100
-80
-60
-40
-20
0
0 10 20 30 40 50 60
P
o
w
e
r

(
d
B
m
)
Distance (meters)
Tag IC power =100uW
Reader min. receive power (-85dBm)
Tag IC power -26dBm (2.5uW)
Tag IC power of -36dBm
Passive tag
Reader
Figure 4.1: Effect of tag circuitry power requirement on the backscattering power and
operating range.
tags of −36dBm (0.25µW). Low IC power translates into longer reading ranges. An
existing high-end RFID reader, with receiving sensitivity of −85dBm (lowest hori-
zontal line), may be able to handle the reflected signal of an existing tag. However,
it is surely incapable of handling the backscattered signal from the anticipated low-
power passive tags as the strength of backscattered signal falls well below (less than
−100dBm) the reader’s receiving sensitivity. In other words, the operating range of
the passive tags would be backward-link constraint. This requires new communica-
tion architectures and protocols be developed which can adapt to new challenges and
constraints.
All of the aforementioned limitations demands that new communication models
and singulation approaches be sought out.
4.3. DR-RFID: DISTRIBUTED RECEIVING SYSTEM FOR RFID 98
4.3 DR-RFID: Distributed Receiving System for RFID
In current RFID systems, the reader acts both as an emanation and an assimilation
point. As an emanation point, it broadcasts the CW signal and interrogation queries.
As an assimilation point, it receives and processes the modulated backscattered signals
from the tags. Such centralized architecture constitutes the master-slave architecture
depicted in Fig. 2.4. As explained earlier, it is the centralized nature of communi-
cation architecture which is the root cause of degradation in system’s performance
metrics, viz. reading rate, reading range, reading reliability, bandwidth, capacity
and data rates. We propose a novel communication paradigm in RFID – distributed
receiving-based communication architecture. The basic idea behind the proposed ar-
chitecture is to disintegrate the transmission and the receiving components of the
existing RFID readers into spatially distributed entities within the concerned RFID
interrogation zone. For this purpose a new component, named fielder, is introduced
in the system. The main tasks of the fielder are collection, processing and relaying of
the received data from RFID tags. Distribution of these tasks empowers the reader to
tackle multiple tags in parallel at spatially isolated clusters managed by the fielders.
Each of the clusters, also known as micro-interrogation zones, is managed by a fielder
and is created by dynamically adjusting the fielder’s signal receiving sensitivity and
the signal reflectivity coefficient of the tags within the prescribed area. This virtu-
ally associates a subset of tags to a specific fielder. This association aids in reducing
collisions as the reflected tags’ signals are now confined within the micro-zones each
of which is managed by a separate fielder. It also increases tags reading range as the
backscattered signal is multi-hop by the fielder hence sustaining the signal integrity.
4.3. DR-RFID: DISTRIBUTED RECEIVING SYSTEM FOR RFID 99
Figure 4.2: Distributed receiving-based RFID architecture.
Furthermore, the autonomous nature of these micro-zones further assists in paral-
lel tags singulation, therefore dramatically increasing the reading rates and reading
readability.
The proposed Distributed Receiving RFID (DR-RFID) system uses multi-hop
communication as depicted in Fig. 4.2. RFID tag data, once available at the fielder,
is replicated and routed to the concerned reader and RFID middleware without ini-
tiating a new interrogation process. This eliminates redundant interrogations of the
same set of tags by readers, which have overlapping interrogation zones. Purging the
system from the overhead interrogations means fewer singulation attempts and thus
savings to the scarce wireless resources. Furthermore, with fielder physically associ-
ated to a location (micro-zone), e.g., at the docking door in a warehouse, a product
tracking application after subscribing to the fielder is notified of items, i.e., with the
concerned tag IDs. This results in better product tracking and visibility without any
extra equipment, e.g., reader or antenna arrays, for tag localization.
For interoperability with existing RFID systems, the DR-RFID system can func-
tionally accommodate both existing and modified components in multiple configu-
rations. However, not all configurations can harness the benefits assured by the
4.3. DR-RFID: DISTRIBUTED RECEIVING SYSTEM FOR RFID 100
Wireless
Transceiver
Microcontroller
de-
modulator
RFID
Interface
non-volatie
memory
Power
Control unit
Power
source
Incoming
RFID signal
Outgoing
wireless
signal
Figure 4.3: Block diagram showing the typical components of fielder in the proposed
DR-RFID system.
DR-RFID system. For instance, the operating range, in a configuration involving
existing readers and tags, may increase for the semi-passive and active RFID tags
but will remain unchanged for the passive RFID tags. This is because an increase
in the operating range entails that the passive tag is equipped with functionality to
change its reflectivity coefficient, a feature lacking in existing tags. As well conven-
tional readers lack the ability to singulate in parallel hence reading rates will not
be increased. Nevertheless, a configuration comprising of a reader with additional
functionality, with existing tags, will enhance the support for tag mobility and ability
to engage in multi-point communication.
In the remainder of this section, we discuss the system components, i.e., reader,
fielder and tag and their comparatively discrete and novel features proposed in the
context of the proposed DR-RFID system.
4.3. DR-RFID: DISTRIBUTED RECEIVING SYSTEM FOR RFID 101
4.3.1 RFID Reader
In a conventional RFID system, source of RF power, task of tag interrogation, process-
ing of backscattered signals, collisions resolution and relaying of data to middleware
are all confined in a single entity — the RFID reader. On the other hand, the DR-
RFID system disperses some of these tasks into the spatially distributed components
of the system, i.e., fielders. The major dispersed task includes dispensation of the
backscattered signals and partial relaying of tag data. This requires modifications at
the hardware and software levels to RFID readers.
In existing systems, the broadcast of the CW signal and collection of reflected
signals is on the same communication channel by the same reader. In the proposed
system however, signal broadcast and its collection is on different channels using
multiple system components. The reader broadcasts the CW signal on one channel,
whereas the fielder relays the received reflected signals on another low power commu-
nication channel, e.g., Zigbee or Wibree. Hence, to cope with such an architectural
change requires enabling the reader, at the hardware level, to transmit and receive
using different communication protocols.
Passive tags within the interrogation zone of multiple readers are singulated, by
each reader, individually in an autonomous manner. However, unlike existing RFID
systems, the singulation process is shared by the multiple overlapping readers. The
tag data processed can potentially be replicated and relayed by the receiving fielders
for the overlapping readers, without execution of their singulation mechanism. To
support this requires revision in reader’s software with ability to subscribe, receive
and process tags data without initiation of the singulation process.
4.3. DR-RFID: DISTRIBUTED RECEIVING SYSTEM FOR RFID 102
4.3.2 RFID Fielder
The fielder, the novel entity of the proposed DR-RFID, is responsible for processing of
the tags’ reflected signals and relaying of the processed data to the subscribed units.
The fielder assists in the formation of micro-zones formation and aid the reader to
resolve tags’ collisions.
The fielder, conceptually similar to a wireless sensor node, is a low-power and
low-cost physical device based on off-the-shelf components. Typical components are
shown in Fig. 4.3 and includes wireless transceiver, memory modules, microprocessor
and RFID signal receiving interface. The fielder adjusts its Receiving Signal Strength
Index (RSSI) threshold to create the virtual micro-zone of a pre-determined circum-
ference. The micro-zone is confined within the reader’s interrogation zone and with
other micro-zone occupies the entire interrogation field. Each fielder is responsible
for processing of the tags that lay within its designated micro-zone.
The functional behavior of the fielders, on receiving of the backscattered signals
from the tags, is defined based on their state, namely silent state or forward. In the
silent state, no data is routed to the reader and the converse holds for the forward
state. These states are set by the reader and assist in collision resolution. For instance,
the reader, upon collision detection, deliberately puts every fielder, except a randomly
picked one, into the silent state. The basic idea is to restrain multiple readers, with
overlapping reading zones, from interrogating the same set of tags at same time. An
overlapping reader not receiving any tag’s reply – the silent fielders will not forward
the data to their respective readers – avoids competing for tags singulation. The
randomly picked fielder, only after successful singulation, replicates and routes the
data to the overlapping readers.
4.3. DR-RFID: DISTRIBUTED RECEIVING SYSTEM FOR RFID 103
DC
power
IC
Logic
unit
Signal
detect
Modulator
Antenna
Power
distributor
Rectifier
Reflectivity
co-eff
Figure 4.4: Block diagram showing typical components of the passive RFID tag along
with highlighted components specific to the DR-RFID system.
4.3.3 RFID Tag
The incident radio waves on the tag antenna facilitate two purposes: first, for the pas-
sive tag, to harvest the DC power and second, for both passive and semi-passive tag,
to communicate with the reader. The passive tag trades off the signal power between
the rectifying unit and the backscattering unit. Existing RFID tags, generally, split
the available power equally between the two units. In special cases, however, signifi-
cant percentage of the power is set for the rectifying unit, hence, tag’s circuitry. This
will result in lesser power available for the backscattering unit hence, low strength
backscattered signal. On the contrary, the tag of the proposed DR-RFID system
always distributes significant percentage of the available power to the circuitry unit,
aided by the introduced power distributor component (see Fig. 4.4) without the need
4.4. MICRO-ZONING IN DR-RFID SYSTEM 104
T
R
T
F1
T
F2
F3
F4
T
T
T
T
Interrogate
Interrogation
zone
T: tags {F1, F2, F3, F4}: Fielders
Micro-zone
Figure 4.5: Micro-zones in the DR-RFID system.
of high-cost and highly sensitive readers. The fielder minimizes the propagation dis-
tance of the reflected signal, therefore, allowing tag’s signal integrity to be valid up to
the spatially closet fielder. Furthermore, shortening of the backscattered path further
facilitate in achieving longer operational range.
Unlike existing RFID tags, the tag in the proposed DR-RFID system has the
ability to dynamically adjust its reflectivity co-efficient. The additional components
are shown in Fig. 4.4. Tag adjusts the strength of the backscatter signal to make it
decodable within the boundary of its micro-zone, hence, does not interfere with tags
from other micro-zones. This significantly reduces tag collisions and enhances the
bandwidth of the backward-link channel, a feature not viable in existing tags.
4.4 Micro-Zoning in DR-RFID System
The basic idea of micro-zones, illustrated in Fig. 4.5, is to cluster the tags into micro-
zones and interrogate these clusters in an autonomous manner. This reduces the
4.4. MICRO-ZONING IN DR-RFID SYSTEM 105
number of tags that needs to be interrogated at one time, reducing collisions and
improving tags reading rate. The micro-zones are formed with support from fielders
and tags. Fielders simply need to adjust their RSSI threshold such that the signal-to-
noise ratio of the spatially close (within the pre-determined area) tags surpasses the
tags which are spatially located afar. Tags need to adjust their reflection co-efficient
such that the integrity of the reflected signal will only be valid within a confined region
and must not be strong enough to interfere outside this region. To this end, the reader,
mainly based on the fielder placement, assists the tag in selecting the right value for
its coefficient in a way that both tags and fielders are able to form the required
number of micro-zones within its interrogation zone. However, due to the nature
of a complex indoor environment (e.g., warehouse), wherein the RF signal suffer
from multi-path phenomenon, it is possible for the micro-zones to have overlapping
interrogation regions. In such a scenario, the tags within these overlapping regions
will interfere will tags of more than one micro-zone. However, such conditions will
only have minor effects on the overall performance of our proposed system. In the
unlikely scenario, wherein the micro-zones are 100% overlapped, our proposed system
will not degrade beyond any existing system. We will address this issue further, in
Section 4.6, when evaluating the performance of the proposed DR-RFID and existing
RFID systems.
4.4.1 Tag Reflectivity Co-efficient
Tag reflectivity depends on the cross-sectional area of its antenna and is modifiable
by varying the antenna impedance mismatch, i.e., by changing either the real part
R
a
or the imaginary part Z for the Amplitude Shift Key (ASK) modulation or the
4.4. MICRO-ZONING IN DR-RFID SYSTEM 106
R
Antenna
Tag Circuitary
Ra
V
Z jX
(a) Current tag schematic
Tag
R
Ra
1
jX
Antenna Tag Circuitary
Ra
2
(b) Revised tag schematic
Figure 4.6: Tag schematic showing the antenna with the added parallel resistor along
with the circuitry imaginary and real part.
Phase Shift Key (PSK) modulation, respectively. The conventional tag with exposes
real part and the imaginary part of the circuit and antenna is shown in Fig. 4.6-a.
The tag backscattered power P
BS
is a function of its distance to the reader d
(r,t)
,
the reader transmitted power P
ERIP
and its radar cross section area σ. Using the
Friis equation we get
P
BS
=
P
ERIP
.G
t
4πD
2
(r,t)
.σ (4.4)
The radar cross section σ of the tag is defined as
σ = A
e
r
.
4R
2
a
[Z
a
+ Z
c
[
2
.G
t
(4.5)
where Z
a
= R
a
, Z
c
= R + jX and A
e
r
is the effective area of tag and is
A
e
r
=
λ
2

.G
t
(4.6)
Dynamic adjustments to the tag’s reflectivity coefficient require to variant the strength
4.4. MICRO-ZONING IN DR-RFID SYSTEM 107
of the backscattered signal P
BS
. To achieve this, we can use a variable resistor R
a2
in
parallel with the antenna resistor R
a1
, as depicted in Fig. 4.6-b, such that the overall
antenna resistance is
R
new
a
=
R
a1
.R
a2
R
a1
+ R
a2
(4.7)
Introducing variable parallel resistor gives flexibility to dynamically adjust the an-
tenna coefficient as needed and is controlled by the tag’s data signal. The coefficient is
adjusted before the execution of singulation algorithm and is made to till is explicitly
reset. The modified radar cross-section σ
new
, after substitution of (4.6) and (4.7) in
(4.5) is given by
σ
new
=
λ
2

.G
t
2
.
4.R
new
a
[R
new
a
[[R + jX[
2
(4.8)
The power density of the reflected signal received by the fielder, with antenna gain
of G
f
, at distance d
(t,f)
is
S
f
=
P
BS
.G
f
4π.D
(t,f)
2

new
(4.9)
Substituting P
BS
and σ
new
from (4.4) and (4.8) we get
S
f
=
P
ERIP
.G
r
.G
f

2
.G
2
t
(4π)
3
.D
2
(r,t)
.D
2
(t,f)
.
_
4.R
new
a
[R
new
a
[[R + jX[
2
_
(4.10)
From (4.10), we infer that the strength of the reflected signal depends on three im-
portant parameters; distance to the reader, distance to the fielder and radar cross
section. Therefore, by adjusting the value of the variable resistor R
a2
, in accordance
with the reader-tag and tag-fielder distance, the strength of the reflected signal can be
modified. For instance, Fig. 4.7-a shows the strength of the backscattered signal, seen
at the fielder (using 4.10), with P
ERIP
4W (North American standard), frequency of
4.4. MICRO-ZONING IN DR-RFID SYSTEM 108
900MHz, antenna gain (G
f
, G
t
and G
r
) of 1.6 and a fixed distance to fielder, D
(t,f)
of 5 meters. The smaller the distance to the reader d
(r,t)
, the lower the reflectivity
coefficient should be.
4.4.2 Dynamic Assignment of Reflectivity Index
To calculate the required tag reflectivity co-efficient needs reader-tag and tag-fielder
distance. The power-level of the reader is varied in order to calculate the reader-tag
distance. Change in the reader’s antenna power level clusters the tags based on their
distances from the reader [10]. The clustering concept is shown in Fig. 4.7-b, where
the reader’s interrogation range is divided into clusters d, d

and d

at respective
distance of d, d +step and d + 2 ∗ step from the reader. At any time, only tags from
same cluster will adjust their reflectivity co-efficient. For instance, assume the reader
has already adjusted the tags from cluster d. Then only the tags from cluster d

,
i.e., tags marked as T

will adjust their coefficients. After these tags, marked as T

,
have been adjusted the reader then increases its range to d

. Now only the tags from
cluster d

,i.e., labeled as T

will act to the reader requests. In other words, the tag
adjusts its reflectivity only once until it is reset. The adjustment process continues
until the reader reaches its maximum interrogation range. The stepping distance and
thus the number of clusters can be optimized for a given tag distribution [12]. The
tag-fielder distance is the Euclidean distance between the boundary intersections of
a variable interrogation range and one of the fielders’ created micro-zone.
The tag reflectivity co-efficient assignment algorithm is shown in Algorithm 3. In
the first step, the intersection points for each fielder’s micro-zone f
k
with the two
4.4. MICRO-ZONING IN DR-RFID SYSTEM 109
-85
-80
-75
-70
-65
-60
-55
-50
-45
-40
1 2 3 4 5 6 7 8 9 10
P
o
w
e
r

(
d
B
m
)
Reader-tags distance (meters)
coeff: 0.1
coeff: 0.3
coeff: 0.5
coeff: 0.7
coeff: 0.9
(a) Effect of reflectivity co-efficient on received signal strength (dBm)
as function of reader-to-tag distance (meters)
(b) Tag clustering
Figure 4.7: Tag reflectivity coefficient and assignment.
4.5. PARALLEL SINGULATION IN DR-RFID 110
boundaries of the current cluster, d and (d −δ), (δ is the stepping parameter) is cal-
culated (lines 4-6). The set of intersection points P
k
are determined for every fielder
k which overlaps with the reader’s current interrogation boundary. The tag-fielder
distance is the maximum Euclidian distance f
k
p, between the fielder f
k
and its in-
tersection points p, amongst all fielders (line 7). The new reflection coefficient σ
new
is calculated by substituting the tag-fielder distance in (4.9) for a pre-defined S
f
and
is subsequently broadcast (lines 8-9). Afterwards, the reader power-level is incre-
mented and the aforementioned steps are repeated until the maximum interrogation
is reached.
Algorithm 3: The reflectivity assignment algorithm.
Set Reflectivity(T) 1
Input: T: set of fielder.
Output: Adjusted tags’ reflectivity
repeat 2
// for the i
th
cluster. 3
foreach fielder f
k
in T do 4
¦P¦
k
← 4-intersection points with f
k
, using d and (d - δ); 5
end 6
d(f,t) ← max ( ∀ f
k
¦F¦ [ max ( ∀ p ¦P¦
f
k
[ f
k
p )); 7
Calculate σ
new
by substituting d(f,t) in (4.9); 8
Send (Broadcast, σ
new
); 9
Increment power-level by δ; 10
until Maximum interrogation range is reached ; 11
4.5 Parallel Singulation in DR-RFID
Existing anti-collision schemes are based on certain assumptions about the under-
lying system. First, the reader in a sequential manner interrogates tags within the
interrogation zone. In other words, it is only after successful singulation of a tag that
4.5. PARALLEL SINGULATION IN DR-RFID 111
the subsequent singulation takes place for the remaining tags. Second, all tags lying
within the interrogation zone will cause collisions at the reader. On the contrary,
DR-RFID systems have the following important characteristics. First, the interroga-
tion zones are divided into multiple non-interfering micro-zones. Second, the reader
only knows and is concerned with the intra-zone collisions. Finally, and most im-
portantly, due to the micro-zones, the sequential singulation is not the only option.
To elaborate, since only the intra-zone collisions are possible, each micro-zone can be
interrogated independently.
To this end, we propose a parallel singulation algorithm where multiple instances
of an existing anti-collision scheme are executed, for each micro-zone, in parallel
and autonomous manner. To determine the effectiveness of the parallel singulation
approach, we use algorithms from the deterministic and probabilistic class of anti-
collision schemes. Although we use simple binary search tree algorithm [42] [95] and
the EPC Q algorithm [40] for deterministic and probabilistic, respectively however, yet
any other anti-collision algorithm can be used. In this section, we describe the system
model, explain the deterministic and probabilistic parallel singulation algorithms and
analyze their running complexities.
4.5.1 System Model
Consider an interrogation area A
R
enclosed by a single RFID reader R with maxi-
mum interrogation range of R
r
. The interrogation area is divided into n micro-zones
f
1
, ..., f
n
of area A
f
1
,..., A
fn
, respectively. A fielder manages each micro-zone. The
fielder selectively relays the received tags’ serial numbers to the reader. Assuming
the disk model, the reader interrogation area is completely covered by the clusters,
4.5. PARALLEL SINGULATION IN DR-RFID 112
i.e., A
R
≤ (A
f
1

...

A
fn
). Assuming a uniform distribution, there are k tags within
the interrogation area A
R
, i.e., t
1
,..., t
k
, where k >> n. Each micro-zone i receives
responses only from the tags within its area A
f
i
. The reader during each interrogation
cycle i broadcasts singulation requests Reqa
(i,j)
intended for the micro-zone f
j
. For
an interrogation cycle, the maximum number of singulation requests transmitted by
the reader equals the number of micro-zone.
4.5.2 The Deterministic Algorithm
The main goal of RFID anti-collision algorithms is to increase the singulation rate
by reducing the tags’ collisions. In the case of deterministic parallel singulation algo-
rithm, this is achieved in two ways. First, execution of an instance of the deterministic
singulation algorithm, for each micro-zone in an autonomous manner. In this section,
we use the conventional binary search tree as the candidate deterministic algorithm.
Second, by making use of reader and fielders synchronization. Pseudocode for the
parallel singulation algorithm is shown in Algorithm 4. The parallel singulation algo-
rithm is composed of four logical components: initialization, concurrent singulation,
communication between the RFID components and communication timeout scenar-
ios.
Initialization: The algorithm begins with housekeeping tasks (lines 2-7) which in-
cludes setting all fielders states to FORWARD, broadcasting the RESET and the ini-
tial serial request (REQA) command. Fielders in the forward state always relay tags’
serial numbers to the reader whereas converse holds in silent state. The reader, after
resetting tags internal state (RESET), broadcasts the initial serial request (REQA)
4.5. PARALLEL SINGULATION IN DR-RFID 113
Algorithm 4: Deterministic parallel singulation algorithm.
Deterministic Parallel Reading (T) 1
Input: T: set of fielder.
Output: Singulated tags.
foreach fielder f
k
in T do 2
s
k
= FORWARD; 3
end 4
Send (broadcast, RESET); 5
Adjust the tags reflectivity index using Algorithm 3; 6
Send (broadcast, REQA
0xffffffff
); 7
8
// Singulation process for the i
th
interrogation cycle. 9
repeat 10
wait until REQA
timeout
; 11
12
foreach fielder f
k
in T do 13
// Let j be the MSB collision bit, if any. 14
if (Collision)
f
k
then 15
REQA
(i,k)
← b
n−1
...b
j
= 0b
j−1
= 1...b
0
= 1; 16
Queue
k
← b
n−1
...b
j
= 1b
j−1
= 1...b
0
= 1; 17
end 18
else if (NoReply)
f
k
then 19
// NULL is returned, if the Queue
k
is empty. 20
REQA
(i,k)
← Pop from Queue
k
; 21
end 22
else if (NoCollision)
f
k
then 23
// Tag is successfully singulated. Perform tag’s read/write 24
operations.
REQA
(i,k)
← Pop from Queue
k
; 25
end 26
end 27
foreach fielders f
k
in T do 28
if REQA
(i,k)
,= NULL and REQA
(i,k)
is Unique for i
th
cycle then 29
Send(f
k
, REQA
(i,k)
); 30
end 31
else if REQA
(i,k)
= NULL then 32
Send(f
k
, SILENT); 33
s
k
= SILENT; 34
end 35
end 36
until ∃k[s
k
= FORWARD ; 37
4.5. PARALLEL SINGULATION IN DR-RFID 114
command using the highest possible 32-bit serial number as its parameter. This ini-
tial interrogation request serves as the root of the singulation tree.
Parallelism: For each micro-zone, the reader maintains an independent instance
of the singulation tree and its associated data structures, i.e., singulation tree and
queue. An instance of the anti-collision algorithm (listed in lines 15-25) is iteratively
executed during each micro-zone’s interrogation cycle. In a collision scenario, between
the tags of the k
th
micro-zone, a new broadcast request REQA
(i,k)
is formed by re-
placing the most significant collision bit by 0 followed by trailing 1’s. The broadcast
request for the k
th
micro-zone is then used for the subsequent interrogation cycle.
The most significant collision bit is also replaced by 1, followed by trailing 1’s, and is
queued on Queue
k
(line 17) which is de-queued under two scenarios. First, when no
response from the tags was received for the last request (line 21) and second, after a
successful tag singulation (line 25).
Reader and Fielder Communication: The tag replies to a reader request if and
only if the requested serial number is greater than its own. Due to broadcasting na-
ture, the tag listens and in some cases may respond to an irrelevant request, i.e., is
not intended for its micro-zone. For instance, consider two micro-zone a and b with
their respective request as REQA
(i,k)
and REQA
(i,j)
. If REQA
(i,k)
≤ REQA
(i,j)
then
a subset of tags from the cluster a may response even if their serial number is greater
than REQA
(i,j)
. Such a situation potentially inhibits the micro-zones singulation tree
with unwanted tree nodes and, hence, an unpredictable outcome. To alleviate such
a data structure corruption, the fielders are kept informed, by the reader, of their
successive interrogation requests. This facilitates in effective filtering (listed in lines
28-36) of the irrelevant tags’ replies.
4.5. PARALLEL SINGULATION IN DR-RFID 115
Reader
Tag
Fielder
REQA-0
REQA-1
REQA-1
Time
out
(a) Redundant cycle
Reader Tag Fielder
(REQA-0, seq=0)
Tag Data
Time
out
(REQA-1, seq=1)
(b) Filtering
Reader
Tag
Fielder
(REQA-0, seq=0)
Time
out
(REQA-1, seq=1)
(c) No tag data
Figure 4.8: Timeout scenarios in parallel singulation.
Timeout: Various scenarios trigger timeouts and retransmissions and are outlined
in Fig. 4.8. The REQA
timeout
time (line 11) is the duration of an interrogation cy-
cle, i.e., the maximum time the reader waits for a tag response. The timeout can
be non-legitimate or legitimate. The non-legitimate case is when the reader, upon
time expiration, broadcasts the queued request (line 21). However, shortly after that,
it receives the delayed tag response (Fig. 4.8-a). In such a case, the reader cancels
the current interrogation cycle, queues the transmitted request and continues with
the singulation process based on the delayed responses. The handling of the delayed
transmission is critical as, by neglecting such cases, we may potentially overlook some
tags. The legitimate case happens when either the fielder filters out all of the tags’
4.5. PARALLEL SINGULATION IN DR-RFID 116
responses (Fig. 4.8-b) or when an empty request exists, i.e., no tag matches with the
request (Fig. 4.8-c). Both scenarios lead to a graceful timeout followed by broadcast-
ing of the successive interrogation request.
4.5.3 The Probabilistic Algorithm
The EPC Q-algorithm [40], i.e., the probabilistic class of anti-collision algorithm, uses
the Q parameter which determine the number of slots in the Query Round. When a
tag receives a Query, it chooses a random number in the range (0, 2
Q
− 1), where Q
is in the range (0, 15), and the value is stored in the slot counter of the tag. If a tag
stores a 0 in its slot counter, it will immediately transmit a 16 bit random number
(RN16). In case of collision, the reader will adjust the Q parameter and will resent
the query.
The main objective of the probabilistic parallel algorithm is to increase the singu-
lation rate by reducing the tags’ collisions. This is achieved by determining the value
for Q parameter that will results into best performance, in terms of idle, collision and
successful slots, over all the micro-zones. Pseudocode for the probabilistic parallel
singulation algorithm is shown in Algorithm 5. The parallel singulation algorithm
is composed of two logical components: initialization and updating the value of Q
parameter.
Initialization: The algorithm begins with housekeeping tasks (lines 2-10). These
tasks start off with setting the initial value of Q (4.0) and C (0.2), broadcasting the
RESET to all tags and estimating the tags cardinality. The tags estimation is use to
terminate the algorithm (line 38), however, in case of conventional Q-algorithm is also
used to optimally set the Q value [72]. The tags reflectivity index is then adjusted
4.5. PARALLEL SINGULATION IN DR-RFID 117
Algorithm 5: Probabilistic (Q-algorithm) parallel singulation algorithm.
Q-algorithm Parallel Reading (T) 1
Input: T: set of fielder.
Output: Singulated tags.
Q=4.0; 2
C=0.2; 3
Send (broadcast, RESET); 4
E
t
← Tags estimation within the reader. 5
Adjust the tags reflectivity index using Algorithm 3; 6
foreach fielder f
k
in T do 7
E
t
[k] ← Determine tag estimation for each micro-zone of f
k
; 8
end 9
Send (broadcast, Q); 10
11
repeat 12
wait for response 13
Q

=1.0; 14
Idx=0; 15
while Q

≤ 15.0 do 16
P=-1; 17
foreach fielders f
k
in T do 18
S[k] ←
Et[k]
2
Q

∗(1−1/2
Q

)
E
t
[k]
;
19
if S[k] > P then 20
P ← S[k]; 21
end 22
end 23
P[Idx] ← P; 24
Q

← Q

+ C; 25
Idx ← Idx + 1; 26
end 27
Q

=15.0; 28
while Idx ≥ 0 do 29
if P[Idx −1] < P[Idx] and P[Idx] ≥ 1 then 30
break; 31
end 32
Idx ← Idx −1; 33
Q

← Q

−C; 34
end 35
Q ← Q

; 36
Send (broadcast, ¸Q|); 37
until all tags are read ; 38
4.5. PARALLEL SINGULATION IN DR-RFID 118
using the Algorithm 3. Afterwards, a rough estimate of tags cardinality, using similar
method as is used in line 5, however, one a time for each fielder. Finally, the initial
query request is sent using Q = 4.0.
Updating Q: In the conventional Q-algorithm, the Q value is either incremented
and decremented, by a constant C, in case of collision and idle slot, respectively. The
flow chart of the Q value updates is illustrated in Fig. 2.11. Generally speaking, if
number of tags is small with large Q value, it will result into idle slots. Similarly, if
the number of tags are large but with small Q value, it will results into collision slots.
Henceforth, to avoid the oscillations between large and small value of Q, and directly
to minimize the idle and collision slots, it is important to quickly reach the optimal
value, which has been observed to be equal to number of tags [72]. However, such
an update approach can not be applied directly in the context of micro-zone. Due
to the non-uniform nature of tags distribution, it is possible that some micro-zone
significantly will have more tags than others. Adjusting the Q value for a micro-zone
may have negative effect on the rest of micro-zones hence, might overshadow the
benefits of parallel singulation. To fix this, the proposed algorithm iterates over all
the Q values, between 1 and 15 (line 16), incrementing each iteration by C (line 25),
calculating the ratio of successful slots for each micro-zone (line 19). The Q with
maximum ratio of successful slots, amongst all micro-zone, (lines 28 −36)is used for
next query (line 37). Due to the nature of Q algorithm, unlike the deterministic al-
gorithm (Algorithm 4), only single query is broadcast for all micro-zones. Due to the
nature of micro-zones, only intra-zone collisions are now possible therefore, increasing
the reading rates.
4.5. PARALLEL SINGULATION IN DR-RFID 119
4.5.4 Delay Analysis of Deterministic Algorithm
In this section, we analyze the running complexity, i.e., average, best and worst
singulation delay, of the proposed deterministic parallel singulation algorithm. Let
f(k) be the number of cycles required to read a set of k tags. For the binary tree
protocol
1
, in general
f(n) = c n −1, (4.11)
where c is a constant and n is total number of tags.
In the parallel singulation algorithm, the reader interrogates the micro-zones’ tags in
parallel. Let E[t
1
],...,E[t
k
] be the estimated number of tags within the micro-zones
f
1
,...,f
k
, respectively. Overall number of cycles is then the maximum number of cycles,
amongst all micro-zones, the singulation process takes, i.e.
f(n) = Max [f(E[t
1
]), f(E[t
2
]), ..., f(E[t
k
])] (4.12)
To determine the best, average and worst case scenarios, based on (4.12), it suffices
to find an estimate of minimum, maximum and average of the number of tags in the
micro-zones.
Best Case: The best case scenario is when all the tags are evenly distributed amongst
the micro-zones
E[t
i
] =
A
i
∗ k
A
R
; for any given micro −zone i. (4.13)
1
The scheme in reference [95] is utilized in this chapter with f(n) = 2 n −1
4.5. PARALLEL SINGULATION IN DR-RFID 120
Worst Case: The worst case scenario is when all the tags are within a single micro-
zone i
E[t
i
] = k and t
j
= 0, (∀i[i ,= j) (4.14)
This corresponds to the case of the existing approaches, i.e., sequential singulation
algorithms. Hence, the parallel singulation algorithm cannot perform any worse than
existing algorithms.
Average Case: The micro-zone, in an attempt to cover the reader’s interrogation
zone, may overlap. In other words, there is a possibility that a tag may belong to
multiple micro-zones and hence, is counted by than one micro-zone. The area covered
by an overlapping region of the two micro-zones A
overlap
, each with equal radius, i.e.,
R
c1
= R
c2
= R
c
, is formulated using simple geometry as
A
overlap
= 2R
2
c
cos
−1
_
d
2R
c
_

1
2
d
_
4R
2
c
−d
2
(4.15)
and therefore, an estimate of the number of tags is
E[t
i
] =
N
A
R
_
1 −
A
overlap
O
_
P
i
(4.16)
Here O is the count of the micro-zone overlapping regions and P
i
is the probability
with which the micro-zone f
i
singulate the overlapping tags. A common value of P
i
,
assuming two overlapping micro-zones, is 0.5 as each micro-zone has equal probability
of tags’ singulations within the overlapping regions. That is, each micro-zone has
equal probability to singulate the overlapping tags. In a typical grid-based deployment
(Fig. 4.9), since each entity has at-most four neighbors, the fielder’s zone is at-most
4.6. SIMULATION SETUP AND RESULTS 121
overlapped by four other zones, i.e., O = 4.
4.5.5 Delay Analysis of Probabilistic Algorithm
In this section, we analyze the running complexity of the proposed probabilistic paral-
lel singulation algorithm. For the Q-algorithm, the running complexity can be defined
in terms of collision slots probability, successful slots probability and idle slots prob-
ability. Lets assume the number of tags is n and the frame size, for the Q algorithm,
is 2
Q
, then the probability of idle (P
i
), probability of successful (P
s
) and probability
of collision (P
c
) is [22]
P
i
=
_
1 −
1
2
Q
_
n
(4.17)
P
c
= 1 −P
i
−P
s
(4.18)
P
s
=
n
2
Q
_
1 −
1
2
Q
_
n−1
(4.19)
In case of the parallel variation of Q-algorithm, the worst case reading rate is defined
by the micro-zone which have the largest ratio of collision slots, amongst others. How-
ever, as the accumulative number of tags can not be smaller than even the largest tags
residing micro-zone, therefore, the computational complexity of the parallel variation
cannot be worst than the original Q-algorithm.
4.6 Simulation Setup and Results
In this section, we analyze and compare the performance of the proposed DR-RFID
system as opposed to conventional RFID systems. We use reading range, reading
rate, reading reliability and mobility as performance metrics. We also compare the
performance of the proposed DR-RFID system with a multi-reader configuration.
4.6. SIMULATION SETUP AND RESULTS 122
Table 4.1: Simulation parameters
Parameter Value
Grid size 20m
2
Max. interrogation range 15m
Reader power level 4W
Tag1 min. IC power 1.0µW
Tag2 min. IC power [34] 2.5µW
4.6.1 Simulation Setup
We have extended the ns-2 network simulator to support both existing RFID and
the proposed DR-RFID system. The major extension involves modification of the
underlying simulator architecture to support the basic RFID functionality, the EPC
Class1 Gen2 MAC protocol, multi-channel interfaces (RFID and IEEE 802.15.4) and
non-EPC singulation protocols from the literature. Minor changes to the simulator
include inheriting of the network node to serve as an RFID tag, reader, and fielder,
single-hop communication model (backscattering modulation) between the tag and
the reader, the reader’s power control, etc. Unless otherwise specified, the simulations
are performed using the following setting and parameters as shown in Table 4.1.
In the setup, tags are uniformly distributed in with single reader located at the
center of the grid. The fielders are deployed in grid fashion, as configurations of 2, 4
and 16, covering the reader’s interrogation zone with minimum possible overlapping.
Simulations are made to run until all the tags are successfully identified. The perfor-
mance metrics are averaged over twenty different experiments of different topologies.
We evaluate the system for its reading range, reading rate, reading reliability and im-
pact of tags mobility. We also compare the system with multiple readers, mimicking
4.6. SIMULATION SETUP AND RESULTS 123
the DR-RFID approach.
X
Y
R
C1
C2
(a) Two
X
Y
R
C1 C2
C3 C4
(b) Four
X
Y
R
C1 C2
C3 C4
C5 C6
C7 C8
C9 C10
C11 C12
C13 C14
C15 C16
(c) Sixteen
Figure 4.9: Grid-based fielders’ deployment scheme.
4.6. SIMULATION SETUP AND RESULTS 124
4.6.2 Simulation Results
Reading Range
Existing RFID systems use direct communication link between the tag and reader.
This limits the reading range for both passive and semi-passive tags. A longer reading
range means fewer readers to cover an area and longer tagged object visibility thus
lower equipments manageability and monetary costs.
In the DR-RFID system, the multi-hop, i.e., relaying of data through fielders,
between the reader and tags, significantly increases the operating range of the semi-
passive tags. The multi-hop link involves one or more wireless-hops, i.e., fielders, to
transmit the data between the tag and the reader. Fig. 4.10-a shows the reading
range of a semi-passive for the reader’s power level of 1W and 4W, for the direct
(existing approach) and 1-hop (1-fielder per micro-zone configuration) system. The
emitting power of 4W and 1W is the maximum allowable power that a reader may
transmit according to the North American and European regulations, respectively. As
expected, by using the DR-RFID system, the reading range for the semi-passive tag
can be increased by an order of magnitude. For instance, for the 4W case, using 1-hop
increases the reading range by a factor of 5. Using more than 1-hop can potentially
increase the reading range of the semi-passive tags.
The reading range of passive tags depends on several factors including minimum
circuitry power requirement, efficiency of the rectifying unit and strength of backscat-
tering signal. The reading range of the two passive tags (tag1 and tag2), see Table 4.1,
is shown in Fig. 4.10-b and Fig. 4.10-c, respectively. The horizontal dashed line in
the graph shows the maximum possible reading range when using a rectifying module
with an efficiency of 35%. The proposed system increases the passive tag’s reading
4.6. SIMULATION SETUP AND RESULTS 125
-85
-80
-75
-70
-65
-60
-55
-50
-45
-40
-35
0 100 200 300 400 500 600 700 800
P
o
w
e
r

(
d
B
m
)
Reading range (meters)
direct (1W)
direct (4W)
1-hops (1W)
1-hops (4W)
(a) Semi-passive tag
12
13
14
15
16
17
18
19
20
21
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
R
e
a
d
i
n
g

r
a
n
g
e

(
m
e
t
e
r
s
)
Ratio of backscattering power
Tag’s maximum possible operating range
direct (2.5uW)
1-hop (2.5uW)
(b) Passive tag (min. IC 2.5µW)
15
20
25
30
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
R
e
a
d
i
n
g

r
a
n
g
e

(
m
e
t
e
r
s
)
Ratio of backscattering power
Tag’s maximum possible operating range
direct (1.0uW)
1-hop (1.0uW)
(c) Passive tag (min. IC of 1.0µW)
Figure 4.10: Improvement in the reading range of the semi-passive and the passive
tags for the DR-RFID system.
4.6. SIMULATION SETUP AND RESULTS 126
range under two scenarios. First, low-power circuitry and second, setting of the sig-
nificant available power portion for the circuitry. For instance, in the case of tag2, the
maximum possible operating range, in the proposed system, is achievable by using a
very small ratio, 1% of the available power, for backscattering. On the contrary, to
maintain similar range in existing systems, 7% of the available power must be avail-
able for backscattering. Using a lower ratio adversely affects integrity of the reflected
signal.
For low-power circuits, the passive tag evolves into being a backward-link con-
straint. For instance, tag1 has a maximum theoretical range of 29.5m. Existing
systems, using up to 9% of available power available for backscattering, cannot go
beyond the 21m mark. Diverting more power for backscattering, at the cost of lesser
power for circuitry, is not helpful either as the tag, in such a situation, will not have
enough power to operate. On the other hand, the proposed DR-RFID system achieves
the maximal operating range.
Reading Rate
The tag singulation rate is a product of the total number of interrogation queries
and the individual query interval. High reading rates imply that more tags can be
singulated by a reader thus supporting item-level tagging. The total number of queries
for BST and Q-algorithm, using the conventional, four micro-zones and sixteen micro-
zones version is shown in Fig. 4.11. The Ideal scheme, in case of BST, represents
ideal number of queries, hence, no collisions, that would be required to interrogate
a given number of tags. Ideal scheme is not devised for the Q-algorithm as it is a
probabilistic approach. As is evident, the DR-RFID parallel singulation approach,
4.6. SIMULATION SETUP AND RESULTS 127
0
500
1000
1500
2000
2500
3000
3500
4000
100 200 300 400 500 600 700 800 900
N
u
m
b
e
r

o
f

Q
u
e
r
i
e
s
Number of Tags
Q-algo (optimal)
4-microzone
16-microzone
Ideal
(a) Q-Algorithm
0
500
1000
1500
2000
2500
3000
100 200 300 400 500 600 700 800 900
N
u
m
b
e
r

o
f

Q
u
e
r
i
e
s
Number of Tags
BST
4-microzone
16-microzone
(b) BST
Figure 4.11: Comparison of number of queries for BST and Q-algorithm, using con-
ventional RFID and DR-RFID systems.
4.6. SIMULATION SETUP AND RESULTS 128
for both deterministic and probabilistic algorithm, outclass the conventional RFID by
many folds. Furthermore, the proposed approach is scalable as the number of queries
increases gracefully with an increase in number of tags. This is because the number
of collision queries, for the DR-RFID system, does not increase significantly with an
increase in tags. The number of collision queries, for both BST and Q-algorithm, is
shown in Fig. 4.12. For instance, the collision queries for 100 tags for the conventional
BST and DR-RFID variation of the BST are 99 and 10.3, respectively, whereas for
800 tags are 799 and 63.2, respectively. This significant performance improvements
is achieved as each micro-zone has very limited number of tags, relative to overall
interrogation area. The performance of DR-RFID will get effected once the number
of tags within each micro-zone exceed significantly large, e.g., 500 and beyond.
Reading Reliability
Reading reliability is the probability a tag is readable regardless of the reader and
tag antennas’ orientations. The readability ratio of a tag, in the conventional (direct-
hop) and the proposed (1-hop) system, using single antenna configuration, at various
distance from the reader is shown in Fig. 4.13. Each point in the graph is an average
of successful tag read, using orientations between 10

and 180

. The readability
of the DR-RFID system, at long distances, is almost three times higher than the
conventional system. Note that 100% readability is achievable at short distances.
Such significant improvements are partly due to tag setting considerable portion of
the available power for its circuitry. Therefore, even under non-friendly surroundings,
e.g., a tag mounted on a steel item, the majority of available power can be used for
tag operations. As well, the integrity of the backscattered signal is to be valid only
4.6. SIMULATION SETUP AND RESULTS 129
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
100 200 300 400 500 600 700 800 900
N
u
m
b
e
r

o
f

C
o
l
l
i
s
i
o
n

Q
u
e
r
i
e
s
Number of Tags
Q-algo (optimal)
4-microzone
16-microzone
Ideal
(a) Q-Algorithm
0
100
200
300
400
500
600
700
800
900
100 200 300 400 500 600 700 800 900
N
u
m
b
e
r

o
f

C
o
l
l
i
s
i
o
n

Q
u
e
r
i
e
s
Number of Tags
BST
4-microzone
16-microzone
(b) BST
Figure 4.12: Comparison of number of collision queries for BST and Q-algorithm,
using conventional RFID and DR-RFID systems.
4.6. SIMULATION SETUP AND RESULTS 130
0
0.2
0.4
0.6
0.8
1
10 20 30 40 50 60 70 80
R
e
a
d
a
b
i
l
i
t
y

r
a
t
i
o
Reader’s Antenna angle (in degree)
15m - direct
15m - 1hop
10m - direct
10m - 1hop
5m - direct
5m - 1hop
Figure 4.13: Reading reliability comparison.
inside its micro-zone. In other words, the tags in the DR-RFID system are capable
of operating under diverse antenna’s orientations, i.e., diverse range of power, thus
meeting the challenges of futuristic applications, e.g., item level tagging.
Mobility
In existing RFID systems, a mobile tag may be miss-read in two scenarios. First, if it
moves out of the reader’s zone before being interrogated. Second, being interrogated,
in a dense environment and thus immense collisions, prevent it from being singulated
in time. Both scenarios lead to miss-read tags. The effect of mobility on tag reading,
using single reader configuration for the conventional and proposed system, for 100
and 600 tags is respectively shown in Fig. 4.14-a and Fig. 4.14-b. To be fair in com-
parison, we have not entertained the long reading range and the multipoint paradigm
of the proposed system, for this experiment.
The reading rate of mobile tags in the conventional RFID system, both at high
4.6. SIMULATION SETUP AND RESULTS 131
84
86
88
90
92
94
96
98
100
10 20 30 40 50
R
e
a
d
i
n
g

p
e
r
c
e
n
t
a
g
e
Speed (m/s)
direct
2-fielders
4-fielders
16-fielders
(a) 100 mobile tags
20
30
40
50
60
70
80
90
100
10 20 30 40 50
R
e
a
d
i
n
g

p
e
r
c
e
n
t
a
g
e
Speed (m/s)
direct
2-fielders
4-fielders
16-fielders
(b) 600 mobile tags
Figure 4.14: Effect of mobility on the tag reading rate.
4.6. SIMULATION SETUP AND RESULTS 132
speeds and dense setting, falls well under 30%. The DR-RFID system, on the other
hand, can handle high-speed and dense-tag environments. The micro-zones facilitate
potentially fewer collisions, whereas parallel singulation algorithm support fast read-
ing rates. Fewer collisions and fast reading rates assists in mobile tags singulation
even at high speed and in dense environments. Furthermore, higher the number of
micro-zone, the better system’s ability is to tackle high speed mobility. For instance,
using 16-fielders, the reading percentage is an order of magnitude higher than when
using of 2-fielders. Reading rates of over 90% is maintainable, regardless of tags speed
and density.
Multi-reader Configuration
Micro-zones may be mimicked by using a set of readers, equal in number and interro-
gation area, deployed at the same locations as the fielders. The motive behind such
an experiment is to compare the performance of multiple readers to the distributed
receiving system with multiple fielders. The total number of reading cycles, with mul-
tiple standalone readers and the fielders, is shown in Fig. 4.15. Interestingly, the total
cycles for the multi-standalone system are only marginally lower than the proposed
system. For instance, the difference between two and four multi-standalone readers
to that of a single distributed reader aided by 2-fielders and 4-fielders is at most 4%
and 11%, respectively.
The minor performance difference between the two configurations is because of
the micro-zones’ overlapping effect. In the case of multiple standalone readers, the
tags within the overlapping region are singulated as many times as there are the
overlapping readers. However, in the DR-RFID system, each tag in the overlapping
4.7. WORK IN PROGRESS 133
0
200
400
600
800
1000
1200
1400
1600
100 200 300 400 500 600 700 800 900 1000
T
o
t
a
l

