Post on 04-Jun-2018
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
1/165
AN ENERGY EFFICIENT DISTRIBUTED CLUSTER
BASED SELF ORGANISING ALGORITHM FOR AD-HOC
DEPLOYED WIRELESS SENSOR NETWORKS
By
Prasanna Sankalpa Gamwarige
A THESIS
This thesis is submitted to the Department of Electronic
and Telecommunication Engineering at the University of
Moratuwa in partial fulfilment of the requirements for
the Degree of Doctor of Philosophy.
October 2010
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
2/165
DECLARATION BY CANDIDATE
I certify that this thesis does not incorporate without acknowledgement any material
previously submitted for a Degree or Diploma in any University; and to the best of my
knowledge and belief it does not contain any material previously published or written by
another person except where due reference is made in the text.
Prasanna Sankalpa Gamwarige
Certified by:
Dr. Chulantha KulasekereThesis Supervisor
October 1, 2010.
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
3/165
UNIVERSITY OF MORATUWA
A thesis submitted to the Department of Electronic and Telecommunication
Engineering at the University of Moratuwa in partial fulfillment of the
requirements for the Degree of Doctor of Philosophy
AN ENERGY EFFICIENT DISTRIBUTED CLUSTER BASED
SELF ORGANISING ALGORITHM FOR AD-HOC DEPLOYED
WIRELESS SENSOR NETWORKS
Prasanna Sankalpa Gamwarige
Approved:
Prof. Saman HalgamugeUniversity of Melbourne
Prof. Keerthi WalgamaUniversity of Peradeniya
Prof. Dileeka DiasUniversity of Moratuwa
Dr. Ajith PasqualUniversity of Moratuwa
Dr. Chulantha KulasekereUniversity of Moratuwa
October 1, 2010.
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
4/165
Abstract
Wireless sensor networks (WSNs) consist of a large number of inexpensive, low-power,
sensors that can be placed in an ad hoc fashion to form a data gathering network. Subse-
quent to the sensor node deployment, the nodes will self-organize themselves to periodically
collect reliable information from the environment to a central location called base station
(BS). Once the nodes are deployed, upgrading and maintaining them is not practical. In
such a scenario, the main concern would be the optimal utilization of the sensor energy,
so that the entire sensor bed lasts as long as possible gathering useful information. Inter
node communication for network organization and information gathering requires the mostenergy. Therefore, it is necessary to manage these activities in an energy efficient manner to
optimize the lifetime of the sensor network. This research focuses on finding energy efficient
methods of operating the sensor bed such that the lifetime is maximally extended.
Distributed clustering provides an effective way for self-organizing the wireless sensor
networks for periodic data gathering applications. The research identifies the most positive
and negative aspects of the currently available distributed clustering algorithms. Based on
these findings, the research proposes a new energy efficient distributed clustering algorithm
where the cluster heads (CHs) are selected based on relative residual energy level of sensors.
Further, the cluster boundary determination and cluster head role rotation is governed by
the cluster heads residual energy level. The algorithm favors more powerful nodes over the
weaker ones thus makes local energy balancing to prolong the lifetime of the entire sensor
network at a very low energy overhead. The proposed algorithm has realized near ideal
local energy balancing. The proposed algorithm is also extended to achieve global energy
balancing by introducing a mix strategy of communication (multi-hop and direct) from
cluster head to base station.
The research shows that the algorithm performance is in line with the desired objectives
using analytical proofs to back the simulation test results. Further, the research proposes
an analytical framework in determining the cluster distribution of the presented algorithm.
Subsequently, the framework was extended to other similar types of distributed clustering
algorithms. Finally, the research proposes an analytical technique in finding optimum al-
gorithm parameters such as the cluster head message broadcasting range and cluster head
role rotation.
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
5/165
To
Arosha
the Love of My Life
iii
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
6/165
Acknowledgements
I am eternally indebted to my research supervisor Dr. Chulantha Kulasekere for his guid-
ance, support, and encouragement during this challenging research. His unique approach in
analyzing problems and finding novel solutions inspired me to widen my intellectual horizon.
My life has been enriched professionally, academically, and personally by working closely
with him.I am also thankful to Prof. Saman Halgamuge, the Assistant Dean of Melbourne School
of Engineering, University of Melbourne; Prof. Keerthi Walgama, the Director of Aca-
demic Affairs and former Head, Department of Engineering Mathematics, University of
Peradeniya; Prof. (Mrs.) Dileeka Dias of University of Moratuwa; and Dr. Ajith Pasqual,
the Research Coordinator of University of Moratuwa for their invaluable suggestions and
support at the thesis evaluation. My special thanks are also due to Prof. (Mrs.) Indra
Dayawansa of University of Moratuwa for her encouragement and support throughout this
research. Further, I like to acknowledge the support given by Eng. Kithsiri Samarasinghe,
Dr. Ranga Rodrio and all other staff of University of Moratuwa.
It was not an easy task to carry out a research of this nature, while being actively engaged
with the industry. Zone24x7 Inc, my employer, provided the support and flexibility I needed
to complete this research as expected. For this, I am grateful to Mr. Llavan Fernando, the
CEO, Mr. Manjula Dissanayake, the Vice President and the entire team of Zone24x7.