C
y
c
l
e
s
Number of Tags
2-Readers
4-Readers
2-Fielders/1-Reader
4-Fielders/1-Reader
Figure 4.15: Reading cycles comparison for the multi-readers and DR-RFID.
zone is singulated only once, hence causing idle cycles for other zones’ singulation
trees. These overlapping micro-zones, and hence their potential negative effects, can
be alleviated using an optimal placement scheme for the fielders. Nevertheless, the
DR-RFID system seems an attractive and low-cost solution over the high monetary
costs associated with using multiple readers, for minimal performance gains.
4.7 Work in Progress
To proof the working concept of the proposed DR-RFID, we have implemented its
newly introduced components namely, fielder and reflectivity adjusting semi-passive
tags. The fielder schematic, using Eagle CADsoft, is shown in Fig. 4.16. The
schematic highlights various functional components of the fielder. These components
include micro-controller, transceiver, power, dual antenna interfaces and RFID signal
de-modulator.
The fielder uses Texas Instrument MSP430F2274 [6] for the micro-controller and
4.7. WORK IN PROGRESS 134
M
S
P
4
3
0
F
2
2
7
4
I
R
H
A
D
e
s
i
g
n
e
d

b
y

K
a
s
h
i
f

A
L
I

@

Q
u
e
e
n
'
s

U
n
i
v
e
r
s
i
t
y
D
R
-
R
F
I
D
:

F
i
e
l
d
e
r
9
1
6
M
H
z

R
F

s
i
g
n
a
l
r
e
d
G
N
D
G
N
D
CC2500
GND GND
G
N
D
G
N
D
G
N
D
G
N
D 1
0
0
n
1
0
0
n
2
2
0
p
2
2
0
p
27p
27p
4
n
7
0
.
4
7
p
0
.
4
7
p
GND
GND
G
N
D
G
N
D
100p
GND
GND
GND
220p
GND
56K
100n
100n
2
6
M
H
z
GND GND
G
N
D
CHIP_ANTENNE (2.4GHz)
1
0
0
p
1
0
0
p
4
n
7
1
p
1p
4n7
100p
GND GND
g
r
e
e
n
G
N
D
1
X
0
6
G
N
D
3
3
0
R
3
3
0
R
0
47k

1
0
0
n
1
0
0
n
G
N
D
G
N
D
B
3
U
-
1
0
0
0
P
GND
K
S
R
2
2
1
J
GND
GND
39p
P
E
4
4
1
2
1
L
T
C
5
5
0
7
3
9
p
1
0
0
p
0
.
1
u
G
N
D
G
N
D
GND
D
1
D
V
S
S
1
X
O
U
T
/
P
2
.
7
2
X
I
N
/
P
2
.
6
3
D
V
S
S
4
R
S
T
/
S
B
W
T
D
I
O
5
P
2
.
0
/
6
P
2
.
1
7
P
2
.
2
8
P
3
.
0
9
P
3
.
1
1
0
P3.2
11
P3.3
12
AVSS
13
AVCC
14
P4.0
15
P4.1
16
P4.2
17
P4.3
18
P4.4
19
P4.5
20
P
4
.
6
2
1
P
4
.
7
2
2
P
3
.
4
2
3
P
3
.
5
2
4
P
3
.
6
2
5
P
3
.
7
2
6
P
2
.
3
2
7
P
2
.
4
2
8
P
1
.
0
2
9
P
1
.
1
3
0
P1.2
31
P1.3
32
P1.4
33
P1.5
34
P1.6
35
P1.7
36
TEST/SBWTCK
37
DVCC
38
DVCC
39
P2.5/ROSC
40
D
I
E
P
A
D
D
SCLK
1
SO(GDO1)
2
GDO2
3
DVDD
4
DCOUPL
5
G
D
O
0
(
A
T
E
S
T
)
6
C
S
N
7
X
O
S
C
_
Q
1
8
A
V
D
D
1
9
X
O
S
C
_
Q
2
1
0
AVDD2
11
RF_P
12
RF_N
13
AVDD3
14
AVDD4
15
G
N
D
1
1
6
R
B
I
A
S
1
7
D
G
U
A
R
D
1
8
G
N
D
2
1
9
S
I
2
0
IC2
C
1
4
1
C
1
1
1
C
1
5
1
C
9
1
C81
C101
L
1
2
3
C
1
2
3
C
1
2
4
C125
C181
R171
C41
C51
G
N
D
4
G
N
D
2
Q
2
P$1
P$1
ANT1
C
1
2
1
C
1
3
1
L
1
3
1
C
1
3
2
C122
L121
C126
D
2
1 2 3 4 5 6
S
T
1
R
6
R
2
R
1
R5
L2
C
1
C
2
1
2
T
S
1
T
S
2
P
$
1
P
$
2
C3
X
1
S
H
D
N
P
$
1
G
N
D
P
$
2
V
O
U
T
P
$
3
R
F
P
$
6
P
C
A
P
P
$
5
V
C
C
P
$
4
I
C
3
C
4
C
6
C
5
P
3
1
_
U
C
B
0
S
I
M
O
P31_UCB0SIMO
P
3
2
_
U
C
B
0
S
O
M
I
P32_UCB0SOMI
P
3
3
_
U
C
B
0
C
L
K
P33_UCB0CLK
V
C
C
_
E
X
T
V
C
C
_
E
X
T
V
C
C
_
E
X
T
VCC_EXT
V
C
C
_
H
F
V
C
C
_
H
F
T
E
S
T
/
S
B
W
T
C
K
T
E
S
T
/
S
B
W
T
C
K
TEST/SBWTCK
#
R
S
T
/
S
B
W
T
D
I
O
#
R
S
T
/
S
B
W
T
D
I
O
#RST/SBWTDIO
R
X
D
0
R
X
D
0
T
X
D
0
T
X
D
0
V
P
3
0
_
U
C
B
0
S
T
E
P30_UCB0STE
P12
TDO
TDI
TMS
TCK GDO2
G
D
O
2
GDO0
G
D
O
0
R
e
m
o
v
e

R
1

i
f

U
1

i
s

s
u
p
p
l
i
e
d

e
x
t
e
r
n
a
l
l
y
1
2
3
4
5
6
1
.
)

M
i
c
o
c
o
n
t
r
o
l
l
e
r




2
.
)

2
.
4
G
H
z

R
F

i
n
t
e
r
f
a
c
e


3
.
)

Z
i
g
b
e
e

t
r
a
n
s
c
e
i
v
e
r




4
.
)

R
F
I
D

d
e
-
m
o
d
u
l
a
t
o
r















5
.
)

P
o
w
e
r
/
d
e
b
u
g

i
n
t
e
r
f
a
c
e

















6
.
)

R
F
I
D

a
n
t
e
n
n
a

i
n
t
e
r
f
a
c
e
Figure 4.16: Fielder schematic for the DR-RFID system.
4.7. WORK IN PROGRESS 135
Zigbee based CC2500 transceiver [4] for the outgoing wireless communication with
reader(s) and other IP-based network devices. The micro-controller and transceiver
are extremely low-power, for instance, the MSP430F2274 controller utilizes only
0.7µA in the standby mode. The micro-controller and transceiver can be put into
deep sleep modes, when idle and can be woken up due to their on-chip remote wakeup
capability, To detect and de-modulate the incoming RFID signal, the fielder make use
of an extremely low power RF detector LTC5507 IC, from Linear Technology [1]. The
LTC5507 consumes 550µA and 2µA during active and shutdown mode. As is evident
from the schematic, the fielder has two RF interfaces. First interface is the 916MHz
patch antenna which is used to receive the reflected signals from the nearby tags. Sec-
ond interface is the 2.4GHz chip antenna which is used by the CC2500 to transmit
any IP packet to an IP-enabled device on the network or Internet. A prototype im-
plementation of the fielder is shown in Fig. 4.17-a. The fielder, for power and debug,
use the 6-pins (with 1.127Mil) spy-bi-wire JTAG interface. Using standard interface
facilitate, the choice of off-shelf power board, e.g., Texas Instrument’s EZ430-rf2500
kit and Cymbet’s solar energy harvesting kit [5] becomes available. A custom power
board with dual power source, i.e., a coin battery and the 6V adapter is also de-
signed for the fielder. The prototype implementation of the power board is shown in
Fig. 4.17-b. Depending upon the anticipated tag cardinality and interrogation rate
the appropriate power source can be selected during deployment phase. For instance,
in a low density environment, wherein the number of tags and readers are low, the
fielder can be powered using on-board battery therefore, allowing complete wireless
solution. On the other hand, the dense scenarios demands larger processing and data
communication volumes therefore requires higher availability, i.e., the 6V adapter.
4.7. WORK IN PROGRESS 136
(a) Fielder
(b) battery board
Figure 4.17: Fielder and power board prototype.
For software, we have enhanced Contiki OS [38] to support RFID de-modulation
using LTC5507. Contiki is an open source, highly portable, multi-tasking operating
system designed for low memory micro-controllers. The choice of Contiki, over other
OS e.g., TinyOS, is due to following reasons. First, it require low memory, i.e., fits
within the available 2KB RAM and 40KB ROM of MSP430F2274 micro-controller.
Second, it is IP enabled and its readily available Zigbee communication stack which
is required to transfer tags information to other networked devices. We have tested
the operational capability of the Fielder using the Skyetek’s M9 RFID reader and
Alien’s RFID tags wherein, the fielder was able to successfully decode the received
4.8. SUMMARY 137
tag signal and transfer it via IP to the host machine.
The tag schematic, using Eagle CADsoft, is shown in Fig. 4.18. The schematic
highlights various functional components of the tag. These components include micro-
controller, power interface, RF antenna interface, backscattering and RF attenuation.
Similar to fielder, the tag use Texas Instrument’s MSP430F2274 and Linear Technol-
ogy LTC5507 as its micro-controller and RFID de-modulator, respectively. The RFID
de-modulator is use to detect incoming signal, from the reader, to wake up the micro-
controller and to decode requests. The tag attenuates the backscatter signal using
the Skywork’s SKY12329 IC [3]. The SKY12329 is a GaAs digital attenuator with
attenuation range of 31 dB and utilizes less than 100µA. Attenuation enable the for-
mation of non-interfering micro-zones which significantly enhances the reading rates.
The tag modulate and backscatter the signal using Analog Device’s ADG902 IC [2].
The tag powers up using the fielder’s battery board, shown in Fig. 4.17-b. Currently,
we are modifying Contiki OS to enable RFID de-modulation, backscattering and RF
attenuation.
4.8 Summary
Current RFID systems suffer many shortcomings; inherently so from adopting the
centralized architecture. They impose a detrimental effect on performance, scalabil-
ity and usability. In this chapter, we introduce the Distributed Receiving paradigm
in RFID systems, namely DR-RFID, as a foundation for a more dynamic RFID ar-
chitecture. This new paradigm utilizes spatially dispersed entities, named fielders, to
collect the modulated signals backscattered from tags. Fielders are, by design, low
cost and low power devices that are deployed within the reader’s interrogation zone.
4.8. SUMMARY 138
M
S
P
4
3
0
F
2
2
7
4
I
R
H
A
D
R
-
R
F
I
D
:

S
e
m
i
-
p
a
s
s
i
v
e

T
a
g

w
i
t
h

r
e
f
e
c
t
i
v
i
t
y

a
d
j
u
s
t
m
e
n
t
D
e
s
i
g
n
e
d

b
y

K
a
s
h
i
f

A
L
I

@

Q
u
e
e
n
'
s

U
n
i
v
e
r
s
i
t
y
This will be replaceable with dipole/patch antenna
I

a
m

n
o
t

s
u
r
e

a
b
o
u
t

t
h
i
s
!
I

a
m

n
o
t

s
u
r
e

a
b
o
u
t

t
h
i
s
.
P
o
w
e
r

c
o
n
n
e
c
t
e
d

w
i
t
h

R
F
?
5
6
p

i
s

n
e
e
d
e
d

b
y

N
J
G
w
h
e
r
e
a
s

5
6
0

b
y

S
K
Y
,

d
a
t
a

s
p
e
c
s
.
R
4
,

R
7
-
R
1
0

a
r
e

s
i
z
e

0
4
0
2
C
3
-
C
4

a
r
e

s
i
z
e

0
4
0
2
r
e
d
G
N
D
G
N
D
GND GND
g
r
e
e
n
G
N
D
1
X
0
6
G
N
D
3
3
0
R
3
3
0
R
0
47k

1
0
0
n
1
0
0
n
G
N
D
G
N
D
B
3
U
-
1
0
0
0
P
GND
K
S
R
2
2
1
J
GND
CHIP_ANTENNA (916MHz)
N
J
G
5
1
9
K
C
1
L
T
C
5
5
0
7
S
K
Y
1
2
3
2
9
-
3
5
0
L
F
A
D
G
9
0
2
50
GND
G
N
D
GND
1
0
0
1
0
0
1
0
0
1
0
0
1
0
0
G
N
D
560pF
560pF
10k
50
G
N
D
GND
1
0
p
G
N
D
1
0
p
1
0
p
1
0
p
G
N
D
G
N
D
G
N
D
5
6
p
56p
GND
56p
GND
56p
39p
39p
0.1u
100p
GND
GND
D
1
D
V
S
S
1
X
O
U
T
/
P
2
.
7
2
X
I
N
/
P
2
.
6
3
D
V
S
S
4
R
S
T
/
S
B
W
T
D
I
O
5
P
2
.
0
/
6
P
2
.
1
7
P
2
.
2
8
P
3
.
0
9
P
3
.
1
1
0
P3.2
11
P3.3
12
AVSS
13
AVCC
14
P4.0
15
P4.1
16
P4.2
17
P4.3
18
P4.4
19
P4.5
20
P
4
.
6
2
1
P
4
.
7
2
2
P
3
.
4
2
3
P
3
.
5
2
4
P
3
.
6
2
5
P
3
.
7
2
6
P
2
.
3
2
7
P
2
.
4
2
8
P
1
.
0
2
9
P
1
.
1
3
0
P1.2
31
P1.3
32
P1.4
33
P1.5
34
P1.6
35
P1.7
36
TEST/SBWTCK
37
DVCC
38
DVCC
39
P2.5/ROSC
40
D
I
E
P
A
D
D
D
2
1 2 3 4 5 6
S
T
1
R
6
R
2
R
1
R5
L2
C
1
C
2
1
2
T
S
1
T
S
2
P
$
1
P
$
2
P$1
P$1
ANT2
P
1
P
$
1
G
N
D
1
P
$
3
V
C
T
L
2
P
$
4
P
C
P
$
8
V
C
T
L
1
P
$
2
P
4
P
$
1
0
I
C
2
V
C
T
L
3
P
$
7
V
C
T
L
4
P
$
9
P
2
P
$
5
P
3
P
$
6
S
H
D
N
P
$
1
G
N
D
P
$
2
V
O
U
T
P
$
3
R
F
P
$
6
P
C
A
P
P
$
5
V
C
C
P
$
4
I
C
3
G
N
D
1
P
$
1
R
F
1
P
$
1
0
N
C
1
P
$
8
G
N
D
2
P
$
2
GND3
P$3
GND4
P$4
GND5
P$5
GND6
P$6
G
N
D
7
P
$
7
N
C
2
P
$
9
R
F
2
P
$
1
1
V
5
P
$
1
6
V4
P$15
V3
P$14
V2
P$13
V1
P$12
I
C
4
G
N
D
4
P
$
5
G
N
D
3
P
$
6
G
N
D
2
P
$
7
R
F
2
P
$
8
V
D
D
P
$
1
C
T
R
L
P
$
2
R
F
1
P
$
4
G
N
D
1
P
$
3
I
C
5
R3
R
4
R
7
R
8
R
9
R
1
0
C3
C4
R11
R12
C
5
C
6
C
7
C
8
C
9
C10
C11
C12
C13
C14
C15
C16
V
C
C
_
E
X
T
V
C
C
_
E
X
T
V
C
C
_
E
X
T
V
C
C
_
E
X
T
V
C
C
_
E
X
T
VCC_EXT
V
C
C
_
H
F
T
E
S
T
/
S
B
W
T
C
K
T
E
S
T
/
S
B
W
T
C
K
TEST/SBWTCK
#
R
S
T
/
S
B
W
T
D
I
O
#
R
S
T
/
S
B
W
T
D
I
O
#RST/SBWTDIO
R
X
D
0
R
X
D
0
T
X
D
0
T
X
D
0
V
P12
R
e
m
o
v
e

R
1

i
f

U
1

i
s

s
u
p
p
l
i
e
d

e
x
t
e
r
n
a
l
l
y
1
2
3
4
5
6
1
.
)

M
i
c
r
o
c
o
n
t
r
o
l
l
e
r












2
.
)

P
o
w
e
r
/
d
e
b
u
g

i
n
t
e
r
f
a
c
e




















3
.
)

A
n
t
e
n
n
a

i
n
t
e
r
f
a
c
e

(
9
1
6
M
h
z
)




4
.
)

B
a
c
k
s
c
a
t
t
e
r
i
n
g

I
C








5
.
)

R
F

s
i
g
n
a
l

a
t
t
e
n
u
a
t
o
r
Figure 4.18: Semi-passive tag schematic for the DR-RFID system.
4.8. SUMMARY 139
As such, each fielder would create micro-interrogation zones, in which tags with ad-
justed reflectivity coefficients would fall under its region, conceptually excluding them
from micro-zones of other fielders. A parallel singulation algorithm, exploiting the
localization and distributed nature of these micro-zones, autonomously interrogates
each zone in parallel. Accordingly, adopting DR-RFID yields significant performance
improvements in achieving high reading rates, low collisions, longer reading range for
semi-passive tags and higher bandwidth.
140
Chapter 5
Coverage in RFID
5.1 Introduction
Coverage is crucial for large scale RFID applications. For an item-level object tagging
application, in a self-checkout counter at a department store, for instance, missing a
tag is simply a loss of revenue. Several factors influence coverage in RFID networks;
some of these factors are density of tags, size of the interrogation area, number of
deployed readers, and overlapping among readers [42]. The complexity and monetary
cost associated with an RFID system strongly depend on the number of readers
being deployed to interrogate tags. Similar to the wireless sensor networks, coverage
in RFID systems is a deployment problem which can be formulated as a deterministic
deployment and a random deployment problem.
In the case of deterministic deployment, the location and number of the readers
are determined in a pre-deployment manner. A typical deployment scenario of a
warehouse application is shown in Fig. 5.1. While deploying too many readers may
guarantee full coverage, it has the side effect of significant interference among readers,
which results in more reader-to-reader collisions and, hence, a lower performance for
5.1. INTRODUCTION 141
the whole system. Furthermore, the number of tags within each reader’s interrogation
zone determines the interrogation delay of that reader, and the overall interrogation
delay of the whole system is determined by the longest delay associated with a single
reader. Therefore, reducing overlapping and balancing the load of readers are essential
to any coverage scheme in RFID systems. Several maximal RFID coverage schemes
have been proposed in the literature, wherein the readers are deployed in a grid
[83, 92, 93], honey grid [110] or at a predetermined set of locations [104]. While these
schemes are able to achieve maximal coverage, that comes at the cost of a large pool
of RFID readers and, hence, results in the previously mentioned problems of having
too many readers.
Figure 5.1: 3D Coverage in RFID Networks.
In the case of random deployment, the readers are deployed in an random manner.
Typical random RFID deployment includes RFID enabled wireless sensor networks
and RFID enabled mobile phones. The nature of random deployment yields over-
lapped interrogation zones among RFID readers. This causes significant wireless
5.1. INTRODUCTION 142
interference in terms of reader and tag collisions without improving coverage. Fur-
thermore, the overlapping readers may also simultaneously interrogate the same set
of tags, resulting into data corruption. These readers collisions severely affects the
performance of the overall system [110]. Furthermore, dense deployment may result in
having redundant readers; a redundant reader is one whose interrogation area is fully
covered by neighboring readers. An example of such a scenario is shown in Fig. 5.2
wherein the readers R2 and R3 are redundant. That is because all tags covered by
R2 or R3 (i.e., T2 and T3) are also covered either by R1 or by R4. However, tag T1
is exclusively covered by R1 and tag T4 is exclusively covered by R4. A redundant
reader is one that does not exclusively cover any tag.
Figure 5.2: An RFID network with redundant readers.
Redundant readers translate into unnecessary energy consumptions due to du-
plicate interrogations. Moreover, it results into extra processing and communica-
tion overheads and does not necessarily bring an enhancement to the tags coverage.
Thereby, in order to extend the lifetime of the network and to improve its perfor-
mance, there is a significant need for light-weight methods that efficiently detect and
5.1. INTRODUCTION 143
eliminate redundant readers. Many schemes exist in the literature for redundant read-
ers elimination in RFID networks. Elimination of readers is mainly based on different
markings, made by covering readers, into the tags memory. These markings include
tags count of the covering readers [16], identity of the first interrogating reader [52],
identity of the first singulating reader [109, 111], count of the neighboring readers [55]
and others. Based on these markings, an algorithm decides on the redundancy of an
RFID reader. These schemes, however, are not light-weight as they require consid-
erable states to be maintained by the tags and involve frequent interrogations and
singulations.
Our main contributions in this chapter are as follows. We address the deterministic
deployment through the set-cover approximation approach. We present an algorithm
that gives the approximate minimum number readers and their locations to cover a
set of tags. By reducing the number of readers, not only is the overall network cost
reduced, but the overall system performance is improved also by reducing interference
and overlapping among readers coverage regions and, hence, having less collisions
among readers. Furthermore, we also enhance the overall system reading rates by
achieving a near-optimal load balancing amongst the deployed readers; by minimizing
the maximum delay encountered by a single reader we minimize the overall delay of
the system. In this work, we first minimize the number of readers than balance the
load amongst readers (the opposite order can be applied depending on the application
scenario). Note that overlapping readers result into readers-to-tag collisions which
may cause tags being undetectable. On the other hand, load imbalance amongst
readers can cause tags reading delay.
5.2. SYSTEM MODEL 144
We address the random deployment using a light-weight greedy algorithm to de-
tect and eliminate redundant readers in RFID networks. The algorithm estimates the
number of tags each reader covers and finds the number of neighboring readers it has.
The motivation is that a reader covering less tags and one that has more neighboring
readers have a higher chance of being redundant. Based on this observation, our al-
gorithm finds the ratio of the number of tags a particular reader covers to the number
of neighboring readers it has, and uses that ratio to predict whether that reader is re-
dundant or not. We evaluate the proposed algorithms via comprehensive simulations
and we compare them with other schemes available currently in the literatures.
The remainder of this chapter is organized as follows. Section 5.2 outline the
adopted model. Section 5.3 outlines the underlying assumptions and explains the
proposed deterministic readers placement algorithm. Section 5.4 formulates and ex-
plains the random deployment as redundant reader elimination algorithm. Section 5.5
presents simulated experiments, evaluation metrics and analyze the obtained results
using multiple scenarios. Section 5.6 summarizes our work.
5.2 System Model
We consider an RFID system of n passive tags and multiple readers. The transmission
range of a passive tag t
i
is modeled as a sphere of radius r
i
, i.e., tag t
i
can be
interrogated by an RFID reader if and only if the distance between the tag and the
reader is at most r
i
. A reader can communicate with its neighboring readers and
also some central middleware. We assume the existence of a central middleware as
depicted by the EPC Gen2 standard [40], where our algorithm is to be executed.
5.3. RFID READERS DEPLOYMENT (DETERMINISTIC
DEPLOYMENT) 145
5.3 RFID Readers Deployment (Deterministic Deployment)
In this section we present our algorithm for RFID readers deployment in rectangular
cuboid. We first outlines the adopted model and its underlying assumptions. Next,
we first describe an algorithm that provides full coverage, we then show how to modify
the algorithm for the purpose of reducing overlapping among readers. And finally we
give a scheme that uses our algorithm to give load-balancing deployment strategies.
5.3.1 Assumptions and Definitions
Tags are assumed to be placed in a 3D space modeled as a rectangular cuboid, and
readers are to be placed on particular sides of the rectangular cuboid. Figure 5.1
illustrates this environment. The problem can be defined as follows. Given the
number of tags and their 3D locations, find a minimum number of RFID readers and
their exact locations such that all tags are covered. While the location of tags is
assumed to be known [92,93], we show later how to accommodate fluctuations in tags
locations.
In the presentation of our algorithms, we use the following notations:
• T = ¦t
1
, t
2
, . . . , t
n
¦ is the set of tags.
• p
i
is a point representing the location of tag t
i
; p
i
.x, p
i
.y, and p
i
.z are the x−,
y−, and z− coordinates, respectively.
• r
i
is the transmission range of tag t
i
.
• [S[ is the cardinality of a set S.
• [p, q[ is the Euclidean distance between two points p and q.
5.3. RFID READERS DEPLOYMENT (DETERMINISTIC
DEPLOYMENT) 146
• For any point q in the 3D space, Cover(q) is the set of all tags that have q
within their transmission ranges (i.e., Cover(q) = ¦t
i
: [p
i
, q[ ≤ r
i
¦).
Our algorithms are based on an approximation algorithm for the set covering
problem which is known to be NP-hard [33]. For the sake of completeness, we give a
formal definition for the set covering problem as follows:
An instance of the set covering problem consists of a set ( of a finite number of
elements and a set T whose elements are subsets of ( with the condition that each
element c ∈ ( belongs to at least one subset f ∈ T; when c ∈ f, f is said to cover
c. The problem is to find a minimum size set ( ⊆ T such that each element in ( is
covered by at least one element in ( [33].
5.3.2 Providing Coverage
In order to provide full coverage, each tag must have at least one RFID reader placed
within its transmission range. And since RFID readers are to be placed only on the
sides of the rectangular cuboid, see Fig. 5.1 we need to find the intersection of the
transmission spheres of tags and the planes representing the sides of the rectangular
cuboid. If a sphere intersects with a plane, their intersection will be a circle on that
plane. For example, the intersection of a sphere centred at the point (a, b, c) with a
radius of r and the plane z = 0 is a circle centred at the point (a, b, 0) with a radius
of

r
2
−c
2
, if r
2
− c
2
≥ 0. Let us call those circles side circles. Each tag will have
at most one side circle on each side, and in order to cover a particular tag t
i
, at least
one reader must be placed within one of the side circles that belong to that tag.
To minimize the number of readers needed to cover all tags, readers should be
placed in areas where several side circles, belonging to several tags, overlap. Instead
5.3. RFID READERS DEPLOYMENT (DETERMINISTIC
DEPLOYMENT) 147
of looking for regions where side circles overlap, it suffices to focus on intersection
points, which are the points where the boundaries of side circles intersect. These
intersection points can be considered as representatives to all possible overlapping
regions (for more details on this, see our earlier work in [11]). Thereby, locations of
RFID readers will be limited to intersection points and, hence, the problem becomes a
discrete optimization problem. Moreover, an instance of the RFID reader deployment
problem can be reduced to an instance of the set covering problem in which ( = T
(i.e., the set of tags) and T = ¦Cover(q) : q is an intersection point¦.
Therefore, the greedy approximation algorithm for the set covering problem can be
used to solve the RFID readers deployment problem; and for the sake of completeness,
the greedy approximation algorithm is shown in Algorithm 6 [33].
Algorithm 6: The greedy set covering algorithm.
Set Cover((,T) 1
Input: (: set of tags.
T: coverage set.
Output: V : set of readers.
U = (; 2
V = φ; 3
while U ,= φ do 4
Find a set S ∈ T that maximizes U ∩ S; 5
U = U −S; 6
V = V ∪ ¦S¦; 7
end 8
return V ; 9
This greedy algorithm is known to have an approximation ratio of ln [C[ + 1 and
can be implemented in O([C[[T[ MIN([C[, [T[)) time.
5.3. RFID READERS DEPLOYMENT (DETERMINISTIC
DEPLOYMENT) 148
5.3.3 Reducing Overlap Among Readers
Overlapping among RFID readers causes reader-to-reader interference which result
in collisions at the readers and, hence, an additional delay to the interrogation pro-
cess. The general set covering problem does not take overlapping among subsets into
account; the sole objective therein is to minimize the number of subsets covering all el-
ements. For example, let ( = ¦1, 2, 3, 4, 5¦ and T = ¦¦1, 2, 3¦, ¦2, 3, 4¦, ¦1, 2, 5¦, ¦4¦,
¦5¦¦. The greedy algorithm may pick the set ¦¦1, 2, 3¦, ¦2, 3, 4¦, ¦1, 2, 5¦¦ to cover
all elements with a minimum number of subsets, which is an optimal solution for this
instance of the set covering problem. However, another optimal solution with no over-
lapping is ¦¦1, 2, 3¦, ¦4¦, ¦5¦¦, which is much more suitable for readers deployment
in RFID networks. In this subsection, we modify the earlier described greedy set
covering algorithm (Algorithm 6) to make solutions with less overlapping preferred.
The greedy set covering algorithm constructs the set of covering subsets incre-
mentally; at each stage, it picks the subset that covers the maximum number of
uncovered elements until all elements are covered. That can be viewed as giving each
remaining subset a weight equal to the number of uncovered elements it covers, and
picking the one with the maximum weight. We can calculate the weight of subsets
in a different way that gives credit to those subsets that do not cover many cov-
ered elements. Let U be the set of uncovered elements and weight(S) be the weight
of a subset S. According to the general greedy algorithm, weight(S) = [U ∩ S[.
We can modify the general algorithm so that among subsets covering the maximum
number of uncovered elements, it picks the one that covers the minimum number
of already covered elements. This can be done using the following weight formula:
weight(S) = [U ∩ S[ −α([S[ −[U ∩ S[), 0 < α ≤ 1/n. The greedy algorithm for set
5.3. RFID READERS DEPLOYMENT (DETERMINISTIC
DEPLOYMENT) 149
covering with less overlapping is shown in Algorithm 7.
Algorithm 7: Weighted set covering algorithm to minimize overlapping
amongst subsets.
Weighted Set Covering((,T,α) 1
Input: (: set of tags.
T: coverage set.
α: weight ratio.
Output: V : set of readers.
Z = T; 2
U = (; 3
V = φ; 4
while U ,= φ do 5
foreach subset S ∈ Z do 6
Weight(S) = [U ∩ S[ −α([S[ −[U ∩ S[); 7
end 8
Find a set S ∈ Z that maximizes Weight(S); 9
if Weight(S) ≤ 0 then 10
/* This means that the set C can not be covered by subsets
in T */
return φ and exit; 11
end 12
U = U −S; 13
Z = Z −¦S¦; 14
V = V ∪ ¦S¦; 15
end 16
return V ; 17
It is obvious that when 0 ≤ α ≤ 1/n, the approximation ratio of ln ( +1 holds for
the weighted set covering algorithm. This is because both the weighted set covering
algorithm and the general set covering algorithm picks the same subset at each stage
except when there is a tie (i.e., several subsets covers the same maximum number of
uncovered elements). When there is such a tie, the weighted set cover algorithm picks
a subset that covers the minimum number of already covered elements. However, the
approximation ratio of the general greedy algorithm holds regardless of how ties are
5.3. RFID READERS DEPLOYMENT (DETERMINISTIC
DEPLOYMENT) 150
dealt with.
5.3.4 Load Balancing
The algorithms we have discussed so far aim at reducing the total number of RFID
readers and reducing overlapping among different readers. However, the issue of load
balancing is not considered. Therefore, we may get solutions in which readers have
significant variations in their workloads (i.e., the number of tags they cover), which
lowers the performance
1
of the whole system. This is because each RFID reader is
associated with a subset of tags and readers interrogate their tags in parallel. Thus,
the reader with the maximum workload determines the overall performance because
it will be the last to finish. For example, a system with two RFID readers each covers
50 tags is much better than a system with two readers one covers 90 tags and one
covers 10 tags.
In this subsection we present a method that distributes the load evenly among
RFID readers for the sake of improving the overall performance of the system. The
main idea is to put a constraint on the maximum number of tags β, which are covered
by a single reader. However, the objective of minimizing the number of readers and
that of minimizing β conflict with each other; the minimum possible value of β is 1,
which means that each reader covers a single tag; and the minimum number of read-
ers may be one reader covering all tags, which results in the maximum value for β.
This can be dealt with either by putting a constraint on β and finding the minimum
number of readers meeting that constraint, or putting a constraint on the maximum
number of readers R and finding the minimum value of β meeting that constraint.
The former option is straight forward and can be done by excluding intersection points
1
The performance herein is defined as the time needed to interrogate all tags.
5.3. RFID READERS DEPLOYMENT (DETERMINISTIC
DEPLOYMENT) 151
Algorithm 8: RFID readers deployment algorithm with load balancing.
Readers Deployment(T,α, R) 1
Input: T : set of tags.
T: set of readers.
α: weight.
Output: V : set of readers.
C = T; 2
Q = the set of all intersection points; 3
T = ¦Cover(q) : q ∈ Q¦; 4
β
max
= MAX
S∈F
[S[; 5
β
min
= 1; 6
while β
max
> β
min
do 7
β = ¸
βmax+β
min
2
|; 8
Z = ¦S : S ∈ T ∧ [S[ ≤ β¦; 9
V =Set Cover((,Z,α); 10
if 0 < [V [ ≤ R then 11
β
max
= β; 12
else 13
β
min
= β + 1; 14
end 15
end 16
Z = ¦S : S ∈ T ∧ [S[ ≤ β
max
¦; 17
V =Set Cover((,Z,α); 18
if 0 < [V [ ≤ R then 19
return V ; 20
else 21
return φ; 22
end 23
whose coverage exceeds β (i.e., when [Cover(q)[ > β, Cover(q) is excluded from T).
The latter option is less trivial. In fact, one may need to try all possible values of β
and take the minimum value that meets the constraint of having at most R readers.
That can be done more efficiently by doing a binary search over all possible values of
β (i.e., [1..MAX
S∈F
[S[]). This method is illustrated in Algorithm 8. The overall com-
putational complexity of this algorithm is O(log (MAX
S∈F
[S[)[C[[T[ MIN([C[, [T[)).
5.4. RFID REDUNDANT READERS ELIMINATION (RANDOM
DEPLOYMENT) 152
5.3.5 Displaced tags
While we assume that exact locations of tags are known, it is expected in practice
to have some tags being displaced from their planned locations. However, this can
be easily accommodated in our scheme; if we reduce the actual transmission ranges
of tags by a value of r units, our scheme becomes resilient to any displacement of at
most r units from the planned locations.
5.4 RFID Redundant Readers Elimination (Random Deployment)
In this section, we present our redundant reader elimination algorithm. We start
with some assumptions and definitions we use in this work and that is followed by
a detailed description of the algorithm we propose. We then show further details
of the algorithm using an illustrative example. Finally, we discuss the algorithm’s
computational and communication complexity.
5.4.1 Problem Definition
Formally, the redundant reader problem is defined as follow. Given a set of RFID
tags and a set of RFID readers covering all tags, find the minimum cardinality subset
of RFID readers that cover all the tags [16]. To this extent we propose the Neighbor
and Tag Estimation (NTE) algorithm.
5.4.2 Neighbors and Tags Estimation (NTE) based algorithm
The redundant reader elimination problem is known to be NP-hard [16]. Therefore,
we propose the NTE algorithm, which is a heuristic greedy algorithm, to find near
optimal solutions in a reasonable time. The main idea is that when a reader R
i
has
5.4. RFID REDUNDANT READERS ELIMINATION (RANDOM
DEPLOYMENT) 153
a large number of neighboring readers, tags covered by R
i
have a high probability
of being also covered by other readers in the neighborhood of R
i
. The algorithm
uses this observation to identify redundant readers. For each active reader R
i
, the
algorithm assigns a weight based on the ratio of the number of active tags covered
by R
i
to the number of active neighboring readers of R
i
. Initially, all tags are active
and all readers are active. At the beginning of the each iteration, weights of active
readers are calculated and then a reader with the maximum weight is deactivated and
all tags covered by that reader are deactivated, too. Any reader with a weight of 0 is
considered to be redundant and deactivated. This process continues until all tags are
inactive, at which point all remaining active readers are considered to be redundant.
The pseudocode for the NTE algorithm is shown in Algorithm 9.
In the presentation of our algorithm, we use the following notations:
• 1 = ¦R
1
, R
2
, . . . , R
n
¦ is the set of active readers.
• R
a
= ¦R
a
1
, R
a
2
, . . . , R
a
m
¦ is the set of non-redundant readers.
• R
r
= ¦R
r
1
, R
r
2
, . . . , R
r
l
¦ is the set of redundant readers.
• E(R
i
) is the estimated number of active tags within the interrogation range of
reader R
i
.
• N(R
i
) is the number of active neighboring (interfering) readers of reader R
i
.
The algorithm begins with basic housekeeping (lines 3 and 4). At the beginning
of each iteration, each active reader R
i
estimates the number of tags it covers (i.e.,
E(R
i
)) and determines the number of active neighboring readers it has (i.e., N(R
i
))
(lines 8 and 9). To estimate the number of tags, the scheme presented in [62] can
5.4. RFID REDUNDANT READERS ELIMINATION (RANDOM
DEPLOYMENT) 154
Algorithm 9: The greedy NTE redundant reader elimination algorithm.
NTE-Reader(1): 1
Input: 1: set of all readers.
Output: 1
r
: set of redundant readers.
// Initially all tags are active; 2
1
a
= φ; 3
1
r
= φ; 4
while 1 , = φ do 5
max ← 0 ; 6
foreach R
i
∈ 1 do 7
E(R
i
)= An estimate to the number of active tags covered by R
i
; 8
N(R
i
) = The number of active neighboring readers for R
i
; 9
if E(R
i
) = 0 then 10
1
r
← 1
r
∪ ¦R
i
¦; 11
1 ← 1−¦R
i
¦; 12
end 13
else 14
if N(R
i
) > 0 then 15
Ratio[i] ← E(R
i
)/N(R
i
); 16
end 17
else 18
Ratio[i] = ∞; 19
end 20
if max < Ratio[i] then 21
max ← Ratio[i]; 22
idx ← i; 23
end 24
end 25
end 26
1
a
← 1
a
∪ ¦R
idx
¦; 27
1 ← ¦1¦ −¦R
idx
¦; 28
All tags covered by R
idx
are deactivated; 29
// using a broadcasted request. 30
end 31
return 1
r
; 32
5.4. RFID REDUNDANT READERS ELIMINATION (RANDOM
DEPLOYMENT) 155
be used as it can estimate the number of tags within 99% accuracy ratio. The
active neighboring readers of a reader R
i
are those readers which are within the
communication range of R
i
and, hence, can be easily counted. Readers that do not
cover any active tags (i.e., E(R
i
) = 0) are considered to be redundant (lines 10-
13). Amongst active readers that still cover at least one active tag, the one with
the maximum weight (i.e., Ratio[i]) is selected and excluded from the redundancy
list (lines 27 and 28), and active tags covered by that reader are deactivated (line
29). Any tie is broken arbitrarily. This can be done by a single message broadcasted
by the selected reader. The selected reader itself is deactivated and not counted
in the number of neighboring readers during subsequent iterations. The algorithm
terminates when the set of readers R is empty (i.e., all readers have been classified
as redundant or not).
A snapshot of an example of an RFID network before and after applying the
NTE algorithm is shown in Fig. 5.3, shows how effective the NTE algorithm is in
eliminating a significant number of redundant readers.
5.4.3 Exemplary Scenario
We show in this subsection how our algorithm finds the redundant readers in the
scenario shown in Fig. 5.2. The set of readers 1 = ¦R
1
, R
2
, R
3
, R
4
¦ is given, as
input, to the NTE algorithm. Tracing the NTE algorithm on this instance is shown
in Table 5.1.
In the first step, all readers in 1 estimate the number of active tags they cover
and the number of active neighboring readers they have. Readers R
1
, R
2
, R
3
and R
4
have 2 tags, 1 tag, 2 tags and 2 tags, respectively. The numbers of active neighboring
5.4. RFID REDUNDANT READERS ELIMINATION (RANDOM
DEPLOYMENT) 156
1
(a) Network before the execution of NTE algorithm
1
(b) Network after the execution of NTE algorithm
Figure 5.3: Snapshot of random RFID network from the simulator.
5.4. RFID REDUNDANT READERS ELIMINATION (RANDOM
DEPLOYMENT) 157
readers R
1
, R
2
, R
3
and R
4
have are 1, 2, 3 and 2, respectively. Accordingly, reader R
1
will have the maximum weight and, hence, will be excluded from the list of redundant
readers and deactivated be removing it from the set 1 . Active tags covered by R
1
(i.e., T
1
and T
2
) are deactivated.
In the second step, active readers (i.e., the set 1) estimate the number of active
tags they cover and the number of active neighboring readers they have. At this
stage, reader R
1
and tags T
1
and T
2
are not counted. Therefore, readers R
2
, R
3
and
R
4
cover 1 tag, 1 tag and 2 tags, respectively. The numbers of active neighboring
readers they have are 2, 2 and 2, respectively. R
4
is picked as it has the maximum
weight.
In the last step, readers R
2
and R
3
will be added to the list of redundant readers
as they do not cover any active tag. As a result, the readers R
1
and R
4
can be used to
interrogate all tags and readers R
2
and R
3
will be turned off as they are redundant.
5.4.4 Computational Complexity
Estimating the number of tags covered by a reader can be done in a constant time [62].
Deactivating tags covered by a particular reader can be also done in a constant time,
as it just requires a single broadcasted message. When a reader is deactivated, it can
broadcast a message to its neighboring readers telling them that it is going into an
inactive mode; upon receiving such a message, reader decrements its number of active
neighboring readers. Thereby, readers manipulate the number of tags they cover and
the number of neighboring readers they have in a constant time. Now, we are left with
the while loop that have, in the worst case, O(N) iterations, where N is the number
of readers. The time complexity of each iteration is dominated by the process of
5.4. RFID REDUNDANT READERS ELIMINATION (RANDOM
DEPLOYMENT) 158
Table 5.1: Tracing the NTE algorithm for the scenario of Fig. 5.2.
(a) Step-I
Input: 1 = ¦R
1
, R
2
, R
3
, R
4
¦ 1
a
= φ
Reader E(R
i
) N(R
i
)
R
1
2 1
R
2
1 2
R
3
2 3
R
4
2 2
Output: 1 = ¦R
2
, R
3
, R
4
¦ 1
a
= ¦R
1
¦
(b) Step-II
Input: 1 = ¦R
2
, R
3
, R
4
¦ 1
a
= ¦R
1
¦
Reader E(R
i
) N(R
i
)
R
1
- -
R
2
1 2
R
3
1 2
R
4
2 2
Output: 1 = ¦R
2
, R
3
¦ 1
a
= ¦R
1
, R
4
¦
(c) Step-III
Input: 1 = ¦R
2
, R
3
¦ 1
a
= ¦R
1
, R
4
¦
Reader E(R
i
) N(R
i
)
R
1
- -
R
2
0 0
R
3
0 0
R
4
- -
Output: 1
r
= ¦R
2
, R
3
¦ 1
a
= ¦R
1
, R
4
¦
finding the reader with the maximum weight which is done in a centralized fashion
and, hence, takes O(M) time, where M is the number of active readers. Therefore,
the overall time complexity of the algorithm is O(N
2
).
5.5. SIMULATION SETUP AND RESULTS 159
5.5 Simulation Setup and Results
In this section, we evaluate the performance of the proposed algorithms and that of
other recently proposed schemes for the readers deployment and redundant readers
elimination in RFID. We use the number of readers, maximum load amongst the
readers and the number of redundant readers as our evaluation criteria.
5.5.1 Simulation Setup
We have implemented the RFID system using an in-house simulator, for both deter-
ministic deployment and random deployment.
Deterministic deployment: In the case of deterministic deployment, for compar-
ison, we use two recent readers deployment schemes, namely the optimal grid cov-
erage [83, 92] and the honey grid coverage [110]. Both schemes were selected based
on their relative performance with other schemes, from the literature. For rest of
the section, the two proposed schemes, i.e., Algorithm 7 and Algorithm 8 are termed
as Approximate Set Cover (ASC) and Approximate Set Cover with Load Balanc-
ing (ASC-LB), respectively. For evaluation, we use the number of readers and the
maximum load, i.e., the maximum number of tags a single reader may cover. Lower
number of readers imply less overlapping and, hence, less reader collisions. And lower
values for the maximum load amongst readers implies a higher overall reading rates.
We also evaluated the schemes for various tags load (sparse and dense) and various
interrogation areas. Unless otherwise mentioned, simulations for the deterministic
deployment are performed using the following parameters. The RFID reader has an
interrogation range of 4 −5m and an interrogation area of 5x5x100m
3
. Tags are as-
sumed to be placed in a 3D space modelled as a rectangular cuboid, and readers are
5.5. SIMULATION SETUP AND RESULTS 160
to be placed on particular sides of the rectangular cuboid, as is illustrated in Figure
5.1. Locations of tags are generated using pseudo-random number generator. The
performance metrics are averaged over twenty different runs generated using distinct
random seeds.
Random deployment: In the case of random deployment, or comparison, we use
recently proposed algorithms such as RRE [16], LEO+RRE [52] and TREE [109].
These schemes were selected based on the their relative performance improvements
over other schemes in the literature. For comparative analysis, our evaluation met-
rics are the number of redundant readers, the number of tag reads and the number
of tag writes. A high number of redundant readers imply the scheme is effective in
detecting and eliminating redundant readers. A low number of tag reads and writes
implies that the scheme is light-weight and does not require state maintenance on
the tags. Furthermore, we evaluated the NTE scheme in terms of resilience to the
reader’s range. Unless otherwise mentioned, simulations for the random deployment
are performed using the following parameters. The RFID reader has an interrogation
range of 5m and an interrogation area of 250x250m
2
. Both tags and readers are as-
sumed to be randomly place within the interrogation area. The results are averaged
over twenty different runs generated using distinct random seeds.
5.5.2 Simulation Results for Deterministic Deployment
Number of Readers: Minimal number of readers with maximal required cover-
age is an important evaluation metric as it translates into minimal communication
overheads, minimal communication interferences and is economically sound. Fig. 5.7
shows the number of readers required to cover the same interrogation area for various
5.5. SIMULATION SETUP AND RESULTS 161
20
40
60
80
100
120
140
160
0 100 200 300 400 500 600 700
N
u
m
b
e
r

o
f

r
e
a
d
e
r
s
Number of tags
Optimal-Grid
Honey-Grid
ASC
ASC-LB
Figure 5.4: Number of readers required for various tag density.
algorithms. Our proposed ASC algorithm is the most prominent as it significantly
reduces the required readers, an average two times reduction compared with Honey-
Grid is obtained. Both Optimal-Grid and Honey-Grid quickly reaches its maximum
allowable readers as the tags density exceed certain limits. Whereas, for the pro-
posed algorithms, ASC and ASC-LB, such a saturation point comes much later hence,
translating into lesser readers while still covering the same tags set. In highly dense
environment, in terms of readers, the two proposed schemes and optimal-grid would
perform same, as after certain number of tags, either scheme requires certain number
of reader to achieve 100% coverage.
Load Balancing: The ASC algorithm guarantees minimum number of readers
however, maybe results in imbalance number of tags between the readers. Fig. 5.5
shows the maximum number of tags, amongst all the readers interrogation range for
various algorithms. As anticipated, having reader deployment scheme, without any
5.5. SIMULATION SETUP AND RESULTS 162
0
5
10
15
20
25
30
35
40
45
0 100 200 300 400 500 600 700
M
a
x
i
m
u
m

l
o
a
d
Number of tags
Optimal-Grid
Honey-Grid
ASC
ASC-LB
Figure 5.5: Maximum load (tags) for various tag density.
load balance, will yield reader(s) having much larger set of tags to interrogate, com-
pared with others. For instance, as the number of tags increases, the loads for all
the schemes increase. Therefore, we may get solutions in which readers have signifi-
cant variations in their workloads (i.e., the number of tags they cover), which lower
the performance of overall system. As expected, the proposed load-aware placement
scheme, i.e., ASC-LB performs the best, as it is distribute the load many-folds effi-
cient compared with others. The effect of increasing the interrogation area, in sparse
(100tags) and dense (500tags) settings, are shown in Fig. 5.6. With an increase in the
interrogation areas and tags cardinality, our proposed algorithms are more scalable
i.e., can handle diverse tag deployments. However, trade-off exists between mini-
mal readers and load balancing as the later would require more readers in order to
maintain the balance amongst them.
5.5. SIMULATION SETUP AND RESULTS 163
0
20
40
60
80
100
50 100 150 200 250
N
u
m
b
e
r

o
f

r
e
a
d
e
r
s
Interrogation area
Optimal-Grid
ASC
ASC-LB
(a) 100 tags
40
60
80
100
120
140
160
180
200
220
240
50 100 150 200 250
N
u
m
b
e
r

o
f

r
e
a
d
e
r
s
Interrogation area
Optimal-Grid
ASC
ASC-LB
(b) 500 tags
Figure 5.6: Effect of tag density and interrogation area on the readers deployment
algorithms.
5.5. SIMULATION SETUP AND RESULTS 164
5.5.3 Simulation Results for Random Deployment
Redundant Readers: The overall network performance, mainly in terms of en-
ergy consumption and reading delay, is influenced by interference among readers [42].
Hence, an effective scheme is one that detects more redundant readers. Compar-
isons of different schemes based on the number of redundant readers they detect for
both dense (50x50m
2
) and sparse (250x250m
2
) networks are depicted in Fig. 5.7. In
dense environments, the NTE algorithm detects redundant readers far beyond other
schemes whereas in a sparse environment the improvement obtained is not signifi-
cant. For instance, in dense environments, the NTE schemes can detect 15% more
redundant readers than the best of the other schemes, i.e., LEO+RRE. However, in
sparse environments only 2% improvement is observed over LEO+RRE. There are
two reasons behind this behavior. First, in a dense environment a reader has a good
chance of having a large number of neighbors and picking a reader with the maxi-
mum tags to neighbors ratio as a non-redundant reader has the potential to render
many neighboring readers redundant, which is not the case in a sparse environment.
Second, in a sparse environment, readers are expected to have a limited variation in
the tags to neighbors ratio they have, and accordingly this ratio is not as effective as
it is in a dense environment.
Tags Reads and Writes: Most of the existing schemes, e.g., RRE and LEO, in the
literature have high communication complexities associated with them. Communica-
tion complexity is defined as the number of reads and writes operation made from
and to tags. Tag singulation is required in order to read-from a tag. Tag singulation
is an expensive process as it is determined by the number of tags within a reader’s
interrogation range and the overlapping among readers. Fig. 5.8 and Fig. 5.9 show
5.5. SIMULATION SETUP AND RESULTS 165
240
260
280
300
320
340
360
380
0 1000 2000 3000 4000 5000
R
e
d
u
n
d
a
n
t

r
e
a
d
e
r
s
Number of Tags
RRE
LEO
LEO + RRE
TRRE
NTE
(a) Simulation area of 50x50m
2
100
200
300
400
500
600
700
800
0 1000 2000 3000 4000 5000
R
e
d
u
n
d
a
n
t

r
e
a
d
e
r
s
Number of Tags
RRE
LEO
LEO + RRE
TRRE
NTE
(b) Simulation area of 250x250m
2
Figure 5.7: Redundant reader detected in sparse and dense network.
5.5. SIMULATION SETUP AND RESULTS 166
0
10000
20000
30000
40000
50000
60000
0 1000 2000 3000 4000 5000
T
a
g

r
e
a
d
s
Number of Tags
RRE
LEO
LEO + RRE
TRRE
NTE
(a) Simulation area of 50x50m
2
0
10000
20000
30000
40000
50000
60000
0 1000 2000 3000 4000 5000
T
a
g

r
e
a
d
s
Number of Tags
RRE
LEO
LEO + RRE
TRRE
NTE
(b) Simulation area of 250x250m
2
Figure 5.8: Number of tag reads in sparse and dense environments.
5.6. SUMMARY 167
comparisons between different schemes in terms of the number of read and write
operations they incur. The number of reads and writes is obviously lower in sparse
environments than it is in dense environments for all schemes. However, the number
of reads and writes in the NTE scheme is many folds less than those in other schemes.
This is because the reader, in the NTE scheme, writes only once to the tags within
its range. Furthermore, each tag receives a single write operation.
Operations in the tags count estimation are not counted as read operations. This
is because the estimation scheme does not require tags singulation. Read operations
that make a difference are ones that require singulation and no such an operation is
needed in the NTE scheme. Therefore, although the number of redundant readers
detected by the LEO+RRE scheme is similar to that in the NTE scheme, in the sparse
environment, the NTE scheme is much more efficient in terms of energy consumption
and delay. In fact the NTE algorithm is at least six times more efficient in accessing
the tags memory.
5.6 Summary
Coverage is crucial for large scale RFID applications wherein large number of readers
can be deployed in deterministic and random manners. In deterministic approach,
minimum readers must be deployed to maintain certain coverage. In random ap-
proach, once the readers are deployed, redundant readers must be detected to avoid
unnecessary interference caused by them. Several schemes have been proposed in
the literature, however, issues such as interrogation zone overlap among readers, load
balancing amongst reader and light-weight redundant readers detection schemes have
received little attention. In this chapter, for deterministic approach, we propose an
5.6. SUMMARY 168
0
5000
10000
15000
20000
25000
30000
0 1000 2000 3000 4000 5000
T
a
g

w
r
i
t
e
s
Number of Tags
RRE
LEO
LEO + RRE
TRRE
NTE
(a) Simulation area of 50x50m
2
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
0 1000 2000 3000 4000 5000
T
a
g

w
r
i
t
e
s
Number of Tags
RRE
LEO
LEO + RRE
TRRE
NTE
(b) Simulation area of 250x250m
2
Figure 5.9: Number of tags write in sparse and dense environments.
5.6. SUMMARY 169
algorithm that approximately determines the minimal number of readers and the loca-
tion of each reader in a 3D space, and guarantees less overlapping among readers and
a balanced distribution of load among them. Also, for random approach, light-weight
greedy algorithm to detect and eliminate redundant readers in RFID networks. By
reducing the number of readers, either to-be deployed or to-be active at one time, not
only is the overall network cost reduced, but the overall system performance is im-
proved also by reducing interference and overlapping among readers coverage regions
and, hence, having less collisions among readers.
170
Chapter 6
Conclusion
The last decade has witnessed a growing interest in RFID due to their unique poten-
tial in a wide range of applications. RFID systems provide low cost and low power
object identification and tracking mechanisms. While being the key requirement
for anticipated ultra large scale applications, e.g., Internet of Things, RFID brings
several design challenges that need to be overcome. One of these challenges is tag
anti-collision and hence, the reading rates of RFID tags. The tags, especially passive
RFID tags, with functional constraints due to limited power cannot adopt conven-
tional medium access control approaches. Another challenge is scaling to large scale
application (a typical scenario in Internet of Things) which involve a large number of
tags within the interrogation zone of single or multiple readers. This scenario results
in a large number of tag collisions and therefore, lowers overall system reading rate.
Furthermore, power requirements and signal attenuation limit the reading range. A
promising solution to overcome these limitation exists at the protocol and system
architecture levels. At protocol level, an optimal power level algorithm which clusters
and isolates tags according to their physical distance can reduce tag collisions. At
the system architecture level, reducing collisions can be achieved by distributing the
6.1. SUMMARY 171
signal receiving task amongst spatially dispersed devices. These devices then selec-
tively collect and relay them to the reader; forming micro-interrogation zones which
are interrogated in parallel and autonomously.
6.1 Summary
Schemes presented in Chapter 3, Chapter 4 and Chapter 5 aim at enabling RFID
systems for ultra large scale of the anticipated applications, such as the Internet
of Things and item-level tagging. These schemes achieve higher reading rates and
range, using sophisticated medium access control and architecture design, and to
provide load-aware and energy-efficient reader deployment to prolong overall network
lifetime.
In Chapter 3, we introduce the use of reader’s power control to mitigate tag
collisions. By interrogating a subset of tags, one at a time, collisions are reduced
and hence higher reading rates can be achieved. To this end, two optimal power
stepping algorithms are proposed. The first scheme targets an ideal setting in which
the transmission power, and hence, the transmission range of an RFID reader can be
tuned with high precision. The second scheme relaxes this assumption and considers
an RFID reader with a finite and discrete transmission range. For each optimization
problem, we present mathematical delay analysis and devise an efficient method to
find the optimal clustering.
In Chapter 4, we introduce a new paradigm in RFID systems, namely Distributed
Receiving RFID. The objective is to lessen the limitations and alleviate bottlenecks
imposed by centralized architectures; with the goal of boosting the operational per-
formance of RFID systems. In the proposed paradigm, the task of collecting the
6.1. SUMMARY 172
modulated backscattered signals is dispersed to spatially distributed entities (devices)
within a reader’s interrogation zone. This model supports multipoint communication
and hence eliminates redundant interrogation of the same set of tags, by overlapping
RFID readers. To this end, we propose two schemes. In the first scheme, using the
spatial devices and by adjusting the tags reflectivity co-efficient, micro-interrogation
zones are formed. Micro-zones spatially isolate tags within the readers interrogation
zone therefore facilitate higher bandwidth communication link and data rates. In the
second scheme, exploiting micro-zones, the tags are interrogated in parallel; namely
parallel singulation. We propose both a deterministic and a probabilistic model of
this parallel singulation mechanism. Parallel singulation assists in achieving a higher
reading rate under both dense and mobile tag environments.
In Chapter 5, we introduce RFID coverage as deterministic and random deploy-
ment problems. The complexity and monetary cost associated with ultra large scale
RFID systems strongly depends on the number of readers being deployed to inter-
rogate tags. To this end, three RFID deployment schemes are proposed. The first
scheme, targeting deterministic deployment, proposes a set-cover approximation al-
gorithm to determine the minimal number of RFID readers required to maintain
the requested coverage while having minimal interrogation overlapping among the
readers. The second scheme then extends the algorithm to facilitate load-balancing
deployment strategies. The third scheme, targeting random and post deployment
coverage, proposes the use of neighbouring readers ratio and covering tags to heuris-
tically turn off redundant readers to achieve the maximal coverage with the least
possible readers and overlapping interrogation zones.
6.2. FUTURE OUTLOOK 173
6.2 Future Outlook
Several future research problems stem from our work thus far. In this section, we
point out some of them.
1. Current proposals for RFID systems assume near-ideal interrogation environ-
ment. A more challenging problem is evaluating the performance of an RFID
system and the proposed algorithms in a testbed setting.
2. The parallel singulation algorithm makes use of the binary search tree and
the query tree algorithm. An efficient and tailored algorithm, aware of the
underlying distributed receiving reader-tag communication architecture, will
yield optimal data rates.
3. In Chapter 4, the fielders are deployed in a grid, thus have problematic overlap-
ping regions. These regions result in idle queries. Optimized fielder deployment
techniques, and formation of micro-zones with minimal overlapping, is an inter-
esting problem as it would to further boost system performance. Novel MAC
protocols, for both fielders and reader, need to be developed in a distributed
context.
4. In Chapter 5, reader deployment algorithms assume that RFID readers have
a fixed transmission range. An interesting problem is optimizing and/or dy-
namically controlling the transmission range of each reader to minimize reader
overlap and balance workload among readers.
BIBLIOGRAPHY 174
Bibliography
[1] LTC5507: 100kHz to 1GHz RF power detector.
[2] ADG901/ADG902: Wideband, 40 db isolation at 1 GHz, CMOS 1.65 v to 2.75
v, SPST switches, rev. b, 2005.
[3] SKY12329-350LF: GaAs digital attenuator 5-bit, 1 db LSB 400 MHz4 GHz,
Sep 2007.
[4] CC2500 single chip low cost low power RF transceiver, May 2008.
[5] eZ430-RF2500-SEH solar energy harvesting development tool, May 2010.
[6] MSP430x22x2, MSP430x22x4 mixed signal microcontroller, March 2010.
[7] The network simulator ns-2, available at http://www.isi.edu/nsnam/ns/, 2011.
[8] www.skyetek.com, 2011.
[9] K. Ali and H. Hassanein. Passive RFID for intelligent transportation systems. In
Proceedings of the 6th Annual IEEE Consumer Communications & Networking
Conference (CCNC), pages 1–2, 2009.
[10] K. Ali, H. Hassanein, and A. M. Taha. RFID anti-collision protocol for dense
passive tag environments. In LCN ’07: Proceedings of the 32nd IEEE Confer-
ence on Local Computer Networks, pages 819–824, 2007.
[11] W. Alsalih, S. Akl, and H. Hassanein. Placement of multiple mobile data
collectors in underwater acoustic sensor networks. In Proceedings of the IEEE
International Conference on Communications (ICC), pages 2113–2118, 2008.
[12] W. Alsalih, K. Ali, and H. Hassanein. Optimal distance-based clustering for
tag anti-collision in RFID systems. In Proceedings of the 33rd IEEE Conference
on Local Computer Networks (LCN), pages 266–273, 2008.
BIBLIOGRAPHY 175
[13] N. Amin, Ng Wen Jye, and M. Othman. A BPSK backscatter modulator design
for RFID passive tags. In Proceedings of the IEEE International Workshop on
Radio-Frequency Integration Technology, pages 262–265, 2007.
[14] R. Angeles. RFID technologies: Supply-chain applications and implementation
issues. Information Systems Management, 22(1):51–65, 2005.
[15] S.M. Birari and S. Iyer. Mitigating the reader collision problem in RFID net-
works with mobile readers. In Proceedings of the 13th IEEE International Con-
ference on Networks, pages 463–468, 2005,.
[16] C. Bogdan, K. R. Murali, K. Mehmet, H. Christoph, and G. Ananth.
Redundant-reader elimination in RFID systems. In Proceedings of the Sec-
ond Annual IEEE Communications Society Conference on Sensor and Ad Hoc
Communications and Networks (SECON), pages 176–184, 2005.
[17] L. Bolotnyy and Robins G. Randomized pseudo-random function tree walking
algorithm for secure radio-frequency identification. In Proceedings of the Fourth
IEEE Workshop on Automatic Identification Advanced Technologies, pages 43–
48, 2005.
[18] J. Capetanakis. Tree algorithms for packet broadcast channels. IEEE Transac-
tions on Information Theory, 25(5):505–515, 1979.
[19] B. Carbunar, M. K. Ramanathan, M. Koyuturk, C. Hoffmann, and A. Grama.
Redundant reader elimination in RFID systems. In Proceedings of the Sec-
ond Annual IEEE Communications Society Conference on Sensor and Ad Hoc
Communications and Networks (SECON), pages 176–184, 2005.
[20] M. Cardei and D-Z. Du. Improving wireless sensor network lifetime through
power aware organization. Wireless Networking, 11:333–340, May 2005.
[21] J. Cha and J. Kim. Novel anti-collision algorithms for fast object identification
in RFID system. In Proceedings of the 11th International Conference on Parallel
and Distributed Systems, pages 63–67, 2005.
[22] J. Cha and J. Kim. Novel anti-collision algorithms for fast object identification
in RFID system. In Proceedings 11th International Conference on Parallel and
Distributed Systems, pages 63–67, 2005.
[23] J. Cha and J. Kim. Dynamic framed slotted aloha algorithms using fast tag
estimation method for RFID system. In Proceedings of the 3rd IEEE Consumer
Communications and Networking Conference (CCNC), pages 768–772, 2006.
BIBLIOGRAPHY 176
[24] H. S. Chae, J. G. Park, J. F. Cui, and J. S. Lee. An adaptive load balancing
management technique for RFID middleware systems. Software Practice and
Experience, 40(6):485–506, 2010.
[25] H. Chen, Y. Zhu, and K. Hu. Multi-colony bacteria foraging optimization
with cell-to-cell communication for RFID network planning. Applied Software
Computing, 10(2):539–547, 2010.
[26] W. Chen, S. Horng, and P. Fan. An enhanced anti-collision algorithm in RFID
based on counter and stack. In Proceedings of the Second International Confer-
ence on Systems and Networks Communications (ICSNC), pages 21–24, 2007.
[27] C. H. Cheng and R. D. Murch. Asymmetric RFID tag antenna. In Proceedings
of the IEEE International Symposium of Antennas and Propagation Society,
pages 1363–1366, 2006.
[28] K. W. Chiang, C. Hua, and T. P. Yum. Prefix-randomized query-tree protocol
for RFID systems. In Proceedings of the IEEE International Conference on
Communications (ICC), pages 1653–1657, 2006.
[29] K.W. Chiang, C. Hua, and P. Yum. Prefix-length adaptation for PRQT pro-
tocol in RFID systems. In Proceedings of the IEEE Global Telecommunications
Conference (Globecom), pages 1–5, 2006.
[30] J. Choi, D. Lee, and H. Lee. Bi-slotted tree based anti-collision protocols for fast
tag identification in RFID systems. IEEE Communications Letters, 10(12):861–
863, 2006.
[31] J. Choi, D. Lee, and H. Lee. Query tree-based reservation for efficient RFID
tag anti-collision. IEEE Communications Letters, 11(1):2007, 85–87.
[32] J.H. Choi, D. Lee, H. Jeon, J. Cha, and H. Lee. Enhanced binary search with
time-divided responses for efficient RFID tag anti-collision. In Proceedings of the
IEEE International Conference on Communications (ICC), pages 3853–3858,
2007.
[33] T. Cormen, C. Leiserson, R. Rivest, and C. Stein. Introduction To Algorithms
(2nd Edition). McGraw-Hill, 2001.
[34] J. Curty, N. Joehl, C. Dehollain, and M.J. Declercq. Remotely powered ad-
dressable UHF RFID integrated system. IEEE Journal of Solid-State Circuits,
40(11):2193–2202, Nov. 2005.
BIBLIOGRAPHY 177
[35] K. Dasgupta, M. Kukreja, and K. Kalpakis. Topology-aware placement and
role assignment for energy-efficient information gathering in sensor networks.
In Proceedings of the 8th IEEE Symposium on Computers and Communication
(ISCC), pages 341–348, 2003.
[36] D. M. Dobkin. The RF in RFID: Passive UHF RFID in Practice. Newnes,
2005.
[37] Q. Dong, A. Shukla, V. Shrivastava, D. Agrawal, S. Banerjee, and K. Kar.
Load balancing in large-scale RFID systems. In Proceedings of the 26th IEEE
International Conference on Computer Communications (INFOCOM), pages
2281–2285, 2007.
[38] A. Dunkels, B. Grnvall, and T. Voigt. Contiki - a lightweight and flexible
operating system for tiny networked sensors. In Proceedings of the 29th Annual
IEEE International Conference on Local Computer Networks (LCN), pages 455–
462, 2004.
[39] D.W. Engels and S. E. Sarma. The reader collision problem. In Proceedings
of the IEEE International Conference on Systems, Man and Cybernetics, pages
641–646, 2002.
[40] EPCglobal. EPC radio-frequency identity protocols class-1 generation-2 UHF
RFID protocol for communications at 860 MHz - 960 MHz version 1.2.0. 2008.
[41] B. Feng, J. Li, J. Guo, and Z. Ding. Id-binary tree stack anticollision algorithm
for RFID. In Proceedings of the 11th IEEE Symposium on Computers and
Communications, pages 207–212, 2006.
[42] K. Finkenzeller. RFID Handbook: Fundamentals and Applications in Contact-
less Smart Cards and Identification. John Wiley & Sons, Inc., 2003.
[43] C. Floerkemeier. Transmission control scheme for fast rfid object identification.
In Proceedings of the Fourth Annual IEEE International Conference on Per-
vasive Computing and Communications (PerCom) Workshops, pages 457–462,
2006.
[44] C. Floerkemeier. Bayesian transmission strategy for framed ALOHA based
RFID protocols. In Proceedings of the IEEE International Conference on RFID
2007, pages 228–235, 2007.
[45] P.R. Foster and R.A. Burberry. Antenna problems in RFID systems. In Pro-
ceedings of the IEE Colloquium on RFID Technology, pages 1–5, 1999.
BIBLIOGRAPHY 178
[46] F. Fuschini, F. Paolazzi C. Piersanti, and G. Falciasecca. Analytical approach to
the backscattering from UHF RFID transponder. IEEE Antennas and Wireless
Propagation Letters, 7:2008, 33–35.
[47] D. W. Gage. Command control for many-robot systems. In Proceedings of the
Nineteenth Annual AUVS Technical Symposium, pages 22–24, 1992.
[48] W. W. Gregg, W. E. Esaias, G. C. Feldman, R. Frouin, S. B. Hooker, C. R.
McClain, and R. H. Woodward. Coverage opportunities for global ocean color
in a multimission era. IEEE Transactions on Geosience and Remote Sensing,
36(5):1620–1627, 1998.
[49] J.D. Griffin and G.D. Durgin. Gains for RF tags using multiple antennas. IEEE
Transactions on Antennas and Propagation, 56(2):563–570, 2008.
[50] Q. Guan, Y. Liu, Y. Yang, and W. Yu. Genetic approach for network planning
in the rfid systems. In Proceedings of the Sixth International Conference on
Intelligent Systems Design and Applications (ISDA), pages 567–572, 2006.
[51] J. Ho, D. W. Engels, and S. E. Sarma. HiQ: a hierarchical Q-learning algorithm
to solve the reader collision problem. In Proceedings of the International Sym-
posium on Applications and the Internet Workshops (SAINT-W), pages 88–91,
2006.
[52] C-H. Hsu, Y-M. Chen, and C-T. Yang. A layered optimization approach for
redundant reader elimination in wireless RFID networks. In Proceedings of the
2nd IEEE Asia-Pacific Service Computing Conference (APSCC), pages 138–
145, 2007.
[53] J-W. Hung, I-H. Li, H-H. Lin, and J-A. Cai. The first search right algo-
rithm for redundant reader elimination in RFID network. In Proceedings of
the 9th WSEAS international conference on Software engineering, parallel and
distributed systems (SEPADS), pages 177–183, 2010.
[54] T. Hwang, B. Lee, Y. S. Kim, D. Y. Suh, and J. S. Kim. Improved anti-
collision scheme for high speed identification in RFID system. In Proceedings of
the First International Conference on Innovative Computing, Information and
Control (ICICIC), pages 449–452, 2006.
[55] N. Irfan and M.C.E. Yagoub. Efficient algorithm for redundant reader elimi-
nation in wireless RFID networks. International journal of computer science
issues, 3(11):1–8, May 2010.
BIBLIOGRAPHY 179
[56] K. Kar and S. Banerjee. Node placement for connected coverage in sensor
networks. In Proceedings of the Modeling and Optimization in Mobile, Ad Hoc
and Wireless Networks (WiOpt), 2003.
[57] U. Karthaus and M. Fischer. Fully integrated passive UHF RFID transponder
IC with 16.7-µw minimum rf input power. IEEE Journal of Solid-State Circuits,
38(10):1602–1608, 2003.
[58] G. Khandelwal, A. Yener, and M. Chen. OPT: Optimal protocol tree for ef-
ficient tag identification in dense RFID systems. In Proceedings of the IEEE
International Conference on Communications (ICC), pages 128–133, 2006.
[59] G. Khandelwal, A. Yener, K. Lee, and S. Serbetli. ASAP : A MAC protocol
for dense and time constrained RFID systems. In Proceedings of the IEEE
International Conference on Communications (ICC), pages 4028–4033, 2006.
[60] J. Kim, W. Lee, E. Kim, D. Kim, and K. Suh. Optimized transmission power
control of interrogators for collision arbitration in uhf rfid systems. IEEE Com-
munications Letters, 11(1):22–24, 2007.
[61] J. Kim, J. Yu, J. Myung, and E. Kim. Effect of localized optimal clustering
for reader anti-collision in RFID networks: fairness aspects to the readers. In
Proceedings of the 14th International Conference on Computer Communications
and Networks (ICCCN), pages 497–502, 2005.
[62] Murali Kodialam and Thyaga Nandagopal. Fast and reliable estimation schemes
in RFID systems. In Proceedings of the 12th Annual International Conference
on Mobile Computing and Networking (MobiCom), pages 322–333, 2006.
[63] C. Law, K. Lee, and K. Siu. Efficient memoryless protocol for tag identifica-
tion. In Proceedings of the 4th international workshop on Discrete algorithms
and methods for mobile computing and communications (DIALM), pages 75–84,
2000.
[64] Ching Law, Kayi Lee, and Kai-Yeung Siu. Efficient memoryless protocol for
tag identification. In Proceedings of the 4th International Workshop on Discrete
Algorithms and Methods for Mobile Computing and Communications (DIALM),
pages 75–84, 2000.
[65] Jeongkeun Lee, Taekyoung Kwon, Yanghee Choi, Sajal K. Das, and Kyung
ah Kim. Analysis of RFID anti-collision algorithms using smart antennas. In
Proceedings of the 2nd international conference on Embedded networked sensor
systems (SenSys), pages 265–266, 2004.
BIBLIOGRAPHY 180
[66] S. Lee, S. Joo, and C. Lee. An enhanced dynamic framed slotted ALOHA
algorithm for RFID tag identification. In Proceedings of the Second Annual
International Conference on Mobile and Ubiquitous Systems: Networking and
Services (MobiQuitous), pages 166–172, 2005.
[67] Y-D. Lee, K-K. Nae, E-J. Lee, H-H. Ron, J-W. Jung, Y-K. Han, J-W. Jung,
S-C. Baek, M-S. Jung, H-G. Cho, H-R. Oh, Y-R. Seong, and J-S. Park. ASK
modulator for RFID system using a novel variable DGS resonator. In Proceed-
ings of the 2007 European Microwave Conference, pages 1660–1663, 2007.
[68] K. S. Leong, M. L. Ng, and P. H. Cole. The reader collision problem in RFID
systems. In Proceedings of the IEEE International Symposium on Microwave,
Antenna, Propagation and EMC Technologies for Wireless Communications
(MAPE), pages 658–661, 2005.
[69] C. Lin and F. Lin. A simulated annealing algorithm for RFID reader networks.
In Proceedings of the IEEE Wireless Communications and Networking Confer-
ence (WCNC), pages 1669–1672, 2007.
[70] H. Liu, M. Bolic, A. Nayak, and I. Stojmenovic. Taxonomy and challenges of the
integration of RFID and wireless sensor networks. IEEE Network, 22(6):26–35,
November-December 2008.
[71] L. Liu, Z. Xie, J. Xi, and S. Lai. An improved anti-collision algorithm in
RFID system. In Proceedings of the 2nd International Conference on Mobile
Technology, Applications and Systems, pages 137–142, 2005.
[72] Y. Maguire and R. Pappu. An optimal Q-algorithm for the ISO 18000-6C
RFID protocol. IEEE Transactions on Automation Science and Engineering,
6(1):16–24, 2009.
[73] M. Marengoni, B. A. Draper, A. Hanson, and R. Sitaraman. A system to
place observers on a polyhedral terrain in polynomial time. Image and Vision
Computing, 18(10):773–780, 2000.
[74] S. Meguerdichian, F. Koushanfar, G. Qu, and M. Potkonjak. Exposure in
wireless ad-hoc sensor networks. In Proceedings of the 7th annual international
conference on Mobile computing and networking (MobiCom), pages 139–150,
2001.
[75] C. Mutti and C. Floerkemeier. CDMA-based RFID systems in dense scenarios:
Concepts and challenges. In Proceedings of the IEEE International Conference
on RFID, pages 215–222, 2008.
BIBLIOGRAPHY 181
[76] J. Myung and W. Lee. Adaptive binary splitting: a RFID tag collision arbitra-
tion protocol for tag identification. Mobile Network Applications, 11(5):711–722,
2006.
[77] J. Myung, W. Lee, J. Srivastava, and T. K. Shih. Tag-splitting: Adaptive
collision arbitration protocols for RFID tag identification. IEEE Transaction
on Parallel and Distributed System, 18(6):763–775, 2007.
[78] V. Namboodiri and L. Gao. Energy-aware tag anti-collision protocols for RFID
systems. In Proceedings of the Fifth Annual IEEE International Conference on
Pervasive Computing and Communications (PerCom), pages 23–36, 2007.
[79] V. Namboodiri and L. Gao. Energy-aware tag anti-collision protocols for RFID
systems. In Proceedings of the Fifth Annual IEEE International Conference
on Pervasive Computing and Communications (PerCom), pages 23–36, March
2007.
[80] M. Nanjundaiah and V. Chaudhary. Improvement to the anticollision protocol
specification for 900MHz class 0 radio frequency identification tag. In Proceed-
ings of the 19th International Conference on Advanced Information Networking
and Applications (AINA), pages 616–620, 2005.
[81] P.V. Nikitin and K.V.S. Rao. Theory and measurement of backscattering from
RFID tags. IEEE Antennas and Propagation Magazine, 48(6):212–218, 2006.
[82] P.V. Nikitin, K.V.S. Rao, and R.D. Martinez. Differential RCS of RFID tag.
Electronics Letters, 43(8):431–432, 2007.
[83] A. Oztekin, F. Mahdavi, K. Erande, Z. Kong, L. K. Swim, and S. T. S. Bukkap-
atnam. Criticality index analysis based optimal RFID reader placement models
for asset tracking. International Journal of Production Research, 48(9):2679–
2698, 2010.
[84] J. ORourke. Art Gallery Theorems and Algorithms. Oxford University Press,
1987.
[85] J. Park, M. Y. Chung, and T. Lee. Identification of rfid tags in framed-slotted
ALOHA with tag estimation and binary splitting. In Proceedings of the First
International Conference on Communications and Electronics (ICCE), pages
368–372, 2006.
[86] J. Park, M. Y. Chung, and T. Lee. Identification of RFID tags in framed-slotted
ALOHA with robust estimation and binary selection. IEEE Communications
Letters, 11(5):452–454, 2007.
BIBLIOGRAPHY 182
[87] N. Pastos and R. Viswanathan. A modified grouped-tag tdma access protocol
for radio frequency identification networks. In Proceedings of the IEEE Wireless
Communications and Networking Conference (WCNC), pages 512–516, 2000.
[88] N. Pastos and R. Viswanathan. A modified grouped-tag TDMA access protocol
for radio frequency identification networks. In Proceedings of the IEEE Wireless
Communications and Networking Conference (WCNC), pages 512–516, 2000.
[89] V. Raghunathan, C. Schurgers, S. Park, and M. B. Srivastava. Energy-aware
wireless microsensor networks. IEEE Signal Processing Magazine, 19:40–50,
2002.
[90] D.C. Ranasinghe, D. W. Engels, and p. H. Cole. Security and privacy solutions
for low-cost RFID systems. In Proceedings of the 2004 Intelligent Sensors,
Sensor Networks and Information Processing Conference, pages 337–342, 2004.
[91] K.V.S. Rao, P. V. Nikitin, and S. F. Lam. Antenna design for UHF RFID
tags: a review and a practical application. IEEE Transactions on Antennas
and Propagation, 53(12):3870–3876, 2005.
[92] A. Reza, T. K. Geok, and K. Dimyati. Tracking via square grid of RFID reader
positioning and diffusion algorithm. Wireless Personal Communications, pages
1–24, 2010.
[93] A. W. Reza and T. K. Geok. Investigation of indoor location sensing via rfid
reader network utilizing grid covering algorithm. Wireless Personal Communi-
cations, 49(1):67–80, 2009.
[94] J. Ryu, H. Lee, Y. Seok, T. Kwon, and Y. Choi. A hybrid query tree proto-
col for tag collision arbitration in RFID systems. In Proceedings of the IEEE
International Conference on Communications (ICC), pages 5981–5986, 2007.
[95] X.-L. Shi, X.-W. Shi, Q.-L. Huang, and F. Wei. An enhanced binary anti-
collision algorithm of backtracking in RFID system. Progress In Electromagnetic
Research, 4:263–271, 2008.
[96] D. Shih, P. L. Sun, D. C.Yen, and S. M. Huang. Taxonomy and survey of RFID
anti-collision protocols. Computer and communication, 29(11):2150–2166, 2006.
[97] W. J. Shin and J. G. Kim. Partitioning of tags for near-optimum rfid anti-
collision performance. In Proceedings of the IEEE Wireless Communications
and Networking Conference (WCNC), pages 1673–1678, 2007.
BIBLIOGRAPHY 183
[98] C-S. Son, J-S. Park, H-G. Cho, and J-B. Lim. A novel variable-impedance mi-
crostrip circuit and its application to direct antenna modulation. In Proceedings
of the European conference on Wireless Technology, pages 403–406, 2005.
[99] P. Vandenameele. Space division multiple access for wireless local area networks.
Kluwer Academic Publishers, 2001.
[100] G. De. Vita, F. Bellatalla, and G. Iannaccone. Ultra-low power PSK backscat-
ter modulator for UHF and microwave RFID transponders. Microelectronics
Journal, 37(7):627–629, 2006.
[101] G. De Vita and G. Iannaccone. Design criteria for the RF section of UHF
and microwave passive RFID transponders. IEEE Transactions on Microwave
Theory and Techniques, 53(9):2978–2990, 2005.
[102] H. Vogt. Efficient object identification with passive RFID tags. In Proceedings of
the First International Conference on Pervasive Computing (Pervasive), pages
98–113, 2002.
[103] J. Waldrop, D. W. Engles, and S. E. Sarma. Colorwave: an anti-collision algo-
rithm for the reader collision problem. In Proceedings of the IEEE International
Conference on Communications (ICC), pages 1206–1210, 2003.
[104] L. Wang, B. A. Norman, and J. Ragopal. Placement of multiple RFID reader
antennas to maximize portal read accuracy. International journal of Radio
Frequency Identification Technology and Applications, 1(3):260–277, 2007.
[105] T. Wang. Enhanced binary search with cut-through operation for anti-collision
in RFID systems. IEEE Communications Letters, 10(4):236–238, 2006.
[106] S. Weis, S. E. Sarma, R. L. Rivest, and D. W. Engels. Security and privacy
aspects of low-cost radio frequency identification systems. Security in Pervasive
Computing, 28(2):201–212, 2004.
[107] C. P. Wong. Grouping based bit-slot aloha protocol for tag anti-collision in
RFID systems. IEEE Communications Letters, 11(12):946–948, 2007.
[108] Q. Xianming and Y. Ning. A folded dipole antenna for RFID. In IEEE In-
ternational Symposium on Antennas and Propagation Society, pages 97–100,
2004.
[109] Z-Y. Yang and J-L. Chen. The simulation and analysis of algorithms for re-
dundant reader elimination in rfid system. In Proceedings of the Third UKSim
European Symposium on Computer Modelling and Simulation (EMS), pages
494–498, 2009.
BIBLIOGRAPHY 184
[110] W. Yoon and N. H. Vaidya. RFID reader collision problem: performance analy-
sis and medium access. Wireless Communications and Mobile Computing, pages
1–24, 2010.
[111] K-M. Yu, C. W. Yu, and Z-Y. Lin. A density-based algorithm for redundant
reader elimination in a RFID network. In Proceedings of Second International
Conference on Future Generation Communication and Networking, pages 89–
92, 2008.
[112] L. Zhang and Z. Wang. Integration of RFID into wireless sensor networks:
Architectures, opportunities and challenging problems. In Proceedings of the
Fifth International Conference on Grid and Cooperative Computing Workshops,
pages 463–469, 2006.
[113] N. Zhang and B. Vojcic. Binary search algorithms with interference cancel-
lation RFID systems. In Proceedings of the IEEE Military Communications
Conference (MILCOM), pages 950–955, 2005.
[114] X. Zhang and S. B. Wicker. How to distribute sensors in a random field? In
Proceedings of the 3rd International Symposium on Information Processing in
Sensor Networks (IPSN), pages 243–250, 2004.
[115] F. Zhou, C. Chen, D. Jin, C. Huang, and H. Min. Evaluating and optimizing
power consumption of anti-collision protocols for applications in RFID systems.
In Proceedings of the 2004 International Symposium on Low Power Electronics
and Design (ISLPED), pages 357–362, 2004.
[116] F. Zhou, D. Jin, C. Huang, and M. Hao. Optimize the power consumption
of passive electronic tags for anti-collision schemes. In Proceedings of the 5th
International Conference on ASIC, pages 1213–1217, 2003.

Abstract

Radio Frequency IDentification (RFID) is growing prominence as an automated identification technology able to turn everyday objects into an ad-hoc network of mobile nodes; which can track, trigger events and perform actions. Energy scavenging and backscattering techniques are the foundation of low-cost identification solutions for RFIDs. The performance of these two techniques, being wireless, significantly depends on the underlying communication architecture and affect the overall operation of RFID systems. Current RFID systems are based on a centralized master-slave architecture hindering the overall performance, scalability and usability. Several proposals have aimed at improving performance at the physical, medium access, and application layers. Although such proposals achieve significant performance gains in terms of reading range and reading rates, they require significant changes in both software and hardware architectures while bounded by inherited performance bottlenecks, i.e., master-slave architecture. Performance constraints need to be addressed in order to further facilitate RFID adoption; especially for ultra large scale applications such as Internet of Things. A natural approach is re-thinking the distributed communication architecture of RFID systems; wherein control and data tasks are decoupled from a central authority and dispersed amongst spatially distributed low-power wireless devices. The i

distributed architecture, by adjusting the tag’s reflectivity coefficient creates micro interrogation zones which are interrogated in parallel. We investigate this promising direction in order to significantly increase the reading rates and reading range of RFID tags, and also to enhance overall system scalability. We address the problems of energy-efficient tag singulations, optimal power control schemes and load aware reader placement algorithms for RFID systems. We modify the conventional set cover approximation algorithm to determine the minimal number of RFID readers with minimal overlapping and balanced number of tags amongst them. We show, via extensive simulation analysis, that our approach has the potential to increase the performance of RFID technology and hence, to enable RFID systems for ultra large scale applications.

ii

Acknowledgments

First of all, I thank ALLAH (God) for his mercy, help and guidance.

It has been a great honour for me to be supervised by Dr. Hossam S. Hassanein. His knowledge and guidance in research, and nice personality have been a great source of support for me throughout my research. I would also like to thank Dr. Waleed Al Salih, Dr. Abd-Elhamid Taha, Dr. Afzal Mawji, Sharief Oteafy, Ed Elkouche, Abdallah Alma’aith and Louai Al-Awami, Hatem Abo Zied, Mervat Abu-Elkheir, Xiangyang Zhang, Ashraf Alfagih, Abdulrahman Abahsain, Umair, Haroon Malik and Pratik Dhrona for their support, collaboration, friendship and fruitful discussions.

I owe to my parents any success I make in my life. Their encouragement and support is a key factor in any achievements I have ever made. I also want to thank my brother, Kamran, sisters, Ambreen and Pinky, sister-in-law, Zara, and niece, Zainab, for their love and support.

I would like to thank all members of the School of Computing, especially Telecommunication Research Labs, at Queens University for their support, thoughtful discussions and friendship. iii

I would like to thank all anonymous conference and journal reviewers for their contributions.

Special thanks to the Natural Science and Engineering Research Council and Ontario Graduate Scholarship for their financial support.

iv

Statement of Originality

I hereby certify that this Ph.D. thesis is original and that all ideas and inventions attributed to others have been properly referenced.

Kashif Ali June 8th 2011

v

List of Acronyms

AFSA ALE ASAP ASC ASC-LB ASK BST CW CDMA CMOS CSMA DCS DNS DFSA DR-RFID dB dBm EBST EDFSA EPC FDMA GSM FSA HF HiQ Hz IC

Advanced Framed Slotted Aloha Application Level Events Adaptive Slotted Aloha Protocol Approximate Set Cover Approximate Set Cover with Load Balancing Amplitude Shift Keying Binary Search Tree Carrier Wave Carrier Division Multiple Access Complementary Metal Oxide Semiconductor Carrier Sense Multiple Access Distributed Color Selection Domain Name Service Dynamic Framed Slotted Aloha Distributed Receiving Radio Frequency Identification Decibel Power ratio in decibel Enhanced Binary Search Tree Enhanced Dynamic Framed Slotted Aloha Electronic Product Code Frequency Division Multiple Access Global System for Mobile Communications Framed Slotted Aloha High Frequency Hierarchical Q-Learning Hertz Integrated Circuit

vi

IFF IoT LF LSB MAC MACA MIMO MHz MSB NTE O-PDC OD-PDC ONS PIE PDC PML PRQT QT RCS RF RFID RRE SDMA SS TDMA UHF URL VDCS µW WSN XML

Identify Friend or Foe Internet of Things Low Frequency Least Significant Bit Medium Access Control Multiple Access with Collision Avoidance Multiple Input Multiple Output Mega Hertz Most Significant Bit Neighbors and Tags Estimation Optimal Power-based Clustering Optimal Discrete Power-based Clustering Object Naming Service Pulse Interval Encoding Power-based Distance Clustering Physical Markup Language Prefix Randomized Query Tree Query Tree Radar Cross Section Radio Frequency Radio Frequency Identification Redundant Reader Elimination Space Division Multiple Access Spread Spectrum Time Division Multiple Access Ultra High Frequency Universal Resource Locator Variable-Maximum Distributed Color Selection Micro-Watts Wireless Sensor Network Extensible Markup Language

vii

Security and Integrations”. 6 pages. 3-30. A-E. S. CRC press. Hassanein. Ali and H. W. W. in proceedings of IEEE International Conference on Communications (ICC). Computer Networks. 2009 7. Protocols. “Using neighbour and tag estimations for redundant reader eliminations in RFID networks”. S. June 2011 2.“DR-RFID: Distributed Receiving RFID System A novel paradigm”. Hassanein. Oct. “Medium access control in Radio Frequency Identification”. accepted. Chen. K. K. “Set cover approximation algorithms for load-aware readers placement in RFID networks”. and H. Taha and H. submitted. Ali. in proceedings of IEEE Global Communications Conference (Globecom). “Parallel singulation in RFID systems”. Hassanein. S. S. 6 pages. Hassanein. accepted. “Energy-efficient parallel singulation in RFID”. Nov. in proceedings of 34th IEEE Conference on Local Computer Networks (LCN). S. 2009 8. 1-6. pp. 5594-5599. K. Alsalih. May 2010 6. Ali and H. edited by Y. IEEE Transaction on Mobile Computing. Ali and H. Ali. Ali and H. S. S. W. “A Power Control Technique for Anticollision Schemes in RFID Systems”. K. Book chapter in “RFID and Sensor Networks: Architectures. Dec. submitted. L. K. pp. 2009 viii . May 2011 3. Yang and J. S. T. pp. Alsalih. to appear in IEEE International Communication Conference (ICC). Hassanein. March 2011 5. K. pp. May 2011 4. Oteafy and H. “Passive RFID tags and vehicle-assisted information dissemination for ITS applications”. Hassanein. to appear in IEEE Wireless Communication and Networking Conference (WCNC). Ali. Hassanein. Zhang. Alsalih and H. 688-694. Hassanein. K. S. K.Co-Authorship 1. Ali.

Ali. K. Hassanein and A-E. 69-76. 819-824. 2009 10. Hassanein. Ali and H. Oct. K. “Distributed receiving in RFID systems”. in proceedings of the 32nd IEEE Conference on Local Computer Networks (LCN). Oct. Taha. Hassanein. 266-273. S. pp. “RFID anti-collision protocol for dense passive tag environments”. Oct. 2008 11. Alsalih. S. in proceedings of 34th IEEE Conference on Local Computer Networks (LCN). pp. pp. 2007 ix . “Optimal distance-based clustering for tag anti-collision in RFID systems”. K. Ali and H. S.9. W. H. in Proceedings of 33rdIEEE Conference on Local Computer Networks (LCN).

. . . . . . . . . . . . . . . . . . . . .1. . . . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x i iii v vi viii x xiii xiv 1 2 4 6 7 7 7 10 12 18 30 30 31 38 52 . . . . . . . . . . . . . . . . . . . . . . . . . . .1 System Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . .1. . . .2 Literature Survey . .1 Motivations . . . . . . . . .3 System Architecture .2 Enabling Technologies . . . . . . . . . . . . . . . 2. . . . . . . . . . 2. . . . .3 Tag Collisions . . . . . . . . 2. . 1. . .4 Reader Collisions . . 2. . . . . . . . .2. . . . . .4 Communication Protocols . . . .Contents Abstract Acknowledgments Statement of Originality List of Acronyms Co-Authorship Contents List of Tables List of Figures Chapter 1: Introduction 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Thesis Contributions . . . . . . . . . . . . . . .2. . . . . . .2 Medium Access Control . . . . . .2. . . . . . . . . . . . . . . Chapter 2: Background and Literature Survey 2. . . . .1 History . . . . . . . . . . .2. . . . . .3 Document Outline . . 2. . . . . . . . . . . . . . . . .1. .1. . . . 2. . 2. . . . . . . . . . . .1 Background . . 1.

. . . . . . .2 Simulation Results . 4. . . . . . . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . .1 Introduction .6 Summary . 3. . . . . . . . 4. . .3. . . . . . . . . . . . . . .5. . . . . . . . . . . 3. . .2 Dynamic Assignment of Reflectivity Index .6 Simulation Setup and Results . . .5. . . . . . . . . . . . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 2. . . . . . . . . Chapter 4: Distributed Receiving in RFID 4. . . . .4 Delay Analysis of Deterministic Algorithm . .3. . . . . . . . . . . . . . . . . 4. . . .1 Introduction . . . . . .4. . . . . .1 Tag Reflectivity Co-efficient . . . .1 System model and problem definition . .3 RFID Tag . . . . . . .3. . . . . . . . . . . . . . . . . . . . . 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 DR-RFID: Distributed Receiving System for RFID 4.3. . . . . 4. . 4. . . . . 4. . .2 The Power-based Distance Clustering (PDC) Anti-Collision Scheme 3. . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . 4. . 3. . . . . . . . . . . . . . . . . . . . . . . . . . . 4. .4 Micro-Zoning in DR-RFID System . . . . . . . . . . . . . . . . . . . . . . .5.1 Simulation Setup .5 Parallel Singulation in DR-RFID . . . . . 4. . . 3. . . . . . . . . . . . . . . . . . . . . . . . 3. . .5. . . . . 3. . . . . .3. . . .2 Assumptions . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . .5. . . . . . . . .1 Simulation Setup . . . 4. . . 4. . . . . . . . . . . .2 Motivation . . . . . . . . . . . . . . . . . . . . . . .2 The Deterministic Algorithm . . . . . . . . . . .1 System Model . 3. . .2. . . . . . . . . 3.5. . . 3. . . . .4. . .4. . . 4. 4. . .4. . . . .3.1 System model and problem definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Delay analysis . xi . . . . . . . . .1 RFID Reader . . 3. . . . . . . . . . . . . . . . . . . . . .2 Simulation Results . . . . . . . . . . . . . . . .3 The Probabilistic Algorithm . .5 Simulation Setup and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6. .4 Optimal clustering algorithm . . . . . . . . . . .5 Coverage . . . . . . . . . . . . .5. . . . . . . . . . . . . . 3. . . . 57 62 64 64 66 68 68 70 70 72 75 75 77 77 78 79 80 81 88 91 91 94 98 101 102 103 104 105 108 110 111 112 116 119 121 121 122 124 Chapter 3: Power Control in RFID 3. . . . . . . . . 3. . . . . . .2. .3 Delay analysis . . . .4 Optimizing the delay . . . . . . .6. . . .4 Optimal Discrete Power-based Distance Clustering (OD-PDC) . . . . . . . . . . . . . . . . 4. . 3. . . . . . . . . .2 Assumptions .2 RFID Fielder . . . . . . . . . . . .3 Optimal Power-based Distance Clustering . . . . . . . . . . . . . . .5 Delay Analysis of Probabilistic Algorithm . . . . . . .