All the guidance, commitment, and perseverance I had, would not have made this thesis
possible, if not for the support, endurance, and understanding of my family. Thus, my
utmost gratitude goes to my loving wife, two daughters, my parents and parents in-law.
Prasanna Sankalpa Gamwarige
University of Moratuwa
October 2010
iv
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
7/165
Contents
Acknowledgements iv
List of Figures ix
List of Tables xii
Nomenclature xiii
CHAPTER 1 Introduction 1
1.1 Current Challenges in Sensor Networks . . . . . . . . . . . . . . . . . . . . . 2
1.2 Direction of the Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Challenges in Cluster based Self Organization . . . . . . . . . . . . . . . . . 5
1.5 Scope of the Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.6 Organization of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
CHAPTER 2 Related Work 10
2.1 Overview of Energy Aware Communication Protocols . . . . . . . . . . . . 10
2.2 Wireless Sensor Network Clustering Algorithms . . . . . . . . . . . . . . . . 12
2.2.1 LEACH: Low Energy Adaptive Clustering Hierarchy . . . . . . . . . 13
2.2.2 LEACH-D: Low Energy Adaptive Clustering Hierarchy with Deter-ministic Cluster Head Selection . . . . . . . . . . . . . . . . . . . . . 15
2.2.3 SEP: A Stable Election Protocol for Clustered Heterogeneous Wire-
less Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.4 HEED: Hybrid Energy Efficient Distributed Clustering . . . . . . . . 16
2.2.5 ANTCLUST based Energy-Efficient Clustering Method for Data Gath-
ering in Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.6 EDAC: Energy Driven Adaptive Clustering Data Collection Protocol
in Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . 17
v
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
8/165
2.2.7 MEDIC: Medium-Contention Based Energy-Efficient Distributed Clus-
tering Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Anatomy of a Wireless Sensor Node . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Ultra Low Power Micro Controllers . . . . . . . . . . . . . . . . . . . 20
2.3.2 Low Power Wireless Transceivers . . . . . . . . . . . . . . . . . . . . 20
2.3.3 Battery and Optional Energy Harvesting Techniques . . . . . . . . . 22
2.4 Sensor Network Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4.2 Energy Consumption Model . . . . . . . . . . . . . . . . . . . . . . . 27
2.4.3 Lifetime of the Sensor Network . . . . . . . . . . . . . . . . . . . . . 28
CHAPTER 3 Proposed Energy Balanced Distributed Clustering Algorithm 30
3.1 Objectives of the EDCR Algorithm . . . . . . . . . . . . . . . . . . . . . . . 30
3.2 Overview of the Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.1 Cluster Head Candidacy . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.2 Cluster Head Selection . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2.3 Data Transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.4 Cluster Head Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 Algorithm Pseudo Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
CHAPTER 4 Performance Analysis of the EDCR Algorithm 40
4.1 Accuracy and Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2 Cluster Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2.2 Probability Density Function of Cluster Area, . . . . . . . . . . . 44
4.2.3 Derivation of Expected Cluster Density . . . . . . . . . . . . . . . . 49
4.2.4 Expected Number of Clusters,E[k] of a Rectangular Deployment Area 51
4.2.5 Expected Number of Clusters,E[k] of a Circular Deployment Area . 52
4.2.6 Average Distance between Neighboring Cluster Heads . . . . . . . . 53
CHAPTER 5 Optimization of the Control Parameters for EDCR Algo-
rithm 54
5.1 Optimum Cluster Head Candidacy Broadcasting Range,Ropt . . . . . . . . 54
5.1.1 Circular Deployment Area . . . . . . . . . . . . . . . . . . . . . . . . 57
5.1.2 Rectangular Deployment Area . . . . . . . . . . . . . . . . . . . . . 58
vi
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
9/165
5.2 Computation of Optimum Cluster Head Rotation Trigger Function Parame-
ter, copt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.3 Estimation of Second Degree Neighborhood Determining Parameter for a
Given Wireless Sensor Network Setup . . . . . . . . . . . . . . . . . . . . . 71
CHAPTER 6 Global Energy Balancing 78
6.1 EDCR in Multi-hop Network Setup . . . . . . . . . . . . . . . . . . . . . . . 79
6.1.1 Identification of Next-hop Cluster Head . . . . . . . . . . . . . . . . 80
6.1.2 Determination ofRopt and copt for EDCR-MH . . . . . . . . . . . . . 81
6.1.3 Limitations of the EDCR-MH Algorithms . . . . . . . . . . . . . . . 84
6.2 EDCR-EB Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.2.1 Determination ofRopt and copt for EDCR-EB . . . . . . . . . . . . . 92
6.2.2 Application Guidelines of EDCR-EB Algorithm . . . . . . . . . . . . 94
CHAPTER 7 Simulation Results 96
7.1 Comparison of Performance of EDCR Algorithm with Similar Class of Algo-
rithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
7.1.1 Results for the Free Space Model . . . . . . . . . . . . . . . . . . . . 98
7.1.2 Results for the Simplified Multi-path Fading Model . . . . . . . . . . 104
7.2 Cluster Distribution and Cluster Head Location in a Cluster . . . . . . . . 1107.3 Applicability of EDCR algorithm in Non Rectangular Deployment Regions 111
7.4 Accuracy of the Analytical Framework Proposed in Finding R for an Ex-
pected Cluster Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.5 Validation of the Analytical Techniques for Determining the EDCR Algo-
rithm Parameters for Maximizing the Network Lifetime . . . . . . . . . . . 120
7.5.1 Validation ofRopt . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
7.5.2 Validation ofcopt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
7.6 Performance Evaluation of EDCR-MH . . . . . . . . . . . . . . . . . . . . . 124
7.7 Performance Evaluation of EDCR-EB . . . . . . . . . . . . . . . . . . . . . 125
CHAPTER 8 Conclusion and Future Direction 129
References 134
APPENDIX A 144
A.1 Expected Distance between Two Immediate Neighboring Nodes . . . . . . . 144A.2 Energy Optimum Cluster Head Location in an Arbitrary Cluster . . . . . . 144
vii