. . . .4 Load Balancing . . 5. . . . .5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 4. . 5.3 RFID Readers Deployment (Deterministic Deployment) .2 Simulation Results for Deterministic Deployment . . . . .2 Future Outlook .2 System Model . . . . . . . . . . . . . . .5. . . . . . . . . . . . . . . . . .4. . . . .1 Problem Definition . . . . . . . . . . . . . . . . . . . . .5 Displaced tags . . . . . . . . . . . . . . . . . . . 5. . . .3 Simulation Results for Random Deployment . . . .4 Computational Complexity .3 Reducing Overlap Among Readers . 5. . . . . . . . .4 RFID Redundant Readers Elimination (Random Deployment) 5. . . . . . . . 133 Summary . . . . . . . . . . . . . . . . 5. .1 Simulation Setup . . .4. . . . . . . .1 Introduction . . . . . . . . . . . .4. 5. . . . . . .3. . . . . . . . .3. . 171 6. . . . . . . . . . .2 Providing Coverage . . . . . . .1 Summary . . . 5. . . . 173 Bibliography 174 xii . . . . . . . . . . . . . . . . . . . . . . . . . 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . .8 Work in Progress . . . .3. . . . Chapter 6: Conclusion 170 6. . . . . . . . . . . . . . .5. . . . 5. . . . . . . . . . . . . . . . . .6 Summary . . . . .1 Assumptions and Definitions . . . . . 5. . .5 Simulation Setup and Results . . . . . . . .3. . . . . . . . .3 Exemplary Scenario . . . . . . . . . . 5. . . . . . . . . . . . . . .2 Neighbors and Tags Estimation (NTE) based algorithm 5. . . . . . . . . . . . . . . .3. . . . . . . . . . . 5. . . . . . . . . .4. . . . . . . . 5. 5. . . 137 140 140 144 145 145 146 148 150 152 152 152 152 155 157 159 159 160 164 167 Chapter 5: Coverage in RFID 5. 5. . .

.2. . . . . . . . . . . 14 51 52 57 Simulation parameters . Comparison of deterministic anti-collision algorithms. Comparison of probabilistic anti-collision algorithms. 158 xiii . . . . . . . .1 World-wide UHF RFID Frequency Allocation. . . . . . . 5. . . . .3 2. . . . . . . .1 5. . . . . . 122 Tracing the NTE algorithm for the scenario of Fig. . . . . . . . . . . .2 2. . . . . . . . . . . . . . . .List of Tables 2. . . . . . . . .1 2. . . . . .4 4. . Comparison of reader collision algorithm.

. .9 Tags and readers collision for a typical RFID system . . . . . . . . .List of Figures 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . Backscattering modulation concept of RFID tags . . 2. . . . 2. .14 Adaptive SDMA using directional antenna . . . . . . . . . . . .2 2. . . Multiple readers-to-tag collisions. . 2. . . . . . . . 2. . . . . . . . . . . Multiple tags-to-reader collisions. . .15 Frequency domain multiple access. .1 2. . . Schematic of an exemplary RFID tag. . . . xiv . . . . . . . . . . . . . . . . . . Master-slave architecture in RFID systems. . . . . .5 2. . History timeline of advancements in RFID technology . . .13 Power-based tags clustering . . . .8 2. . . . . . Execution trace of the binary search tree anti-collision . . . . . . 2 9 11 12 13 16 18 19 22 23 25 28 29 33 35 36 37 41 2. . . . .7 2. . . .6 2.1 2.3 2. . . .10 Execution trace of the slotted Aloha anti-collision . . . . . . . .12 Counter execution flow chart of the Q-algorithm . . . .4 2. . . . EPC Network Architecture . . . .17 Classification of the deterministic TDMA schemes. . . . . . . . .11 Query round for a single RFID tag . . . . Singulation tree for the example RFID system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .16 Classification of the time division multiple access procedures.

. .2 3. . . . . . .10 Ratio of collision queries with the QT scheme. . . Ratio of successful queries with the EBST scheme.9 Flow diagram of the PDC scheme. . 106 Tag reflectivity coefficient and assignment. . . . . . . . . . . . . . . . . . .6 3. . . . . . . . . . . . . . . . . . . . .1 3. . . . . . . 3. . . . . . . . . . . . . . . . . . . . . . . . 4. . . . . . .7 Effect of tag power requirements on the operating range . . . . .11 Ratio of collision queries with the EBST scheme. . . . . . . . . . . . . .15 Ratio of idle queries with the Q-Algorithm scheme. . . . . 46 53 68 69 76 82 82 83 84 85 85 86 87 87 88 89 89 97 99 3. .13 Ratio of idle queries with the QT scheme. 109 xv . .7 3. . . . 2. . . . . . . Ratio of successful queries with the QT scheme. . . . Ratio of successful queries with the Q-Algorithm scheme. . . . . . . . . . . .14 Ratio of idle queries with the EBST scheme. .2. . .4 3. . . . . . . . . .12 Ratio of collision queries with the Q-Algorithm scheme. . . . . . Distributed receiving-based RFID architecture. . . . .5 3. . .6 4. .2 4. . . .3 3. . . .5 4. . . . . . . . . .4 4. . . . . . .8 3. . . . . . . . . . . . Total number of queries with the EBST scheme. . . . .18 Taxonomy of the probabilistic TDMA-based schemes. . . . . . . . Total number of queries with the Q-Algorithm scheme. . . . . . . . . . . . . .1 4. . . . . . . . . . . . . . . . . . . . . . . . . 3. . . . .19 Taxonomy of reader collisions schemes. . . . . . . . . . . . . . . . . . . . . An example of 3 transmission ranges and 2 clusters. . . 100 Block diagram of passive tag . Total number of queries with the QT scheme. . . . . . . . . 3. . . . . 3. . . . .3 4. . . . . . . . . . . . 3. . . . . . . . 3. . 104 Tag schematic . . . . . . . . . . . . . Block diagram of fielder . . . . . . . . . . . . . . 103 Micro-zones in the DR-RFID system. An example of an interrogation zone of 3 clusters. . . . . . . . . . .

. . . . 129 4. . . .7 5. .4 5. . .9 Timeout scenarios in parallel singulation. . . . . . .16 Fielder schematic for the DR-RFID system. . . . . . . . 115 Grid-based fielders’ deployment scheme. . . . . . . . . . . . .11 Overall query’s comparison between RFID and DR-RFID . . . . .5 5. . 141 An RFID network with redundant readers. .13 Reading reliability comparison. . 138 5. . . . . . . . . . . . . 127 4. . . . . 131 4. . . . . . . . . . . .14 Effect of mobility on the tag reading rate. .6 5. 161 Maximum load (tags) for various tag density. . . . . . . . .4. . . . . . . . 130 4. . . . . 125 4. . . . . . 165 Number of tag reads in sparse and dense environments. .17 Fielder and power board prototype. . . 163 Redundant reader detected in sparse and dense network. . . . . . 134 4. . . . . . . .8 4. . . . . . . . . .1 5. 133 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18 Semi-passive tag schematic for the DR-RFID system. . . . . . . . . . . . . . . . . . . . . . . .3 5. . . . . . . . . . .2 5.8 5. 162 Effect of tag density and interrogation area . . . 156 Number of readers required for various tag density.10 Reading range improvement of DR-RFID . . . . . 168 xvi . . . 142 Snapshot of random RFID network from the simulator. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 4.12 Collision comparison between RFID and DR-RFID . . . . . . . 166 Number of tags write in sparse and dense environments. . 136 4. . . .15 Reading cycles comparison for the multi-readers and DR-RFID.9 3D Coverage in RFID Networks. . . .

Coping with these challenges and hence . respectively. e. and a set of passive tags.1 Chapter 1 Introduction Radio Frequency IDentification (RFID) is a prominently emerging automated identification technology. performance of an RFID system. biometrics (voice. reader. RFID seamlessly turns everyday objects into mobile network nodes that can be tracked. allows simultaneous identification. is robust to harsh physical environments.g. . RFID has an edge over other identification systems such as barcode systems. in that it requires no line-of-sight for communication. defined in terms of reading range. in addition to being cost and power efficient. A passive tag has neither a physical power source nor a transceiver as it utilizes the energy-harvesting and reflection based communication. However. reading rate and reading coverage is affected significantly by the magnitude of ultra large scale applications. Such performance constraints need to be addressed in order to facilitate efficient RFID continuing adoption. optical character recognition systems. This emerging and rapidly evolving technology has the potential to revolutionize the way we interact with physical objects. An RFID system is typically composed of an application.. smart cards. for power and communication. traced and monitored. Internet of Things (IoT). fingerprinting. retina scanning) .. . etc.

anti-collision algorithms are needed for both tag collisions and reader collisions.1: Tags and readers collision for a typical RFID system enabling ultra large scale RFID systems is the main focus of this research. This means that neither collision avoidance nor collision resolution mechanism can be implemented in tags. MAC in RFID systems is confined to resolving collisions only at the reader.1. MOTIVATIONS 2 Figure 1. Hence. . highlight the thesis contributions and outline the thesis document. detect collisions. 1. In the rest of this chapter.1.1 Motivations The nature of a tag’s reflection based communication dictates failure of the usage of the conventional wireless Medium Access Control (MAC) techniques in RFID systems. as tags cannot sense the medium. we present the motivations behind our research. or sense the presence of other channel traffic.

1 as tags T 4 and T 5 are within reader R4’s interrogation zone. Tag collisions and reader collisions result in reduced reading rates. especially when passive tags. are involved. Considering tag simplicity. are capped by theoretical upper bounds on reading rates. as illustrated in Fig. mobility limit the number of tags that can be physically monitored within the interrogation zone of an RFID reader. Therefore. also limit the operational range of a passive tag to a few meters. In addition. because of the centralized reader and tag communication architecture and the probabilistic nature of their executions. However. these algorithms.1. Tag collisions are the most devastating. back and forth.1 wherein the tag T 1 is interrogated by readers R2. existing anti-collision algorithms do not scale to ultra large RFID applications having 10-100 thousands of tags and thousands of readers. in addition to attenuation. Existing . eventually affecting the network lifetime.1. reflection based communication also imposes limitations on the reading range of RFID tags. 1. R3 and R5. The operational power requirement of upcoming passive tag circuitry and the maximal allowable Carrier Wave (CW) signal strength. the existing anti-collision algorithms usually employ the Aloha-based contention avoidance schemes. Furthermore. waste of reader’s available power. 1. with their limited communication functionalities. This is illustrated in Fig. MOTIVATIONS 3 Tag collisions occur when more than one tag is within the reader’s interrogation zone and attempt to reply to the reader’s requests at the same time. Reader collisions occur when one tag is interrogated by more than one reader. The reflected signal strength faces significant attenuation as it propagates twice the distance between the tag and reader. hence capping the upper bound on the operational range of passive tags to some tens of meters.

1.2 Thesis Contributions Our research aims primarily at enabling RFID for ultra large scale applications and improving the performance of existing RFID systems. readers R1 and R2. both of which cover other tags as well. are both redundant readers. and hence. THESIS CONTRIBUTIONS 4 approaches fail to provide significant enhancements to reading rates of RFID tags. R2 is redundant as the only tag it covers (T 1) is also covered by its neighboring readers R3 and R5. in Fig. The centralized reader and tag communication architecture of an RFID system . The main contributions of this thesis are: 1. individually by each reader.2. reading rate and reading coverage. 1. R1 is redundant as it does not cover any tag. mainly to cover as many tags as possible. This conflicts with the objective of minimizing tag collisions. We overcome this problem by devising an optimal power stepping scheme that minimizes tag collisions while covering the required interrogation area.1. We formulate the problem as a closed-form optimization problem. Whereas. T 2 and T 3. 2.1. jeopardize their readability. Tags within the interrogation zone of a reader cause immense “in-the-air” data collisions. For instance. Such redundant interrogations translate into unnecessary energy consumptions. Deploying more readers to cover the required interrogation area will result in redundant readers. In general. respectively. mainly in reading range. a reader interrogates tags within its zone using maximal power. hence lower reading rates thereby affecting network lifetime. The overlapping tags need to be interrogated multiple times. therefore forcing more readers to maintain required coverage which further escalates the collisions.

e. We propose parallel singulation algorithms for the DR-RFID system. We extend the DR-RFID system and introduce the concept of a micro-zone. We overcome these limitations by assigning the existing centralized design of RFID system − the nominal task of receiving tags replies − to spatially distributed devices within the interrogation zones of readers. we propose an approximation algorithm to determine the minimal number of RFID readers required to maintain the requested coverage while having minimal overlapping interrogation zones amongst the readers. with the objective of confining its readability within a smaller subset. i. we also propose both deterministic and probabilistic variations of existing tag anti-collision algorithms. 4. We devise an algorithm which determines the reader’s power level and alters the strength of the tag’s reflected signal. For the former. is enhancing the reading reliability and increasing the reading range and reading rate in a scalable manner. The goal of the proposed system. In addition. THESIS CONTRIBUTIONS 5 places fundamental constraints on its performance. namely reading rate and reading range. Our parallel singulation alleviates the reading rate bottlenecks. named Distributed Receiving RFID (DR-RFID). a micro-zone. We propose a heuristic algorithm to eliminate redundant readers in . In it.2. We also extend the algorithm to give a load-balancing deployment strategy. the reader’s interrogation zone is divided into smaller zones. They facilitate autonomous interrogation of tags. 3.1. each of which is managed by a set of distributed devices.. We define and solve RFID coverage as a deterministic as well as random deployment problem.

Chapter 4 presents the distributed design for RFID system. closed-form optimization scheme and discrete greedy optimization scheme. This is useful for large scale. Chapter 6 concludes this document by highlighting the main issues addressed in this thesis and outlining some of the future research directions.1.3. It includes the approximation algorithms for load-aware reader placement and the greedy heuristic algorithm for redundant reader elimination. DOCUMENT OUTLINE 6 random deployments. in deterministic and random deployments.3 Document Outline The remainder of this document is organized as follows. i. Chapter 2 presents the background and literature survey of RFID systems. respectively. This includes the basic heuristic scheme.e. Finally. This also includes our micro-zone concept and parallel singulation algorithms. The proposed design uses distributed receiving entities to facilitate reading range and reading rates. adopted for both deterministic and probabilistic classes of anti-collision schemes. 1. . dense and random deployments of RFID systems. The goal is to achieve maximal coverage with the minimal possible readers by turning off redundant readers. readers without additional coverage benefits. Chapter 5 presents RFID coverage both as a deterministic and a random deployment problem.. Chapter 3 presents the power control scheme for tag anti-collision scheme in RFID systems.

2. we briefly present the history of RFID technology. the enabling technologies. developed the first active Identify Friend or Foe (IFF) system which involves deploying an active transmitter on each of the British plane.7 Chapter 2 Background and Literature Survey 2.1 History The roots of RFID can be traced back to World War II when Germans enhanced upon the Scottish physicist Sir Robert Alexander Watson-Watt radar to differentiate between the enemy and their own planes. This method would allow the ground crew to differentiate between the German planes and the Allied planes. fundamental system architecture and prominent underlying communication protocols. when it .1. The transmitter. under Watson-Watt secret project. the British. The approach was to enable pilots to roll their planes as they returned to base. During same era.1 Background In this section. allowing the reflected signal to alter in its amplitude. This was conceptually the first single IDbit passive RFID system.

Today. due to unregulated and unused spectrum. Later on. In the early 1990s. the HF 13. These two systems were conceptually the first single bit passive and active RFID system. was then adopted in various other applications. begins broadcasting signal which helps to identify it as a friendly plane. This identification system. suffers from wide adoption-scale adoption due its high associated cost. Los Alamos National Laboratory developed the first transponder reader system to detect the trucks carrying nuclear materials in and out of the secure facilities. High-Frequency (HF) 13. In 1970s. This system was commercialized in mid 1980s. depending on the successful payment.1.g. e. payment systems (e. These high costs were mainly due to low sales volume. is either on or off to detect any forgery or theft.2.. advanced with a passive Low-Frequency (LF) 125 KHz RFID tag. The early RFID technology also found its way into military applications.56 MHz RFID systems are used for access control. still use in goods theft protection. These advancements aided into commercialization of the 1-bit RF based identification resulted in to electronic article surveillance tags. The reader wakes up the transponder which then will respond with its serial number and stored information. Advances in radar and RF communications systems continued throughout the 1950s and 1960s. BACKGROUND 8 received signals from radar stations on the ground.g. since then it is adopted for the automated toll payment system for roads. animal tracking and building access. IBM engineers developed an Ultra-High Frequency (UHF) RFID system which offered longer range and faster data transfers however. speed-pass) and contact-less smart cards for secure access control. uncertainty in return-of-investment and lack . These HF tags offered relatively faster data transfer rates. bridges and tunnels across the globe.56 MHz RFID tags were introduced.. The bit.

the Auto-ID Centre. The standardization of RFID. of international standardizations. developed two important air interface protocols (Class 0 and Class 1). highlighting the key applications and enabling technologies.2. Between 1999 and 2003. and U. BACKGROUND 9 Figure 2.S.1: History timeline of advancements in RFID technology. This approach enabled low-cost RFID tags which can then be placed on products to be tracked throughout the numerous manufacturers supply chain. this changed with the establishment of Auto-ID centre. which introduced storing the serial number on a small microchip.1. with support of more than 100 large end-user companies. However. the Electronic Product Code (EPC) numbering scheme and a network architecture to lookup tag data over the network. at the Massachusetts Institute of Technology. key RFID vendors. especially with evident of the second . Department of Defence.

Other industries such as pharmaceutical. aviation. Using this process. These incident radio waves. The major RFID related historical events are summarized in Fig.2 Enabling Technologies Advancements in energy harvesting. The energy harvesting yields battery-less RFID tags which enables minimal maintenance. backscatter modulation and micro electro mechanical systems present the foundation of low-cost identification solutions provided by RFID. 2.1. Tesco. Metro. Target. defence..1. the tag processed the reader’s query request and acts accordingly. after interacting with the antenna. currently known as EPCGlobal Gen2. produce high-frequency voltage which is rectified using a diode. Energy scavenging (harvesting) technique harness incident radio waves and facilitate the operation of battery-less RFID tags. 2. automotive. BACKGROUND 10 generation standard. The rectified signal is smoothed out using a storage capacitor to provide constant voltage.S. longer lifetime and low cost RFID solutions. Department of Defence are evaluating the EPC Gen2 technology for goods tracking in their supply chain. Wal-Mart. and many more are also deploying prototype systems to evaluate the technical feasibilities of RFID systems within their business process. enable large-scale deployments and adoption of RFID solutions. e.g.2. known as envelope detection. The voltage is then used to power the tag’s logic unit and nonvolatile memory space.1. A similar circuitry with a smaller capacitance is used to de-modulate the data transmitted by the reader. and U. Some of the world biggest retailers. The schematic of a simple energy harvesting scheme and envelope detection of a typical RFID tag is . The backscattering yields transceiver-less RFID tags which enable minimal power and low cost RFID solutions.

open circuit and short circuit. shown in Fig.1. by the reader. namely normal operation.2. as is explained above. BACKGROUND 11 Figure 2.g. synthesizer. This power. The tag is guaranteed a continuous unmodulated carrier wave. Radio frequency based energy harvesting enables a cost effective solution as the tag circuitry does not require sophisticated signal processing blocks. is used by the rectifier to charge up the tag’s circuitry. for the modulated backscattering. to harvest upon. To modulate the backscattering signal the tag transits between short and open state. the tag’s antenna is matched with the IC load to allow maximum possible power delivery to the IC.3. 2. e. Conceptually. In . power amplifier. the RFID tag iterates between three different states.2: Schematic of an exemplary RFID tag.2. crystal frequency reference. The tag modulates the signal scattered by the antenna to communicate with the reader. These three states are depicted in Fig. as it stays in this state for a maximum allowable time. Backscattering reflects and modulates incident radio waves and facilitates transceiverless RFID tags. low-noise amplifier and batteries. 2.. In the normal state.

1. in the open state. A passive tag has no physical power source and no transceiver as it uses the energy-harvesting and backscattering modulation techniques for power source and communication. A tag is designed to store certain information. Figure 2. respectively.. These two state. Backscattering modulation enables a cost effective solution as the tag does not require a transceiver to communicate with the reader. respectively. BACKGROUND 12 the short state. i.1. the passive tag has limited . minimal radiation is reflected back. This results in no current flow in the antenna and hence. semi-passive or active. the size of which varies between 96 bits and 64K bytes. allow a tag to modulate the bit as ’0’ or ’1’.3 System Architecture An RFID system is typically composed of an application host. the signal path to the ground is blocked. The flow of the signal to the ground. On the other hand. scatters with minimal resistance.e. 2. as is not used by the rectifying unit. While most common in the RFID market. a reader and a set of tags. Tags can be either passive. the current flow steadily to the ground plane. resulting in detectable power signal at the reader.3: Backscattering modulation concept of RFID tags [36]. short and open.2.

BACKGROUND 13 Master Application Interrogation and tag reply Slave Reader Tag Tag Tag Master Slave Figure 2. Specifically. Tags targeted by a certain reader are said to be with in that reader’s interrogation zone. A semi-passive tag has an on-board battery. has both on-board battery and transceiver and may also carry certain sensing capabilities for example temperature or pressure. RFID reader operates in wide UHF frequency ranges. an interrogation zone is the physical distance within which the strength of the electromagnetic waves. Upon application request. receive the tags’ signals. an RFID reader will interrogate the tags within its interrogation zone. 2.4.2.e. An active tag. An RFID reader acts as a master for the tag and a slave for the application host. use to power its circuitry. and successfully decode them — a process known as singulation. It processes simple state machines and has no medium sensing capabilities. is able to power the tags.4: Master-slave architecture in RFID systems. but no transceiver hence. The masterslave architecture of the RIFD system is shown in Fig. on the other hand. The RFID tags are listen-first devices. At any arbitrary instant. The read tag data is then passed onto the middleware application where it is filtered accordingly. generated by the reader. they does not initiate communication unless are interrogated by the reader..1. a reader can only read one tag within its interrogation zone and a tag can only be read by one reader. relies on backscattering modulation for communication with the reader. between 860-960 MHz . functionality for processing and communication. i.e. i..

.4-2. The worldwide UHF band allocation for RFID systems is summarized in Table 2. The 2.4-2. RFID readers collect large amounts of data. in North America and European Union the allowed RFID frequency is between 902-928 MHz and 865-868 MHz. Chile. managing the .1.1. For instance.5-914 MHz 866-869 MHz & 919-923 MHz 915 MHz 864-868 923-925 MHz & 866-869 MHz 865. and 2. An RFID middleware filter out the redundant and irrelevant data to be allowed as input for localization and tracking. 2-10m at 900 MHz).A China EU and Russia Hong Kong Japan Korea Malaysia Mexico New Zealand Singapore South Africa Taiwan Thailand Radio frequency band 902-928 MHz 918-926 MHz 902-907. However. the different RFID spectrum varies for different region.4 GHz vs. Costa Rica.g.S. Dominican Republic.6-867. as it competes with cellular hence.5 MHz & 915-928 MHz 902-928 MHz 917-922 MHz 865-868 MHz 920-925 MHz & 865-868 MHz 952-954 MHz 908. Zibgee and WiFi.45 GHz band is an attractive alternate because of its unlicensed spectrum. most of which are redundant or irrelevant for the time instance. respectively.2. e.6 MHz & 917-921 MHz 922-928 MHz 920-925 MHz Table 2. The worldwide spectrum regulates within the 860-960 MHz is complex.1: World-wide UHF RFID Frequency Allocation. Uruguay Australia Brazil Canada and U. it receives high interference from other devices. BACKGROUND 14 Region Argentina.45 GHz. and has reduce reading range (1-3m at 2.

EPC Middleware: The middleware facilitate the exchange of data between the RFID reader. returns the network-addressable endpoints mapping to a particular EPC code data. 3. along with components and layers. The . within the interrogation zone of the reader. or a network of readers. For instance. Due to frequent nature of these polling. 2. 4. Object Naming Service: The Object Naming Service (ONS) translates the bytes in the EPC tag data into the web address (URL) of the object data.1. accumulates and filters the raw EPC data. i. Application Level Events: The Application Level Events (ALE) is an Interface definition that defines the interface between the EPC-enabled RFID application and the EPC-enabled RFID reader or a network of readers. is shown in Fig. after initial interrogation. Electronic Product Code: Electronic Product Code (EPC) is the tag data which frequently is polled by the reader(s) on a regular basis. BACKGROUND 15 tagged-object historical data.e. to pass on to the reader. and an RFID application.. respectively. The EPC network architecture. so it can be stored in proper format within the tag memory. The middleware also encodes the data.5. the middleware needs only to maintain the database for those existing or new tags which are either leaving or entering. 2. trigger associated events and so forth. According to the EPCglobal standardization efforts.2. The processed and useful EPC data is then passed onto the RFID application. The middleware instructs reader(s) for singulation. the EPC data needs to be filtered by the middleware to remove duplicate codes. 1. the following components are the required modules to properly associate the middleware applications with the tagged object.

BACKGROUND 16 Figure 2.1.2. .5: EPC Network Architecture (reproduced from [40]).

etc.100734. 5.2. BACKGROUND 17 ONS is an automated network service. which is the typical. In order to use DNS to find information about an item.g. The EPC Discovery Service is the search engine for the EPC related data which returns the URL locations that have the requested data.1245). allowing the end-user to execute Domain Name Service (DNS) like queries. dot delimited. the data needs to be available to multiple participating organizations. The information service is one of the core service and allow the business partners to access and search for EPC data..1. and it makes sense to store it in an XML-based format that can be read consistently by all parties. web services based data exchange mechanism. urn:epc:id:sgtin:0614141. facility wide. The ONS resolution process requires that the EPC being asked about is in its pure identity URI form as defined by the EPCglobal Tag Data Standard [EPC] (e. left to right form of all domainnames. 6. The PML. Multiple discovery service may be run. The Physical Markup Language (PML) provide this service to the participating application. based on XML. The Physical Markup Language: In a supply chain application. regardless of their information systems. provides a format for storing and retrieving data about objects that can be parsed by multiple RFID applications. the items EPC must be converted into a format that DNS can understand. local.) . each with limited scope (regional. EPC Information Services: The EPC Information Service is an inter-application data exchange standard that defines a secure.

Specific to RFID systems.1. Avoidance mechanisms also carry the risk of increasing the overall cost of tags and shrinking the potential interrogation zone for a reader.6: Multiple tags-to-reader collisions. Proactively. i. As with other radio-based systems. be it in a proactive or a reactive manner. collisions are avoided by distributing sufficient information about the access requirements of elements sharing the medium. Reactive mechanisms respond to collisions and attempt to speed the system’s recovery from a collision stall.1.. Conventional collision avoidance methods. is to reduce collisions. such as Carrier Sense Multiple Access (CSMA) cannot be adopted for RFID systems. especially when passive tags are used due to power limitations and its basis of reflection-based communication. use of backscattering modulation. collisions can be classified based on the type of entities involved: . efficient mechanisms for Medium Access Control (MAC) are required. BACKGROUND 18 Tag Tag Reader Tag Tag Figure 2. Proposals for RFID systems have therefore favoured reactive approaches to alleviate collisions.e.4 Communication Protocols The simultaneous wireless medium access results in collisions that undermine a RFID system’s overall performance. the main objective of mechanisms aimed at regulating medium access. 2.2. To sustain system operability.

These anti-collisions schemes are classified in literature to be either deterministic or probabilistic [42. multiple readers try to singulate a single tag which results in corruption of the tag’s internal state.1. dynamic frequency allocation and dynamic power adjustment are utilized to hinder these collisions. As a result. Readers-to-tag Collisions: Occur when one tag is interrogated by more than one reader.7: Multiple readers-to-tag collisions. 96].7. Existing mechanisms such as frequency-hopping.. and increased delay. 2. as shown in Fig. Tags-to-reader collisions are the most devastating.6. the tag may not be detectable. Reader-to-reader Collisions: Are result of the conventional frequency interferences. are the most common source of collisions in RFID systems and are resolved by the reader utilizing techniques collectively known as anti-collisions schemes.e. both Tags-to-reader and Readers-to-tag. multiple readers within each other’s interference zones are locked on the same frequencies. Tag collisions.2. i. wasted resources.2. . They result in reduced reading rates. especially when passive tags are involved. In such a scenario. BACKGROUND 19 Reader1 Tag Reader2 Figure 2. Tags-to-reader Collisions: Occur when more than one tag within a reader’s interrogation zone attempts to reply to the reader’s requests at the same time. depicted in Fig.

which are . normally. The splitting of the tree. the location of which is obtained from the previous interrogation cycle. The tree-based algorithm is a 2-way handshake algorithm involving sequences of interaction between the reader and the tags. Splitting is based on contention information obtained from the previous interrogation cycle and attempts to reduce contention for the next cycle. mostly binary trees.1. Obtaining collision information at the bit level requires that the precise bit position of collision is identifiable by the reader. we trace the execution of the conventional binary search tree algorithm [102]. The objective of these interrogation cycles is to split the tags. known as the interrogation cycle. BACKGROUND 20 Deterministic anti-collision scheme In deterministic mechanisms the reader splits and identifies a set of tags to respond in a given time. into a more manageable set of tags. The deterministic mechanisms utilize either tags’ serial numbers (identification codes) or randomly generated numbers to be used for splitting of the tree branches. into two branches (leaves) is based on the bit collisions and their respective positions. For this particular reason.2. The 2-way handshakes during each interrogation cycle (iteration) are the commands that the reader broadcasts based on its previous iteration for the following iteration. Deterministic anti-collision mechanisms fall under the general algorithmic classification of tree-based algorithms because of their splitting approach. The set of commands constitutes four main commands. either the Manchester coding or the NRZ (non-return-to-zero) bit coding is used by the RFID reader. The objective of using the binary search algorithm with multiple interrogation cycles is to singulate a tag from a larger set. using their serial numbers (IDs) or randomly generated numbers. In order to understand the deterministic tree-based algorithm.

starting from the Least Significant Bit (LSB). e. Beyond this. the first iteration of the algorithm begins with the reader transmitting the REQUEST command with the serial number 1111b as an argument. 1C1C.e. the unselected tag becomes completely inactive and does not respond to further REQUEST commands until is reset by the reader. UNSELECT: Cancels the selection of an up-until-now active tag. shown in Fig. the tag sends its own serial number back to the reader. Otherwise. 2. all three tags reply back with their IDs.g. The serial number 1111b is the highest possible in this case. 2. Only a tag with an identical serial number is selected for the processing of other commands. reading and writing data. .1. SELECT: Carries a serial number to the tag as a parameter. Let us assume that there are three RFID tags within the interrogation zone of the reader with 4-bit IDs of 1010b . 4. If a tag’s own serial number is less than or equal to the received serial number. i. This first iteration results in collisions (C) at position 0 and 2.2. In our example. 3.8. REQUEST: Carries a serial number to the tag as a parameter. mutes the tag. Only the selected tag will continue to respond to the reader commands. The purpose is to make all tags respond so that the exact bit collisions among all the tags’ ID’s could be determined. the tag will not respond. and 1110b . 1. 1011b . READ DATA: The selected tag sends stored data to the reader. i. BACKGROUND 21 as follows. The singulation process starts with the RFID reader sending the highest possible serial number. As the serial number of the tags in the interrogation zone is less than the requested serial number.e..

i. Two tags fulfill the command’s criteria. The reader. This implies that there is at least one tag between 1100b and 1011b . tag1 and tag2. This request is fulfilled by only . now chooses 1010b as the requesting string for the subsequent iteration. The reader now broadcasts the command REQUEST with argument 1011b . BACKGROUND 22 Figure 2. The response back from these two tags causes collision at the bit position 0. i. The binary search algorithm at this point splits the search into two subsets in an attempt to limit the search zone for subsequent interaction... i.. The algorithm sets the third bit of the request to 0. 1011b .e. The least significant bits after the third bit are all set to 1 in an attempt to capture all the tags whose two most significant bits are 10.e.1.8: Execution trace of the binary search tree anti-collision algorithm using three tags and a single reader. 101C. repeating the same splitting procedure.e. i. The third bit is the highest valued bit at which the collision has occurred during the first iteration.e.2.. the second iteration.

2. The node splits into two children nodes. is shown in Fig. other tags are silent as the READ DATA command is selective.9. appending 0 and 1 at the Most Significant Bit (MSB) collision. to the left and right child node. tag1 is selected using the detected tag ID address and can now be read or written by the reader without interference from other tags. The tree. which is distinguished in the figure by a double concentric circle.1. therefore resulting in lower collisions and less iterations. whereas the string next to each circle is the received response. The muted tag is completely inactive and does not reply to any further REQUEST commands. as the reader has successfully detected a single tag without collision. After the completion of the required read/write operation. from the request point of view. At this point.9: Singulation tree for the example RFID system. This process continues until tag 1010b is singulated. one tag tag1 which is now singulated. The string shown in the tree nodes are requested and sent by the reader. respectively. no further iterations are required. the selected tag (tag1) is muted by the use of the UNSELECT command. For now.2. Muting the tags reduces the number of responding tags in the subsequent iteration of the singulation process. BACKGROUND 23 1C1C 0 1111 1 1111 1 1011 101C 0 1011 1010 1010 Figure 2. Referring back to . Using subsequent SELECT commands.

1. For our sampler RFID system. Reader controlled synchronization is necessary as the tag has to transmit within its slotted time frame. it may suffers from tag starvation syndrome. thereby reducing idle and collision frames. The probabilistic approach is faster in comparison to the deterministic approach because of its low overhead. This encourages frame adaptability according to tag density and distribution. the singulation of the tag (tag2) requires one less iteration. REQUEST: Synchronizes and prompts all tags to transmit their serial number to the reader using one of the time slots that follow. the probabilistic approach is a chit-chat protocol that uses a sequence of interactive commands. The muting of tag1 has created a positive impact by yielding low collision as its serial number. Probabilistic anti-collision scheme In the probabilistic mechanisms the reader communicates the frame length and the tag picks a particular slot in the frame for transmission. Similar to the deterministic approach. The set of commands are as follows: 1. it would have collided with the tag2 serial number. compared with tag1’s singulation. The reader repeats this process until all tags have been successfully transmitted at least once. there are three available time slots. Consider again three tags with IDs of 1010b . idle and successful frame information from a previous interrogation cycle for the subsequent cycle.8. 2.2. BACKGROUND 24 Fig. However. We will now illustrate the details of the conventional slotted Aloha anti-collision procedure [42] using the last example. 1011b and 1110b . . The frame size may be adjusted based on the collision. otherwise.

are still responsive to the REQUEST command. i. SELECT: Sends a specific serial number to the tag as a parameter. the tag whose serial number matches with the passed parameter. After a certain time interval. Downlink (R => T) Uplink (T => R) Tag1 Tag2 Tag3 REQUEST 1 2 3 SELECT 1110 READ/ WRITE Collision 1010 1011 1110 1110 1110 Figure 2. BACKGROUND 25 2. Only tag3 is successful in transmitting its serial number back to the reader. there is a collision between tag1 and tag2 in slot1.1. allowing it to read and write. The tag uses the slot to transmit its serial number back to the reader. Due to a random slot selection method.2.10: Execution trace of the conventional slotted Aloha protocol. using example RFID system of three tags and a single reader.. After receiving the REQUEST command. the tag randomly selects one of the three available slots. 2. Only the matching tag. The execution trace of the slotted Aloha protocol is shown in Fig.10. The successful tag is selected for subsequent reading and writing after it is been selected . is flagged for further operation. 3. Tags with conflicting serial numbers however. READ DATA: The selected tag sends stored data to the reader.e. In real RFID systems. the reader periodically transmits the REQUEST command. other commands will also be available but they are omitted here for the purpose of simplicity.

If no transmission was successful. BACKGROUND 26 using the SELECT command.2. When the number of tags is small. EPC Class-1 Gen-2 RFID standard The EPCglobal UHF Class-1 Gen-2 protocol defines the standard physical and medium access requirements for the passive RFID tags. Therefore. in the probabilistic anti-collision algorithms. The backward-link rate is determined partially by the . The specification defines a number of physical layer options for both forward-link and backward-link. however. As the number of tags increases. The tags receives its operating energy. and Pulse Interval Encoding (PIE). the probability of collision becomes higher and the time used to identify the tags increases rapidly. the probability of tag collision is low. from the reader’s RF signal.7 kbps to 128 kbps. and the reader can choose pulse durations that result in down-link rates ranging from 26. The time between low amplitude periods differentiates a zero or a one. both down-link (forward) and up-link (backward). using the unmodulated Continue RF Wave (CW). and the time used to identify all tags is relatively short. All forward-link communication uses Amplitude Shift Keying (ASK). the tag randomly selects a slot number in the frame and responds to the reader using the slot it selected. the reader communicate with tag using RF modulation. the probabilistic algorithm performance depends significantly on the number of tags within the interrogation zone of the reader. and communication link. We will now explain the physical and medium access layer as is described the EPCglobal. where bits are indicated by brief periods of low amplitude. the REQUEST command is iteratively repeated until the serial number can be received successfully and no collision is observed in the frames.1. According to the defined standard. As we’ve witnessed.

The specification also defines the medium access control protocol. the individual frame is known as Query Round. A Query command is then transmitted which contains the fields that determine the backward-link frequency and data encoding. Miller-2. Query. initialize the number of frames and slots within each each frame. i. 4. The tag interrogation process starts off by reader transmitting the CW to power-up the tags within its interrogation range. However. the reader also sets one of four up-link encodings. and a target parameter.11. each frame has a fixed number of slot and the tags response by randomly picking the slot from each frame. Ack. is shown in Fig. Miller-4.e. or Miller-8. depending on the encoding. allowing for the backward-link frequency ranging from 40 kHz to 640 kHz. 2. The sequence of interrogation commands. based on Framed Slotted Aloha (FSA).. their link rates are reduced by a factor of 2. FM0 is highly susceptible to noise and interference which motivated the addition of the Miller encodings. When using FM0. the Q parameter which determines the number of slots in the Query Round. etc. When a . Query Rep. BACKGROUND 27 forward-link preamble and partially by a bit field set in the Query command which starts each query round. In FSA scheme. for a single tag. one bit is transmitted during each cycle and the data rate is equivalent to the link frequency. At the beginning of a Query Round. the reader can optionally transmit a Select command which limits the number of active tags by providing a bit mask. as only tags with IDs (or memory locations) that match this mask will respond in the subsequent round.1.2. or 8. The reader. afterwards. While these are more robust to errors with a higher number being more robust. namely FM0. that make up the Query Round. In literature. The number of frames and slots are dynamically selected according to tags population. Along with setting the backward-link frequency.

it will immediately transmit ais16 bit random two of which choose an initial random value of 0 resulting in termines the number of slots in the Query Round. ried flags are set to B. seen only in the upper In the Query command the Target parameter is set shown that the optimum isand when the number of slots is set equal to the number of tags. as all successfully received by the If the ID command tags have set their Inventoried flags to B. as only tags this collision occurs with ID’s (or memory locations) that match this mask will rewhere Q is in the range (0. the reader changes the Q parameter for every Query If a tag stores a 0 in its slot counter. 2 number inicantag receives a Query command whichit chooses a random number choose the same random Q − 1). if there are no tags with a slot counter range (0. a random number in the range (0. and QueryInventoID. this configuration option is not considered in this will not toggle the Inventoried flag at the next QueryRepeat. a subsequent command command will consists theseries ofto toggle tag their Rounds with the target set to A. When Query elicits no tag a NACK not its Inventoried flag. and the value is stored in the slot counter of the of 0. it is to either A or B. sistence time is defined to be anywhere between 500 ms to even if a NAK is lost the tag will be reset by the timeout and 5 seconds. After transmits its a subsequent QueryRepeatQueryRepeat Query Cycle generally cause of a read. the tag will quently be active during next round. where they will choose a new random number. a scenario shown with three tags. reader. it chooses number (RN16). the appropriate Q parameter can be set for the next ferred to as EPC in Figure 1). where Q is in the As well as collisions. tag reply during the course of the Query Round. While there are a number of strategies for reading a tag set. the Invenwillcommandthe flag being changed. As this perresponses and subsequent reader commands. A Query command is then next round. In number of active tags by providing a bit mask. the tag will backscatter its ID (reformation. it will immediately number. Consequently. On receiving Q parameter. the tag toried flag of each tag is reset to A.the stored in the slot counter of not be ACKed during the these tags will active in spond in the subsequent round.1. with the number of QueryRepeats being equal to the number of slots set using the the command the remaining tags will decrement their slot of the slot. 2. As tags are a tag transmits its ID. the Q parameter which de. and only tags with a a matching Inventoried flag will respond during the round.11). Of course. and all tags will consewill be active in the next round. Using this inthe correct random number. a QueryRepeat signals the end process then repeats. the lower half of Figure 1. 2Q . if the tag successful receives the ACK withorder to reduce empty slots and collisions. a slot may be empty. optionally transmit a Select command. The flag of a tag. and slots with only a single powered up. of tags transmit a 16 bitthe tag will (RN16). if the tags maintain anthe ACK with tag successful receives Inventoried flag whichslots with collisions. 15). tags may in the range (0. random number backscatter its ID Round in to to EPC in Fig. half of Figure 1. When a tag receives a Query command. around. Upon receiving (referred orderas best accommodate the number Whileremaining to be read. the reader the first slot. In the correct random tag. In this the tag so that a subsequentdown. . and approxWhile powered up. and one which in an ACK value of 2 and so responds after the second QueryRepeat. InInventoried flag will imately 35% during the round. the Query Cycle ends sent which resets the tag so that a subsequent QueryRepeat and the reader powers QueryRepeat down. This is done by determing the number of the RN16. A at which point. On receiving the be active in the the next Query Cycle. In [16].2. a single taga tag be in with a matching the Query command the respond of the slots would see only After response. This message sequence is shown or B. 15). to change. limits the tially with the result that multiple tags reply during a slot. can be in one of two states. round. tags maintain an Inventoried flag which cantagsone of two states. BACKGROUND 28 Figure 1: C1G2 a single RFID EPCglobal) Figure 2. and the tags willbethe tag.1 2 Along with toggling the Inventoried flag of a tag. Upon receiving the RN16. In slot. 2 Note that a reader can be configured such that the state of a 1 The C1G2 standard specifies a timeout period between tag tag’s Inventoried flag will persist across cycles. the reader will echo the RN16 in an ACK packet empty slots.1). and the value iscase. link frequencytag data encoding.packet at which point. When powered will cause not result in is sent which resets case. and a Target a collision in will echo the RN16 chooses an initial parameter.11: Query round for Protocol (Courtesy oftag (reproduced from [40]). QueryRepeat command the remaining tags will decrement their slot counter Along with if their slot the Inventoried and respond with a RN16 toggling counter is set to 0. a the flagsignals the end of thethis case. If the ID was Query Rounds. study. a NAK was notissuccessfully receiveda by the reader. Target parameter is set to either A or B. they become inactive during subsequent will cause the tag to toggle its Inventoried flag. responses. A or B. In transmitted which contains the fields that determine the upIf a and stores a 0 in its slot counter. However.

2.1. BACKGROUND

29

counter and respond with a RN16 if their slot counter is set to 0. The process then repeats, with the number of QueryRepeats being equal to the number of slots set using the Q parameter. Of course, tags may choose the same random number initially with the result that multiple tags reply during a slot. In this case, a collision occurs and the tags will not be ACK’ed during the round. However, these tags will be active in the next round, where they will choose a new random number. In the lower half of Fig. 2.11, a scenario is shown with three tags, two of which choose an initial random value of 0 resulting in a collision in the first slot, and one which chooses an initial value of 2 and so responds after the second QueryRepeat. If there are no tags with a slot counter of 0, a slot will go empty.

Figure 2.12: Counter execution flow chart of the Q-algorithm (reproduced from [40]). To reduce empty slots and collisions, the reader changes the Q parameter for every Query Round in order to best accommodate the number of tags remaining to be read. The algorithm, known as Q-algorithm, achieves this by determining the number of

2.2. LITERATURE SURVEY

30

empty slots, slots with collisions, and slots with only a single tag reply during the course of the Query Round. The execution flow chart of the Q-algorithm is shown in Fig. 2.12. The Q-algorithm evaluates the tags’ replies, for the current frame, to determine the next frame size. The Q-algorithm classifies the replies as idle, collision, and successful slots and accordingly update the Q value for the subsequent query cycles. The value of Q parameter is added and subtracted by a fixed constant C if current slot resulted in collision and idle, respectively. In case of successful slot, the Q parameters remains unchanged. The slot are updated, incremented or decremented, to avoid mis-match between the number of slots and number of tags. For instance, in case of idle slot, the number of slots are decremented as it imply that the number of available slots exceed the number of tags. Similarly, in case of collision slot, the number of tags exceeds the number of slots. In [72], it is shown that the optimum is seen when the number of slots is set equal to the number of tags, and approximately 35% of the slots would see only a single tag response.

2.2

Literature Survey

In this section, we will survey RFID literature for system designs, tags anti-collision and readers anti-collision protocols.

2.2.1

System Design

RFID is a dynamic and active research area. System level development includes antenna designs, RFID IC with enhanced RF sensitivity, ultra low power tag IC, lowpower backscattering modulation technique and so forth. The antenna design involves tradeoff between antenna gains, impedance and bandwidth while satisfying numerous

2.2. LITERATURE SURVEY

31

requirements such as size, form, reading range, support for mobility, cost and reading reliability [91]. Various antenna configurations such as patch antenna [42], folded dipole antenna [108], asymmetric antenna [27] and use of multiple tag antenna [49] have been proposed in literature. A detailed comparative performance evaluation of various antenna configurations can be found in [45, 91, 101]. Low-power IC design for passive tags [34, 36, 42, 57], as low as 2.7µW , has been proposed in the literature to enhance the operating range of the passive RFID tags. Improvements in the backscattering and modulation techniques are also proposed in the literature. Various analytical studies on the Radar Cross Section (RCS), an important parameter of in determining the quality of the backscattering communication, for the passive tags [81], its differential scalar scattering [82] and to enhance the bit error rate [46], have been performed. Energy-aware modulation techniques [13, 67, 98] also been proposed to reduce tag’s energy consumption while using backscattering communication. Wireless Sensor Network (WSN) and RFID integration provides multi-hop communication to the network of RFID readers [70]. In such a setting, the WSN node and RDID reader individually sense, read tags and perform multi-hop data communication with other interrogating readers.

2.2.2

Medium Access Control

In a radio based system, MAC protocols provide mechanisms for channel access control, allowing multiple devices to share the same physical medium. The most prominent MAC techniques are based on the Carrier Sense Multiple Access (CSMA), Multiple Access with Collision Avoidance (MACA), or the Aloha protocol. RFID systems use half-duplex, point-to-point communication links in which communication between

2.2. LITERATURE SURVEY

32

the reader and the tag is based on backscatter modulation. In backscattering, the tag sends its serial number by adjusting the antenna reflectivity to modulate the reflected signal. The nature of reflection based communication results in failure of conventional MAC techniques; as tags cannot sense the medium, detect collisions, or sense the presence of other channel traffic. This means that no collision avoidance or resolution mechanisms could be implemented at the tags. Hence, MAC in RFID systems is confined to resolving of collisions only at the reader, resulting in anti-collision algorithms for both tag collisions and reader collisions, mainly employing Aloha-based contention avoidance schemes. RFID systems exhibit very short durations of very high bursty activity followed by relatively long durations of inactivity. Generally speaking, four different approaches exist in the literature in handling multi-access issues in radio technologies. The approaches regulate access based on time, space (location), frequency, or code. Code Division Multiple Access (CDMA) supports high-rate data multiplexing using the spread-spectrum (SS) technique. In conventional SS, each user encodes data packets using an orthogonal spreading code, allowing successful transmission by multiple users. However, complex receiver design and higher computational and energy requirements restrained adaptation of CDMA technology in RFID systems. A combination of TDMA and CDMA based schemes has been investigated recently [75]. However, the CDMA scheme is a largely unexplored area in RFID systems. Space Division Multiple Access (SDMA) [42,99] spatially reuses the channel. The philosophy behind the SDMA procedure is that practically at any given time there will be a limited number of RFID tags spatially co-located at the same position, independent of the total number of tags that may be present in the overall interrogation

2. 12.13. 60]. An example of such a partitioning is shown in Fig. Therefore. 12] divides the interrogation zone into smaller clusters based on the distance to the reader by adjusting the reader’s power levels. When the reader sends a request. using reader’s power levels to adjust its interrogation zone. only . Numerous spatial isolation techniques exist with the most notable ones being: adjusting the reader power levels (power control) [10. d . by spatially isolating the tags. zone. 2. collisions are minimized.2.13: Clustering based on the distance between the reader and the tags. where the interrogation zone is divided into three clusters: d. Tags in each cluster are read separately. and d . Since the number of tags in a cluster is less than that in the whole interrogation zone. LITERATURE SURVEY 33 T" T" T’ T’ T" T’ Reader d d’ T" d’’ T’ T" Figure 2. as well as the use of electronically controlled directional antenna [42]. the interference caused by other tags can be minimized. using adaptive arrays and Multiple Input Multiple Output (MIMO) antennae [65]. The power control cluster-based algorithm [10.

and is therefore restricted to a few specialized .2. will respond to that request. in the latter case. To read a specific tag or all the tags within the interrogation zone.. the reader’s zone is increased to d and a new request is sent. These two techniques lower tag collisions. they are put into a sleep mode. where reader’s power levels are adjusted to reduce the interrogation overlaps. The SDMA technique is effective in reducing collisions but with a relatively high implementation cost because of sophisticated antenna systems.14. The antenna array adaptively nullifies interferences from other array elements. and so on.e. tags marked as T . These differences are used to separate multiple streams with signal processing at the receiver.2. one element receives the desired signal whereas other array elements are used to remove the interference in order to maximize the received signal strength. LITERATURE SURVEY 34 those tags in the current cluster respond. i. For instance. one at a time. when all the tags have been read by the reader. i. but with significant increase in the reader complexity and monetary cost. the reader scans the area using the directional beam until the desired tag is detected or.. assume the reader has read the tags from cluster d and has just sent a request to cluster d . each antenna element receives a superposition of the multiple transmitted streams with different spatial signatures.e. 2. Electronically controlled directional antennas [42] are used to adaptively adjust the directional beam of the RFID reader at subset of tags. however. After all tags marked as T have been read. Smart antenna techniques such as adaptive array antenna and MIMO antenna [65] are used to maximize system throughput by reducing the collisions. Then. In that instance only tags from cluster d . Such techniques are commonly known as adaptive space division multiple access and are depicted in Fig. Similar transmission control schemes have been used for reader collision avoidance [60]. With MIMO.

The FDMA .14: Adaptive SDMA with an electronically controlled directional antenna. LITERATURE SURVEY 35 Tag Tag Tag Tag Reader Tag Tag Tag Interrogation Zone of reader Figure 2. shown as f1 . applications. The tags reply to the reader using one of the many available frequencies.2. this implies the availability of broadcasting frequency for the reader and several frequencies for the tags that they can lock on to. In the context of RFID systems. Frequency Domain Multiple Access (FDMA) proposes the use of multiple channels. One exceptional approach is the aforementioned power control scheme [10]. f2 . synchronizes and issues the interrogation commands.15. The reader. f3 and f4 . which is low in cost and is effective in reducing both reader and tag collisions. 2. in Fig. using the broadcasting frequency. each with distinct carrier frequencies for communication. The advantage of FDMA is the availability of non-interfering frequencies for concurrent communication by multiple tags and readers.2.

This is because of TDMA simplicity. restricting the use of FDMA for limited and specialized applications. GSM. is used in other prominent networks.2. in a modified and mostly hybrid manner. processing. The TDMA technique is by far the most dominant medium access protocols in RFID systems. TDMA. low processing overhead for passive tags and low complexity (computational. LITERATURE SURVEY 36 Broadcast/synchronisation fB f1 Tag Tag Tag Reader f2 f3 Tag f4 Tag Interrogation Zone of the reader Figure 2. The reason is the impracticality for tags and relatively high cost of the readers as a dedicated receiver must be provided for every reception channel.15: Frequency domain multiple access.. Bluetooth and IEEE 802.g. e.2. along the time dimension. between potential participants. and monetary cost) compared with other available procedures such as .16 (WiMax). Time Domain Multiple Access (TDMA) relates to techniques where the available channel is divided. technique has not been significantly utilized in RFID systems.

amongst the tag population. LITERATURE SURVEY 37 Figure 2. is known as the singulation process.e. The reader-driven approach is used to tackle the collisions in RFID systems. 2. i. In the context of the RFID system.. will transmit its data at any given time.2. the TDMA procedure is further classified into tag-driven and reader-driven. i.16: Classification of the time division multiple access procedures. as is shown in Fig. control when the tags transfer data and selects which ones.2. SDMA. CDMA. The tag-driven procedure operates in an asynchronous fashion as the reader does not control the data transfer. FDMA. Most of the TDMA procedures are therefore based on the reader-driven approach. isolating an individual tag from a large group of tags. The reader-driven approach is a synchronous mechanism.. Tag-driven procedures are naturally very slow and inflexible with limited applicability.e. with all the tags’ data transfer handled by the reader. etc. . This selection process.16.

16. into a more manageable set of tags. Therefore. Reader collisions employ conventional contention resolution approaches such as scheduling.2..2. Deterministic anti-collision mechanisms fall under the general algorithmic classification of tree-based algorithms because of their splitting approach. 96]. The various collision resolution schemes for tags and readers’ collisions are classified based on their underlying techniques. and are discussed next. In deterministic mechanisms the reader splits and identifies a set of tags to respond in a given time. The deterministic mechanisms utilize either tags’ serial numbers (identification codes) or randomly generated numbers to be used for splitting of the tree branches. interference learning. tag collisions are resolved by the reader utilizing techniques collectively known as tags anti-collision schemes. as depicted in Fig. in the following sections. . in a sequential manner. Splitting is based on contention information obtained from the previous interrogation cycle and attempts to reduce contention for the next cycle. and coloring schemes to tackle reader collisions.3 Tag Collisions Tag collisions are the most common source of collisions in RFID systems. These anti-collisions schemes are classified in literature. They take place when multiple tags are within the interrogation zone of a reader and simultaneously reply to the reader’s commands. LITERATURE SURVEY 38 Tag collision schemes usually involve a logical partitioning of the tag population using tree-based algorithms or probabilistic-framed Aloha schemes. 2. to be either deterministic or probabilistic [42. i.2. reader-to-reader and readers-to-tags collisions. 2. The passive RFID tag is by design readerdriven. Under certain circumstances.e. as it does not have carrier sensing nor inter-tag communication capabilities.

We further classify the deterministic anti-collision mechanisms. The objective of these interrogation cycles is to split the tags. In probabilistic mechanisms. based on the information from the previous interrogation cycle. as shown in Fig. and in the worst case. randomly transmits a particular slot in the frame. The tree-based algorithm is a 2-way handshake algorithm involving sequences of interaction between the reader and the tags. the location of which is obtained from the previous interrogation cycle. might not be able to be read at all. but suffers from tag starvation syndrome. . a tag may not be identifiable for a long time. into two branches (leaves) is based on the bit collisions and their respective positions. LITERATURE SURVEY 39 the deterministic approach may take a considerably longer time. mostly binary trees. The splitting of the tree. However. it does not suffer from the tag starvation problem. encouraging adaptability according to tag density and distribution. based on whether they use a collision tracking or a collision detection approach. known as the interrogation cycle. The frame process is repeated until all the tags have been identified. Obtaining collision information at the bit level requires that the precise bit position of collision is identifiable by the reader. and the tag. 2. due to its low overhead. into a more manageable set of tags. either the Manchester coding or the NRZ (non-return-to-zero) bit coding is used by the RFID reader. normally.2. The frame size may be adjusted. The probabilistic approach is fast. For this particular reason. the reader communicates the frame length. using their serial numbers (IDs) or randomly generated numbers.2. Deterministic Schemes Deterministic anti-collision mechanisms are essentially tree-based anti-collision algorithms.17. In tag starvation.

i. 105].. reducing memory usage.e. both from communication and processing perspectives. collision bit.e. 41. The conventional binary tree-based algorithm recognizes only a single tag at a time through setting the most significant collision bit to either 1 or 0 during the interrogation process. reducing processing overhead or eliminating the unnecessary iterations. LITERATURE SURVEY 40 Collision Tracking: In the collision tracking method. shortening execution time. 113. 80. The variants of conventional binary tree algorithms differ in their enhancement objectives. 71] notes that this results in low performance and claims that without any fundamental change in the tree navigation process two . In the conventional binary tree algorithm.e.. both the reader and the tag maintain a certain amount of collision information from the previous interrogation cycle. This results in unnecessary overhead. is singulated and unselected. 43. 54. 116]). This translates to selecting either the parent node of the current query node or its sibling node [54. The work in [32. the singulation process initiates from the root for the subsequent tag. i. i. 106. for instance. in our example above where the reader sends out the initial query of 1111b . 71. read. The information. mostly as pointer to the most recent query. fall under the collision tracking class of the deterministic anti-collision algorithms. when a tag is successfully singulated. after successful singulation. 32. by setting the starting query point to the queued query from the last interrogation cycle. is used to send a subsequent query for the next interrogation cycle.2. or node in the tree.. 18. This simple modification speeds up the interrogation process and reduces overhead by utilizing the saved (queued) information. and muted. after the first tag. The binary search tree algorithm (explained above using an example RFID system) and its variants [17. tag1. the highest possible serial number. This was seen. 105. The overhead can be eliminated.2.

when the least significant bit collision has been resolved.. therefore speeding up interrogation process. or energy aware. depth-first search algorithms with tag state support [60] and secure variation of the tree algorithms [17. hybrid. utilizing the similarities among the tags’ IDs [58].2. 76. as is shown in Fig. 2. i.17. 106] have also been investigated. We further classify the collision tracking approach to be either statistical. LITERATURE SURVEY 41 Deterministic Collision detecting methods Collision tracking methods Memoryless Prefix overhead reduction Binary Tree Quary Tree Hybrid Statistical Energy aware Figure 2. with collision bits of 0 and 1. 77] initiate the conventional tree .e.. Adaptive binary splitting algorithms [26. tags are identifiable at one time.2.e. i. All these approaches enhance the binary tree algorithm without deterring significantly from the conventional algorithm. The statistical approach exploits information stored in previous interrogation cycles (iterations) to assist in the forthcoming cycle. This takes place during the last bit of the collision sequence. This is evident. using height-oriented. The use of other mechanisms such as interference cancellations [112].17: Classification of the deterministic TDMA schemes. as the least significant bit collision indicates the presence of two tags.

2. the hybrid approach increases the tags’ cost as it requires additional functionality. LITERATURE SURVEY 42 splitting procedure by using the information obtained from the last reading cycle.2. this approach utilizes the information from previous reading cycles to avoid collision for the static tags and starts the querying process from the root in order to accommodate the tag mobility. where a significant portion of tags are mobile. the adaptive splitting algorithms will perform no better than the conventional binary algorithm. In such scenarios. or has moved out of the interrogation zone of the reader. The fixed tags. back-off timer. with slotted back-off aimed at reducing the number of unwanted and unnecessary query commands. i.. e. It combines the tree-based protocol with a slotted back-off mechanism. and therefore was not present during the last reading cycle.g. If there is a collision the reader can partially deduce how the IDs of the tags are distributed and potentially reduce the unwanted idle cycles. which have been identified in the previous cycles will be re-identified again by the reader in the next reading cycle.e. The algorithm uses a 4-ary tree instead of the conventional binary tree.. static tags. a tag sets its back-off timer using a part of its ID. These adaptive splitting algorithms significantly reduce the number of collisions while supporting mobility but at the cost of substantial storage requirements. The mobile tag is defined as the tag just coming in. The MAC scheme’s objective of reducing collisions usually comes at the expense . whereas conventional methods are used to identify the mobile tags. Upon responding to the reader. As the reader knows about the fixed tags it can avoid collisions. Like the statistical approach of [76]. The hybrid tree approach [94] is a combination of the deterministic and probabilistic approach. In addition to inheriting the drawbacks of the statistical approaches.

Relative to collision tracking. The approach in [116]. 115.2. as is shown in Fig. These two mechanisms help to read more tags with fewer queries. indirectly reducing energy consumption. by reducing communication overhead. 87. mainly the tag energy consumption. 2. thereby avoiding collisions. The tags are allowed to transmit responses within a slotted time frame. 116] address the energy issues.77. Their main advantage is in their non-storage requirements. Collision Detection: Collision detection methods are stateless algorithms in that they do not exercise tracking. We classify the collision detection methods into memoryless and prefix reduction. thereby reducing the energy consumption at the tags. detection algorithms may take longer to read all tags within an interrogation zone. More efficient schemes [90] use multiple time slots per tree node and three different anti-collision protocols. while maintaining low collisions.116] are prominent examples of . The energy-aware approaches [78.2. This is important in the conventional binary tree algorithm as it needs to determine which sub-tree to query in the subsequent interrogation cycle. Query tree protocol (QT) [63] and variants [31. This energy consumption occurs because of additional processing and communication overhead imposed by the MAC protocols. while utilizing the conventional binary tree algorithm. involves the reader taking action if bit collision is detected. The motive is that in order to detect the tags. the binary tree algorithm relies on the collision as it transmits redundant queries to the tags. This translates to the reader sending a special symbol to the tags to stop sending data back. Fewer bits are thus sent by the tags. especially for the tags and their comparatively low communication overhead. The queries are used to find colliding sub-trees as well as to read the tags. LITERATURE SURVEY 43 of increasing energy consumption.17.

i. a QT imposes simple functions and requires no state or pointer maintenance by the tag. At the beginning of the frame. This means that no storage is required for random serial numbers. LITERATURE SURVEY 44 the memoryless approach in which each tag does not need additional memory beyond storing its serial number. it remains silent.. Contrary to binary trees. from most significant to least significant. in that the reading delay is increased when tags have minimal difference in their serial numbers. The reader broadcasts the query. The reader pops a bit string from the queue and broadcasts to all tags in its interrogation zone. pointers or states. However. the string bits. By expanding the query until there is either a response or no response — all the tags will respond eventually — the query bit string traverses from the most significant to the least significant bits for the possible serial numbers.e. The reason is the string bits have to traverse. using a string of bits as an argument. the queue is initialized with two 1-bit strings. 0 and 1. . the reader pushes two 1-bit longer bit strings. If there is a match the tag transmits the remaining least significant portion of its serial number. compared to the last bit string transmitted. The reader maintains the queue for the transmitted strings of bits. The tag receiving the query matches the most significant bit of its own serial number with the broadcasted serial number. If the tag responses collide. as it was required by the binary tree of the collision tracking algorithms. the tag reading delay for the QT protocol is significantly affected by the numerical distribution of the tag serial number. into the queue. The QT and its variants has noticeable overhead in terms of the number of bits that are sent as query argument. otherwise. equal to the length of the serial number.2. The QT uses the tag’s serial number for the tree splitting mechanism. This results in communication overhead as redundant bits are transmitted.2.

during the interrogation cycles. The initial prefix length is then increased repeatedly until the collision ratio satisfies a prescribed condition. Tag estimation is proposed in the adaptive optimal incremental prefix length algorithm [29]. falls under the prefix reduction mechanism and overcomes the limitation of the QT-based algorithm. Prefix overhead reduction algorithms [28. The frame size may be adjusted based on the collision. idle and successful . i. create longer reading cycles due to tag length and its distributions. The reader repeats this process until all tags have been successfully transmitted at least once. Reader controlled synchronization is necessary as the tag has to transmit within its slotted time frame. the collision detection mechanisms enjoy a faster reading rate and lower overheads with marginal need of additional memory at the tags.29. Prefix Randomized Query Tree (PRQT) [113]. The algorithm begins by setting a small initial prefix length followed by the polling of all possible prefixes..2. LITERATURE SURVEY 45 both by tags and readers.80. PRQT similar to QT is a stateless approach however. The optimal length prefix depends on an estimate of the number of tags within the interrogation zone.e.113] maintain prefix and iteration overhead reduction approaches which enhance the performance of the memoryless mechanisms. thereby providing additional memory space. Probabilistic Schemes In the probabilistic mechanisms the reader communicates the frame length and the tag picks a particular slot in the frame for transmission. By reducing the prefix overhead. Each tag generates a random length prefix which is then used during the interrogation process.2. and it differs from QT in that it uses prefixes chosen randomly by tags (rather than using their ID-based prefixes).

thereby reducing idle and collision frames. Generally speaking. we classify the probabilistic mechanisms into two main groups: static and dynamic. For example.18: Taxonomy of the probabilistic TDMA-based schemes. The probabilistic approach is faster in comparison to the deterministic approach because of its low overhead. However.2.18. and as shown in Fig.2. This encourages frame adaptability according to tag density and distribution. making it mostly used in instances of low tag density. it may suffer from tag starvation syndrome. frame information from a previous interrogation cycle for the subsequent cycle. An example of static algorithms includes those based on Aloha and framed slotted Aloha. LITERATURE SURVEY 46 Probabilistic Static Dynamic Non-adaptive Adaptive Reservation Estimation Statistical Figure 2. or vice . 2. While a fixed frame size leads to a simple implementation. The number of slots in the static approach is fixed. long frames used with a small number of tags. static algorithms risk certain inefficiencies.

the frame size is increased when the number of collision slots exceeds a preset threshold. results in delays and resource underutilization. when the number of tags is small the reader can identify all tags without much collision.2. in the second variation above. In another variation. That is to say. LITERATURE SURVEY 47 versa. the frame size is adjusted based on some predefined threshold. For instance. As the collisions decrease. For example. as the frame size cannot be infinitely increased. the Dynamic Framed Slotted Aloha (DFSA) algorithm [42] determines the frame size based on information such as the number of slots used to identify the tag and the number of the slots with collisions. Such issues are addressed in dynamic approaches in which the frame size is adjusted based on the collision rate from tags within an interrogation zone. If at least one tag is successfully identified. Variations of the dynamic framed slotted Aloha algorithm differ in frame size by changing approach. and vice versa. The frame size is incrementally increased with unsuccessful transmissions until at least one tag transmits successfully. however. For example. changing the frame size only may not be sufficient given the number of tags. In one such variation. If the number of tags is large. This dynamic adjustment of frame size allows readers to change its frame size depending on the tag density. the frame size is increased as with an increase in number of tags increases the collisions. the current reading process is aborted and reinitiated from the beginning at its initial frame size.2. the number of slots needs to exponentially increase since the reader always . Despite the general performance of the dynamic framed slotted Aloha and its variants. the frame size is reduced to its previous value. the interrogation cycle begins by setting the initial frame size to two or four.

23.59.86.2. The Advanced Framed Slotted Aloha (AFSA) algorithm [102] estimates the number of tags. Estimation based approaches are the most important Aloha based schemes [21. estimation.97. A tag receiving the request generates a new number from the received random number and the tag’s own serial number and divides the new number by the number of tag groups.18).85. Only the tags having the remainder of zero respond .e. i. However. Dynamic algorithms (both adaptive and nonadaptive) can be further classified based on their use of reservation. LITERATURE SURVEY 48 starts with the initial minimum frame size.66.. which is impractical with high tag density. prior to initiating the reading process.102] and utilize tag count estimation techniques to adjust the frame accordingly. or statistical techniques. Adaptive algorithms accommodate tag mobility more than non-adaptive algorithms because they reduce the probability of tag collisions while simultaneously expediting the identification of the RFID tags.2. and adjusts the frame size based on the estimation. Only one group of tags is allowed to respond at one time. setting an upper bound on the frame size. We further classify the dynamic algorithms as adaptive or non-adaptive (see Fig. Adaptive algorithms use statistical information for frame size adjustment. EDFSA initially estimates the number of unread tags using the AFSA estimation function and then partitions the unread tags into a number of groups. The reader transmits the number of tag groups and a random number to the tags when it broadcasts a request. AFSA has to increase the frame size indefinitely. The Enhanced Dynamic Framed Slotted Aloha (EDFSA) [66] overcomes the limitation of AFSA by bounding the estimation based on maximum frame size. The number of groups that give output to the maximum throughput for the subsequent reading cycle is calculated and adjusted after every interrogation cycle. 2.

This increase. overhead communication and processing requirements that may increase the cost of a tag. for example. Although EDFSA overcomes the limitation of AFSA protocol.2. from previous interrogation cycles and reading processes. The mobility is supported by accounting for tag arrival and departure rate. LITERATURE SURVEY 49 to the request. 44. Speed: the rate at which tags can be read. In what follows. Discussion Most of the anti-collision protocols discussed so far in this chapter attempt to maximize specific gains at the expense of other performance merits. This is a general objective that is sought by most proposals.2. while initiating the estimation and frame adjustment at the beginning of every interrogation cycle. to estimate the number of tags presently within the interrogation zone of the reader. The statistical algorithm resembles the deterministic anti-collision category and shares the same pros and cons. Adaptive Slotted Aloha Protocol (ASAP) [59] utilizes information relative to the tag population. we discuss some of the performance merits traded off in the design of anti-collision algorithms. however. The ML-based estimation algorithm is used for this purpose. The estimation and grouping of the unread tags are performed after each read cycle until all the tags have been read. 59] exploit statistical information to improve the read time of the RFID systems. an algorithm increases the reading rate. comes at the expense of additional memory. The frame size is adjusted optimally to reflect the tag estimation. 1. Statistical algorithms [30. In minimizing collision. . it has longer reading cycles and additional functionality at the tag. 43.

The communication overhead includes sending additional bits which otherwise would not be sent. Scalability: the ability to accommodate high tag deployment densities. 3. Overhead: the communication and processing overhead for the tags and the readers.2. A reader may maintain statistical information but the storage cost at the reader is significantly less than at the tag.2. The processing overhead includes additional interrogation cycles or data crunching beside the nominal. and the tradeoffs. depicted by a down arrow (↓). depicted by a checkmark ( ). Mobility: the ability to accommodate tags which enter and leave an interrogation zone during the interrogation process. the prefix reduction scheme’s main objective is to reduce the communication overhead which indirectly reduces the time it requires to interrogate the tags. A scheme may not affect certain metric but may do so indirectly. Increase in the cost also reflects an increase in the device monetary value. State: the amount of state that can be reliably stored on the tag. for the deterministic and probabilistic anti-collision algorithms. thereby accelerating the tag reading rate. 4. For instance. respectively. 5. 6. Table 2. Additional storage means additional memory.2 and Table 2. with diverse sets of RIFD applications. come with a tradeoff and the suitability of each varies under different circumstances. Therefore. Cost: in terms of additional functionality and memory requirements at both tags and reader. no .3 show the scheme improvements. It is evident from the tables that each of the schemes. either deterministic or probabilistic. LITERATURE SURVEY 50 2. This shown by −.

63. 80. with its limited scalability or memoryless. 29.2. or possibly thousands with item-level tagging. LITERATURE SURVEY 51 Approach Energy-aware [78. docking doors) to another (e. it is of utmost importance to be able to read all the tags encompassing a certain performance metric.2: Comparison of deterministic anti-collision algorithms. 116] Hybrid [94] Memoryless [31.. An RFID based access control system does not require scalability and high reading rates. in a typical warehouse scenario. 61. as they move from one section (e.g. with its low support for mobility. For such scenarios. single existing scheme fulfills all performance metrics. therefore any generic tree-based algorithm is sufficient. For example. sorting) section of the warehouse. 76. 77.g.2. compared to other high-speed reading schemes such as hybrid. mobility and scalability matter more than overhead reduction. the reader deployed at the docking doors or the conveyor belts is to interrogate hundreds of mobile tags. the statistical approach is the most promising. . 77] Speed Overhead Stateful Mobility Scalability Cost ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ Table 2. 113] Statistical [26.. as speed. 116] Prefix [28. 87. Therefore.

19. 2. Multiple mechanisms based on scheduling. 43. 102] Statistical [30. RFID tags are limited functionality devices that are incapable of differentiating between multiple readers and they cannot be mass produced to communicate using multiple frequencies. We classify the reader collision schemes.2. and can support . While the use of multiple frequencies may effectively minimize reader collisions. 2. In the scheduling based approach.4 Reader Collisions Readers with intersecting interrogation zones may interfere with each other’s operation to the extent of barring a reader from identifying tags within its reach. Scheduling can be centralized or distributed. 66. shown in Fig. coverage and learning are adopted to resolve collisions. LITERATURE SURVEY 52 Approach Static [42] Reservation [107] Estimation [21. 59] Speed Overhead Stateful Scalability Cost ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓ Table 2. 68]. 85. 86.2. a reader collision results [39. 58. In either case. a reader’s operation may also be affected by other readers even if their interrogation zone does not overlap. 44. 23.3: Comparison of probabilistic anti-collision algorithms. As with other radio systems. frequency and an associated time slot is scheduled for the reader.2.

Algorithm such as HiQ [51] falls under this category. simply known as colorwave. An objective of both algorithms. genetic algorithms. Occasionally. Alternate approaches such as the use of beacon channels [15]. and neural network methods. LITERATURE SURVEY 53 Reader Collision Scheduling Coverage Learning With Tag support Without Tag support Online Learning Neural Network Genetic Algorithms Figure 2. Both DCS and VDCS fall under the scheduling category. The learning approach attempts to minimize collision by learning the collision patterns and assigning frequencies based on the learned pattern. scheduling without tag support is the most common approach. additional data is stored within tags at the expense of additional memory requirement. a pair of distributed algorithms called Distributed Color Selection (DCS) and Variable-Maximum Distributed Color Selection (VDCS) are introduced [103]. is to color each node of the multiple reader networks using multiple colors in such a manner that .2. i. both static and dynamic frequency assignments. central cooperators [105]. Learning based approaches.2. However. are based on hierarchical online learning..e. Algorithms such as colorwave [103] fall under this category.19: Taxonomy of reader collisions schemes. and handling reader collisions as coverage problems [19] also exist in literature. on the other hand. the scheduling is enhanced with tag support. To achieve optimal frequency channel assignment.

as required. HiQ. tries to minimize the reader-to-reader collisions by assigning frequencies to each reader over time.2. The process of selecting frequency and scheduling time slot is repeated. as in practice only a limited numbers of frequencies are available that can be utilized. The colorwave reduces the reader collision up to certain network scales. The RFID reader communicates using prescheduled frequency in the reserved time . R-servers. the reader selects a new time slot and informs its adjacent nodes of the selection. namely: readers. Upon collision. With color representing different frequencies. selects a new frequency and informs its neighbor of its actions.2. Colorwave is a TDMA based approach. and Q-servers. i. collision patterns and estimation. wherein each reader randomly selects a time slot to transmit using its assigned frequency. it reinitiates itself. Colorwave optimizes the minimum number of colors and therefore frequencies required. HiQ utilizes three basic hierarchical tiers. LITERATURE SURVEY 54 the possibility of having two adjacent nodes with the same color is minimized. similar to Colorwave. At the lowest tier are the RFID readers. The maximum allowable color is increased if the successful transmission rate drops below the configured value. The optimal or near-optimal frequency assignment is achieved by repetitive environment interaction. online learning algorithm that finds the dynamic solution to reader collision problems. If any neighbor is scheduled to use the same frequency (color). is based on learning the collision patterns among neighbors and then assigning the optimal frequency. in some form already present in the existing RFID systems. The frequency assignment however.e. the difference between an adjacent node’s color could indicate the availability of non-interfering frequencies.. up to a predefined maximum value while maintaining a configured successful transmission rate by the readers. The Hierarchical Q-Learning (HiQ) algorithm [51] is a hierarchical.

Genetic algorithms are a form of blind local search. the decision being made by upper tiers. The reader detects the collisions amongst the neighboring readers. This statistical information. which is repeated until a solution is found with no interference.2. A neural network approach is a fixed channel solution where each neuron represents an RFID reader. the reader-level server (R-server). as explained earlier. Other learning based approaches include neural networks and genetic algorithms.e. by finding an optimum scheduling policy based on the underlying readers’ learning experience or collision information. a single solution is an assignment scheme for all RFID readers. the readers with overlapping interrogation zones. i.. The Q-learning server (Q-server) is the highest tier and is responsible for resource allocation. In the genetic algorithm approach to the channel assignment. The HiQ learning algorithm is distributed at lower-level (readers) with centralized scheduling at the top-most tier. The genetic algorithm takes a set of solutions as the population and begins the evolutionary process. i.2. The distributed . which the neural network tries to minimize. A one-on-one relation exists between the low-level reader and the R-server. Alternate procedures such as beacon channels [15] and central cooperators [105] are used to tackle multiple readers-to-tag collisions. but at the cost of a centralized solution.e. LITERATURE SURVEY 55 slots.. The learning and scheduling approaches are mainly targeted towards reader-toreader collision. resulting in higher throughput. is utilized for optimal frequency assignments. HiQ provides optimal scheduling. The constraint and optimization criteria of the system are translated into an energy function. The outcome of these neurons determines if the channel may be usable for the represented RFID reader.

Table 2.2. is inclined toward eliminating the collision caused by mobile RFID readers. periodically broadcasts a beacon using the control channel. The pulse protocol however. Any other reader that wants to read the tags. the present ’multiple points to multiple points’ collision problem (caused when multiple readers are trying to access multiple tags) translates into the classical problem of ’multiple point to one point’ collision. a centralized device is responsible for the multiplexing of tags’ data to multiple readers. Using a control cooperator. However. is based on the beaconing mechanism. The algorithms vary and have fundamental differences. The central cooperator provides an affective approach in handling multiple readers-to-tag collisions. The central cooperator multiplexes multiple reader requests into a single request for the tags.. tags within the overlapping interrogation zone. The reader.. it requires a device with all the functionality of a conventional RFID reader. i. The reader collision avoidance schemes include frequency assignment scheduling and online learning approaches to avoid reader collisions.e. while reading the tags within its interrogation zone. Pulse [15]. using central authority for channel assignment and multiplexing the reader queries. i. has to ensure that no beacon has been transmitted on the control channel before it can initiate its interrogation process.4 shows a contrast between various algorithms for resolving the collision problem. some are centrally controlled. In the central cooperative approach [105]. LITERATURE SURVEY 56 protocol. The responses from the tags are de-multiplexed to the individual readers separately.e. For instance. . The latter reader has to wait until the first reader has executed its interrogation process because the control channel is idle for a specific time period before triggering its own beacon.2.

especially reader-to-reader interference. especially passive tags. are unable to differentiate between various readers. the tag shall provide support for up to four sessions.5 Coverage Existing schemes handle coverage and deployment issues for RFID networks in two approaches: planned or random. The readers. this requires extra memory at the tag. the fundamental difference is that RFID tags.2. However. increasing the cost. According to the EPC Gen2 standard. . is similar to a frequency assignment problem in conventional wireless communication. two readers communicating with the tag must communicate at different time slots or make use of sessions. Therefore.4: Comparison of reader collision algorithm. This allows multiple readers to interrogate common populations of tags. selectively singulate the tags into their respective sessions and move them into other respective sessions upon completion. allowing two or more readers to independently interrogate the tag. 2. LITERATURE SURVEY 57 Method Approach Centralized Distributed Fixed control control Channel - Dynamic Channel - DCS [103] scheduling VDCS [103] scheduling HiQ [51] learning Neural learning Annealing [69] learning Genetic learning Redundancy [19] coverage CC [105] coverage - Table 2.2. Although effective. Reader collision. step-by-step.2.

The scheme proposed in [104] finds an optimal number of readers. each reader broadcasts a message carrying its own identity and its tag count. Access to a particular tag is granted to the reader covering that . Another similar approach is the honey grid [110] in which the readers are deployed in rings of different sizes and centred at the centre of the interrogation zone. readers not covering any tags can be turned off. and under some constraints on where readers can be located [25. Some reader deployment schemes that take into account tags and readers orientation have been proposed [104].2. 83.2. LITERATURE SURVEY 58 In the planned approach. Each tag stores the identity of the reader having the maximum tag count it received.53]. the ith ring contains 6i readers. In the random deployments. In the RRE scheme [16]. after deployment. like other schemes in the literature. Similar to the regular grid approach. After deployment. readers are deployed in a grid where the distance between two neighboring readers is determined by the interrogation range of readers. When rings are numbered based on their radii. 110]. algorithms are devised to find redundant readers in a given deployment strategy [16. which is the number of tags within its region. the reader(s) not covering any tag can be turned off. It has been proven that both schemes provide maximal possible coverage. 50. 92. it is only being proposed for portal accuracy in supply chain management scenarios.52. However. In the optimal grid coverage approach [83]. The scheme. 104. is not generic. it considers only placement without load balancing. Furthermore. algorithms are developed to find optimal placement of readers in terms of maximal coverage and minimum number of readers. however. their locations out of a set of pre-fixed locations and their antennas orientation to maximize readability. that comes at the cost of large number of readers. 93. significant readers collisions and unfair load distribution.

The art gallery problem for the 2D and 3D case is linear and NP-hard. the RRE algorithm is then used to further eliminate any redundant readers. 111]. has also been formulated as art gallery problem. inside a polygonal art gallery. Oceans coverage problem. Coverage.. coverage for robots and coverage in wireless sensor networks. Readers that are not granted access to any tag are marked redundant and are turned off. In the art gallery problem [84]. Various randomized algorithms have been proposed to solve the art gallery problem. The reader that manages to singulate the tag and writes the identity first is granted access. wherein an approximation algorithm is presented for the later case [73]. The art gallery based coverage problem has many real world applications.e. the LEO and the LEO+RRE algorithms [52] introduce the concept of ”first-read first-own” principle: an RFID reader only write its own identity into a tag memory. such as placement of antennas for cellular telephone . The basic idea is that each reader writes its tags count into tag memory space. The RRE scheme has a lot of communication overhead as it requires frequent write operations to the tag. Various other middleware-based load balancing schemes have also been proposed in the literature [24]. and they try to make enhancements upon the RRE and LEO schemes. To mitigate this overhead. respectively. A load balancing scheme utilizing tag storage space has also been proposed [37]. LITERATURE SURVEY 59 tag with the maximum tag count (i. and the tag picks and responses back to the reader with the lowest load. upon completion of LEO execution.2. the reader of the identity stored in the tag). in context of node placement.2. the problem translate into finding the minimum set of locations for security guard. Various other redundant reader elimination schemes have been proposed in the literature [53. In case of the LEO+RRE algorithm [52]. such that the gallery is visible by at least one of the guard.

In barrier coverage the goal is to achieve a static arrangement of nodes that minimizes the probability of undetected penetration through the barrier. the goal is to achieve a static arrangement of sensors that maximizes the total detection area. is justifiable only for large coverage (e. the results show that merging data from three satellites can increase ocean coverage by 58% for one day and 62% for four days. due to large associated cost. is an open research direction. Using numerical analysis..2.e. In the blanket coverage. using satellites based remote sensing. and sweep coverage. can not applied directly to RFID systems. whereas the sweep coverage is more or less equivalent to a moving barrier. The author herein define three types of coverage: blanket coverage.g. The exceptional coverage. deployment of nodes are subject to multiple objectives. the analogy of art gallery problem has limited applications for RFID systems as they have limited coverage. extending the network .2. These solution can be adopted to mobile RFID readers however. The main objectives are to minimize the number of nodes. the authors are interested in monitoring the Oceans pigments concentrations. Oceans) and hence. i.. In wireless sensor networks. reading range. LITERATURE SURVEY 60 companies. the nodes can spread out such that the area covered by the network is maximized and can relocate to handle sensor failures. barrier coverage. However. New applications arise in the context of mobile WSNs. and placement of cameras for security purposes in banks and supermarkets. Thus. Such a multi-sensing solutions. The coverage concept with regard to the many-robot systems was introduced in [47]. In ocean coverage problem [48]. using limited satellites was made possible as the satellite nodes had different orbits and therefore was able to sense the same ocean location at different times of day. where sensors have locomotion capabilities.

deterministic coverage and k-coverage. 89].e. The authors prove that the disjoint set coverage problem is NP-complete and and propose an efficient heuristic for set covers computation using a mixed integer programming formulation. These objectives are addressed in the context of area coverage. A large number of sensors are dispersed randomly in close proximity to the targets. point coverage and barrier coverage. the objective is to determine a minimum number of sensor nodes and their location such that the given points are covered and the sensors deployed are connected. mainly is used for surveillance applications. authors consider the scenario of deterministic point coverage. the sensor network. The most studied coverage problem is the area coverage problem.2. where the main objective of the sensor network is to cover (monitor) an area. i. These variation utilizes diverse techniques such as node duty cycling and node scheduling.. Given these set of points. .e. Such coverage involve three tiers mobility. In [20] authors proposed a solution wherein the disjoint sets are modelled as disjoint set covers.. using both centralized and distributed algorithms. i. such that every cover completely monitors all the target points.2. LITERATURE SURVEY 61 lifetime [35] and increasing the network fidelity [114]. requirement is that every target must be monitored at all times by at least one sensor. Various energy efficient variations of the area coverage also have been proposed in the literature. the goal is to minimize the probability of undetected penetration through the barrier.e.. In such applications. The main objective is to minimize the number of nodes required to maintain certain coverage while maintaining network connectivity. connected random coverage. i. In the barrier coverage [74. target mobility. the possibility for explicit placement of set of sensor nodes at certain points. In [56]. The point coverage scenario considers a limited number of points (targets) with known location that need to be monitored.

the reader splits and identifies a set of tags to respond in a given time. 2. The energy harvesting yields battery-less RFID tags which enables minimal maintenance. Therefore the solutions from the wireless sensor network coverage can be adopted for RFID systems however. coverage. The backscattering yields transceiver-less RFID tags which enable minimal power and low cost RFID solutions. In the deterministic algorithms. within its reading range. it is not possible to use the conventional wireless medium access and collision avoidance protocols. Energy harvesting and backscattering modulation are the main foundation technologies of low-cost identification solutions provided by RFID. longer lifetime and low cost RFID solutions.3. similar to RFID.2. Due the simplicity of the RFID tags. Splitting is based on contention information obtained from the previous interrogation cycle and attempts to reduce contention for the next cycle. and data collector mobility. RFID system employs a centralized communication architecture wherein the reader simultaneously interrogates and process the tags. SUMMARY 62 sensor mobility. However.3 Summary In this chapter we present the background and literature survey on medium access control and coverage in RFID systems. is very application dependant. RFID employes deterministic slotted aloha and probabilistic framed slotted aloha anti-collision algorithms. Unlike RFID. In the probabilistic algorithms. To this extend. the reader communicates the frame length. This centralized and simultaneous interrogations result into collisions. is a largely unexplored research area. the most common goal is to reduce the number of deployed sensor node while maintaining an acceptable coverage of the event sensing area. the WSNs node deployment and hence. and .

based on the information from the previous interrogation cycle. the deterministic algorithms may take a considerably longer time. On the other hand. Under certain circumstances. . but may suffers from tag starvation syndrome. SUMMARY 63 the tag. encouraging adaptability according to tag density and distribution. The frame size may be adjusted. due to its low overhead. the probabilistic algorithms are fast. randomly transmits a particular slot in the frame.3.2.

In PDC. Tags which are being interrogated within the current cluster will not respond to the reader’s queries for the subsequent .e. This causes immense data collisions at the reader. one cluster at a time). To the best of our knowledge. the interrogation delay is still a problem for some applications that involve dense and/or fast moving passive tags. the proposed Power-based Distance Clustering (PDC) scheme [10] is the first to exploit such an avoidance approach. Any existing anti-collision scheme can be used to resolve collisions in a single cluster. Such a scheme should reduce the number of collisions as it reduces the number of tags that may respond at the same time. and to have tags in each cluster being read separately (i.1 Introduction The focus of our work in this chapter is to overcome tags-to-reader collisions.64 Chapter 3 Power Control in RFID 3.. the reader tunes transmission power so that tags within the interrogation zone are clustered based on their distance from the reader. A promising opportunity to avoid a significant amount of collisions is to partition the interrogation zone spatially into smaller clusters. While several schemes have been proposed to deal with tag collisions in RFID systems.

hence. We formulate two optimization problems for finding the clustering scheme that minimizes the number of collisions and. it was not clear how to find the best partitioning scheme. we relax this assumption and consider an RFID reader with finite. Indeed. these tags are forced into a sleep mode. In the second problem.4 presents a mathematical analysis and an optimal algorithm for clustering the interrogation zone of an RFID reader with a finite number of transmission ranges. Section 3. The rest of this chapter is organized as follows. reading rates. The first problem targets an ideal setting in which the transmission power and. both situations affect the performance of that approach significantly. the transmission range of the RFID reader can be tuned with high precision.3 presents a mathematical analysis and an optimal algorithm for clustering the interrogation zone of an RFID reader whose transmission range can be tuned with high precision. Section 3. Our proposed methods have been designed to adapt to different system environments such as the number of tags and their distribution.1. we present a mathematical delay analysis and devise an efficient method to find the optimal clustering scheme. INTRODUCTION 65 clusters. as once read. having too many clusters may result in many empty clusters. hence. However.5 presents the results of our experiments. We also show the results of several experiments.3. which is an extra overhead. and delay. Section 3. We studied the viability and efficiency of the PDC scheme in [10]. For each optimization problem. which verify the effectiveness and performance improvements of the PDC scheme in terms of number of collisions. and few clusters may result in having crowded clusters.2 introduces our PDC scheme. . reduces the interrogation delay. using the ns-2 simulator [7]. The contributions of this chapter are as follows. Section 3. discrete transmission ranges.

2.GT ag (4π)2 R4 S= (3.Preader . Preader is the power supplied to the reader’s antenna.2) that the reading distance between the tag and the reader can be changed by changing the power supplied to the reader’s antenna while maintaining Pback .GReader .G2 Reader .GT ag . . This has the potential to reduce the number of tags which can be concurrently active.1) that the power density reflected back by the antenna is proportional to the fourth root of the power transmitted by the reader. It follows from (3. the interrogation range of the reader can be reduced by lowering Preader . The reflected power density and the reader range can be computed using the following formulas [42] [115]: λ2 .1) and R= 2 λ 4 k.3.6 summarizes our work. 4π Pback (3. and. GReader and GT ag are respectively the antenna gain for the reader and the tag.Preader . lower the collision probability. hence. and Pback is the power received by the reader from the tag. expedite the interrogation process. Hence. R is the distance between the reader and the tag. Partitioning the interrogation zone can be achieved by controlling the readers antenna power level. 66 3. Section 3.2 The Power-based Distance Clustering (PDC) Anti-Collision Scheme The PDC is a divide-and-conquer anti-collision scheme in which tags are divided into clusters based on their distance to the reader. λ is the wavelength of the emitted electromagnetic wave. Also it follows from (3. Tags in different clusters are then read separately (one cluster at a time).2) where S is the reflected power density. THE POWER-BASED DISTANCE CLUSTERING (PDC) ANTI-COLLISION SCHEME Finally.

Then only tags from cluster D2 (i. 3. The scheme is sensitive to the size and the number of clusters.e.2. On the other hand. A large number of clusters in a sparse tag environments yields longer delays as a result of many empty clusters and idle cycles. especially in dense tag environments. D2 . 3. This identification process continues until the reader reaches its maximum interrogation range. divides the interrogation zone into clusters based on a fixed stepping value d. The flow diagram of the PDC scheme is illustrated in Fig. Any anti-collision scheme can be used to resolve collisions within a single cluster.5.1. assume that the reader just sent a request (query) to cluster D2 after it had finished reading tags from cluster D1 .2. and D3 . When the reader sends a request to a particular cluster. where the interrogation zone is divided into three clusters: D1 . Then. For example. this scheme does not divide the interrogation zone into equal-size clusters and does not adapt to different tags densities. The Static PDC.3. which we proposed in [10]. the reader transmission range is increased to cover cluster D3 and a new request is sent. THE POWER-BASED DISTANCE CLUSTERING (PDC) ANTI-COLLISION SCHEME 67 An example of the partitioning is shown in Fig. it has the ability to be integrated with any anti-collision scheme and it has shown acceptable improvements as shown in Section 3.e. This means that the transmission range of a cluster is more than that of the previous cluster by a fixed value d. Now only tags from cluster D3 (i. After all tags marked as t2 have been read successfully. tags marked as t3 ) will respond to the reader requests... a small number of clusters may result in having too many tags in one cluster which renders the scheme ineffective. While very simple. . they are put into a sleep mode. only those tags in that particular cluster may respond. However. tags marked as t2 ) will respond to that request.

3.1 System model and problem definition We consider an RFID system consisting of an RFID reader and n passive tags.3. 3. all tags located within . 3.3. This scheme targets RFID readers whose transmission power can be tuned with a high precision so that the transmission range can be controlled with high accuracy.3 Optimal Power-based Distance Clustering In this section we present our theoretical analysis and methodology for finding the Optimal Power-based Distance Clustering (O-PDC) scheme. OPTIMAL POWER-BASED DISTANCE CLUSTERING 68 Figure 3.1: Flow diagram of the PDC scheme. the RFID reader can communicate with. and read. The interrogation zone is modelled as a circle centred at the RFID reader with a radius of R units.

one. OPTIMAL POWER-BASED DISTANCE CLUSTERING 69 T’’ T’’ T’ T T’ T’’ T T T’’ T’ Reader d T d’ T’ d’’ T’’ T’’ Figure 3. and tags in each cluster will be read separately (existing protocols use k = 1). its interrogation zone. An idle cycle is one in which no tag responds.3. Now. This process will be repeated until all tags are read successfully. the problem can be defined as follows: .3. and a collision cycle is one in which two or more tags respond. Eventually. In general.2: An example of an interrogation zone of 3 clusters. Upon detecting a collision. a tag will be read successfully. a successful cycle is one in which exactly one tag responds. in each cycle the RFID reader sends a request and zero. or more tags respond to the request by sending their IDs. the RFID reader will send a more restricted request which excludes some of the colliding tags. The RFID reader and tags will go through several cycles during the reading process. The interrogation zone is divided into k equal size clusters. any anti-collision protocol can be used to resolve collisions in a particular cluster.

3) To generalize our analysis to other tree-based anti-collision schemes (e. the recent work in [95]1 ). For the QT protocol.881n − 1 ≤ E[f (n)] ≤ 2.887n − 1 (3. Let g(n. any existing tree-based protocol can be equally used for that purpose. 2. and tags are uniformly distributed over the interrogation zone.. Nevertheless. [62]) are able to find a precise estimation to the number of tags in a negligible time.g.4) The scheme proposed in [95] has f (n) = 2 n − 1.3. 3.3 Delay analysis Let f (n) denote the number of cycles required to read a set of n tags.3. we use the Query Tree (QT) protocol [64]. Now.g.3.2 Assumptions To resolve collisions within a single cluster. we will have: E[f (n)] = c n − 1. OPTIMAL POWER-BASED DISTANCE CLUSTERING 70 Find the optimal number of clusters that minimizes the total number of cycles required to read all tags.3. assume that the interrogation zone is divided into k equal size clusters. k) denote the number of cycles required to read a set of n tags uniformly distributed over k clusters and 1 (3. it has been shown in [64] that for n ≥ 4. and let E[f (n)] denote the expected value for f (n). where c is a constant.. . We also assume that the total number of tags is known to the reader. 3. We also assume that tags are uniformly distributed over the interrogation zone. several schemes in the literature (e.