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
10/165
A.3 Global Re-clustering or Local Cluster Head Role Delegation . . . . . . . . . 145
viii
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
11/165
List of Figures
1.1 Browsing physical environment over the Internet . . . . . . . . . . . . . . . 2
2.1 Non uniform cluster formation in LEACH . . . . . . . . . . . . . . . . . . . 14
2.2 Anatomy of a wireless sensor node . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 Current consumption of CC1101 transceiver for different Tx power output
levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4 Radio energy dissipation model . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.5 Number of live sensor nodes at the end of each round . . . . . . . . . . . . 29
3.1 Second degree neighborhood . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 Re-clustering sequence of cluster headi when Etresi i . . . . . . . . . . 373.3 State change of a sensor node . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.1 Smallest possible cluster size3R2
2 . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 Largest possible closed packed cluster size 33R2
2 . . . . . . . . . . . . . . . 46
4.3 Cluster area more than 33R2
2 create uncovered region shaded in gray . . . 46
4.4 Proof ofPB( > 33R2
2 ) 0 . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.1 Different scenarios of rectangular area with BS at the centre . . . . . . . . . 58
5.2 Different scenarios of rectangular area with BS at the centre of the long side
of perimeter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.3 Typical Jtotal vs R curve for a given WSN . . . . . . . . . . . . . . . . . . . 61
5.4 Lifetime of WSN with respect to the change ofc . . . . . . . . . . . . . . . 63
5.5 Round robin CH selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.6 Constraints for maximum inter CH distance . . . . . . . . . . . . . . . . . . 72
5.7 Movement ofa and b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.1 Single-hop cluster based WSN organization . . . . . . . . . . . . . . . . . . 78
6.2 Multi-hop cluster based WSN organization . . . . . . . . . . . . . . . . . . 79
ix
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
12/165
6.3 Area where CHs would never relay through another CH . . . . . . . . . . . 81
6.4 Determination of . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.5 Expected number of relay packets by an average CH, p= A2A1 . . . . . . . . 85
6.6 WSN deployment regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
7.1 Number of Live Nodes vs. Data Transmission Rounds for the simulation
experiments using free space communication model - FS1 . . . . . . . . . . 100
7.2 Number of Live Nodes vs. Data Transmission Rounds for the simulation
experiments using free space communication model - FS2 . . . . . . . . . . 101
7.3 Number of Live Nodes vs. Data Transmission Rounds for the simulation
experiments using free space communication model - FS3 . . . . . . . . . . 101
7.4 Number of Live Nodes vs. Data Transmission Rounds for the simulation
experiments using free space communication model - FS4 . . . . . . . . . . 102
7.5 Number of Live Nodes vs. Data Transmission Rounds for the simulation
experiments using simplified multi-path fading communication model - MF1 105
7.6 Number of Live Nodes vs. Data Transmission Rounds for the simulation
experiments using simplified multi-path fading communication model - MF2 106
7.7 Number of Live Nodes vs. Data Transmission Rounds for the simulation
experiments using simplified multi-path fading communication model - MF3 107
7.8 Number of Live Nodes vs. Data Transmission Rounds for the simulation
experiments using simplified multi-path fading communication model - MF4 108
7.9 Node distribution among all clusters . . . . . . . . . . . . . . . . . . . . . . 110
7.10 EDCR performance under Case NR1 . . . . . . . . . . . . . . . . . . . . . . 112
7.11 EDCR performance under Case NR2 . . . . . . . . . . . . . . . . . . . . . . 113
7.12 EDCR performance under Case NR3 . . . . . . . . . . . . . . . . . . . . . . 114
7.13 E[k]A vs for different R - 200 200m2 square deployment area . . . . . 118
7.14 E[k]A
vs for different R - 100m radius circular deployment area . . . . . 1187.15 Typical Average Lifetime vs R curve for a given sensor network requirement 121
7.16 Different Lifetime curves of a WSN for different c: Case 1 . . . . . . . . . . 123
7.17 Different Lifetime curves of a WSN for different c: Case 2 . . . . . . . . . . 123
7.18 Lifetime comparison EDCR and EDCR-MH . . . . . . . . . . . . . . . . . . 124
7.19 Far end nodes die first with EDCR . . . . . . . . . . . . . . . . . . . . . . . 126
7.20 Nodes close to BS die first with EDCR-MH . . . . . . . . . . . . . . . . . . 126
7.21 Lifetime comparison between EDCR, EDCR-MH and EDCR-EB . . . . . . 127
7.22 First set of nodes die irrespective of node location in EDCR-EB . . . . . . . 127
x
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
13/165
A.1 Cluster head location in global re-clustered cluster . . . . . . . . . . . . . . 146
A.2 Cluster head location in local cluster head role rotation . . . . . . . . . . . 147
xi
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
14/165
List of Tables
7.1 Summary of results for the free space model (Unit: Number of data trans-
mission rounds) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.2 Summary of results for the multi-path fading model (Unit: Number of data
transmission rounds) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1097.3 Distribution of member nodes among different clusters . . . . . . . . . . . . 110
7.4 Sample network deployment requirements for EDCR . . . . . . . . . . . . . 116
7.5 Comparison of actual average number of clusters and expected value of it for
EDCR algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.6 Comparison of actual average neighbor cluster head distance and theoreti-
cally expected value, DCHCH . . . . . . . . . . . . . . . . . . . . . . . . . 119
7.7 Comparison of the average lifetime with Ropt against actual maximum lifetime121
xii
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
15/165
Nomenclature
ampfs Radio propagation attenuation constant in Free Space model.