We have s empty clusters and t nonempty clusters. If we have n tags. then k−1 k n E[s] = k (3. (3. Now. The expected number of cycles required to read tags in the ith nonempty cluster is c ni − 1. If tags are uniformly distributed over the interrogation zone. Let ni denote the number of tags in the ith nonempty cluster.. Lemma 2. then E[g(n. clusters which do not contain any tag). the expected total number of cycles is c n − t + s. we deduce the following lemma for the expectation of the number of empty clusters. For each empty cluster.6) Proof. so we get c n − k + 2s. we deduce the following lemma. and the interrogation zone is divided into k equal size clusters. k)] = c n − k + 2E[s]. this results in a total of s cycles. the expected number of cycles required to read tags in all nonempty clusters is c n − t. Let Pi denote the probability that the ith disk is empty (i.3. Therefore. Thereby.3.e. no tag is located . Proof. OPTIMAL POWER-BASED DISTANCE CLUSTERING 71 using distance-based clustering. Therefore. we will have a single idle cycle.5) where s is the number of empty clusters (i.e.. Lemma 1. but t = k − s.

∞).3.10) 3.8) (3. Proof.7) = i=1 (3.3.e. k)] is convex over the interval [1. OPTIMAL POWER-BASED DISTANCE CLUSTERING 72 in that cluster). We start with the following lemma. k)] = c n − k + 2k (3. E[g(n. k)] is a convex function over the interval [1. k)] = 3 (n − 1) 2 dk k Since n ≥ 1 and k ≥ 1. Therefore. k)]).3. the value of k that minimizes E[g(n. E[g(n. ..9) = k The following theorem is a direct result from Lemma 1 and Lemma 2. k)] is non-negative.4 Optimizing the delay The objective of this sub-section is to find the optimal number of clusters (i. then k−1 k n E[g(n. d2 dk2 k−1 k n−2 (3. and 2n d2 E[g(n. ∞). k E[s] = i=1 k Pi k−1 k k−1 k n n (3. If we have n tags uniformly distributed over k equal size clusters. ∞).11) E[g(n. k)] is twice differentiable with respect to k over the interval (0. Lemma 3. E[g(n. Theorem 1. Then.

Thus. n)] (3. it suffices to show that the expected number of cycles with n clusters is less than that with n + i clusters.e. The following lemma shows that 1 ≤ xint ≤ n − 1. k)] is convex and the optimal number of clusters is at most n. the optimal number of clusters is either xint or xint + 1. In general. there is no closed-form solution to equation 3.12. where i ≥ 1. Therefore. the optimal number of clusters is at most n.3. k)] dk is a non-decreasing function. · · · . we know that the optimal value for k is the one at which 0. n}. one can find the optimal solution through binary search over the set {1. d E[g(n. OPTIMAL POWER-BASED DISTANCE CLUSTERING 73 From lemma 3. With n + i clusters. n + i)] = E[g(n. However. n + i)]). Lemma 4. Therefore. 2.12) Let xopt denote the solution to equation 3. n)] < E[g(n. When there are n tags. we need to show that E[g(n. such that xint ≤ xopt ≤ xint + 1. which results in i idle cycles. E[g(n. (i. To prove lemma 4.3. n)] > E[g(n. n + i)|at least i clusters are empty] = i + E[g(n.13) Since E[g(n. we just need to find an integer xint .. since the number of clusters is integer. Proof. k−1 k n d E[g(n. k)] dk = −1 + 2 + 2n k k−1 k n−1 =0 (3.12. Moreover. we need to solve the following equation. we are certain (with probability 1) that at least i clusters are empty. it can be solved by numerical methods. Algorithm 1 finds the optimal number of .

lef t+right ) < 0 then 2 lef t = lef t+right . OPTIMAL POWER-BASED DISTANCE CLUSTERING 74 clusters in O(log n). 2 else right = lef t+right .3. i i i 4 end 5 6 7 8 9 10 11 Function FindOptimal(n) Input: n: the number of tags. k)] at k = i. else optimal = right. Output: the optimal number of clusters. 1 Function Derivative(n. right = n. end 12 13 14 15 16 17 18 19 20 21 22 . i: an integer between 1 and n.i) Input: n: the number of tags. while right-left>1 do if Derivative(n. begin lef t = 1. 2 begin n n−1 3 return −1 + 2 i−1 + 2n i−1 . end return optimal .3. lef t)| < |Derivative(n. Algorithm 1: Optimal Clustering. 2 end end if |Derivative(n. d Output: dk E[g(n. right)| then optimal = lef t.

a finite number of transmission ranges).. To the best of our knowledge. this applies to all today’s commercial RFID readers (e.e. OPTIMAL DISCRETE POWER-BASED DISTANCE CLUSTERING (OD-PDC) 3. and . We may also have two clusters if only T r1 and T r3 are active.. the scheme we present here is not limited to any particular distribution for the tags nor to any particular anti-collision protocol to resolve collisions in single clusters. We present a dynamic-programming algorithm that finds a near-optimal set of clusters for this setting. For example. Furthermore. all tags located within its interrogation zone. used) transmission ranges as shown in Fig.3. T r2 . . the RFID reader can communicate with. 3. and read. .4.1 System model and problem definition We consider an RFID system consisting of an RFID reader and n passive tags. 3. we can have 3 clusters if all transmission ranges are active. In this section we present a near-optimal scheme that suits RFID readers with discrete transmission ranges (i. SkyeTek’s M10 RFID reader [8]).3. The RFID reader has k discrete transmission ranges T r1 . T rk . . Each transmission range may and may not be used in interrogating RFID tags.4 Optimal Discrete Power-based Distance Clustering (OD-PDC) 75 In the previous section.g. The interrogation zone is modelled as a circle centred at the RFID reader with a radius of R units.. we presented an exact optimal scheme that is suitable for an ideal setting in which RFID tags are uniformly distributed and the RFID reader has a precisely tuneable transmission range. when k = 3. this makes it possible to divide the interrogation zone into up to k clusters. We also use T r0 = 0 to denote a virtual transmission range. where T r1 < T r2 < · · · < T rk = R.e. A cluster is the area between two consecutive active (i.4. .

such that the total number of cycles required to read all tags is minimized.4. Now. we may have only one cluster if only T r3 is active. Note that T rk must be active in order to cover the whole interrogation zone. OPTIMAL DISCRETE POWER-BASED DISTANCE CLUSTERING (OD-PDC) 76 Inactive transmission range RFID reader RFID tag Active transmission range A cluster Figure 3. . A clustering scheme is defined by a subset of transmission ranges to be active.3.3: An example of 3 transmission ranges and 2 clusters. A clustering scheme defines the set of clusters used in reading all tags in the interrogation zone. the problem can be defined as follows: Find a clustering scheme that covers the whole interrogation zone.

3 Delay analysis This subsection explains how each cluster decreases or increases the total number of cycles. we can for example add more clusters until each cluster has at most one tag and. Based on this observation.4. increasing the number of clusters arbitrarily results in having many empty clusters and. we should maximize the number of non-empty clusters and minimize the number of empty clusters.2 Assumptions 77 To resolve collisions within a single cluster.4.3. To make a general optimization scheme.14) where α is the number of non-empty clusters and β is the number of empty clusters. yet we do not assume a particular distribution.g. It is obvious that more clusters results in less collisions. the cluster (i. For some anti-collision schemes. the objective is: MAX E[α]d − E[β]. a detailed distribution does not have to be available. However. quantifying the positive effect of a non-empty cluster depends heavily on the anti-collision scheme used within single clusters. We assume that the distribution of tags is known. . OPTIMAL DISCRETE POWER-BASED DISTANCE CLUSTERING (OD-PDC) 3. additional idle cycles. hence. 3. In fact all what is required by this scheme is the probability that a particular cluster is empty of tags. we assume that an empty cluster adds one extra idle and a non-empty cluster saves d cycles. While the negative effect of an empty cluster is straight forward (which is one additional idle cycle). A cluster is defined by a pair of transmission ranges that bound it (e. (3. any anti-collision scheme can be used. there will be no collisions. Therefore. finding the exact value of d is trivial. j) is the area between T ri and T rj ). For example. We also assume that the total number of tags is known to the reader.4. hence.

(3. j) can be computed as follows. j). OPTIMAL DISCRETE POWER-BASED DISTANCE CLUSTERING (OD-PDC) 78 the binary anti-collision algorithm with backtracking has d = 1 [95]. is the sum of the ranks of clusters composing cl. We present a dynamic programming . When the rank of a cluster is negative. which can be computed based on the distribution of tags. h(i. j) denote the rank of a cluster (i. The optimal clustering scheme is one with the maximum rank. it means that the cluster is expected to add an extra cycle rather than to save cycles.j) is the number of tags located in the cluster (i.j) > 0) ∗ d − P (n(i. and let h(i.4. On the other hand. we can use d = 1 to reflect the objective of maximizing the number of non-empty clusters and minimizing the number of empty clusters regardless of the anti-collision scheme being used to resolve collisions within single clusters. This is based on the fact that a non-empty cluster saves d cycles and an empty one adds an extra idle cycle. which is denoted by H(cl). Nevertheless.4.15) where n(i. there will be 2k−1 possible clustering schemes that cover the whole interrogation zone. In general when there are k transmission ranges.j) = 0). Let’s give each cluster a rank based on its expected contribution to the objective function. Each cluster will contribute to the objective function in (3. j) = P (n(i.3. h(i. j) and P (e) is the probability of an event e. it is not easy to find the exact value of d for some probabilistic anti-collision schemes.4 Optimal clustering algorithm In this sub-section we present an algorithm that finds the optimal clustering scheme (a clustering scheme is defined by a subset of transmission ranges to be active). it is actually equivalent to the number of all subsets of a set of k − 1 elements. The rank of a clustering scheme cl.14). 3.

1 Function Find Optimal (n. CL[k] will belong to the optimal clustering scheme. k: the number of transmission ranges. 2 M [0] = 0. It is obvious that when we have k transmission ranges. j). Output: The optimal clustering. Algorithm 2: Optimal Discrete Clustering. The algorithm is described in Algorithm 2. we present the results of a simulation based study we conducted to evaluate the performance of the PDC schemes. 10 end 11 end 12 end 13 return (M [k].3.. SIMULATION SETUP AND RESULTS 79 algorithm to solve this problem. 3 for i = 1 to k do 4 N ext[i] = 0. Let M [i] denote the maximum rank of a clustering scheme in CL[i] (i. Let CL[i] denote the set of all clustering schemes in which T ri is active and T rj is inactive for all j > i. We start with some definitions. k) Input: n: the number of tags. 9 N ext[i] = j. N ext[ ]).5 Simulation Setup and Results In this section. 3. j) > M [i] then 8 M [i] = M [j] + h(i. 6 for j = 1 to i − 1 do 7 if M [j] + h(i. M [i] = MAXcl∈CL[i] H(cl)).e. This algorithm finds the clustering scheme whose rank is maximum. 0). 5 M [i] = h(i.5. We investigate the performance improvements that can be made by the three PDC schemes presented: the static PDC .

We generate random RFID topologies in which tags are uniformly distributed in a grid of 20 × 20 m2 . we have a stepping value of 0. For the OD-PDC. SIMULATION SETUP AND RESULTS 80 which is presented in Section 3. d = 0. A single reader is located at the centre of the grid. without PDC) with that of the PDC scheme integrated with the same classical scheme. the EBST scheme.1 Simulation Setup We extended the ns-2 simulator [7] to implement RFID. To compare the performance . we have 30 transmission ranges. The reader has a maximum transmission range of 10 m.e. the ratio of collision queries. an anti-collision scheme is applied and its operation continues until all tags are successfully identified by the reader.5 m). In each RFID network. The results are averaged over 20 randomly generated topologies. compare the performance of the pure QT scheme with that of the O-PDC integrated with the QT scheme (i.2.3.5 m (i... which is a direct indicator to the performance of different schemes. We may.3.. and the Optimal Discrete PDC (OD-PDC) which is presented in Section 3.e.5. and the ratio of idle queries. the Optimal PDC (O-PDC) which is presented in Section 3. and the EPC Q-algorithm. 3. Our PDC schemes are compared with three existing schemes. for example. the O-PDC scheme withe the QT scheme being used to resolve collisions within single clusters). For the PDC scheme.e. We use several performance metrics to evaluate and compare different schemes.4. We also consider the ratio of successful queries. namely the QT scheme.5. We study the performance of these schemes by generating random RFID networks (topologies) and comparing the performance of a particular pure classical anti-collision scheme (i. The main metric is the total number of queries (cycles) needed to identify all tags. Tags have randomly generated IDs.

3. we show. Fig. The O-PDC scheme saves around 35% of the queries. Fig.5. 3. On the other hand. The OD-PDC scheme saves around 20% of th queries. SIMULATION SETUP AND RESULTS 81 of our PDC schemes with these existing schemes. the result of the existing scheme without any clustering and the results of the PDC schemes when integrated with that particular existing scheme.6 shows the results of the PDC schemes when integrated with the Q-Algorithm scheme. This clearly shows the improvements that can be achieved using different PDC schemes.3. and the Q-Algorithm scheme.2 Simulation Results Total number of queries The main indicator for the efficiency of an anti-collision scheme is the total number of queries needed to identify all tags. The OD-PDC scheme saves around 10% of the queries. Fig. 3. for each performance metric. The static PDC scheme saves around 10% of the queries.5 shows the results of the PDC schemes when integrated with the EBST scheme. the PDC scheme and the OD-PDC scheme do not have any assumptions on the classical anti-collision scheme being used to resolve collisions in single clusters and.4 shows the improvements made by the PDC schemes when integrated with the QT scheme as compared to the pure QT scheme. hence. the EBST scheme. The ODPDC scheme saves around 25% of the queries. they are compared with the QT scheme. The O-PDC scheme is compared with the QT and EBST schemes only because they meet the assumptions made for the O-PDC scheme. . The O-PDC scheme saves around 25% of the queries.5. 3.

1800 1600 Number of Queries 1400 1200 1000 800 600 400 200 100 200 300 EBST PDC(EBST) O-PDC(EBST) OD-PDC(EBST) 400 500 600 Number of Tags 700 800 900 Figure 3.3.5.5: Total number of queries with the EBST scheme.4: Total number of queries with the QT scheme. . SIMULATION SETUP AND RESULTS 82 4000 3500 Number of Queries 3000 2500 2000 1500 1000 QT PDC(QT) O-PDC(QT) OD-PDC(QT) 500 100 200 300 400 500 600 Number of Tags 700 800 900 Figure 3.

this metric helps to see how the behaviour of different schemes changes with different tag densities. for these two schemes. The nice observation here is the . Fig. SIMULATION SETUP AND RESULTS 83 3000 2500 Number of Queries 2000 1500 1000 500 0 100 Q-Algo PDC(Q-Algo) OD-PDC(Q-Algo) 200 300 400 500 600 Number of Tags 700 800 900 Figure 3. It is not a surprise to see constant ratios for the pure QT scheme and for the pure EBST scheme.5.3. 3. This is because the total number of successful queries should be the same for all schemes. this metric is inversely proportional to the total number of queries. we look into the ratio of successful queries to the total number of queries in different schemes. However. it is actually the same as the number of tags. The reason is that. In fact. 3. the total number of queries is a linear function of the number of tags.8 shows the ratio of successful queries achieved by the PDC schemes when integrated with the EBST scheme and that of the pure EBST scheme.6: Total number of queries with the Q-Algorithm scheme. Ratio of successful queries To understand the behaviour of different schemes. Fig.7 shows the ratio of successful queries achieved by the PDC schemes when integrated with the QT scheme and that of the pure QT scheme.

5. 3. The probabilistic nature of the Q-Algorithm makes the ratio of successful queries less stable than those of the QT and the EBST schemes. this means that the improvements made by the O-PDC scheme are the same regardless of the number of tags. 3.9 shows the ratio of successful queries achieved by the PDC schemes when integrated with the Q-Algorithm scheme and that of the pure Q-Algorithm scheme.28 0.9.29 0. the improvements made by the PDC schemes stay significant regardless of the tags densities.31 0.24 0.26 0. as shown in Fig.3 Ratio of Successful Queries 0. and specially the O-PDC scheme. 3. This directly affects the stability of the same ratio for the PDC schemes when integrated with the Q-Algorithm. Fig.22 100 200 300 400 500 600 Number of Tags 700 800 900 QT PDC(QT) O-PDC(QT) OD-PDC(QT) Figure 3. SIMULATION SETUP AND RESULTS 84 0.23 0. However.10 shows the ratio of collision queries observed using the PDC schemes when integrated with the QT scheme and that of the pure QT scheme.7: Ratio of successful queries with the QT scheme. 3. Indeed.11 shows . stable behaviour of the PDC schemes. Ratio of collision queries Fig.25 0. Fig.27 0.3.

3.5. SIMULATION SETUP AND RESULTS

85

0.62 0.6 0.58 0.56 0.54 0.52 0.5 100 EBST PDC(EBST) O-PDC(EBST) OD-PDC(EBST)

Ratio of Successful Queries

200

300

400 500 600 Number of Tags

700

800

900

Figure 3.8: Ratio of successful queries with the EBST scheme.

0.42 0.41 Ratio of Successful Queries 0.4 0.39 0.38 0.37 0.36 0.35 0.34 0.33 100 200 300

Q-Algo PDC(Q-Algo) OD-PDC(Q-Algo)

400 500 600 Number of Tags

700

800

900

Figure 3.9: Ratio of successful queries with the Q-Algorithm scheme.

3.5. SIMULATION SETUP AND RESULTS

86

0.34 0.33 Ratio of Collision Queries 0.32 0.31 0.3 0.29 0.28 0.27 100 QT PDC(QT) O-PDC(QT) OD-PDC(QT)

200

300

400 500 600 Number of Tags

700

800

900

Figure 3.10: Ratio of collision queries with the QT scheme.

the ratio of collision queries achieved by the PDC schemes when integrated with the EBST scheme and that of the pure EBST scheme. These ratios for the pure QT and EBST schemes are almost constant for the same reason mentioned earlier for the constant ratio of successful queries. The PDC schemes also show stable ratios over varying tag densities. Fig. 3.12 shows the ratio of collision queries achieved by the PDC schemes when integrated with the Q-Algorithm scheme and that of the pure Q-Algorithm scheme. Values for this metric has nothing to do with the performance because they are relative to the total number of queries achieved by different schemes.

Ratio of idle queries Fig. 3.13 shows the ratio of idle queries achieved by the PDC schemes when integrated with the QT scheme and that of the pure QT scheme. Fig. 3.14 shows the ratio of idle queries achieved by the PDC schemes when integrated with the EBST scheme

3.5. SIMULATION SETUP AND RESULTS

87

0.5 0.48 Ratio of Collision Queries 0.46 0.44 0.42 0.4 0.38 0.36 0.34 0.32 0.3 100 200 300 400 500 600 Number of Tags 700 800 900 EBST PDC(EBST) O-PDC(EBST) OD-PDC(EBST)

Figure 3.11: Ratio of collision queries with the EBST scheme.

0.45 0.44 Ratio of Collision Queries 0.43 0.42 0.41 0.4 0.39 0.38 0.37 0.36 100 200 300

Q-Algo PDC(Q-Algo) OD-PDC(Q-Algo)

400 500 600 Number of Tags

700

800

900

Figure 3.12: Ratio of collision queries with the Q-Algorithm scheme.

3.6. SUMMARY

88

0.445

0.44 Ratio of Idle Queries

QT PDC(QT) O-PDC(QT) OD-PDC(QT)

0.435

0.43

0.425

0.42 100

200

300

400 500 600 Number of Tags

700

800

900

Figure 3.13: Ratio of idle queries with the QT scheme.

and that of the pure EBST scheme. As explained in Chapter 2, the EBST scheme does not have any idle queries. This makes these ratios for the PDC schemes, when integrated with the EBST scheme, very low; idle collisions come only from empty clusters. When the PDC schemes are compared with the QT and the Q-Algorithm schemes, no significant changes are observed between the PDC schemes and the pure classical ones. As is the case of the ratio of collision queries, values for this metric has nothing to do with the performance because they are relative to the total number of queries achieved by different schemes.

3.6

Summary

In this chapter we introduce the power-based distance clustering scheme for tag collision resolution in RFID systems. The main idea in our approach is to divide tags

0. .26 Ratio of Idle Queries 0.14: Ratio of idle queries with the EBST scheme.2 0.05 0.28 0.02 0.08 0.24 0.06 0.16 0.04 0.03 0.15: Ratio of idle queries with the Q-Algorithm scheme.14 100 Q-Algo PDC(Q-Algo) OD-PDC(Q-Algo) 200 300 400 500 600 Number of Tags 700 800 900 Figure 3.22 0.07 Ratio of Idle Queries 0. SUMMARY 89 0.3.01 0 100 200 300 400 500 600 Number of Tags 700 800 900 EBST PDC(EBST) O-PDC(EBST) OD-PDC(EBST) Figure 3.6.18 0.

Moreover. the number of clusters has to be selected carefully. Since the number of tags in a single cluster is less than that in the whole interrogation zone. Our approach finds a balance between reducing the likelihood of collisions and reducing the number of empty clusters by finding an optimal number of clusters. our approach can be integrated with any existing anti-collision scheme and improve its performance. the likelihood of a collision is reduced.6. and read tags in each cluster separately. which causes a significant number of idle cycles. . Theoretical analysis and simulations have been presented to verify performance improvements of our approach. However. SUMMARY 90 into clusters based on their distance to the reader. as having too many clusters results in having many empty clusters.3.

57] and the maximal allowable CW signal power. . 42.5µW − 100µW [34. a single reader broadcasts the un-modulated signal (on forward link) and receives the backscattered modulated signal from the RFID tags (on backward link). in addition to attenuation.91 Chapter 4 Distributed Receiving in RFID 4.1 Introduction In RFID systems. in response to the queries. the RFID reader broadcasts un-modulated Carrier Wave (CW) signals and interrogation queries. limit the operational range of a passive tag to a few meters. would harvest its circuitry power from the CW signal incident on its antenna. anywhere between 2. In the backward-link tags communicate with the reader by reflecting and modulating the incident CW signal on their respective antennas. The reflected signal strength faces significant attenuation as it propagates twice the distance between the tag and reader. 36. hence caps the upper bound on the operational range of the semi-passive tags at some tens of meters. Each passive tag. In the forward-link. The operational power requirement of a passive tag circuitry.

Imminent RFID applications (e. In addition. to attain maximal operational performance for RFID systems. numerous solutions have been proposed. provoking even stronger advocates to tackle the . operational range [42]. These limitations place upper bounds on performance. have been proposed to enhance the operational range of RFID tags. at the protocol-level. higher gain antennas [36].g. 100]. directional antennas [42]. 116]. it will also open the door to challenges that require new initiatives. design of low-power circuits [34]. data rates and jeopardize system operational capacity. have been proposed to overcome collisions.4. the inherited centralized architecture coupled with RFID tags’ inability to sense carrier channels create collisions at the reader. both at the system-level and protocol-level. Currently. etc. item-level tagging [14] and mobility demand performance beyond these bounds. 96]. passive tags will be constrained by the backward-link channel performance. energy-aware modulation schemes [13. reducing the tag circuit power requirements by at least a factor of ten [36]. and hence. existing and related proposed solutions are based on centralized communication architectures. at the system-level. 88. item-level tagging and mobile tags (potentially numbered in thousands [14]) will also prove impossible to interrogate using the existing centralized systems. decoupling passive tags from the performance of the forward-link channel. energy-aware anti-collision protocols [79. such as reading rate [91]. Collisions are the most prominent hindrance to performance in RFID systems. However. bandwidth [36] and reading reliability [104]. INTRODUCTION 92 Furthermore. For instance.1. sophisticated anti-collision schemes [12. However. Similarly. It is anticipated that promising trends in CMOS technologies may help in mitigation of some root causes. 42. For instance. they lower reading rates. intelligent transportation [9]). therefore inherit their limitations.. etc.

Indeed. via comprehensive simulations. Parallel singulation assists in achieving higher reading rate under both dense and mobile tag environments. in a parallel and autonomous manner. and demonstrate . by overlapping RFID readers.4. there is a need for new communication models and singulation approaches for RFID systems. with the goal of boosting the operating performance of RFID systems to meet both existing and future requirements of ultra large scale RFID applications. In the proposed paradigm. using the proposed DR-RFID and existing RFID systems. namely Distributed Receiving RFID (DR-RFID) system. we also devise a parallel singulation algorithm which exploits distributed receiving and micro-zones to mitigate tag collisions. we propose the distributed receivingbased communication architecture. The micro-zones are formed by dynamically adjusting reflectivity coefficients of RFID tags. We introduce our distributed receiving paradigm in RFID systems. This model supports multipoint communication and hence eliminates the redundant interrogation of same set of tags. INTRODUCTION 93 aforementioned limitations. hence facilitating high bandwidth and data rates. from each micro-zone. We evaluate the proposed system. We also present the novel notion of micro-interrogation zones. Furthermore. the task of collecting the modulated backscattered signals is dispersed to spatially distributed entities within the reader interrogation zone. Our main contributions in this chapter are as follows.1. The focus of our work in this chapter is to lessen the limitations and alleviate the bottlenecks imposed by centralized architectures. To this end. Micro-zones spatially isolate tags within the reader’s interrogation zone. The results validate the proposed communication architecture and singulation algorithm. The algorithm singulate tags. implementation on the ns-2 simulator [7].

6 provides detailed information about the simulation methodology and performance analysis.t) = λ 4π PERIP . Section 4.4 explains the concept of micro-zones and the algorithm formalizing its formation. Section 4. usability and enhanced scalability under diverse distribution of tags. MOTIVATION 94 substantial improvement in system performance.1) . 2.2 Motivation Communication in RFID follows a centralized-control model. in the case of passive tags.4.Gr . which hinder overall system performance.5 explains the parallel singulation algorithm and analyzes its running complexity. we elaborate on the limitations inherently imposed by existing RFID communication architectures. usually termed as a master-slave.2. the operating distance between a reader and a passive tag is given by d(r. The rest of the chapter is organized as follows.2 elaborates on the motivation behind the proposed system.7 outlines the work-in-progress on the proposed system.Gt . it has significant drawbacks and limitations. Although the centralized approach has been widely employed in RFID systems. Section 4. In this section. Reading Range: The least available power for tag circuitry and the strength of backscattered signal are two key requirements in determining the reading range for passive and semi-passive tags. In addition. Section 4. Section 4. Analytically. as is depicted in Fig.4. Section 4.8 summarizes our work. other major parameters include reader operating power PERIP and tag mismatch factor τ . respectively.3 explains the distributed receivingbased architecture and details its components.τ Pmin (4. Section 4. 4.

First. a tag with a modest power requirement Pmin of 16µW [57] and a mismatch factor τ of 0. Lower tags’ collisions and the ability of the RFID reader to handle multiple tags at one time imply high reading rates. However. MOTIVATION 95 As such. such as item-level tagging for the supply chain and retail industry. causes immense data collisions at the reader [42. The reading range of the semi-passive tag is a function of the strength of the modulated reflected signal λ2 . are restricted only to limited applications. the broadcast nature of the backscattered signal from the tags and reader inability to process them . it is then constrained by the integrity of the received signal (4.2.2) it is evident that the tag reading range can be increased by adjusting following variables. by improving the reader’s antenna sensitivity.Gr 2 . generally involve dense and mobile passive tag distributions. is 15 meters. However. Prcvd Second.1) and (4. Reading Rate: RFID applications. In doing so the tag will be operative at a longer distance however. and as a consequence. ( PERIP ) by using arrays of antennas.t) )4 Prcvd = (4. higher gain antennas [36] or directional antennas [42]..5.Gt (4π)2 (d(r. 96]. Such environment constitutes potentially thousands of mobile tag within the interrogation zone of a single reader.6) operating at 900M Hz. following North American regulations.PERIP . such solutions. i. Reading rate is defined as the number of tags that can be singulated per time unit.4. under the assumption of a dipole antenna (Gr = Gt = 1.e. due to their high associated monetary costs. by decreasing the power requirements for the tag circuitry Pmin .2) From (4.2) on by the backward-link. the maximum theoretical operating distance of the passive tag.

e. 4. considering its antenna orientation.2. These reading rates (which may be lower in practice) are not adequate for typical item-level applications involving thousands of moving tags.e. φr ) and the tag (θt . The power received by the RFID tag.3) The tag orientation proportionally affects the readability and in some cases. MOTIVATION 96 in parallel affect the reading rates of current RFID systems. −10dBm (100µW ) [36] and −26dBm (2. The effect of IC power requirements on the tag reading range is illustrated in Fig. φt ) [104]. Reading Reliability: Reading reliability is the probability a tag is readable regardless of the reader and tag antennas’ orientations..1. from top to bottom. Due to the current centralized communication architecture. Furthermore. i.4. The three horizontal lines. Having high reading reliability is of utmost importance for certain applications. For instance. Forward-link: Emerging CMOS fabrication technology promises to significantly reduce the tag’s IC power requirements. are the minimum power (in dBm) requirement of existing tags.. φr ).t) )2 (4.g. theoretically.Gt (θt .Gr (θr . is tag Prcvd = PERIP . φt ). perpendicular orientation between the tag and the reader antenna. which thins out the probability of having decent strength signal for different antenna’s orientations and hence throttles readability. existing RFID reader can only support reading rates of up to 700 tags/second [91]. by a factor of ten.5µW ) [34] and anticipated . can result in zero readability. the CW signal is collected and processed by the emitting reader. self-checkout counter at a department store – a miss-read tag is lost revenue. For instance. at least.λ2 (4λd(r. azimuth and latitude angles of the reader (θr . limited available bandwidth and low data rates further reduce the reading rates.

the operating range of the passive tags would be backward-link constraint. Low IC power translates into longer reading ranges. All of the aforementioned limitations demands that new communication models and singulation approaches be sought out.1: Effect of tag circuitry power requirement on the backscattering power and operating range.5uW) Tag IC power of -36dBm Power (dBm) Reader min. MOTIVATION 97 0 -20 -40 -60 -80 -100 -120 0 Tag IC power =100uW Passive tag Reader Tag IC power -26dBm (2. . with receiving sensitivity of −85dBm (lowest horizontal line).25µW ).2. may be able to handle the reflected signal of an existing tag. In other words. However. it is surely incapable of handling the backscattered signal from the anticipated lowpower passive tags as the strength of backscattered signal falls well below (less than −100dBm) the reader’s receiving sensitivity. This requires new communication architectures and protocols be developed which can adapt to new challenges and constraints. receive power (-85dBm) 10 20 30 40 50 60 Distance (meters) Figure 4. An existing high-end RFID reader. tags of −36dBm (0.4.

bandwidth. Such centralized architecture constitutes the master-slave architecture depicted in Fig. It also increases tags reading range as the backscattered signal is multi-hop by the fielder hence sustaining the signal integrity. The basic idea behind the proposed architecture is to disintegrate the transmission and the receiving components of the existing RFID readers into spatially distributed entities within the concerned RFID interrogation zone. is introduced in the system. reading rate.3 DR-RFID: Distributed Receiving System for RFID In current RFID systems. is managed by a fielder and is created by dynamically adjusting the fielder’s signal receiving sensitivity and the signal reflectivity coefficient of the tags within the prescribed area. Each of the clusters.3. it broadcasts the CW signal and interrogation queries. the reader acts both as an emanation and an assimilation point. reading range. . As an emanation point. it receives and processes the modulated backscattered signals from the tags. Distribution of these tasks empowers the reader to tackle multiple tags in parallel at spatially isolated clusters managed by the fielders. DR-RFID: DISTRIBUTED RECEIVING SYSTEM FOR RFID 98 4.4. As an assimilation point. processing and relaying of the received data from RFID tags. named fielder. This virtually associates a subset of tags to a specific fielder. viz. As explained earlier. The main tasks of the fielder are collection. 2. capacity and data rates. it is the centralized nature of communication architecture which is the root cause of degradation in system’s performance metrics. For this purpose a new component. We propose a novel communication paradigm in RFID – distributed receiving-based communication architecture. reading reliability.4. This association aids in reducing collisions as the reflected tags’ signals are now confined within the micro-zones each of which is managed by a separate fielder. also known as micro-interrogation zones.

reader or antenna arrays. i. is replicated and routed to the concerned reader and RFID middleware without initiating a new interrogation process. However. Furthermore. the DR-RFID system can functionally accommodate both existing and modified components in multiple configurations. for tag localization.2. with the concerned tag IDs.e. which have overlapping interrogation zones.g. RFID tag data.. the autonomous nature of these micro-zones further assists in parallel tags singulation. The proposed Distributed Receiving RFID (DR-RFID) system uses multi-hop communication as depicted in Fig.g. 4. DR-RFID: DISTRIBUTED RECEIVING SYSTEM FOR RFID 99 Figure 4.2: Distributed receiving-based RFID architecture.. e. therefore dramatically increasing the reading rates and reading readability. This eliminates redundant interrogations of the same set of tags by readers. e. Purging the system from the overhead interrogations means fewer singulation attempts and thus savings to the scarce wireless resources. a product tracking application after subscribing to the fielder is notified of items.3. at the docking door in a warehouse. For interoperability with existing RFID systems.. Furthermore. This results in better product tracking and visibility without any extra equipment. with fielder physically associated to a location (micro-zone). once available at the fielder. not all configurations can harness the benefits assured by the .4.

we discuss the system components.3: Block diagram showing the typical components of fielder in the proposed DR-RFID system. DR-RFID system. As well conventional readers lack the ability to singulate in parallel hence reading rates will not be increased.3. may increase for the semi-passive and active RFID tags but will remain unchanged for the passive RFID tags. DR-RFID: DISTRIBUTED RECEIVING SYSTEM FOR RFID 100 Incoming RFID signal Outgoing wireless signal non-volatie memory Wireless Transceiver RFID Interface Power Control unit Microcontroller demodulator Power source Figure 4. Nevertheless. the operating range. reader. in a configuration involving existing readers and tags.. will enhance the support for tag mobility and ability to engage in multi-point communication. In the remainder of this section. For instance. with existing tags. . This is because an increase in the operating range entails that the passive tag is equipped with functionality to change its reflectivity coefficient.4. a configuration comprising of a reader with additional functionality. i. fielder and tag and their comparatively discrete and novel features proposed in the context of the proposed DR-RFID system. a feature lacking in existing tags.e.

The reader broadcasts the CW signal on one channel. whereas the fielder relays the received reflected signals on another low power communication channel. individually in an autonomous manner. the DRRFID system disperses some of these tasks into the spatially distributed components of the system. source of RF power. i..g. On the other hand. collisions resolution and relaying of data to middleware are all confined in a single entity — the RFID reader.3. unlike existing RFID systems.4. to cope with such an architectural change requires enabling the reader. The tag data processed can potentially be replicated and relayed by the receiving fielders for the overlapping readers. processing of backscattered signals. signal broadcast and its collection is on different channels using multiple system components. The major dispersed task includes dispensation of the backscattered signals and partial relaying of tag data. However. This requires modifications at the hardware and software levels to RFID readers. To support this requires revision in reader’s software with ability to subscribe. by each reader. receive and process tags data without initiation of the singulation process. In the proposed system however. Hence. Passive tags within the interrogation zone of multiple readers are singulated. e.1 RFID Reader In a conventional RFID system. fielders.3. to transmit and receive using different communication protocols. Zigbee or Wibree. task of tag interrogation. without execution of their singulation mechanism. In existing systems. the broadcast of the CW signal and collection of reflected signals is on the same communication channel by the same reader. the singulation process is shared by the multiple overlapping readers..e. at the hardware level. . DR-RFID: DISTRIBUTED RECEIVING SYSTEM FOR RFID 101 4.

no data is routed to the reader and the converse holds for the forward state. on receiving of the backscattered signals from the tags.2 RFID Fielder The fielder. DR-RFID: DISTRIBUTED RECEIVING SYSTEM FOR RFID 102 4. is a low-power and low-cost physical device based on off-the-shelf components. deliberately puts every fielder. is responsible for processing of the tags’ reflected signals and relaying of the processed data to the subscribed units. The randomly picked fielder. replicates and routes the data to the overlapping readers. In the silent state. memory modules. with overlapping reading zones. The fielder.3.4. The fielder adjusts its Receiving Signal Strength Index (RSSI) threshold to create the virtual micro-zone of a pre-determined circumference. . is defined based on their state. These states are set by the reader and assist in collision resolution. The micro-zone is confined within the reader’s interrogation zone and with other micro-zone occupies the entire interrogation field. 4. namely silent state or forward. the reader. except a randomly picked one. microprocessor and RFID signal receiving interface. upon collision detection. into the silent state. only after successful singulation.3. The basic idea is to restrain multiple readers. conceptually similar to a wireless sensor node. Typical components are shown in Fig.3 and includes wireless transceiver. from interrogating the same set of tags at same time. An overlapping reader not receiving any tag’s reply – the silent fielders will not forward the data to their respective readers – avoids competing for tags singulation. The functional behavior of the fielders. The fielder assists in the formation of micro-zones formation and aid the reader to resolve tags’ collisions. the novel entity of the proposed DR-RFID. Each fielder is responsible for processing of the tags that lay within its designated micro-zone. For instance.

3.3 RFID Tag The incident radio waves on the tag antenna facilitate two purposes: first. the tag of the proposed DR-RFID system always distributes significant percentage of the available power to the circuitry unit. On the contrary. split the available power equally between the two units. low strength backscattered signal. 4. significant percentage of the power is set for the rectifying unit. to harvest the DC power and second. Existing RFID tags. for both passive and semi-passive tag. The passive tag trades off the signal power between the rectifying unit and the backscattering unit.4: Block diagram showing typical components of the passive RFID tag along with highlighted components specific to the DR-RFID system.4.3. DR-RFID: DISTRIBUTED RECEIVING SYSTEM FOR RFID 103 Antenna Power distributor Rectifier DC power Signal detect Logic unit IC Modulator Reflectivity co-eff Figure 4. generally.4) without the need . This will result in lesser power available for the backscattering unit hence. In special cases. however. for the passive tag. to communicate with the reader. aided by the introduced power distributor component (see Fig. hence. 4. tag’s circuitry.

the tag in the proposed DR-RFID system has the ability to dynamically adjust its reflectivity co-efficient.4. The fielder minimizes the propagation distance of the reflected signal. does not interfere with tags from other micro-zones.4.5: Micro-zones in the DR-RFID system. F3.4 Micro-Zoning in DR-RFID System The basic idea of micro-zones. The additional components are shown in Fig. F4}: Fielders F2 F4 T T T Figure 4.4. allowing tag’s signal integrity to be valid up to the spatially closet fielder. is to cluster the tags into microzones and interrogate these clusters in an autonomous manner. of high-cost and highly sensitive readers. MICRO-ZONING IN DR-RFID SYSTEM 104 Interrogate F1 T T F3 T T R Interrogation zone Micro-zone T: tags {F1. 4. Tag adjusts the strength of the backscatter signal to make it decodable within the boundary of its micro-zone. illustrated in Fig.5. 4. This reduces the . Unlike existing RFID tags. This significantly reduces tag collisions and enhances the bandwidth of the backward-link channel. a feature not viable in existing tags. shortening of the backscattered path further facilitate in achieving longer operational range. hence. F2. therefore. 4. Furthermore.

. 4.6.4.e. in Section 4. Tags need to adjust their reflection co-efficient such that the integrity of the reflected signal will only be valid within a confined region and must not be strong enough to interfere outside this region. warehouse). assists the tag in selecting the right value for its coefficient in a way that both tags and fielders are able to form the required number of micro-zones within its interrogation zone.4.1 Tag Reflectivity Co-efficient Tag reflectivity depends on the cross-sectional area of its antenna and is modifiable by varying the antenna impedance mismatch. reducing collisions and improving tags reading rate. i. wherein the micro-zones are 100% overlapped. In the unlikely scenario. our proposed system will not degrade beyond any existing system. However.g. However. such conditions will only have minor effects on the overall performance of our proposed system. the reader. by changing either the real part Ra or the imaginary part Z for the Amplitude Shift Key (ASK) modulation or the . it is possible for the micro-zones to have overlapping interrogation regions. the tags within these overlapping regions will interfere will tags of more than one micro-zone. In such a scenario.. when evaluating the performance of the proposed DR-RFID and existing RFID systems. due to the nature of a complex indoor environment (e.4. To this end. The micro-zones are formed with support from fielders and tags. MICRO-ZONING IN DR-RFID SYSTEM 105 number of tags that needs to be interrogated at one time. Fielders simply need to adjust their RSSI threshold such that the signal-tonoise ratio of the spatially close (within the pre-determined area) tags surpasses the tags which are spatially located afar. wherein the RF signal suffer from multi-path phenomenon. mainly based on the fielder placement. We will address this issue further.

Gt |Za + Zc |2 σ = Ae .4.Gt 4π (4.t) . r (4. the reader transmitted power PERIP and its radar cross section area σ.σ 2 4πD(r. Using the Friis equation we get PBS = PERIP .t) (4.6-a.6) Dynamic adjustments to the tag’s reflectivity coefficient require to variant the strength .4) The radar cross section σ of the tag is defined as 2 4Ra .Gt . 4.4. Phase Shift Key (PSK) modulation. respectively. Zc = R + jX and Ae is the effective area of tag and is r Ae = r λ2 .6: Tag schematic showing the antenna with the added parallel resistor along with the circuitry imaginary and real part. The conventional tag with exposes real part and the imaginary part of the circuit and antenna is shown in Fig. The tag backscattered power PBS is a function of its distance to the reader d(r. MICRO-ZONING IN DR-RFID SYSTEM 106 Ra Tag R Ra 1 jX Z R V jX Ra 2 Tag Circuitary Antenna Tag Circuitary Antenna (a) Current tag schematic (b) Revised tag schematic Figure 4.5) where Za = Ra .

as depicted in Fig. at distance d(t. The coefficient is adjusted before the execution of singulation algorithm and is made to till is explicitly reset. new a 4π |Ra ||R + jX|2 (4. 4.σnew 4π. by adjusting the value of the variable resistor Ra2 .7) in (4.f ) 2 (4.10).7) Introducing variable parallel resistor gives flexibility to dynamically adjust the antenna coefficient as needed and is controlled by the tag’s data signal. For instance. we can use a variable resistor Ra2 in parallel with the antenna resistor Ra1 . Fig. the strength of the reflected signal can be modified.Ra2 Ra1 + Ra2 (4. after substitution of (4. in accordance with the reader-tag and tag-fielder distance.f ) is Sf = PBS .9) Substituting PBS and σnew from (4.4.G2 4.D(t. MICRO-ZONING IN DR-RFID SYSTEM 107 of the backscattered signal PBS .Gt 2 .10).4) and (4.Gf . Therefore. such that the overall antenna resistance is new Ra = Ra1 .Gf .4.Rnew . distance to the fielder and radar cross section.8) we get new PERIP .6) and (4. distance to the reader.t) .6-b.Ra t . frequency of .7-a shows the strength of the backscattered signal.D(t. with PERIP 4W (North American standard). 4.10) From (4. with antenna gain of Gf . The modified radar cross-section σnew . 2 2 new (4π)3 .D(r.8) The power density of the reflected signal received by the fielder. we infer that the strength of the reflected signal depends on three important parameters.Gr .λ2 . seen at the fielder (using 4.f ) |Ra ||R + jX|2 Sf = (4. To achieve this.5) is given by σnew = λ2 4.

tags marked as T will adjust their coefficients. where the reader’s interrogation range is divided into clusters d. Then only the tags from cluster d .4. Now only the tags from cluster d . d + step and d + 2 ∗ step from the reader. The stepping distance and thus the number of clusters can be optimized for a given tag distribution [12]. only tags from same cluster will adjust their reflectivity co-efficient.t) . In other words. 4. At any time. have been adjusted the reader then increases its range to d . The smaller the distance to the reader d(r. The adjustment process continues until the reader reaches its maximum interrogation range. Gt and Gr ) of 1. The tag-fielder distance is the Euclidean distance between the boundary intersections of a variable interrogation range and one of the fielders’ created micro-zone.6 and a fixed distance to fielder. The tag reflectivity co-efficient assignment algorithm is shown in Algorithm 3. the intersection points for each fielder’s micro-zone fk with the two . The power-level of the reader is varied in order to calculate the reader-tag distance.. labeled as T will act to the reader requests. D(t. After these tags. the tag adjusts its reflectivity only once until it is reset.e.4. antenna gain (Gf . The clustering concept is shown in Fig. d and d at respective distance of d..4. marked as T . For instance. i.e.2 Dynamic Assignment of Reflectivity Index To calculate the required tag reflectivity co-efficient needs reader-tag and tag-fielder distance. the lower the reflectivity coefficient should be. MICRO-ZONING IN DR-RFID SYSTEM 108 900MHz. Change in the reader’s antenna power level clusters the tags based on their distances from the reader [10].i. assume the reader has already adjusted the tags from cluster d. 4.f ) of 5 meters.7-b. In the first step.

7: Tag reflectivity coefficient and assignment.1 coeff: 0.4. MICRO-ZONING IN DR-RFID SYSTEM 109 -40 -45 -50 Power (dBm) -55 -60 -65 -70 -75 -80 -85 1 2 3 4 5 6 coeff: 0.3 coeff: 0. .9 7 8 9 10 Reader-tags distance (meters) (a) Effect of reflectivity co-efficient on received signal strength (dBm) as function of reader-to-tag distance (meters) (b) Tag clustering Figure 4.7 coeff: 0.4.5 coeff: 0.

t) in (4. (δ is the stepping parameter) is calculated (lines 4-6). amongst all fielders (line 7). Output: Adjusted tags’ reflectivity 2 repeat 3 // for the ith cluster.9) for a pre-defined Sf and is subsequently broadcast (lines 8-9).5.5 Parallel Singulation in DR-RFID Existing anti-collision schemes are based on certain assumptions about the underlying system.t) ← max ( ∀ fk {F} | max ( ∀ p {P }fk | fk p )). In other words. Afterwards. The set of intersection points Pk are determined for every fielder k which overlaps with the reader’s current interrogation boundary. First. 6 end 7 d(f. 8 Calculate σnew by substituting d(f. using d and (d . The new reflection coefficient σnew is calculated by substituting the tag-fielder distance in (4. 9 Send (Broadcast. 11 until Maximum interrogation range is reached . between the fielder fk and its intersection points p. The tag-fielder distance is the maximum Euclidian distance fk p. the reader in a sequential manner interrogates tags within the interrogation zone. 10 Increment power-level by δ.4. 1 Set Reflectivity(F) Input: F: set of fielder. 4 foreach fielder fk in F do 5 {P }k ← 4-intersection points with fk . d and (d − δ).δ). it is only after successful singulation of a tag that . PARALLEL SINGULATION IN DR-RFID 110 boundaries of the current cluster. σnew ).9). the reader power-level is incremented and the aforementioned steps are repeated until the maximum interrogation is reached. Algorithm 3: The reflectivity assignment algorithm. 4.

. To determine the effectiveness of the parallel singulation approach. yet any other anti-collision algorithm can be used. PARALLEL SINGULATION IN DR-RFID 111 the subsequent singulation takes place for the remaining tags. Although we use simple binary search tree algorithm [42] [95] and the EPC Q algorithm [40] for deterministic and probabilistic. explain the deterministic and probabilistic parallel singulation algorithms and analyze their running complexities. . respectively however. Second. all tags lying within the interrogation zone will cause collisions at the reader. since only the intra-zone collisions are possible. the sequential singulation is not the only option. Assuming the disk model.. To this end. the reader interrogation area is completely covered by the clusters. for each micro-zone. the reader only knows and is concerned with the intra-zone collisions. in parallel and autonomous manner.5. The interrogation area is divided into n micro-zones f1 .. Afn . fn of area Af1 . Finally. 4. and most importantly. we describe the system model. the interrogation zones are divided into multiple non-interfering micro-zones. due to the micro-zones. each micro-zone can be interrogated independently. To elaborate... Second. The fielder selectively relays the received tags’ serial numbers to the reader. DR-RFID systems have the following important characteristics. we propose a parallel singulation algorithm where multiple instances of an existing anti-collision scheme are executed. A fielder manages each micro-zone.1 System Model Consider an interrogation area AR enclosed by a single RFID reader R with maximum interrogation range of Rr . In this section.. respectively.. First. . On the contrary. we use algorithms from the deterministic and probabilistic class of anticollision schemes.5.4.

.2 The Deterministic Algorithm The main goal of RFID anti-collision algorithms is to increase the singulation rate by reducing the tags’ collisions. tk .j) intended for the micro-zone fj . where k >> n. Initialization: The algorithm begins with housekeeping tasks (lines 2-7) which includes setting all fielders states to FORWARD. The parallel singulation algorithm is composed of four logical components: initialization. Second. broadcasting the RESET and the initial serial request (REQA) command. by making use of reader and fielders synchronization... First. there are k tags within the interrogation area AR .5.. Afn ).4. Each micro-zone i receives responses only from the tags within its area Afi . Assuming a uniform distribution.. The reader during each interrogation cycle i broadcasts singulation requests Reqa(i. For an interrogation cycle. execution of an instance of the deterministic singulation algorithm. communication between the RFID components and communication timeout scenarios. The reader. after resetting tags internal state (RESET). for each micro-zone in an autonomous manner. broadcasts the initial serial request (REQA) . 4. In the case of deterministic parallel singulation algorithm. i. Fielders in the forward state always relay tags’ serial numbers to the reader whereas converse holds in silent state. concurrent singulation. AR ≤ (Af1 . the maximum number of singulation requests transmitted by the reader equals the number of micro-zone...e. Pseudocode for the parallel singulation algorithm is shown in Algorithm 4. this is achieved in two ways.5. In this section. PARALLEL SINGULATION IN DR-RFID 112 i..e. we use the conventional binary search tree as the candidate deterministic algorithm. t1 .

Queuek ← bn−1 .. foreach fielder fk in F do // Let j be the MSB collision bit.k) ← Pop from Queuek . 1 Deterministic Parallel Reading (F) Input: F: set of fielder. if (Collision)fk then REQA(i..k) is Unique for ith cycle then Send(fk . if the Queuek is empty. REQA(i. 2 foreach fielder fk in F do 3 sk = FORWARD.k) ← bn−1 . 6 Adjust the tags reflectivity index using Algorithm 3. REQA(i.bj = 1bj−1 = 1. 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 // Singulation process for the ith interrogation cycle.k) = NULL then Send(fk . Perform tag’s read/write operations.. RESET). end else if REQA(i..5.k) ). repeat wait until REQAtimeout .. end else if (N oCollision)fk then // Tag is successfully singulated.k) ← Pop from Queuek . end end until ∃k|sk = FORWARD . SILEN T ). sk = SILENT. end end foreach fielders fk in F do if REQA(i. if any.4. Output: Singulated tags.bj = 0bj−1 = 1.. 7 Send (broadcast.. PARALLEL SINGULATION IN DR-RFID 113 Algorithm 4: Deterministic parallel singulation algorithm.k) = NULL and REQA(i.b0 = 1. REQA0xf f f f f f f f ).. REQA(i.b0 = 1. end else if (N oReply)fk then // NULL is returned. 4 end 5 Send (broadcast. 25 26 27 28 29 30 31 32 33 34 35 36 37 .

after a successful tag singulation (line 25). Parallelism: For each micro-zone. Such a situation potentially inhibits the micro-zones singulation tree with unwanted tree nodes and.j) . of their successive interrogation requests. The broadcast request for the k th micro-zone is then used for the subsequent interrogation cycle. If REQA(i. First. To alleviate such a data structure corruption. This facilitates in effective filtering (listed in lines 28-36) of the irrelevant tags’ replies.k) ≤ REQA(i. PARALLEL SINGULATION IN DR-RFID 114 command using the highest possible 32-bit serial number as its parameter. when no response from the tags was received for the last request (line 21) and second. the tag listens and in some cases may respond to an irrelevant request. In a collision scenario.e. the reader maintains an independent instance of the singulation tree and its associated data structures. is not intended for its micro-zone. an unpredictable outcome. i.k) is formed by replacing the most significant collision bit by 0 followed by trailing 1’s. i. between the tags of the k th micro-zone.. a new broadcast request REQA(i. singulation tree and queue.j) then a subset of tags from the cluster a may response even if their serial number is greater than REQA(i..j) .4. Due to broadcasting nature. The most significant collision bit is also replaced by 1. This initial interrogation request serves as the root of the singulation tree. Reader and Fielder Communication: The tag replies to a reader request if and only if the requested serial number is greater than its own. the fielders are kept informed. by the reader.e. .k) and REQA(i.5. consider two micro-zone a and b with their respective request as REQA(i. For instance. An instance of the anti-collision algorithm (listed in lines 15-25) is iteratively executed during each micro-zone’s interrogation cycle. followed by trailing 1’s. hence. and is queued on Queuek (line 17) which is de-queued under two scenarios.

seq=0) Time out (REQA-1. upon time expiration. seq=1) Tag Data REQA-1 (a) Redundant cycle Reader Fielder (b) Filtering Tag (REQA-0. The non-legitimate case is when the reader. The REQAtimeout time (line 11) is the duration of an interrogation cycle. by neglecting such cases. the maximum time the reader waits for a tag response.. broadcasts the queued request (line 21). shortly after that. The handling of the delayed transmission is critical as. 4. queues the transmitted request and continues with the singulation process based on the delayed responses. i. PARALLEL SINGULATION IN DR-RFID 115 Reader REQA-0 Time out Fielder Tag Reader Fielder Tag (REQA-0. However. In such a case. Timeout: Various scenarios trigger timeouts and retransmissions and are outlined in Fig.8. 4. seq=0) REQA-1 Time out (REQA-1.8: Timeout scenarios in parallel singulation. The legitimate case happens when either the fielder filters out all of the tags’ . we may potentially overlook some tags. seq=1) (c) No tag data Figure 4. the reader cancels the current interrogation cycle.4. it receives the delayed tag response (Fig.e. The timeout can be non-legitimate or legitimate.8-a).5.

The tags reflectivity index is then adjusted . The tags estimation is use to terminate the algorithm (line 38). no tag matches with the request (Fig.8-c). 2Q − 1). it will immediately transmit a 16 bit random number (RN16). The parallel singulation algorithm is composed of two logical components: initialization and updating the value of Q parameter. In case of collision.5.3 The Probabilistic Algorithm The EPC Q-algorithm [40]. 4. i.2).5.8-b) or when an empty request exists. Pseudocode for the probabilistic parallel singulation algorithm is shown in Algorithm 5. When a tag receives a Query.e.e. in terms of idle. These tasks start off with setting the initial value of Q (4.. i. The main objective of the probabilistic parallel algorithm is to increase the singulation rate by reducing the tags’ collisions. If a tag stores a 0 in its slot counter. 4. the probabilistic class of anti-collision algorithm. 15). in case of conventional Q-algorithm is also used to optimally set the Q value [72]. This is achieved by determining the value for Q parameter that will results into best performance. however. it chooses a random number in the range (0..0) and C (0. over all the micro-zones. broadcasting the RESET to all tags and estimating the tags cardinality. PARALLEL SINGULATION IN DR-RFID 116 responses (Fig.4. the reader will adjust the Q parameter and will resent the query. Both scenarios lead to a graceful timeout followed by broadcasting of the successive interrogation request. collision and successful slots. where Q is in the range (0. uses the Q parameter which determine the number of slots in the Query Round. Initialization: The algorithm begins with housekeeping tasks (lines 2-10). and the value is stored in the slot counter of the tag. 4.

7 foreach fielder fk in F do 8 Et [k] ← Determine tag estimation for each micro-zone of fk .0.0. end Q ← Q∗ . Q). Idx ← Idx + 1. 1 Q-algorithm Parallel Reading (F) Input: F: set of fielder.0 do P=-1. Q∗ ← Q∗ − C. end Q∗ =15.0.5.4. 6 Adjust the tags reflectivity index using Algorithm 3. if S[k] > P then P ← S[k]. 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 repeat wait for response Q∗ =1. RESET). Output: Singulated tags. 9 end 10 Send (broadcast. 4 Send (broadcast. until all tags are read .2. Q ). foreach fielders fk in F do Et [k] S[k] ← 2Q∗ ∗(1−1/2Q∗ )Et [k] . 3 C=0. 5 Et ← Tags estimation within the reader. PARALLEL SINGULATION IN DR-RFID 117 Algorithm 5: Probabilistic (Q-algorithm) parallel singulation algorithm. 2 Q=4. Idx=0. while Q∗ ≤ 15. end Idx ← Idx − 1. Send (broadcast. end end P [Idx] ← P . . Q∗ ← Q∗ + C. while Idx ≥ 0 do if P [Idx − 1] < P [Idx] and P [Idx] ≥ 1 then break.

respectively. however. 2. Due to the non-uniform nature of tags distribution. if number of tags is small with large Q value. Adjusting the Q value for a micro-zone may have negative effect on the rest of micro-zones hence. increasing the reading rates.5. if the number of tags are large but with small Q value. Due to the nature of Q algorithm. Generally speaking. amongst all micro-zone. (lines 28 − 36)is used for next query (line 37). Updating Q: In the conventional Q-algorithm. However. The Q with maximum ratio of successful slots. Similarly. Henceforth. it will result into idle slots. using similar method as is used in line 5. calculating the ratio of successful slots for each micro-zone (line 19).4. incrementing each iteration by C (line 25). between 1 and 15 (line 16). it is possible that some micro-zone significantly will have more tags than others. to avoid the oscillations between large and small value of Q. such an update approach can not be applied directly in the context of micro-zone. in case of collision and idle slot.11. . The flow chart of the Q value updates is illustrated in Fig. Finally. the Q value is either incremented and decremented. a rough estimate of tags cardinality. Due to the nature of micro-zones. Afterwards. which has been observed to be equal to number of tags [72]. one a time for each fielder. might overshadow the benefits of parallel singulation. and directly to minimize the idle and collision slots.0. by a constant C. unlike the deterministic algorithm (Algorithm 4). PARALLEL SINGULATION IN DR-RFID 118 using the Algorithm 3. the proposed algorithm iterates over all the Q values. only intra-zone collisions are now possible therefore. it is important to quickly reach the optimal value. only single query is broadcast for all micro-zones. the initial query request is sent using Q = 4. To fix this. it will results into collision slots.

it suffices to find an estimate of minimum. In the parallel singulation algorithm..fk .E[tk ] be the estimated number of tags within the micro-zones f1 . AR E[ti ] = 1 (4.. f (E[tk ])] (4. i.. amongst all micro-zones.. in general f (n) = c n − 1. the singulation process takes. f or any given micro − zone i.13) The scheme in reference [95] is utilized in this chapter with f (n) = 2 n − 1 . f (E[t2 ])... average and worst case scenarios. Overall number of cycles is then the maximum number of cycles. PARALLEL SINGULATION IN DR-RFID 119 4. where c is a constant and n is total number of tags..5. maximum and average of the number of tags in the micro-zones. of the proposed deterministic parallel singulation algorithm. Best Case: The best case scenario is when all the tags are evenly distributed amongst the micro-zones Ai ∗ k .e.. . we analyze the running complexity.5. Let f (k) be the number of cycles required to read a set of k tags. (4.11) f (n) = M ax [f (E[t1 ]).. based on (4.4 Delay Analysis of Deterministic Algorithm In this section. Let E[t1 ]. the reader interrogates the micro-zones’ tags in parallel. best and worst singulation delay. respectively. i. For the binary tree protocol 1 .e.12) To determine the best...12)..4. average.

each micro-zone has equal probability to singulate the overlapping tags. i. sequential singulation algorithms..14) This corresponds to the case of the existing approaches. i.5.16) Here O is the count of the micro-zone overlapping regions and Pi is the probability with which the micro-zone fi singulate the overlapping tags. In a typical grid-based deployment (Fig.9). may overlap. assuming two overlapping micro-zones. the fielder’s zone is at-most .e. In other words. is counted by than one micro-zone. (∀i|i = j) (4. That is.e.5 as each micro-zone has equal probability of tags’ singulations within the overlapping regions. 4. Rc1 = Rc2 = Rc . PARALLEL SINGULATION IN DR-RFID 120 Worst Case: The worst case scenario is when all the tags are within a single microzone i E[ti ] = k and tj = 0. A common value of Pi . is formulated using simple geometry as 2 Aoverlap = 2Rc cos−1 d 1 2 − d 4Rc − d2 2Rc 2 (4.4.. an estimate of the number of tags is N Aoverlap 1− Pi AR O E[ti ] = (4.15) and therefore. Hence. The area covered by an overlapping region of the two micro-zones Aoverlap . Average Case: The micro-zone. there is a possibility that a tag may belong to multiple micro-zones and hence. is 0. since each entity has at-most four neighbors. in an attempt to cover the reader’s interrogation zone. each with equal radius. the parallel singulation algorithm cannot perform any worse than existing algorithms.

SIMULATION SETUP AND RESULTS 121 overlapped by four other zones. for the Q algorithm. O = 4. 4. However.. reading rate. Lets assume the number of tags is n and the frame size.6 Simulation Setup and Results In this section. as the accumulative number of tags can not be smaller than even the largest tags residing micro-zone. then the probability of idle (Pi ). For the Q-algorithm. amongst others. probability of successful (Ps ) and probability of collision (Pc ) is [22] Pi = 1 1− Q 2 n (4.4. we analyze and compare the performance of the proposed DR-RFID system as opposed to conventional RFID systems.17) (4.5 Delay Analysis of Probabilistic Algorithm In this section.18) (4. is 2Q . the running complexity can be defined in terms of collision slots probability. 4. the worst case reading rate is defined by the micro-zone which have the largest ratio of collision slots.e. i. . reading reliability and mobility as performance metrics.19) Pc = 1 − P i − Ps 1 n Ps = Q 1 − Q 2 2 n−1 In case of the parallel variation of Q-algorithm. We use reading range. the computational complexity of the parallel variation cannot be worst than the original Q-algorithm. We also compare the performance of the proposed DR-RFID system with a multi-reader configuration. successful slots probability and idle slots probability. therefore.5.6. we analyze the running complexity of the proposed probabilistic parallel singulation algorithm.

The performance metrics are averaged over twenty different experiments of different topologies.4. the reader’s power control. single-hop communication model (backscattering modulation) between the tag and the reader. mimicking . 4 and 16.1. reading reliability and impact of tags mobility. SIMULATION SETUP AND RESULTS 122 Table 4.0µW 2. as configurations of 2. the simulations are performed using the following setting and parameters as shown in Table 4. multi-channel interfaces (RFID and IEEE 802.15. interrogation range Reader power level Tag1 min. In the setup. The fielders are deployed in grid fashion.5µW We have extended the ns-2 network simulator to support both existing RFID and the proposed DR-RFID system. tags are uniformly distributed in with single reader located at the center of the grid. and fielder. Simulations are made to run until all the tags are successfully identified. reader.4) and non-EPC singulation protocols from the literature. The major extension involves modification of the underlying simulator architecture to support the basic RFID functionality. We evaluate the system for its reading range. Minor changes to the simulator include inheriting of the network node to serve as an RFID tag. the EPC Class1 Gen2 MAC protocol. etc. Unless otherwise specified. IC power Tag2 min.1: Simulation parameters Parameter Grid size Max. covering the reader’s interrogation zone with minimum possible overlapping. We also compare the system with multiple readers. IC power [34] 4. reading rate.1 Simulation Setup Value 20m2 15m 4W 1.6.6.

4. .9: Grid-based fielders’ deployment scheme. SIMULATION SETUP AND RESULTS 123 the DR-RFID approach.6. Y Y C1 C1 X X C2 R R C2 C3 C4 (a) Two Y C1 C2 C9 (b) Four C10 C3 C4 R C11 C12 X C13 C14 C5 C6 C7 C8 C15 C16 (c) Sixteen Figure 4.

e. 4. between the reader and tags. The reading range of the two passive tags (tag1 and tag2). fielders.10-c. A longer reading range means fewer readers to cover an area and longer tagged object visibility thus lower equipments manageability and monetary costs.4. The horizontal dashed line in the graph shows the maximum possible reading range when using a rectifying module with an efficiency of 35%. is shown in Fig. i..1. for the direct (existing approach) and 1-hop (1-fielder per micro-zone configuration) system. The reading range of passive tags depends on several factors including minimum circuitry power requirement.. For instance. to transmit the data between the tag and the reader. using 1-hop increases the reading range by a factor of 5. efficiency of the rectifying unit and strength of backscattering signal. 4.6.e. respectively. Using more than 1-hop can potentially increase the reading range of the semi-passive tags. by using the DR-RFID system. see Table 4.2 Simulation Results Reading Range Existing RFID systems use direct communication link between the tag and reader. the reading range for the semi-passive tag can be increased by an order of magnitude. The proposed system increases the passive tag’s reading . for the 4W case. In the DR-RFID system. the multi-hop. The multi-hop link involves one or more wireless-hops. This limits the reading range for both passive and semi-passive tags. respectively. SIMULATION SETUP AND RESULTS 124 4.10-b and Fig. As expected. significantly increases the operating range of the semipassive tags. relaying of data through fielders.6. Fig. 4.10-a shows the reading range of a semi-passive for the reader’s power level of 1W and 4W. i. The emitting power of 4W and 1W is the maximum allowable power that a reader may transmit according to the North American and European regulations.

04 0.07 Ratio of backscattering power 0.05 0.01 0. IC of 1.03 0.03 0.09 (c) Passive tag (min. IC 2.5uW) 0.01 0.0uW) 0. SIMULATION SETUP AND RESULTS 125 -35 -40 -45 -50 Power (dBm) -55 -60 -65 -70 -75 -80 -85 0 100 200 300 400 direct (1W) direct (4W) 1-hops (1W) 1-hops (4W) 500 600 700 800 Reading range (meters) (a) Semi-passive tag 21 20 Reading range (meters) 19 18 17 16 15 14 13 12 0.02 direct (2.6.08 0.04 0.08 0.5µW ) 30 Reading range (meters) Tag’s maximum possible operating range 25 20 15 direct (1.10: Improvement in the reading range of the semi-passive and the passive tags for the DR-RFID system.5uW) 1-hop (2.0µW ) Figure 4.06 0.05 0.0uW) 1-hop (1.07 Ratio of backscattering power 0.06 0.09 Tag’s maximum possible operating range (b) Passive tag (min.02 0. .4.

represents ideal number of queries. Existing systems. for backscattering. Reading Rate The tag singulation rate is a product of the total number of interrogation queries and the individual query interval. to maintain similar range in existing systems. Using a lower ratio adversely affects integrity of the reflected signal. The total number of queries for BST and Q-algorithm. On the other hand. the passive tag evolves into being a backward-link constraint. is achievable by using a very small ratio. For instance. Diverting more power for backscattering. As is evident. in the proposed system. cannot go beyond the 21m mark. 4. tag1 has a maximum theoretical range of 29. in such a situation. the maximum possible operating range. will not have enough power to operate. First. in the case of tag2. hence. 1% of the available power. On the contrary.5m. . High reading rates imply that more tags can be singulated by a reader thus supporting item-level tagging. the proposed DR-RFID system achieves the maximal operating range. using the conventional. 7% of the available power must be available for backscattering. at the cost of lesser power for circuitry. The Ideal scheme. the DR-RFID parallel singulation approach. is not helpful either as the tag. in case of BST. Ideal scheme is not devised for the Q-algorithm as it is a probabilistic approach. no collisions.6. low-power circuitry and second.11. that would be required to interrogate a given number of tags. For low-power circuits. SIMULATION SETUP AND RESULTS 126 range under two scenarios. four micro-zones and sixteen microzones version is shown in Fig. setting of the significant available power portion for the circuitry. using up to 9% of available power available for backscattering. For instance.4.

6. using conventional RFID and DR-RFID systems.11: Comparison of number of queries for BST and Q-algorithm. .4. SIMULATION SETUP AND RESULTS 127 4000 3500 Number of Queries 3000 2500 2000 1500 1000 500 0 Q-algo (optimal) 4-microzone 16-microzone Ideal 100 200 300 400 500 600 700 800 900 Number of Tags (a) Q-Algorithm 3000 2500 Number of Queries 2000 1500 1000 500 0 100 200 300 400 500 600 700 800 900 Number of Tags BST 4-microzone 16-microzone (b) BST Figure 4.

the collision queries for 100 tags for the conventional BST and DR-RFID variation of the BST are 99 and 10. The readability ratio of a tag. Note that 100% readability is achievable at short distances. the integrity of the backscattered signal is to be valid only .. a tag mounted on a steel item.4. relative to overall interrogation area.12. in the conventional (directhop) and the proposed (1-hop) system. the majority of available power can be used for tag operations. at long distances. Furthermore. the proposed approach is scalable as the number of queries increases gracefully with an increase in number of tags. at various distance from the reader is shown in Fig. is almost three times higher than the conventional system. for the DR-RFID system. This is because the number of collision queries.13. 4. does not increase significantly with an increase in tags. Reading Reliability Reading reliability is the probability a tag is readable regardless of the reader and tag antennas’ orientations. SIMULATION SETUP AND RESULTS 128 for both deterministic and probabilistic algorithm. whereas for 800 tags are 799 and 63. is shown in Fig. As well. Each point in the graph is an average of successful tag read..6. using orientations between 10◦ and 180◦ . e.g. Therefore. The performance of DR-RFID will get effected once the number of tags within each micro-zone exceed significantly large. 4. The number of collision queries. respectively.3.g. outclass the conventional RFID by many folds. for both BST and Q-algorithm. Such significant improvements are partly due to tag setting considerable portion of the available power for its circuitry. respectively. The readability of the DR-RFID system. e. 500 and beyond. This significant performance improvements is achieved as each micro-zone has very limited number of tags. using single antenna configuration. even under non-friendly surroundings.2. For instance.

12: Comparison of number of collision queries for BST and Q-algorithm.6.3 0. .1 0.35 0.25 0.4 Number of Collision Queries 0. using conventional RFID and DR-RFID systems.15 0.2 0. SIMULATION SETUP AND RESULTS 129 0.4.45 0.05 0 100 200 300 400 500 600 Number of Tags 700 800 900 Q-algo (optimal) 4-microzone 16-microzone Ideal (a) Q-Algorithm 900 800 Number of Collision Queries 700 600 500 400 300 200 100 0 100 200 300 400 500 600 Number of Tags 700 800 900 BST 4-microzone 16-microzone (b) BST Figure 4.

6 0. in a dense environment and thus immense collisions. 4. First. we have not entertained the long reading range and the multipoint paradigm of the proposed system.8 Readability ratio 0. 4. the tags in the DR-RFID system are capable of operating under diverse antenna’s orientations.1hop 10m .1hop 10 20 30 40 50 60 70 Reader’s Antenna angle (in degree) 80 0. thus meeting the challenges of futuristic applications.direct 15m . Mobility In existing RFID systems. i. e. a mobile tag may be miss-read in two scenarios. inside its micro-zone. for 100 and 600 tags is respectively shown in Fig. The reading rate of mobile tags in the conventional RFID system. The effect of mobility on tag reading. for this experiment.14-a and Fig. if it moves out of the reader’s zone before being interrogated. diverse range of power. item level tagging.. using single reader configuration for the conventional and proposed system. In other words. SIMULATION SETUP AND RESULTS 130 1 0. Both scenarios lead to miss-read tags.6.13: Reading reliability comparison.4.. being interrogated. both at high .4 15m . To be fair in comparison.direct 5m .g.direct 10m .e. Second.1hop 5m . prevent it from being singulated in time.2 0 Figure 4.14-b.

SIMULATION SETUP AND RESULTS 131 100 98 Reading percentage 96 94 92 90 88 86 84 10 direct 2-fielders 4-fielders 16-fielders 20 30 Speed (m/s) 40 50 (a) 100 mobile tags 100 90 Reading percentage 80 70 60 50 40 30 20 10 direct 2-fielders 4-fielders 16-fielders 20 30 Speed (m/s) 40 50 (b) 600 mobile tags Figure 4.6. .14: Effect of mobility on the tag reading rate.4.

on the other hand. The motive behind such an experiment is to compare the performance of multiple readers to the distributed receiving system with multiple fielders. whereas parallel singulation algorithm support fast reading rates. higher the number of micro-zone. the total cycles for the multi-standalone system are only marginally lower than the proposed system. Fewer collisions and fast reading rates assists in mobile tags singulation even at high speed and in dense environments. with multiple standalone readers and the fielders.15. in the DR-RFID system. SIMULATION SETUP AND RESULTS 132 speeds and dense setting. the tags within the overlapping region are singulated as many times as there are the overlapping readers. respectively. Furthermore. The total number of reading cycles. regardless of tags speed and density. The DR-RFID system. The minor performance difference between the two configurations is because of the micro-zones’ overlapping effect.6. is shown in Fig. can handle high-speed and dense-tag environments. the difference between two and four multi-standalone readers to that of a single distributed reader aided by 2-fielders and 4-fielders is at most 4% and 11%. using 16-fielders. equal in number and interrogation area. Multi-reader Configuration Micro-zones may be mimicked by using a set of readers. However. The micro-zones facilitate potentially fewer collisions. 4.4. the reading percentage is an order of magnitude higher than when using of 2-fielders. For instance. the better system’s ability is to tackle high speed mobility. Reading rates of over 90% is maintainable. In the case of multiple standalone readers. deployed at the same locations as the fielders. Interestingly. each tag in the overlapping . For instance. falls well under 30%.

is shown in Fig. using Eagle CADsoft. and hence their potential negative effects. zone is singulated only once.4. The fielder uses Texas Instrument MSP430F2274 [6] for the micro-controller and . 4.16.7 Work in Progress To proof the working concept of the proposed DR-RFID. WORK IN PROGRESS 133 1600 1400 1200 Total Cycles 1000 800 600 400 200 0 100 2-Readers 4-Readers 2-Fielders/1-Reader 4-Fielders/1-Reader 200 300 400 500 600 700 800 900 1000 Number of Tags Figure 4. power. transceiver. These overlapping micro-zones. can be alleviated using an optimal placement scheme for the fielders. the DR-RFID system seems an attractive and low-cost solution over the high monetary costs associated with using multiple readers.7. Nevertheless. 4. we have implemented its newly introduced components namely. hence causing idle cycles for other zones’ singulation trees. The fielder schematic. dual antenna interfaces and RFID signal de-modulator. for minimal performance gains.15: Reading cycles comparison for the multi-readers and DR-RFID. fielder and reflectivity adjusting semi-passive tags. The schematic highlights various functional components of the fielder. These components include micro-controller.

) Zigbee transceiver 4.) RFID antenna interface 2 Figure 4.) Power/debug interface 6.) RFID de-modulator 5.) 2.4GHz RF interface 3.134 1 5 4 3 4.7. WORK IN PROGRESS 6 1.16: Fielder schematic for the DR-RFID system. .) Micocontroller 2.

Using standard interface facilitate.e.. A prototype implementation of the fielder is shown in Fig. 4. i. the dense scenarios demands larger processing and data communication volumes therefore requires higher availability. Second interface is the 2. e. The fielder.4GHz chip antenna which is used by the CC2500 to transmit any IP packet to an IP-enabled device on the network or Internet. the MSP430F2274 controller utilizes only 0. i. a coin battery and the 6V adapter is also designed for the fielder. WORK IN PROGRESS 135 Zigbee based CC2500 transceiver [4] for the outgoing wireless communication with reader(s) and other IP-based network devices. First interface is the 916MHz patch antenna which is used to receive the reflected signals from the nearby tags. from Linear Technology [1]. for power and debug. in a low density environment. use the 6-pins (with 1.17-b.. the fielder can be powered using on-board battery therefore. allowing complete wireless solution. the choice of off-shelf power board. wherein the number of tags and readers are low. the 6V adapter.g. On the other hand. Depending upon the anticipated tag cardinality and interrogation rate the appropriate power source can be selected during deployment phase.4. . when idle and can be woken up due to their on-chip remote wakeup capability. To detect and de-modulate the incoming RFID signal. The LTC5507 consumes 550µA and 2µA during active and shutdown mode.7µA in the standby mode. the fielder make use of an extremely low power RF detector LTC5507 IC. 4. The micro-controller and transceiver are extremely low-power.7. The prototype implementation of the power board is shown in Fig. The micro-controller and transceiver can be put into deep sleep modes. Texas Instrument’s EZ430-rf2500 kit and Cymbet’s solar energy harvesting kit [5] becomes available. for instance. A custom power board with dual power source.e..17-a.127Mil) spy-bi-wire JTAG interface. As is evident from the schematic. the fielder has two RF interfaces. For instance.

over other OS e. For software.g. it require low memory. we have enhanced Contiki OS [38] to support RFID de-modulation using LTC5507. the fielder was able to successfully decode the received . The choice of Contiki.4.e. We have tested the operational capability of the Fielder using the Skyetek’s M9 RFID reader and Alien’s RFID tags wherein. Contiki is an open source.17: Fielder and power board prototype.7. i. TinyOS. First. fits within the available 2KB RAM and 40KB ROM of MSP430F2274 micro-controller. highly portable. WORK IN PROGRESS 136 (a) Fielder (b) battery board Figure 4. Second. multi-tasking operating system designed for low memory micro-controllers. it is IP enabled and its readily available Zigbee communication stack which is required to transfer tags information to other networked devices.. is due to following reasons..

Fielders are. backscattering and RF attenuation.8. using Eagle CADsoft. The schematic highlights various functional components of the tag. This new paradigm utilizes spatially dispersed entities.8 Summary Current RFID systems suffer many shortcomings.18. The SKY12329 is a GaAs digital attenuator with attenuation range of 31 dB and utilizes less than 100µA. power interface. the tag use Texas Instrument’s MSP430F2274 and Linear Technology LTC5507 as its micro-controller and RFID de-modulator. Currently. 4. named fielders. as a foundation for a more dynamic RFID architecture. The tag attenuates the backscatter signal using the Skywork’s SKY12329 IC [3]. backscattering and RF attenuation. The tag modulate and backscatter the signal using Analog Device’s ADG902 IC [2]. SUMMARY 137 tag signal and transfer it via IP to the host machine. The tag powers up using the fielder’s battery board. The tag schematic. namely DR-RFID. from the reader. 4. They impose a detrimental effect on performance. inherently so from adopting the centralized architecture. Similar to fielder. . low cost and low power devices that are deployed within the reader’s interrogation zone. shown in Fig. The RFID de-modulator is use to detect incoming signal. we introduce the Distributed Receiving paradigm in RFID systems.17-b.4. These components include microcontroller. we are modifying Contiki OS to enable RFID de-modulation. 4. by design. RF antenna interface. is shown in Fig. In this chapter. scalability and usability. Attenuation enable the formation of non-interfering micro-zones which significantly enhances the reading rates. respectively. to wake up the microcontroller and to decode requests. to collect the modulated signals backscattered from tags.

) Power/debug interface 3.) Antenna interface (916Mhz) 4.) Backscattering IC 5. 2 .) RF signal attenuator Figure 4.) Microcontroller 2. SUMMARY 1.18: Semi-passive tag schematic for the DR-RFID system.138 1 4 3 5 6 4.8.

. low collisions. Accordingly. adopting DR-RFID yields significant performance improvements in achieving high reading rates. longer reading range for semi-passive tags and higher bandwidth. SUMMARY 139 As such. A parallel singulation algorithm.8. in which tags with adjusted reflectivity coefficients would fall under its region.4. exploiting the localization and distributed nature of these micro-zones. each fielder would create micro-interrogation zones. autonomously interrogates each zone in parallel. conceptually excluding them from micro-zones of other fielders.

missing a tag is simply a loss of revenue. 5. some of these factors are density of tags. hence. for instance. The complexity and monetary cost associated with an RFID system strongly depend on the number of readers being deployed to interrogate tags. it has the side effect of significant interference among readers. a lower performance for . number of deployed readers. the location and number of the readers are determined in a pre-deployment manner. and overlapping among readers [42].140 Chapter 5 Coverage in RFID 5. size of the interrogation area. For an item-level object tagging application. A typical deployment scenario of a warehouse application is shown in Fig. Similar to the wireless sensor networks. While deploying too many readers may guarantee full coverage. coverage in RFID systems is a deployment problem which can be formulated as a deterministic deployment and a random deployment problem. Several factors influence coverage in RFID networks. in a self-checkout counter at a department store. In the case of deterministic deployment.1 Introduction Coverage is crucial for large scale RFID applications.1. which results in more reader-to-reader collisions and.

that comes at the cost of a large pool of RFID readers and. the number of tags within each reader’s interrogation zone determines the interrogation delay of that reader. Furthermore.5. The nature of random deployment yields overlapped interrogation zones among RFID readers. 93]. In the case of random deployment. INTRODUCTION 141 the whole system. the readers are deployed in an random manner.1. and the overall interrogation delay of the whole system is determined by the longest delay associated with a single reader. hence.1: 3D Coverage in RFID Networks. Typical random RFID deployment includes RFID enabled wireless sensor networks and RFID enabled mobile phones. 92. wherein the readers are deployed in a grid [83. Therefore. Several maximal RFID coverage schemes have been proposed in the literature. results in the previously mentioned problems of having too many readers. reducing overlapping and balancing the load of readers are essential to any coverage scheme in RFID systems. honey grid [110] or at a predetermined set of locations [104]. Figure 5. This causes significant wireless . While these schemes are able to achieve maximal coverage.

tag T 1 is exclusively covered by R1 and tag T 4 is exclusively covered by R4. T 2 and T 3) are also covered either by R1 or by R4. Figure 5. Furthermore. Furthermore. That is because all tags covered by R2 or R3 (i. An example of such a scenario is shown in Fig.1.e. Redundant readers translate into unnecessary energy consumptions due to duplicate interrogations. Moreover.. there is a significant need for light-weight methods that efficiently detect and . INTRODUCTION 142 interference in terms of reader and tag collisions without improving coverage.5. in order to extend the lifetime of the network and to improve its performance. These readers collisions severely affects the performance of the overall system [110]. dense deployment may result in having redundant readers. the overlapping readers may also simultaneously interrogate the same set of tags. A redundant reader is one that does not exclusively cover any tag. resulting into data corruption.2: An RFID network with redundant readers. However. 5. a redundant reader is one whose interrogation area is fully covered by neighboring readers. Thereby. it results into extra processing and communication overheads and does not necessarily bring an enhancement to the tags coverage.2 wherein the readers R2 and R3 are redundant.

identity of the first interrogating reader [52]. Elimination of readers is mainly based on different markings. count of the neighboring readers [55] and others.111].1. These markings include tags count of the covering readers [16]. we also enhance the overall system reading rates by achieving a near-optimal load balancing amongst the deployed readers. load imbalance amongst readers can cause tags reading delay. having less collisions among readers. Many schemes exist in the literature for redundant readers elimination in RFID networks. identity of the first singulating reader [109. Furthermore. into the tags memory. By reducing the number of readers. . In this work. but the overall system performance is improved also by reducing interference and overlapping among readers coverage regions and. On the other hand. Note that overlapping readers result into readers-to-tag collisions which may cause tags being undetectable. Our main contributions in this chapter are as follows. not only is the overall network cost reduced. These schemes. Based on these markings. hence. an algorithm decides on the redundancy of an RFID reader. INTRODUCTION 143 eliminate redundant readers.5. made by covering readers. are not light-weight as they require considerable states to be maintained by the tags and involve frequent interrogations and singulations. We present an algorithm that gives the approximate minimum number readers and their locations to cover a set of tags. however. we first minimize the number of readers than balance the load amongst readers (the opposite order can be applied depending on the application scenario). We address the deterministic deployment through the set-cover approximation approach. by minimizing the maximum delay encountered by a single reader we minimize the overall delay of the system.

. i.. and uses that ratio to predict whether that reader is redundant or not. Section 5. evaluation metrics and analyze the obtained results using multiple scenarios. Based on this observation.6 summarizes our work. tag ti can be interrogated by an RFID reader if and only if the distance between the tag and the reader is at most ri . our algorithm finds the ratio of the number of tags a particular reader covers to the number of neighboring readers it has.e. Section 5. 5. The motivation is that a reader covering less tags and one that has more neighboring readers have a higher chance of being redundant.2 System Model We consider an RFID system of n passive tags and multiple readers.3 outlines the underlying assumptions and explains the proposed deterministic readers placement algorithm.2 outline the adopted model. The remainder of this chapter is organized as follows.5. We evaluate the proposed algorithms via comprehensive simulations and we compare them with other schemes available currently in the literatures.5 presents simulated experiments. Section 5. The transmission range of a passive tag ti is modeled as a sphere of radius ri . SYSTEM MODEL 144 We address the random deployment using a light-weight greedy algorithm to detect and eliminate redundant readers in RFID networks. We assume the existence of a central middleware as depicted by the EPC Gen2 standard [40]. where our algorithm is to be executed. Section 5.4 formulates and explains the random deployment as redundant reader elimination algorithm. A reader can communicate with its neighboring readers and also some central middleware. The algorithm estimates the number of tags each reader covers and finds the number of neighboring readers it has.2. Section 5.

we use the following notations: • T = {t1 . . The problem can be defined as follows. q| is the Euclidean distance between two points p and q.93]. and readers are to be placed on particular sides of the rectangular cuboid. t2 . and z− coordinates. pi . • pi is a point representing the location of tag ti .1 illustrates this environment. . • ri is the transmission range of tag ti . . and pi .3. we first describe an algorithm that provides full coverage. In the presentation of our algorithms. • |S| is the cardinality of a set S.x. pi . Figure 5. • |p.3 RFID Readers Deployment (Deterministic Deployment) 145 In this section we present our algorithm for RFID readers deployment in rectangular cuboid.5.1 Assumptions and Definitions Tags are assumed to be placed in a 3D space modeled as a rectangular cuboid. .3.y.z are the x−. Given the number of tags and their 3D locations. find a minimum number of RFID readers and their exact locations such that all tags are covered. we show later how to accommodate fluctuations in tags locations. tn } is the set of tags. We first outlines the adopted model and its underlying assumptions. . Next. While the location of tags is assumed to be known [92. And finally we give a scheme that uses our algorithm to give load-balancing deployment strategies. RFID READERS DEPLOYMENT (DETERMINISTIC DEPLOYMENT) 5. respectively. 5. we then show how to modify the algorithm for the purpose of reducing overlapping among readers. y−.

we give a formal definition for the set covering problem as follows: An instance of the set covering problem consists of a set C of a finite number of elements and a set F whose elements are subsets of C with the condition that each element c ∈ C belongs to at least one subset f ∈ F. 5. see Fig.3. The problem is to find a minimum size set G ⊆ F such that each element in C is covered by at least one element in G [33].e. belonging to several tags. overlap. To minimize the number of readers needed to cover all tags. q| ≤ ri }). when c ∈ f . Cover(q) is the set of all tags that have q within their transmission ranges (i. if r2 − c2 ≥ 0. Let us call those circles side circles. Our algorithms are based on an approximation algorithm for the set covering problem which is known to be NP-hard [33]. And since RFID readers are to be placed only on the sides of the rectangular cuboid. readers should be placed in areas where several side circles.2 Providing Coverage In order to provide full coverage. and in order to cover a particular tag ti . the intersection of a sphere centred at the point (a. their intersection will be a circle on that plane. If a sphere intersects with a plane.5. b. f is said to cover c. For the sake of completeness.3. RFID READERS DEPLOYMENT (DETERMINISTIC DEPLOYMENT) 146 • For any point q in the 3D space. at least one reader must be placed within one of the side circles that belong to that tag. Instead . Each tag will have at most one side circle on each side. each tag must have at least one RFID reader placed within its transmission range.. 5. c) with a radius of r and the plane z = 0 is a circle centred at the point (a. 0) with a radius √ of r2 − c2 . b. Cover(q) = {ti : |pi .1 we need to find the intersection of the transmission spheres of tags and the planes representing the sides of the rectangular cuboid. For example.

Moreover. These intersection points can be considered as representatives to all possible overlapping regions (for more details on this. 1 Set Cover(C. locations of RFID readers will be limited to intersection points and.3.F) Input: C: set of tags. Output: V : set of readers. the problem becomes a discrete optimization problem. Thereby. see our earlier work in [11]).5. 4 while U = φ do 5 Find a set S ∈ F that maximizes U ∩ S. 3 V = φ. . an instance of the RFID reader deployment problem can be reduced to an instance of the set covering problem in which C = T (i. hence. which are the points where the boundaries of side circles intersect. it suffices to focus on intersection points. This greedy algorithm is known to have an approximation ratio of ln |C| + 1 and can be implemented in O(|C||F| MIN(|C|. the set of tags) and F = {Cover(q) : q is an intersection point}. RFID READERS DEPLOYMENT (DETERMINISTIC DEPLOYMENT) 147 of looking for regions where side circles overlap.e. Therefore. 8 end 9 return V . 7 V = V ∪ {S}. and for the sake of completeness.. Algorithm 6: The greedy set covering algorithm. the greedy approximation algorithm is shown in Algorithm 6 [33]. F: coverage set. |F|)) time. the greedy approximation algorithm for the set covering problem can be used to solve the RFID readers deployment problem. 2 U = C. 6 U = U − S.

{1. 2. it picks the one that covers the minimum number of already covered elements. In this subsection. 3. We can calculate the weight of subsets in a different way that gives credit to those subsets that do not cover many covered elements. another optimal solution with no overlapping is {{1. weight(S) = |U ∩ S|. 5} and F = {{1. This can be done using the following weight formula: weight(S) = |U ∩ S| − α(|S| − |U ∩ S|). which is an optimal solution for this instance of the set covering problem. {5}}. We can modify the general algorithm so that among subsets covering the maximum number of uncovered elements. 3}. 2. 0 < α ≤ 1/n. {2.3. 5}. 2. Let U be the set of uncovered elements and weight(S) be the weight of a subset S. RFID READERS DEPLOYMENT (DETERMINISTIC DEPLOYMENT) 5. 2. {4}.3. an additional delay to the interrogation process. The greedy set covering algorithm constructs the set of covering subsets incrementally. 3. The greedy algorithm may pick the set {{1. 2. it picks the subset that covers the maximum number of uncovered elements until all elements are covered. and picking the one with the maximum weight. let C = {1. we modify the earlier described greedy set covering algorithm (Algorithm 6) to make solutions with less overlapping preferred. 4}. The general set covering problem does not take overlapping among subsets into account. For example. hence. {4}. However. 4}. 4. {5}}. 5}} to cover all elements with a minimum number of subsets. the sole objective therein is to minimize the number of subsets covering all elements.3 Reducing Overlap Among Readers 148 Overlapping among RFID readers causes reader-to-reader interference which result in collisions at the readers and. 3.5. According to the general greedy algorithm. 3}. {2. {1. That can be viewed as giving each remaining subset a weight equal to the number of uncovered elements it covers. 2. which is much more suitable for readers deployment in RFID networks. at each stage. 3}. The greedy algorithm for set .

8 end 9 Find a set S ∈ Z that maximizes W eight(S). the weighted set cover algorithm picks a subset that covers the minimum number of already covered elements.α) Input: C: set of tags. RFID READERS DEPLOYMENT (DETERMINISTIC DEPLOYMENT) covering with less overlapping is shown in Algorithm 7. α: weight ratio. It is obvious that when 0 ≤ α ≤ 1/n. However. F: coverage set. 3 U = C.5.F.3.. 1 Weighted Set Covering(C. 15 V = V ∪ {S}. 12 end 13 U = U − S. 2 Z = F. This is because both the weighted set covering algorithm and the general set covering algorithm picks the same subset at each stage except when there is a tie (i. 16 end 17 return V .e. 14 Z = Z − {S}. When there is such a tie. the approximation ratio of ln C + 1 holds for the weighted set covering algorithm. Output: V : set of readers. several subsets covers the same maximum number of uncovered elements). 4 V = φ. 149 Algorithm 7: Weighted set covering algorithm to minimize overlapping amongst subsets. the approximation ratio of the general greedy algorithm holds regardless of how ties are . 5 while U = φ do 6 foreach subset S ∈ Z do 7 W eight(S) = |U ∩ S| − α(|S| − |U ∩ S|). 10 if W eight(S) ≤ 0 then /* This means that the set C can not be covered by subsets in F */ 11 return φ and exit.

which results in the maximum value for β. the objective of minimizing the number of readers and that of minimizing β conflict with each other. Thus. However.3. However.. The main idea is to put a constraint on the maximum number of tags β. This is because each RFID reader is associated with a subset of tags and readers interrogate their tags in parallel. the number of tags they cover). RFID READERS DEPLOYMENT (DETERMINISTIC DEPLOYMENT) dealt with. This can be dealt with either by putting a constraint on β and finding the minimum number of readers meeting that constraint. the reader with the maximum workload determines the overall performance because it will be the last to finish. . the minimum possible value of β is 1. which are covered by a single reader. which lowers the performance1 of the whole system.5.e. For example.4 Load Balancing 150 The algorithms we have discussed so far aim at reducing the total number of RFID readers and reducing overlapping among different readers. In this subsection we present a method that distributes the load evenly among RFID readers for the sake of improving the overall performance of the system. The former option is straight forward and can be done by excluding intersection points 1 The performance herein is defined as the time needed to interrogate all tags. 5.3. the issue of load balancing is not considered. a system with two RFID readers each covers 50 tags is much better than a system with two readers one covers 90 tags and one covers 10 tags. we may get solutions in which readers have significant variations in their workloads (i. Therefore. and the minimum number of readers may be one reader covering all tags. which means that each reader covers a single tag. or putting a constraint on the maximum number of readers R and finding the minimum value of β meeting that constraint.

|F|)). Cover(q) is excluded from F). 18 V =Set Cover(C. The latter option is less trivial.5. .e.3.α. 2 C = T... 21 else 22 return φ. 15 end 16 end 17 Z = {S : S ∈ F ∧ |S| ≤ βmax }. The overall computational complexity of this algorithm is O(log (MAXS∈F |S|)|C||F| MIN(|C|. 11 if 0 < |V | ≤ R then 12 βmax = β.Z. 3 Q = the set of all intersection points. [1.α). 2 9 Z = {S : S ∈ F ∧ |S| ≤ β}. That can be done more efficiently by doing a binary search over all possible values of β (i. 10 V =Set Cover(C. one may need to try all possible values of β and take the minimum value that meets the constraint of having at most R readers. when |Cover(q)| > β.MAXS∈F |S|]). This method is illustrated in Algorithm 8. 4 F = {Cover(q) : q ∈ Q}.e. 6 βmin = 1. In fact. 23 end whose coverage exceeds β (i. α: weight.α). R) Input: T : set of tags. 13 else 14 βmin = β + 1. RFID READERS DEPLOYMENT (DETERMINISTIC DEPLOYMENT) 151 Algorithm 8: RFID readers deployment algorithm with load balancing. 7 while βmax > βmin do 8 β = βmax +βmin . F: set of readers. 19 if 0 < |V | ≤ R then 20 return V .. 5 βmax = MAXS∈F |S|. Output: V : set of readers.Z. 1 Readers Deployment(T .