ampmp Radio propagation attenuation constant in Multi-path Fading model.
d0 The distance differentiates the Free Space propagation effect and Multi-path
Fading propagation effect in Simplified Multi-path Fading model. This is
given by
d0= ampfsamp
mp
.
Eelec The energy spent on transmitter and receiver circuits in signal processing of
one bit.
EDA Energy cost of data aggregation.
Bit length of a data packet.
Pi [0, 1] represents the relative position of the nodeiwith respect to the othernodes in its neighborhood in terms of its residual energy level.
Nj The set of sensor nodes within a neighborhood of radius from node jexcluding the node j .
H The set of all cluster heads at a give moment.Mi The set of member nodes in a cluster headed by cluster head i including
itself.
SNi Cluster head node is second degree neighborhood.Et
resi Residual energy of nodei at any given time instance t.
Etresit=
Residual energy of nodei at time t = .
i Dynamic energy threshold value of a given cluster head nodeiwhich becomes
a cluster head at time t = . When its residual energy drops below this
value, it calls for a new cluster head selection phase with the help of the base
station.
dist(x, y) Distance between nodes x and y .
PRxi,j The received signal strength of the signal transmitted by nodei at the nodej.
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
16/165
PTxi The transmitted signal strength of a data packet by node i.
R Cluster head candidacy broadcasting range.
Ropt The value ofR which will minimize the total data gathering energy of one
round.c The cluster head role rotation triggering dynamic energy threshold level
calculation parameter.
copt The value ofc which will maximize the sensor network lifetime.
E[k] Expected (average) number of clusters for a planned wireless sensor network
setup.
DCHCH Expected (average) distance between two neighboring cluster heads.
(di,j) Compressibility of the data of nodej at nodei due to the correlation of data
of node i and j . 1 (di,j) 0. Exponential data correlation model coefficient such that(di,j) = 1 edi,j . Deployed sensor node density. Assumed these nodes are uniform randomly
deployed in a given area resulting a Poisson point distribution with density
.
N Total number of nodes deployed in a given areaA. N =A.
NHi Set of all the neighbor cluster heads of a given cluster head node i.dTH Cluster head nodes whose distance to base station is less thandTHwould not
relay through another cluster head. This is used with EDCR-MH algorithm
to save the energy of cluster head closest to base station by reducing the
burden of serving closest cluster heads who can directly reach base station
without incurring much energy cost.
ECH The total energy spent by a cluster head in a given data transmission round
for useful work.
EnonCH The total energy spent by a non cluster head node in a given data transmis-
sion round for useful work.
ECHoh The energy overhead that a cluster head node has to spend in each cluster
setup phase.
EnonCHoh The energy overhead that a non cluster head node has to spend in each
cluster setup phase.
xiv
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
17/165
Ti This is an energy level calculated at the beginning of any new cluster forma-
tion phase in EDCR-EB algorithm. Given cluster head i stops forwarding
any incoming relay packets at this pre calculated energy level.
ai
This is used in calculatingTi
for each cluster headi making global energy
balancing.
xv
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
18/165
CHAPTER 1
Introduction
Sensor networks have become a popular research area during the past decade with the
increase in availability of sensor nodes in the market [1]. This is a direct result of recent
developments in inexpensive and low power micro sensor technology, radio communication
electronics and processors. Currently these sensor nodes (also known as motes) are produced
targeting various military and civilian applications [1]-[2]. Intruder detection, replacement
for anti-personnel land mines, and sniper localization systems are some identified potential
applications in the military domain. DARPA SensIT [3], DARPA NEST [4] are some
projects sponsored by Department of Defense of USA in this area. Hazard environment
monitoring [5], habitat monitoring (E.g. Micro climate monitoring at James Reserve, bird
nest monitoring at Great Duck Island)[6], browsing physical environment (Sensor Network
Macroscope [7], Figure 1.1), disaster monitoring systems [8], structural health monitoring
[9], building indoor environment monitoring [10], road traffic monitoring [11] and agriculture
[12] are some possible areas of sensor networks in civil applications. Major Semiconductor
companies like Intel have collaborated with many leading research institutions in exploring
different aspects of sensor networks including potential applications and unsolved issues
[13]. At the same time, giant software companies like Microsoft have shown interest in the
sensor networks field as seen in [14].
In general, sensor nodes used in large wireless sensor networks are less reliable and in-
accurate compared to their high end macro sensor counterparts. However, reliability and
usefulness of such sensor nodes can be improved using data aggregation from multiple sen-
sors [15]. This has led to applications using wireless sensor networks with hundreds to
thousands of nodes being used to achieve high reliability. These sensor nodes are equipped
with low energy batteries. This research considers an extreme use of wireless sensor net-
works. For such an application environments, once sensor nodes are deployed, upgrading or
servicing of a malfunctioning node due to technical faults would be prohibited. Similarly,
1
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
19/165
Figure 1.1: Browsing physical environment over the Internet
the replacement of batteries is also not possible [16]. As a result, lifetime of a sensor node
will come to an end as it exhausts its available battery energy. Hence, conserving battery
energy is a critical need in a wireless sensor network. There are many other factors that
limit the lifetime of a sensor node. However, this research focuses on the battery energy op-
timization and the factors that influence such optimization. Radio communication is found
to be the largest energy consuming factor of a sensor node. Therefore, if one requires to
prolong the lifetime of a sensor node, an energy efficient data transmission mechanism is the
key as explained in [17]. In what follows, current challenges that are faced by researchers
in the wireless sensor network area are discussed.
1.1 Current Challenges in Sensor Networks
Researchers have identified many unsolved issues in wireless sensor networks [1],[17]-[18].
However, the key concept that many research have focused on is the limited energy in a
sensor node and the rapid consumption of energy depending on the network design [19]-
[41]. In general, existing sensor nodes equipped with 2 AA batteries, each one with about
2000 mAhr at 1.5 V energy, at the time of deployment, will result in the average life time
ranging from a couple of days to an year which depends on the mode of operation. However,
in reality, these sensors are placed in remote locations and the life time depends on activities
2
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
20/165
such as the inter node communication which drains the battery power. Increase of battery
energy density, harness of reusable energy and minimization of energy usage are different
ways of addressing the lack of adequate battery power during operation. Increasing battery
density will increases the physical size of the sensor node and harnessing reusable energy
will make the sensor less cost effective. Hence, both these options are deemed undesirable.
The only viable option remaining would be to redesign the network and the communication
protocol within this network. This is more cost effective and an attractive solution.
Most sensor network applications require hundreds or even thousands of nodes being
deployed in ad-hoc fashion to collect data from the distributed nodes to a central location.
This ad-hoc deployment of nodes prevents the pre-planning of the network organization,
requiring it to self organize. These self organizing algorithms must be distributed and
scalable to accommodate dynamic changes in the network such as death of existing nodes
and addition of new nodes. Further, the self organizing algorithms must also be aware of
the energy limitation of the network and minimize energy overhead. The direction of this
research is presented in the next section.
1.2 Direction of the Research
The current research attempts to address the energy consumption problem and the self
organization issue of ad-hoc deployed large wireless sensor network to periodically collect
data from the sensor field to a central location called base station (BS). Energy consumption
problem is addressed by minimizing the use of energy, especially in radio communication.
In the current research context of wireless sensor networks existing radio communication
protocols such as IEEE 802.11 failed to be applicable for the following reasons. These pro-
tocols do not address unnecessary drain of energy which can result due to over hearing,
packet collisions, cross communication, and overhead of control, duplicate, and redundant
data packets. Additionally, communication protocols such as 802.11 are developed targeting
applications with one-to-one (unicast) communication, one-to-many (multicast) communi-
cation, and one-to-all (broadcast) communication. On the other hand, most of wireless
sensor application requirement is based on collecting data from the distributed sensors to
one central location i.e. many-to-one. This is known as convergecast communication [42].
There are different data transferring techniques found in literature which addresses the
convergecast communication scenario. Direct Transmission [19]-[20], Minimum Transmis-
sion Energy Multi-hop Routing [19]-[20] and Clustering [19]-[35],[43] are the main tech-niques. Direct Transmission tends to deplete the energy of nodes which are at a distance
3
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
21/165
from the base station far more rapidly compared to those nodes that are closer to the base
station resulting long range transmission. Direct transmission also suffers with hidden node
problem [44], considerable collisions and retransmissions resulting in waste of energy. On
the other hand, in Minimum Transmission Energy Multi-hop Routing, a node transmits to
its closest neighbor node in the direction of the base station. This transmission is continued
via closest neighbor in the direction of base station until this packet reaches the final des-
tination. Here, the nodes that are in close proximity to the base station tend to die much
faster than nodes found far away. This is a direct result of the increased load handling of
a sensor node closer to the base station in relaying large amount of data packets to the
base station. Further, multi-hop routing requires all nodes to keep idle listening resulting
in additional burden being placed on all nodes.
A final conventional protocol for wireless sensor networks is clustering, where nodes are
organized into disjoint clusters, in such a way, that each cluster consists of one cluster head
(CH) and multiple member nodes [16]. These member nodes communicate with the local
cluster head and these cluster heads transmit the data to the global base station where it
is accessed by the end-user. This greatly reduces the distance that nodes need to transmit
their data, since typically the cluster head is close to all the nodes in the cluster. The
current research will primarily concern itself with clustering and the protocols related to
clustering, to arrive at an energy efficient communication algorithm that will prolong thelifetime of the entire sensor network.