5. we present our redundant reader elimination algorithm.4.2 Neighbors and Tags Estimation (NTE) based algorithm The redundant reader elimination problem is known to be NP-hard [16].3. this can be easily accommodated in our scheme. to find near optimal solutions in a reasonable time. To this extent we propose the Neighbor and Tag Estimation (NTE) algorithm. Finally. Given a set of RFID tags and a set of RFID readers covering all tags. The main idea is that when a reader Ri has .5. RFID REDUNDANT READERS ELIMINATION (RANDOM DEPLOYMENT) 152 5. 5.4. Therefore. find the minimum cardinality subset of RFID readers that cover all the tags [16].4.4 RFID Redundant Readers Elimination (Random Deployment) In this section. However. 5. the redundant reader problem is defined as follow. We start with some assumptions and definitions we use in this work and that is followed by a detailed description of the algorithm we propose.1 Problem Definition Formally. we discuss the algorithm’s computational and communication complexity. which is a heuristic greedy algorithm. if we reduce the actual transmission ranges of tags by a value of r units. our scheme becomes resilient to any displacement of at most r units from the planned locations. it is expected in practice to have some tags being displaced from their planned locations.5 Displaced tags While we assume that exact locations of tags are known. we propose the NTE algorithm. We then show further details of the algorithm using an illustrative example.

e.. R2 . • N (Ri ) is the number of active neighboring (interfering) readers of reader Ri . too. R2 . • E(Ri ) is the estimated number of active tags within the interrogation range of reader Ri . r r r • Rr = {R1 . weights of active readers are calculated and then a reader with the maximum weight is deactivated and all tags covered by that reader are deactivated. N (Ri )) (lines 8 and 9). Any reader with a weight of 0 is considered to be redundant and deactivated. all tags are active and all readers are active. . To estimate the number of tags. a a a • Ra = {R1 .e. . At the beginning of each iteration.5. . The algorithm begins with basic housekeeping (lines 3 and 4).. . Initially. we use the following notations: • R = {R1 . tags covered by Ri have a high probability of being also covered by other readers in the neighborhood of Ri . . .4. . Rn } is the set of active readers. In the presentation of our algorithm. This process continues until all tags are inactive. . R2 . The algorithm uses this observation to identify redundant readers. the algorithm assigns a weight based on the ratio of the number of active tags covered by Ri to the number of active neighboring readers of Ri . For each active reader Ri . At the beginning of the each iteration. The pseudocode for the NTE algorithm is shown in Algorithm 9. each active reader Ri estimates the number of tags it covers (i. . . the scheme presented in [62] can . RFID REDUNDANT READERS ELIMINATION (RANDOM DEPLOYMENT) 153 a large number of neighboring readers. Rm } is the set of non-redundant readers. . at which point all remaining active readers are considered to be redundant. . Rl } is the set of redundant readers. E(Ri )) and determines the number of active neighboring readers it has (i.

9 N (Ri ) = The number of active neighboring readers for Ri . 2 // Initially all tags are active. 1 NTE-Reader(R): Input: R: set of all readers. 13 end 14 else 15 if N (Ri ) > 0 then 16 Ratio[i] ← E(Ri )/N (Ri ). 10 if E(Ri ) = 0 then 11 Rr ← Rr ∪ {Ri }. 5 while R = φ do 6 max ← 0 . r 4 R = φ. 7 foreach Ri ∈ R do 8 E(Ri )= An estimate to the number of active tags covered by Ri . 28 R ← {R} − {Ridx }. 20 end 21 if max < Ratio[i] then 22 max ← Ratio[i].4. 30 // using a broadcasted request. a 3 R = φ. RFID REDUNDANT READERS ELIMINATION (RANDOM DEPLOYMENT) 154 Algorithm 9: The greedy NTE redundant reader elimination algorithm. 29 All tags covered by Ridx are deactivated. 31 end r 32 return R .5. 23 idx ← i. Output: Rr : set of redundant readers. 17 end 18 else 19 Ratio[i] = ∞. 24 end 25 end 26 end 27 Ra ← Ra ∪ {Ridx }. 12 R ← R − {Ri }. .

E(Ri ) = 0) are considered to be redundant (lines 1013). all readers in R estimate the number of active tags they cover and the number of active neighboring readers they have. Tracing the NTE algorithm on this instance is shown in Table 5.. RFID REDUNDANT READERS ELIMINATION (RANDOM DEPLOYMENT) 155 be used as it can estimate the number of tags within 99% accuracy ratio. R2 . The numbers of active neighboring . and active tags covered by that reader are deactivated (line 29). can be easily counted. This can be done by a single message broadcasted by the selected reader. The set of readers R = {R1 . R4 } is given.e. R3 and R4 have 2 tags. Amongst active readers that still cover at least one active tag. all readers have been classified as redundant or not).4.1.. A snapshot of an example of an RFID network before and after applying the NTE algorithm is shown in Fig. 2 tags and 2 tags. Readers that do not cover any active tags (i. to the NTE algorithm. 1 tag.e. as input.2. 5. Any tie is broken arbitrarily.5. R3 . R2 . The algorithm terminates when the set of readers R is empty (i.3. The selected reader itself is deactivated and not counted in the number of neighboring readers during subsequent iterations. the one with the maximum weight (i.3 Exemplary Scenario We show in this subsection how our algorithm finds the redundant readers in the scenario shown in Fig.e.. 5. hence. shows how effective the NTE algorithm is in eliminating a significant number of redundant readers. 5.4. respectively. Ratio[i]) is selected and excluded from the redundancy list (lines 27 and 28). Readers R1 . The active neighboring readers of a reader Ri are those readers which are within the communication range of Ri and. In the first step.

3: Snapshot of random RFID network from the simulator.4.5. . RFID REDUNDANT READERS ELIMINATION (RANDOM DEPLOYMENT) 156 (a) Network before the execution of NTE algorithm 1 (b) Network after the execution of NTE algorithm Figure 5.

upon receiving such a message. 2. reader R1 will have the maximum weight and. R2 . 5.4. T1 and T2 ) are deactivated. Accordingly. reader R1 and tags T1 and T2 are not counted. readers R2 . respectively.e. in the worst case. where N is the number of readers. 3 and 2. reader decrements its number of active neighboring readers. respectively. R4 is picked as it has the maximum weight. Deactivating tags covered by a particular reader can be also done in a constant time. respectively.e. Now. it can broadcast a message to its neighboring readers telling them that it is going into an inactive mode.5. Therefore. At this stage. R3 and R4 cover 1 tag. the set R) estimate the number of active tags they cover and the number of active neighboring readers they have.4.4 Computational Complexity Estimating the number of tags covered by a reader can be done in a constant time [62]. hence. the readers R1 and R4 can be used to interrogate all tags and readers R2 and R3 will be turned off as they are redundant.. R3 and R4 have are 1. The numbers of active neighboring readers they have are 2. we are left with the while loop that have. As a result. Thereby. readers manipulate the number of tags they cover and the number of neighboring readers they have in a constant time. When a reader is deactivated. Active tags covered by R1 (i. The time complexity of each iteration is dominated by the process of . readers R2 and R3 will be added to the list of redundant readers as they do not cover any active tag. 2 and 2. as it just requires a single broadcasted message. active readers (i. 1 tag and 2 tags. In the second step. will be excluded from the list of redundant readers and deactivated be removing it from the set R . RFID REDUNDANT READERS ELIMINATION (RANDOM DEPLOYMENT) 157 readers R1 . In the last step.. O(N ) iterations.

Therefore.4. R4 } E(Ri ) 2 1 2 2 R = {R2 . where M is the number of active readers. R4 } Ra = {R1 . R2 . R3 . 5. (a) Step-I Input: Reader R1 R2 R3 R4 Output: Input: Reader R1 R2 R3 R4 Output: Input: Reader R = {R1 . R4 } (b) Step-II Ra = φ N (Ri ) 1 2 3 2 a R = {R1 } Ra = {R1 } N (Ri ) R = {R2 . hence. R4 } E(Ri ) 1 1 2 R = {R2 . R3 . R3 . the overall time complexity of the algorithm is O(N 2 ). takes O(M ) time. R3 } Ra = {R1 .1: Tracing the NTE algorithm for the scenario of Fig. R3 } (c) Step-III 2 2 2 Ra = {R1 .5. . RFID REDUNDANT READERS ELIMINATION (RANDOM DEPLOYMENT) 158 Table 5. R3 } E(Ri ) R1 R2 0 0 R3 0 0 R4 Output: Rr = {R2 . R4 } finding the reader with the maximum weight which is done in a centralized fashion and. R4 } N (Ri ) R = {R2 .2.

Both schemes were selected based on their relative performance with other schemes. from the literature. we use two recent readers deployment schemes. and readers are ..5 Simulation Setup and Results In this section. respectively.5. The RFID reader has an interrogation range of 4 − 5m and an interrogation area of 5x5x100m3 . we evaluate the performance of the proposed algorithms and that of other recently proposed schemes for the readers deployment and redundant readers elimination in RFID. for both deterministic deployment and random deployment. the two proposed schemes. Unless otherwise mentioned.5. hence. less reader collisions. For evaluation. And lower values for the maximum load amongst readers implies a higher overall reading rates.e. For rest of the section. Deterministic deployment: In the case of deterministic deployment. 92] and the honey grid coverage [110]. We also evaluated the schemes for various tags load (sparse and dense) and various interrogation areas.. 5. Lower number of readers imply less overlapping and. maximum load amongst the readers and the number of redundant readers as our evaluation criteria. SIMULATION SETUP AND RESULTS 159 5.e. for comparison. the maximum number of tags a single reader may cover. Tags are assumed to be placed in a 3D space modelled as a rectangular cuboid. Algorithm 7 and Algorithm 8 are termed as Approximate Set Cover (ASC) and Approximate Set Cover with Load Balancing (ASC-LB). i. we use the number of readers and the maximum load. i. namely the optimal grid coverage [83. We use the number of readers.1 Simulation Setup We have implemented the RFID system using an in-house simulator. simulations for the deterministic deployment are performed using the following parameters.5.

minimal communication interferences and is economically sound.2 Simulation Results for Deterministic Deployment Number of Readers: Minimal number of readers with maximal required coverage is an important evaluation metric as it translates into minimal communication overheads.5. These schemes were selected based on the their relative performance improvements over other schemes in the literature.7 shows the number of readers required to cover the same interrogation area for various . we evaluated the NTE scheme in terms of resilience to the reader’s range. SIMULATION SETUP AND RESULTS 160 to be placed on particular sides of the rectangular cuboid. we use recently proposed algorithms such as RRE [16]. Furthermore. our evaluation metrics are the number of redundant readers. the number of tag reads and the number of tag writes. or comparison. A low number of tag reads and writes implies that the scheme is light-weight and does not require state maintenance on the tags. A high number of redundant readers imply the scheme is effective in detecting and eliminating redundant readers. simulations for the random deployment are performed using the following parameters. Locations of tags are generated using pseudo-random number generator. The performance metrics are averaged over twenty different runs generated using distinct random seeds. For comparative analysis. Random deployment: In the case of random deployment. Unless otherwise mentioned.1.5. 5. Both tags and readers are assumed to be randomly place within the interrogation area. LEO+RRE [52] and TREE [109].5. The results are averaged over twenty different runs generated using distinct random seeds. The RFID reader has an interrogation range of 5m and an interrogation area of 250x250m2 . as is illustrated in Figure 5. 5. Fig.

ASC and ASC-LB. As anticipated. having reader deployment scheme. Whereas. in terms of readers.5. as after certain number of tags. translating into lesser readers while still covering the same tags set.4: Number of readers required for various tag density. the two proposed schemes and optimal-grid would perform same. Our proposed ASC algorithm is the most prominent as it significantly reduces the required readers. for the proposed algorithms. algorithms. Both Optimal-Grid and Honey-Grid quickly reaches its maximum allowable readers as the tags density exceed certain limits. Fig. either scheme requires certain number of reader to achieve 100% coverage. amongst all the readers interrogation range for various algorithms. an average two times reduction compared with HoneyGrid is obtained. Load Balancing: The ASC algorithm guarantees minimum number of readers however. In highly dense environment. maybe results in imbalance number of tags between the readers.5. SIMULATION SETUP AND RESULTS 161 160 140 Number of readers 120 100 80 60 40 20 0 100 200 Optimal-Grid Honey-Grid ASC ASC-LB 300 400 Number of tags 500 600 700 Figure 5. 5. without any . such a saturation point comes much later hence.5 shows the maximum number of tags.

.e. as the number of tags increases. compared with others. ASC-LB performs the best. i. As expected.5.. we may get solutions in which readers have significant variations in their workloads (i. are shown in Fig.6.. in sparse (100tags) and dense (500tags) settings. load balance.e.5: Maximum load (tags) for various tag density. Therefore. the number of tags they cover).5. The effect of increasing the interrogation area. which lower the performance of overall system. the loads for all the schemes increase. With an increase in the interrogation areas and tags cardinality. trade-off exists between minimal readers and load balancing as the later would require more readers in order to maintain the balance amongst them. as it is distribute the load many-folds efficient compared with others. However. can handle diverse tag deployments. For instance. will yield reader(s) having much larger set of tags to interrogate. 5. SIMULATION SETUP AND RESULTS 162 45 40 35 Maximum load 30 25 20 15 10 5 0 0 Optimal-Grid Honey-Grid ASC ASC-LB 100 200 300 400 Number of tags 500 600 700 Figure 5.e. our proposed algorithms are more scalable i. . the proposed load-aware placement scheme.

6: Effect of tag density and interrogation area on the readers deployment algorithms. SIMULATION SETUP AND RESULTS 163 100 80 60 40 20 0 50 100 150 Interrogation area Number of readers Optimal-Grid ASC ASC-LB 200 250 (a) 100 tags 240 220 200 Number of readers 180 160 140 120 100 80 60 40 50 100 150 Interrogation area 200 250 Optimal-Grid ASC ASC-LB (b) 500 tags Figure 5.5.5. .

Second. 5. in a sparse environment. LEO+RRE.e.7.5. the NTE algorithm detects redundant readers far beyond other schemes whereas in a sparse environment the improvement obtained is not significant. SIMULATION SETUP AND RESULTS 164 5. First. readers are expected to have a limited variation in the tags to neighbors ratio they have. Tag singulation is an expensive process as it is determined by the number of tags within a reader’s interrogation range and the overlapping among readers.9 show . RRE and LEO. which is not the case in a sparse environment.5. Tags Reads and Writes: Most of the existing schemes. Tag singulation is required in order to read-from a tag. and accordingly this ratio is not as effective as it is in a dense environment. Comparisons of different schemes based on the number of redundant readers they detect for both dense (50x50m2 ) and sparse (250x250m2 ) networks are depicted in Fig. mainly in terms of energy consumption and reading delay. 5. e. In dense environments. is influenced by interference among readers [42]..3 Simulation Results for Random Deployment Redundant Readers: The overall network performance. in sparse environments only 2% improvement is observed over LEO+RRE. Fig. in dense environments. an effective scheme is one that detects more redundant readers. in the literature have high communication complexities associated with them. However. For instance.8 and Fig.g. the NTE schemes can detect 15% more redundant readers than the best of the other schemes. Hence..5. Communication complexity is defined as the number of reads and writes operation made from and to tags. in a dense environment a reader has a good chance of having a large number of neighbors and picking a reader with the maximum tags to neighbors ratio as a non-redundant reader has the potential to render many neighboring readers redundant. 5. There are two reasons behind this behavior. i.

SIMULATION SETUP AND RESULTS 165 380 360 Redundant readers 340 320 300 280 260 240 0 1000 RRE LEO LEO + RRE TRRE NTE 2000 3000 Number of Tags 4000 5000 (a) Simulation area of 50x50m2 800 700 Redundant readers 600 500 400 300 200 100 0 1000 2000 3000 Number of Tags 4000 5000 RRE LEO LEO + RRE TRRE NTE (b) Simulation area of 250x250m2 Figure 5. .7: Redundant reader detected in sparse and dense network.5.5.

5. .5. SIMULATION SETUP AND RESULTS 166 60000 50000 40000 Tag reads 30000 20000 10000 0 0 RRE LEO LEO + RRE TRRE NTE 1000 2000 3000 Number of Tags 4000 5000 (a) Simulation area of 50x50m2 60000 50000 40000 Tag reads 30000 20000 10000 0 0 1000 2000 3000 Number of Tags 4000 5000 RRE LEO LEO + RRE TRRE NTE (b) Simulation area of 250x250m2 Figure 5.8: Number of tag reads in sparse and dense environments.

writes only once to the tags within its range. redundant readers must be detected to avoid unnecessary interference caused by them. in the NTE scheme. Read operations that make a difference are ones that require singulation and no such an operation is needed in the NTE scheme. In this chapter. In fact the NTE algorithm is at least six times more efficient in accessing the tags memory. issues such as interrogation zone overlap among readers. in the sparse environment. load balancing amongst reader and light-weight redundant readers detection schemes have received little attention. Furthermore. The number of reads and writes is obviously lower in sparse environments than it is in dense environments for all schemes.5. the NTE scheme is much more efficient in terms of energy consumption and delay. 5. This is because the reader. we propose an . SUMMARY 167 comparisons between different schemes in terms of the number of read and write operations they incur. once the readers are deployed. Therefore. In deterministic approach. In random approach. although the number of redundant readers detected by the LEO+RRE scheme is similar to that in the NTE scheme. However. the number of reads and writes in the NTE scheme is many folds less than those in other schemes. for deterministic approach. Operations in the tags count estimation are not counted as read operations. This is because the estimation scheme does not require tags singulation.6 Summary Coverage is crucial for large scale RFID applications wherein large number of readers can be deployed in deterministic and random manners. minimum readers must be deployed to maintain certain coverage. each tag receives a single write operation. Several schemes have been proposed in the literature.6. however.

5.6. SUMMARY 168 30000 25000 20000 Tag writes 15000 10000 5000 0 0 RRE LEO LEO + RRE TRRE NTE 1000 2000 3000 Number of Tags 4000 5000 (a) Simulation area of 50x50m2 10000 9000 8000 7000 Tag writes 6000 5000 4000 3000 2000 1000 0 0 1000 2000 3000 Number of Tags 4000 5000 RRE LEO LEO + RRE TRRE NTE (b) Simulation area of 250x250m2 Figure 5.9: Number of tags write in sparse and dense environments. .

and guarantees less overlapping among readers and a balanced distribution of load among them. for random approach. light-weight greedy algorithm to detect and eliminate redundant readers in RFID networks. . SUMMARY 169 algorithm that approximately determines the minimal number of readers and the location of each reader in a 3D space. Also. but the overall system performance is improved also by reducing interference and overlapping among readers coverage regions and. having less collisions among readers. not only is the overall network cost reduced. either to-be deployed or to-be active at one time.5. By reducing the number of readers.6. hence.

RFID brings several design challenges that need to be overcome.170 Chapter 6 Conclusion The last decade has witnessed a growing interest in RFID due to their unique potential in a wide range of applications. the reading rates of RFID tags. especially passive RFID tags. While being the key requirement for anticipated ultra large scale applications. e. At protocol level.. The tags. Furthermore.g. One of these challenges is tag anti-collision and hence. RFID systems provide low cost and low power object identification and tracking mechanisms. Internet of Things. with functional constraints due to limited power cannot adopt conventional medium access control approaches. At the system architecture level. reducing collisions can be achieved by distributing the . an optimal power level algorithm which clusters and isolates tags according to their physical distance can reduce tag collisions. Another challenge is scaling to large scale application (a typical scenario in Internet of Things) which involve a large number of tags within the interrogation zone of single or multiple readers. lowers overall system reading rate. power requirements and signal attenuation limit the reading range. A promising solution to overcome these limitation exists at the protocol and system architecture levels. This scenario results in a large number of tag collisions and therefore.

To this end. namely Distributed Receiving RFID. two optimal power stepping algorithms are proposed. with the goal of boosting the operational performance of RFID systems. In the proposed paradigm. such as the Internet of Things and item-level tagging. and hence. For each optimization problem. collisions are reduced and hence higher reading rates can be achieved. SUMMARY 171 signal receiving task amongst spatially dispersed devices.1 Summary Schemes presented in Chapter 3. These devices then selectively collect and relay them to the reader. we introduce the use of reader’s power control to mitigate tag collisions.1. Chapter 4 and Chapter 5 aim at enabling RFID systems for ultra large scale of the anticipated applications. using sophisticated medium access control and architecture design. In Chapter 4. The first scheme targets an ideal setting in which the transmission power. the transmission range of an RFID reader can be tuned with high precision. the task of collecting the . and to provide load-aware and energy-efficient reader deployment to prolong overall network lifetime.6. one at a time. we present mathematical delay analysis and devise an efficient method to find the optimal clustering. By interrogating a subset of tags. forming micro-interrogation zones which are interrogated in parallel and autonomously. The objective is to lessen the limitations and alleviate bottlenecks imposed by centralized architectures. In Chapter 3. These schemes achieve higher reading rates and range. we introduce a new paradigm in RFID systems. The second scheme relaxes this assumption and considers an RFID reader with a finite and discrete transmission range. 6.

The third scheme. The first scheme. by overlapping RFID readers. In the second scheme. . In Chapter 5. the tags are interrogated in parallel.1. Parallel singulation assists in achieving a higher reading rate under both dense and mobile tag environments. micro-interrogation zones are formed. targeting deterministic deployment. In the first scheme. proposes a set-cover approximation algorithm to determine the minimal number of RFID readers required to maintain the requested coverage while having minimal interrogation overlapping among the readers. Micro-zones spatially isolate tags within the readers interrogation zone therefore facilitate higher bandwidth communication link and data rates. The second scheme then extends the algorithm to facilitate load-balancing deployment strategies. targeting random and post deployment coverage. To this end. using the spatial devices and by adjusting the tags reflectivity co-efficient. We propose both a deterministic and a probabilistic model of this parallel singulation mechanism. we propose two schemes. SUMMARY 172 modulated backscattered signals is dispersed to spatially distributed entities (devices) within a reader’s interrogation zone. To this end. This model supports multipoint communication and hence eliminates redundant interrogation of the same set of tags. we introduce RFID coverage as deterministic and random deployment problems.6. namely parallel singulation. The complexity and monetary cost associated with ultra large scale RFID systems strongly depends on the number of readers being deployed to interrogate tags. proposes the use of neighbouring readers ratio and covering tags to heuristically turn off redundant readers to achieve the maximal coverage with the least possible readers and overlapping interrogation zones. three RFID deployment schemes are proposed. exploiting micro-zones.

2 Future Outlook Several future research problems stem from our work thus far. Novel MAC protocols. aware of the underlying distributed receiving reader-tag communication architecture. Optimized fielder deployment techniques. FUTURE OUTLOOK 173 6. Current proposals for RFID systems assume near-ideal interrogation environment. need to be developed in a distributed context. A more challenging problem is evaluating the performance of an RFID system and the proposed algorithms in a testbed setting. 4. for both fielders and reader.2. In Chapter 4. 3. thus have problematic overlapping regions. An interesting problem is optimizing and/or dynamically controlling the transmission range of each reader to minimize reader overlap and balance workload among readers. In this section. and formation of micro-zones with minimal overlapping. An efficient and tailored algorithm. is an interesting problem as it would to further boost system performance. The parallel singulation algorithm makes use of the binary search tree and the query tree algorithm. . 1. These regions result in idle queries. we point out some of them. the fielders are deployed in a grid. 2.6. will yield optimal data rates. In Chapter 5. reader deployment algorithms assume that RFID readers have a fixed transmission range.

In Proceedings of the 33rd IEEE Conference on Local Computer Networks (LCN). 2008. rev. pages 819–824. 2005. [10] K. 2011. Optimal distance-based clustering for tag anti-collision in RFID systems. available at http://www. [4] CC2500 single chip low cost low power RF transceiver. 40 db isolation at 1 GHz. Passive RFID for intelligent transportation systems.edu/nsnam/ns/. Hassanein. 1 db LSB 400 MHz4 GHz. Hassanein. pages 266–273. . RFID anti-collision protocol for dense passive tag environments.BIBLIOGRAPHY 174 Bibliography [1] LTC5507: 100kHz to 1GHz RF power detector.isi. and H. 2007. Hassanein. Ali. May 2010. pages 2113–2118. pages 1–2.com. 2009. Placement of multiple mobile data collectors in underwater acoustic sensor networks. Hassanein. SPST switches. K.skyetek. In Proceedings of the 6th Annual IEEE Consumer Communications & Networking Conference (CCNC). Alsalih. H. Ali. In Proceedings of the IEEE International Conference on Communications (ICC). S. Taha. 2008. Alsalih. MSP430x22x4 mixed signal microcontroller. In LCN ’07: Proceedings of the 32nd IEEE Conference on Local Computer Networks. May 2008. [11] W. M. March 2010. 2011.65 v to 2. Sep 2007.75 v. [5] eZ430-RF2500-SEH solar energy harvesting development tool. and A. [6] MSP430x22x2. and H. b. [8] www. CMOS 1. Ali and H. [3] SKY12329-350LF: GaAs digital attenuator 5-bit. [2] ADG901/ADG902: Wideband. [9] K. [12] W. [7] The network simulator ns-2. Akl.

In Proceedings 11th International Conference on Parallel and Distributed Systems. Mehmet. and G. Redundant-reader elimination in RFID systems. 2005. [22] J. Capetanakis. 2007. Cardei and D-Z. K. Koyuturk. Information Systems Management. 25(5):505–515. Kim. Kim. Du. Ng Wen Jye. [14] R. May 2005. IEEE Transactions on Information Theory. [19] B. A BPSK backscatter modulator design for RFID passive tags. M. 2005. 2005. Ramanathan. H. In Proceedings of the Second Annual IEEE Communications Society Conference on Sensor and Ad Hoc Communications and Networks (SECON). Dynamic framed slotted aloha algorithms using fast tag estimation method for RFID system. 2005. In Proceedings of the 13th IEEE International Conference on Networks. Angeles. Cha and J. 1979. Christoph. C. [17] L. Carbunar. Grama.M. R. Bogdan. Improving wireless sensor network lifetime through power aware organization. and A. pages 262–265. M. pages 463–468. Redundant reader elimination in RFID systems. pages 43– 48. Iyer. Novel anti-collision algorithms for fast object identification in RFID system. Novel anti-collision algorithms for fast object identification in RFID system. pages 176–184. Hoffmann. K. K. [15] S. 2005. Ananth. pages 63–67. In Proceedings of the 11th International Conference on Parallel and Distributed Systems. Cha and J. 22(1):51–65. [23] J. In Proceedings of the IEEE International Workshop on Radio-Frequency Integration Technology. Tree algorithms for packet broadcast channels. Murali. pages 63–67. Othman. 2006. pages 176–184. Cha and J. In Proceedings of the Second Annual IEEE Communications Society Conference on Sensor and Ad Hoc Communications and Networks (SECON). RFID technologies: Supply-chain applications and implementation issues. In Proceedings of the Fourth IEEE Workshop on Automatic Identification Advanced Technologies. 11:333–340. [18] J. Randomized pseudo-random function tree walking algorithm for secure radio-frequency identification. [20] M. 2005. [21] J.BIBLIOGRAPHY 175 [13] N. Kim. Bolotnyy and Robins G. and M. Birari and S.. . In Proceedings of the 3rd IEEE Consumer Communications and Networking Conference (CCNC). Mitigating the reader collision problem in RFID networks with mobile readers. pages 768–772. Amin. [16] C. 2005. Wireless Networking.

Bi-slotted tree based anti-collision protocols for fast tag identification in RFID systems. McGraw-Hill. 85–87. In Proceedings of the IEEE International Conference on Communications (ICC). Lee. Applied Software Computing. Horng. Joehl. 11(1):2007. pages 1653–1657. Prefix-randomized query-tree protocol for RFID systems. Park. Introduction To Algorithms (2nd Edition). pages 1363–1366. Cormen. Lee. 10(12):861– 863. [29] K. C. D. and J. 2010. 2006. Hu. and H. R. and M. Yum. C. 2010. C. Zhu. Cui. Choi. and P. Declercq. 2006. Stein. Asymmetric RFID tag antenna. D. An adaptive load balancing management technique for RFID middleware systems. In Proceedings of the IEEE International Symposium of Antennas and Propagation Society. In Proceedings of the IEEE Global Telecommunications Conference (Globecom). pages 1–5. J. D. W. Chiang. Chiang. 2001. Leiserson. 2007. 40(6):485–506. Yum. J. Lee. Cheng and R. [28] K.J. Fan. Chae. Multi-colony bacteria foraging optimization with cell-to-cell communication for RFID network planning. IEEE Journal of Solid-State Circuits. 2006. Cha. S. Lee. Lee. IEEE Communications Letters. pages 21–24. F. D. S. [27] C. Nov. [26] W.H. 10(2):539–547. Query tree-based reservation for efficient RFID tag anti-collision. G. Chen. and P. Curty. Choi. 2006. and H. In Proceedings of the Second International Conference on Systems and Networks Communications (ICSNC). C. An enhanced anti-collision algorithm in RFID based on counter and stack. [33] T. Lee. Jeon. H. IEEE Communications Letters. pages 3853–3858. In Proceedings of the IEEE International Conference on Communications (ICC). H. Choi. P. Rivest. Prefix-length adaptation for PRQT protocol in RFID systems. Software Practice and Experience. [25] H.W. S. Chen. and T. 2007. Y. [32] J. Remotely powered addressable UHF RFID integrated system. . Hua. [31] J. [30] J. N. 40(11):2193–2202. Dehollain. J. Hua.BIBLIOGRAPHY 176 [24] H. and C. [34] J. Enhanced binary search with time-divided responses for efficient RFID tag anti-collision. Lee. and H. Murch. 2005. and K.

M. 2006.2. The RF in RFID: Passive UHF RFID in Practice. Feng. Dobkin. pages 641–646. In Proceedings of the 8th IEEE Symposium on Computers and Communication (ISCC).BIBLIOGRAPHY 177 [35] K. Li. [45] P. The reader collision problem. Burberry. 2005. [41] B. 2007. Grnvall.W. J. Transmission control scheme for fast rfid object identification. 2003. Engels and S. Kalpakis. Dong. In Proceedings of the 11th IEEE Symposium on Computers and Communications. . EPC radio-frequency identity protocols class-1 generation-2 UHF RFID protocol for communications at 860 MHz . V. Foster and R.. Finkenzeller. In Proceedings of the 29th Annual IEEE International Conference on Local Computer Networks (LCN).a lightweight and flexible operating system for tiny networked sensors. John Wiley & Sons. [42] K. pages 455– 462. pages 2281–2285.960 MHz version 1. E. Newnes. [36] D. In Proceedings of the IEEE International Conference on RFID 2007.0. D. S. and T. Kukreja. Floerkemeier. M. Agrawal. A. In Proceedings of the IEE Colloquium on RFID Technology. 1999. B.A. [40] EPCglobal. Kar. 2007. Sarma.R. Contiki . Guo. [38] A. [37] Q. Dunkels. Bayesian transmission strategy for framed ALOHA based RFID protocols. [43] C. Antenna problems in RFID systems. Banerjee. pages 341–348. pages 1–5. [44] C. pages 207–212. 2006. [39] D. In Proceedings of the IEEE International Conference on Systems. 2003. Floerkemeier. and K. and Z. Dasgupta. Topology-aware placement and role assignment for energy-efficient information gathering in sensor networks. In Proceedings of the Fourth Annual IEEE International Conference on Pervasive Computing and Communications (PerCom) Workshops. Shukla. Load balancing in large-scale RFID systems. 2004. RFID Handbook: Fundamentals and Applications in Contactless Smart Cards and Identification. Inc. pages 228–235. 2002. Shrivastava. pages 457–462. 2008. Voigt. Man and Cybernetics. and K. Id-binary tree stack anticollision algorithm for RFID. Ding. J. In Proceedings of the 26th IEEE International Conference on Computer Communications (INFOCOM).

33–35. In Proceedings of the Sixth International Conference on Intelligent Systems Design and Applications (ISDA). Yagoub. R. and J-A. E. Piersanti. E. In Proceedings of the International Symposium on Applications and the Internet Workshops (SAINT-W). 3(11):1–8. Y. [53] J-W. H. and C-T. Genetic approach for network planning in the rfid systems. D. Lee. Lin. G. W. Y. The first search right algorithm for redundant reader elimination in RFID network. Esaias. Chen. pages 449–452. Gains for RF tags using multiple antennas.D. Woodward. Frouin. parallel and distributed systems (SEPADS). Engels. Hung. 1992. Cai. [55] N. Yu. pages 177–183. B. Griffin and G. . International journal of computer science issues. Li. [47] D. F. Analytical approach to the backscattering from UHF RFID transponder. and J. Efficient algorithm for redundant reader elimination in wireless RFID networks. and G. pages 567–572. W. H-H. Irfan and M. 56(2):563–570. pages 22–24. [49] J. Yang. C. and W. Coverage opportunities for global ocean color in a multimission era. In Proceedings of the Nineteenth Annual AUVS Technical Symposium. Command control for many-robot systems. May 2010. 2007. 36(5):1620–1627. IEEE Transactions on Geosience and Remote Sensing. Suh. Hwang. Fuschini. Y-M.E. [54] T. [51] J. Hsu. In Proceedings of the First International Conference on Innovative Computing. A layered optimization approach for redundant reader elimination in wireless RFID networks. S. Gregg. Liu.D. pages 88–91. Feldman. [52] C-H. Information and Control (ICICIC). 7:2008. S.C. 2008. Yang. Kim. McClain. Kim. R. Hooker. pages 138– 145. W. Ho. Sarma. Y. D. B. I-H. 2006. and R. Gage. 2006. IEEE Antennas and Wireless Propagation Letters. Durgin. Y. Improved anticollision scheme for high speed identification in RFID system. 1998. S. Paolazzi C. 2010. HiQ: a hierarchical Q-learning algorithm to solve the reader collision problem. and S. 2006. [48] W. Guan. Falciasecca. C. In Proceedings of the 2nd IEEE Asia-Pacific Service Computing Conference (APSCC). In Proceedings of the 9th WSEAS international conference on Software engineering.BIBLIOGRAPHY 178 [46] F. W. IEEE Transactions on Antennas and Propagation. [50] Q.

J. Node placement for connected coverage in sensor networks. . A. Sajal K. Fischer. and S. and E. Yener. Lee. 2000. pages 4028–4033. In Proceedings of the 14th International Conference on Computer Communications and Networks (ICCCN). 2003. Siu. Efficient memoryless protocol for tag identification. Effect of localized optimal clustering for reader anti-collision in RFID networks: fairness aspects to the readers. Myung. Fully integrated passive UHF RFID transponder IC with 16. W. Das. In Proceedings of the 2nd international conference on Embedded networked sensor systems (SenSys). D. Serbetli. J. [64] Ching Law. Karthaus and M. 2006. pages 497–502. and Kai-Yeung Siu. Lee. 2000. pages 265–266. In Proceedings of the IEEE International Conference on Communications (ICC). Taekyoung Kwon. In Proceedings of the IEEE International Conference on Communications (ICC). 2006. [57] U. OPT: Optimal protocol tree for efficient tag identification in dense RFID systems. 2003. Banerjee. pages 75–84. IEEE Communications Letters. Khandelwal. [61] J. ASAP : A MAC protocol for dense and time constrained RFID systems. Kayi Lee.BIBLIOGRAPHY 179 [56] K. IEEE Journal of Solid-State Circuits. 2007. K. K. Optimized transmission power control of interrogators for collision arbitration in uhf rfid systems. Lee. 2004.7-µw minimum rf input power. 38(10):1602–1608. [58] G. In Proceedings of the 12th Annual International Conference on Mobile Computing and Networking (MobiCom). 2005. [59] G. In Proceedings of the 4th international workshop on Discrete algorithms and methods for mobile computing and communications (DIALM). [60] J. Kim. Suh. pages 322–333. Yanghee Choi. [65] Jeongkeun Lee. In Proceedings of the Modeling and Optimization in Mobile. Kim. Kim. Kim. [62] Murali Kodialam and Thyaga Nandagopal. Fast and reliable estimation schemes in RFID systems. Yener. and M. Law. 2006. Yu. A. Kim. Efficient memoryless protocol for tag identification. 11(1):22–24. pages 128–133. E. and Kyung ah Kim. Chen. pages 75–84. and K. and K. Ad Hoc and Wireless Networks (WiOpt). [63] C. Analysis of RFID anti-collision algorithms using smart antennas. Khandelwal. In Proceedings of the 4th International Workshop on Discrete Algorithms and Methods for Mobile Computing and Communications (DIALM). Kar and S.

L. In Proceedings of the IEEE Wireless Communications and Networking Conference (WCNC). Qu. M-S. Propagation and EMC Technologies for Wireless Communications (MAPE). ASK modulator for RFID system using a novel variable DGS resonator. Lee. 18(10):773–780. IEEE Network. 2007. [73] M. [72] Y. H. In Proceedings of the Second Annual International Conference on Mobile and Ubiquitous Systems: Networking and Services (MobiQuitous). Z. CDMA-based RFID systems in dense scenarios: Concepts and challenges. G. A system to place observers on a polyhedral terrain in polynomial time. An enhanced dynamic framed slotted ALOHA algorithm for RFID tag identification. Nayak. Applications and Systems. pages 1660–1663. Park. and S. Jung. B. and R. . Image and Vision Computing. In Proceedings of the IEEE International Symposium on Microwave. Xi. Liu. Draper. Pappu. A. The reader collision problem in RFID systems. [69] C. 22(6):26–35. Joo. H-G. [68] K. Jung. Lee. A. pages 1669–1672. and I. Maguire and R. Exposure in wireless ad-hoc sensor networks. Mutti and C. Lin and F. [71] L. Cho. F. A. 2008. M. Leong. Baek. H-R. Bolic. Xie. 2005. Meguerdichian. Floerkemeier. Oh. 2009. Marengoni. M. Y-R. Hanson. pages 215–222. [74] S. and P. S. Ron. Lai. Potkonjak. Lee. J-W. and J-S. S-C. H-H. Lin. Jung. Y-K. Antenna. Koushanfar. In Proceedings of the IEEE International Conference on RFID. S. [75] C. IEEE Transactions on Automation Science and Engineering. 2000. November-December 2008. In Proceedings of the 7th annual international conference on Mobile computing and networking (MobiCom). A simulated annealing algorithm for RFID reader networks. Stojmenovic. Sitaraman. Taxonomy and challenges of the integration of RFID and wireless sensor networks. An optimal Q-algorithm for the ISO 18000-6C RFID protocol. [70] H. Liu. pages 166–172. Cole. In Proceedings of the 2007 European Microwave Conference. In Proceedings of the 2nd International Conference on Mobile Technology. 6(1):16–24. K-K. 2007. and C. E-J. An improved anti-collision algorithm in RFID system. Seong.BIBLIOGRAPHY 180 [66] S. pages 139–150. J-W. and M. 2001. pages 658–661. pages 137–142. Lee. 2005. J. [67] Y-D. Han. Nae. 2005. Ng.

M. 48(6):212–218.V. Y.S. Identification of RFID tags in framed-slotted ALOHA with robust estimation and binary selection. In Proceedings of the Fifth Annual IEEE International Conference on Pervasive Computing and Communications (PerCom). M. Art Gallery Theorems and Algorithms. and T. Nikitin and K. [82] P. pages 616–620. Nanjundaiah and V. 43(8):431–432. Rao. In Proceedings of the 19th International Conference on Advanced Information Networking and Applications (AINA). Lee. K. 2007. 2010. and S. [77] J. Chung. IEEE Transaction on Parallel and Distributed System. [83] A. S. 1987. 2006. ORourke. Energy-aware tag anti-collision protocols for RFID systems. [81] P. 48(9):2679– 2698. Lee. Chaudhary. 2007. Theory and measurement of backscattering from RFID tags. Rao. Y. T. Park. pages 368–372. Myung and W. Adaptive binary splitting: a RFID tag collision arbitration protocol for tag identification. J. 2006. Mobile Network Applications. Namboodiri and L. 2007. W. Improvement to the anticollision protocol specification for 900MHz class 0 radio frequency identification tag. Chung. March 2007. Kong. Park. Energy-aware tag anti-collision protocols for RFID systems. Swim.S. [79] V. Shih. L. and R.D. In Proceedings of the First International Conference on Communications and Electronics (ICCE). Criticality index analysis based optimal RFID reader placement models for asset tracking.BIBLIOGRAPHY 181 [76] J. 11(5):452–454. [85] J. and T. Oztekin. Namboodiri and L. Lee. Gao. In Proceedings of the Fifth Annual IEEE International Conference on Pervasive Computing and Communications (PerCom). 2006. K. and T. Nikitin.V. 2007. Gao. [84] J. Electronics Letters. 18(6):763–775. Tag-splitting: Adaptive collision arbitration protocols for RFID tag identification.V. . Mahdavi. IEEE Communications Letters. [86] J. Differential RCS of RFID tag. [78] V. International Journal of Production Research. F. Srivastava. Identification of rfid tags in framed-slotted ALOHA with tag estimation and binary splitting. 2005. IEEE Antennas and Propagation Magazine. Lee. pages 23–36. Myung. [80] M.V. K. pages 23–36. Oxford University Press. Martinez. Erande. Bukkapatnam. K. Z. 11(5):711–722.

[95] X. Tracking via square grid of RFID reader positioning and diffusion algorithm. 2006. Y. [89] V. In Proceedings of the IEEE International Conference on Communications (ICC). Huang. [90] D. Kim. In Proceedings of the IEEE Wireless Communications and Networking Conference (WCNC). and S. Shi. Seok. Srivastava.C.-L. 2000. Reza and T. pages 337–342. L. H. Reza. A modified grouped-tag TDMA access protocol for radio frequency identification networks. [93] A. H. [92] A. 2004. Energy-aware wireless microsensor networks. Investigation of indoor location sensing via rfid reader network utilizing grid covering algorithm. Computer and communication. J. Nikitin. G.S. Pastos and R.BIBLIOGRAPHY 182 [87] N. 29(11):2150–2166. K. Viswanathan. Choi. [97] W. Shin and J. W. and F. pages 512–516. Wei. Schurgers. P. Park. Progress In Electromagnetic Research.V. Wireless Personal Communications. B. IEEE Transactions on Antennas and Propagation. Dimyati. Pastos and R. Engels. 2007. Sensor Networks and Information Processing Conference. In Proceedings of the IEEE Wireless Communications and Networking Conference (WCNC). Q. T. In Proceedings of the IEEE Wireless Communications and Networking Conference (WCNC). Cole. Rao. 19:40–50. and Y. P. S. 53(12):3870–3876. A hybrid query tree protocol for tag collision arbitration in RFID systems. Ranasinghe. An enhanced binary anticollision algorithm of backtracking in RFID system. Shi. Lam. A modified grouped-tag tdma access protocol for radio frequency identification networks. V. . [88] N. Geok. Kwon. and p. 2000. M. In Proceedings of the 2004 Intelligent Sensors. [94] J. F. 2008. Geok. X. Lee. 2005. Sun. [91] K. K. Partitioning of tags for near-optimum rfid anticollision performance. and K. pages 1673–1678. Viswanathan. pages 512–516. Shih. W. Taxonomy and survey of RFID anti-collision protocols. D. Antenna design for UHF RFID tags: a review and a practical application. [96] D. pages 1–24. Huang.Yen. D. 2009. and M. T. 49(1):67–80. C. 2010. and S. Wireless Personal Communications. C. IEEE Signal Processing Magazine. Security and privacy solutions for low-cost RFID systems. pages 5981–5986. 2002.-L. Ryu.-W. Raghunathan. 4:263–271. 2007.

2005. Engels. IEEE Communications Letters. 53(9):2978–2990. Xianming and Y. 2009. A novel variable-impedance microstrip circuit and its application to direct antenna modulation. Engles. Efficient object identification with passive RFID tags. Wong. W. A folded dipole antenna for RFID. [103] J. Design criteria for the RF section of UHF and microwave passive RFID transponders. and J. Enhanced binary search with cut-through operation for anti-collision in RFID systems. Kluwer Academic Publishers. 2003. pages 494–498. Ragopal. Iannaccone. and D. A. R. Vogt. Colorwave: an anti-collision algorithm for the reader collision problem. IEEE Transactions on Microwave Theory and Techniques. Lim. Bellatalla. J-S. In IEEE International Symposium on Antennas and Propagation Society. 28(2):201–212. Waldrop. [101] G. The simulation and analysis of algorithms for redundant reader elimination in rfid system. Iannaccone. International journal of Radio Frequency Identification Technology and Applications. 2006. pages 98–113. Yang and J-L. [105] T. 2004. pages 1206–1210. In Proceedings of the European conference on Wireless Technology. P. Cho. Microelectronics Journal. [109] Z-Y. 2006. Sarma. In Proceedings of the Third UKSim European Symposium on Computer Modelling and Simulation (EMS). Vandenameele. pages 403–406. 2007. 2005. E.BIBLIOGRAPHY 183 [98] C-S. W. Rivest. Ultra-low power PSK backscatter modulator for UHF and microwave RFID transponders. Son. D. Space division multiple access for wireless local area networks. F. and G. pages 97–100. Ning. 2007. In Proceedings of the First International Conference on Pervasive Computing (Pervasive). Sarma. Security and privacy aspects of low-cost radio frequency identification systems. 2004. Wang. E. Norman. [106] S. Security in Pervasive Computing. IEEE Communications Letters. B. 10(4):236–238. Vita. [108] Q. [107] C. 2001. Wang. Weis. L. De Vita and G. 11(12):946–948. and J-B. Park. . Chen. H-G. In Proceedings of the IEEE International Conference on Communications (ICC). and S. 1(3):260–277. 2002. [102] H. Grouping based bit-slot aloha protocol for tag anti-collision in RFID systems. 37(7):627–629. [104] L. [100] G. [99] P. De. S. Placement of multiple RFID reader antennas to maximize portal read accuracy.

Yoon and N. Vaidya. Zhang and S. 2004. In Proceedings of the IEEE Military Communications Conference (MILCOM). Huang. C. 2005. Jin. Min. B. Zhou. Chen. and Z-Y. C. W. and M. Optimize the power consumption of passive electronic tags for anti-collision schemes. pages 243–250. Lin. In Proceedings of the 5th International Conference on ASIC.BIBLIOGRAPHY 184 [110] W. Jin. 2010. [114] X. D. D. pages 89– 92. Huang. Vojcic. [113] N. [112] L. In Proceedings of Second International Conference on Future Generation Communication and Networking. 2008. In Proceedings of the Fifth International Conference on Grid and Cooperative Computing Workshops. [111] K-M. A density-based algorithm for redundant reader elimination in a RFID network. C. H. pages 463–469. pages 950–955. In Proceedings of the 2004 International Symposium on Low Power Electronics and Design (ISLPED). Yu. and H. 2004. Zhang and Z. Zhang and B. [115] F. pages 1–24. C. pages 357–362. . RFID reader collision problem: performance analysis and medium access. Wicker. Yu. Binary search algorithms with interference cancellation RFID systems. 2003. Wang. pages 1213–1217. Evaluating and optimizing power consumption of anti-collision protocols for applications in RFID systems. Integration of RFID into wireless sensor networks: Architectures. 2006. opportunities and challenging problems. Wireless Communications and Mobile Computing. [116] F. How to distribute sensors in a random field? In Proceedings of the 3rd International Symposium on Information Processing in Sensor Networks (IPSN). Hao. Zhou.

Sign up to vote on this title
UsefulNot useful

Master Your Semester with Scribd & The New York Times

Special offer for students: Only $4.99/month.

Master Your Semester with a Special Offer from Scribd & The New York Times

Cancel anytime.