1.3 Clustering
Most of the recent research has identified cluster based node organization and data aggrega-
tion techniques as the best methods to answer issues related to energy aware self-organizing
ad-hoc networks due to the following reasons [16]. The first reason is, the energy consump-
tion in wireless data transmission scales proportionally to the nth power of the distance
between transmission and receiving nodes. Therefore, cluster-based data gathering mech-
anisms effectively saves energy by reducing the required transmission distance of most of
the nodes. Member nodes of a cluster will expend a small amount of energy to transmit
information to the nearby cluster head who will aggregate this information and undertake
to transmit it to the distant base station. The extra energy expended by the cluster head in
aggregating is minimal, compared to the collective energy that would be expended by each
node if they were to directly contact the base station. The second reason is, the clusterbased data gathering scenario can also save energy by reducing data collisions. The next
4
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
22/165
reason is, clustering coupled with Time Division Multiple Access (TDMA) protocol can
reduce the energy needed to keep the regular member node receiver circuits on for idle lis-
tening [45]. In general, physically close sensors have highly correlated data. Hence, it is also
possible to reduce the transmission of redundant data by cluster based data aggregation
with relatively high data compression of correlated data. This is also a reason for clustering.
Further, clustering helps to route aggregated data of each cluster to base station through
an overlay among cluster heads which require lesser amount of total transmission energy.
Cluster head nodes consume more energy than other nodes, as a result of its data
aggregation function. These functions include receiving data from its cluster members,
fusing data to reduce the size of a packet, and sending the aggregated data to a base station.
As a result of rapid energy dissipation of cluster heads, they tend to die much faster than
non cluster head nodes. One can extend the lifetime of a cluster head node by specifically
supplying them with more energy or evenly rotating the cluster head role based on the
residual energy. In literature, these high energy nodes are called advanced nodes. It is
possible to incorporate several advanced nodes in the sensor bed at the time of deployment.
However, selection of advanced nodes as cluster heads during normal operation is a much
more difficult task. In practice, the sensor networks are setup in an ad-hoc fashion, hence
conventional clustering algorithms would perform poorly. The second choice is to rotate the
cluster head role among all sensors. This would help to evenly distribute the burden of highenergy requirement of the cluster heads task among all nodes which will result in an even
lifetime for all nodes. This has led to most wireless sensor network organizing protocols to
use dynamic clustering, where the cluster head role is rotated.
1.4 Challenges in Cluster based Self Organization
Any wireless sensor network clustering algorithm faces two challenges [46]. The first is
how should clusters be formed?, the second is how many clusters are required? The first
question includes two aspects: how to select the cluster heads? and how to associate a
non cluster head node to a cluster head? Based on how these questions are answered,
existing clustering schemes can be classified as follows. Clustering scheme can operate as
centralized (e.g. BCDCP [25], EGSOM [47]) or distributed (e.g. LEACH [20]); static or
dynamic (e.g. ANTCLUST based [26]); a scheme can be applicable only for homogeneous
energy networks (i.e. all nodes in the network have same level of energy at deployment) (e.g.
LEACH) or even for a heterogeneous energy network (i.e. during initial deployment nodeshave different amounts of energy. For example, application of a new clustering algorithm
5
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
23/165
to an existing network or addition of new nodes to an existing network. e.g. HEED [27]);
the cluster head selection is weight 1 independent (i.e. randomized, e.g. LEACH) or weight
associated (e.g. HEED); procedure for cluster head selection can be finalized in one step
(e.g. LEACH) or iteratively (e.g. HEED, MEDIC [48]). Each of the above categories has
their own advantages and disadvantages. In general, any algorithm with complex control
messaging will have overhead of control and coordination mechanism which will increase
the energy consumption. However as a result of this complex control messaging, all nodes
will die at the same time but relatively sooner.
Due to the low complexity, good feasibility, and high effectiveness, the class of dy-
namic, distributive, and randomized (DDR) clustering algorithms are promising in provid-
ing energy-efficient, load balanced, scalable and robust communication in wireless sensor
networks. This is the main reason that LEACH [19] and its derivatives (such as SEP [24])
have attracted immense attention and have become a well studied and popularly referred
baseline in the current research context [46]. However, DDR class of algorithms such as
LEACH have following issues.
1. Poor performance in heterogeneous energy networks
2. Non-uniform cluster formation
3. High degree of uncertainty in producing required number of cluster heads
4. Issue related to single cluster head serving the entire wireless sensor network
5. A node with insufficient residual energy may be chosen as a cluster head when neigh-
boring nodes with more battery power is available
Complex cluster setup algorithms like HEED has addressed most of the above issues at a
cost of high energy overhead resulting a negative impact to life time. In addition, complex
algorithms like ANTCLUST base have unacceptable assumptions such as location awareness
of nodes using GPS or some form of localization technique which would increase the cost
and energy requirement.
Most of the existing algorithms such as LEACH, HEED and ANTCLUST use time
driven cluster head role rotation mechanisms. In this method, the role of cluster head will
be changed after a constant predetermined number of data gathering rounds i.e. the cluster
head role is rotated after every T period which is predetermined. Determination of the
1Typically weight is computed based on residual energy, distance, size of neighborhood etc. dependingon the algorithm.
6
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
24/165
optimum time period T is crucial. TypicallyT is found at the design stage and does not
change at run time. Since T is fixed, re-clustering cannot be done adaptively to accom-
modate unforeseen environmental changes which would influence the depletion of battery
energy of a cluster head node. As a result, this method is more vulnerable to environmental
changes. However, use of a heuristic approach with dynamic local information in deciding
the cluster head role rotation would result in a more robust system. As a result, researchers
feel, residual energy based cluster head role rotation algorithms would perform better. In
this method, cluster head rotation trigger threshold level is different for different nodes,
and different for the same node at different instances, depending on the node energy level.
The dynamic nature of energy based cluster head rotation ensures that even if all the nodes
have only little energy left, the system still works. Algorithms like EDAC [35] proposes an
energy driven cluster head role rotation mechanism. Further, authors of [35] have shown
that energy based cluster head role rotation is far superior to predetermined time based
method, especially in heterogeneous energy sensor networks, and networks with varying
traffic patterns including variable size data packets. EDAC is an algorithm which extends
the LEACH with energy driven cluster head rotation instead of predetermined time driven
method. However EDAC has only addressed this issue from LEACH. Hence, EDAC also
suffers with the problems found in LEACH resulting in bad life time performance of the
EDAC algorithm most cases. According to the authors knowledge, existing literature doesnot provide a method of finding a suitable energy level to rotate the cluster head role to
optimize the wireless sensor network lifetime. This too is an important parameter if one
wants to extend the lifetime of a wireless sensor network.
Its a common design level requirement to plan and set the parameters of the clustering
algorithm of an ad-hoc wireless sensor network to produce on average E[k] number of
clusters in the given area of interest, where each cluster would have E[n] number of nodes.
This requires N(=E[k]
E[n]) number of nodes to be uniform randomly deployed in the
area of interest. This indicates the importance of determining the cluster distribution with
respect to algorithm parameters at the system planning stage of a given application.
Selective replacement of batteries or nodes is not practical for a randomly deployed
ad-hoc wireless sensor network. Therefore, it is highly desirable to ensure all nodes deplete
their energy at the same pace, so that all nodes die together and a new wireless sensor
network could be deployed as a replacement. In addition, it is highly desirable to maximize
the time gap between such deployments to reduce the total cost. In what follows, the scope
of this research is defined.
7
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
25/165
1.5 Scope of the Research
A brief discussion of the research challenges in the ad-hoc deployed energy constrained
wireless sensor networks was given in Sections 1.1-1.4. Given this background, the scope of
this research can be defined as follows:
Improved energy efficient distributed clustering algorithm
This research proposes an improved energy efficient distributed cluster based self organi-
zation algorithm for ad-hoc deployed wireless sensor network for periodic data gathering.
This algorithm will address the issues identified in existing algorithms.
The proposed dynamic distributed clustering algorithm selects the node with the highest
residual energy in a given local neighborhood as the cluster head by using local information.
Rest of the nodes join the closest and the highest energy available cluster head from its
neighborhood. The cluster head role is rotated based on residual energy level of cluster
heads. Thereby the system guarantees local energy balancing. Further, proposed cluster
head role rotation should be done efficiently to reduce the overall energy overhead while
guaranteeing fair local energy balancing.
The required number of clusters should be well distributed in the field while each cluster
head should be positioned at a location in a given cluster which would minimize the overall
energy cost of the cluster. These should be realized while keeping the complexity of the
algorithm minimum, as it would help to reduce the overhead of control and coordination
messages of the entire wireless sensor network.
Further, the proposed local energy balancing algorithm will be extended to realize global
energy balancing, enabling it as an ideal algorithm for large ad-hoc deployments in unreach-
able locations.
Analytical framework to prove the effectiveness of the above algorithm
This research proposes an analytical framework to prove the effectiveness and behavior of
the proposed algorithm backed by simulation experiment results. This framework includes
the analysis of cluster head distribution. The proposed framework can be easily extended
to analyze the behavior of similar classes of other algorithms too.
Design framework to optimize the algorithm parameters to increase lifetime
Further, this research proposes an analytical framework in finding optimum values of thealgorithm parameters which maximize the entire wireless sensor network lifetime. These
8
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
26/165
analytical results are backed by simulation results.
1.6 Organization of the Thesis
The rest of the thesis is organized as follows. Chapter 2 presents related literature in
energy aware self organization of ad-hoc deployed wireless sensor networks for periodic data
gathering applications. This chapter also discusses the sensor network model to be used in
this work. Chapter 3 presents the core of the proposed energy aware distributed clustering
algorithm, which achieves perfect energy balancing among nodes in a local neighborhood
with minimum energy overhead. The performance of the proposed algorithm is analyzed in
Chapter 4. This analysis covers the correctness, complexity and behavior of the algorithm
including an analysis of the cluster density for given set of algorithm parameters. Chapter 5presents analytical techniques in finding the proposed algorithm parameters to optimize the
sensor network lifetime. Chapter 6 extends the proposed algorithm to achieve global energy
balancing. Chapter 7 presents the simulation results of the proposed algorithm. Finally,
Chapter 8 concludes the thesis and provides directions for future research.
9
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
27/165
CHAPTER 2
Related Work
This chapter will first summarize the related research work in energy aware ad-hoc deployed
wireless sensor network organization targeting an application, which requires periodic data
gathering from sensor nodes to base station. Based on these related work, a wireless sensor
network model will be built. This model will be used in the rest of the research.
2.1 Overview of Energy Aware Communication Protocols
The current research considers an application of field monitoring with an ad-hoc deployed
stationary wireless sensor network, which gathers data from all sensor nodes to a base
station at regular intervals using low power, low bit rate radio modules [42] such as TI
CC1100 [49]. These sensor networks consist of hundreds to thousands of ad-hoc deployed
sensor nodes, each powered by an energy limited batteries. It is assumed that sensor nodes
are not maintained after deployment, meaning, energy depleted nodes are not selectively
replaced. However in such a situation, additional sensor nodes can be deployed if necessary
in an ad-hoc fashion. Structural health monitoring, habitat monitoring, hazard environment
monitoring and micro weather stations are some applications that fall into this type of
wireless sensor networks.
Such wireless sensor networks share many similarities of communication technologies
with conventional ad-hoc deployed wireless networks. Yet, there are some vital differences
between these two networks. Dense deployment, energy constraint and many to one com-
munication are some of the notable differences. As a result, the protocols developed for
traditional ad-hoc deployed wireless networks are not necessarily well suited to the unique
features of wireless sensor networks [48]. Therefore, energy efficient self organizing and data
gathering are the major aspects to be addressed in a wireless sensor network protocol of
this nature [50].
10
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
28/165
A variety ofenergy aware self organization and data communication protocols are given
in literature. One such technique is direct transmission where all nodes directly communi-
cate with the base station. This is a sensible method as low power radio transceivers such as
CC1100 has a comparable energy consumption for both Receiving (Rx) and Transmitting
(Tx) modes (CC1100 Rx 15 mA, Tx 0 dBm 15 mA / +10 dBm 30 mA [49]). Unfortunately
this method is not suitable for densely deployed wireless sensor network applications, as
it would increase the packet collisions which will result in high energy dissipation due to
retransmissions, hidden node problem [44] etc. Further, as a result of long range transmis-
sion, this method has a negative impact on nodes located far away from the base station
compared to the nodes located close to the base station.
An alternative to direct transmission is minimum transmission energy, which is an ex-
tension to traditional shortest path first or minimum hops routing [48]. In this method,
multi-hop is preferred to a single-hop, if the multiple short distance transmissions costs less
energy than a single long distance transmission. In minimum transmission energy method,
the nodes that are in close proximity to the base station tend to die much faster than nodes
found far away from the base station [19]. This is a direct result of the increase in load
handling by a sensor node closer to the base station, due to assisting all other nodes in
relaying their data packets to the base station. Further, multi-hop communication requires
all nodes to keep idle listening, resulting in additional undue burden for all nodes.Third technique is chain based node organisation method. PEGASIS [50] is one of the
well referred derivatives of this technique. Formation of the optimum chain of nodes is
similar to well known traveling salesman problem [51]. PEGASIS produces a near optimal
chain by sensor nodes themselves in distributed manner using a greedy algorithm [52] based
on node location information and global knowledge of the network. Even though chain
formation algorithm is complex, it is done only once for the entire life time. Once this
chain is formed, each node is only communicating with its neighbor in the chain. All nodes
will aggregate their data with the incoming data packet from the neighboring node before
sending to the next node in the chain. At any given round, there will be one leader in the
chain, whose responsibility is to directly transfer the aggregated final data packet to the base
station. In other words, PEGASIS uses only one node in the chain to transmit to the base
station, instead of multiple nodes with previous two types. Leadership is rotated among all
nodes, such that, each node gets only one chance per round. Following issues are identified
with PEGASIS. It requires all nodes to have a global knowledge of the entire network.
Hence, the complexity of the algorithm increases with the increase size of the network.
11
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
29/165
Therefore, there is a clear scalability issue. Similarly, the algorithm depends on the location
information. This would increase the hardware complexity of nodes. PEGASIS algorithm
expects all nodes to be perfectly synchronized for its operation. Further, algorithm assumes
perfect data compressibility i.e. all bit data packets can be compressed to a single data
packet of. However, this may not be a reasonable assumption. When this assumption is
relaxed, the total transmission energy of one bit would be significant, compared to direct
or minimum transmission energy method with perfect synchronization.
A final conventional energy aware protocol for wireless sensor networks is based on
clustering, where nodes are organized into disjoint clusters in such a way, that each cluster
consists of one cluster head and multiple member nodes [19]-[31]. These member nodes
communicate with the local cluster head and these cluster heads transmit the data to the
base station. Some of the advantages of cluster based wireless sensor network protocols as
given in [19]-[31] are as follows. Clustering greatly reduces the distance that a node needs
to transmit their data, since the cluster head is located close to all nodes in the cluster
compared to the base station. In general, physically close sensors have highly correlated
information. Further, clustering helps to reduce the transmission of redundant data by
cluster based data aggregation with relatively high compressibility of correlated data. In
addition, cluster based data gathering can save energy by reducing data collisions [16]. More
importantly, clustering couple with TDMA can reduce the energy consumption in keepingthe regular member node receiver circuits on for idle listening [45]. Finally, clustering helps
to routing through an overlay among cluster heads, which have a relatively small network
diameter.
2.2 Wireless Sensor Network Clustering Algorithms
As discussed in the previous chapter, cluster head consumes relatively large amount of
energy compared to regular nodes. Hence, a capable node has to be elected as the cluster
head. Further, cluster head role should be rotated among all sensor nodes. This would
help to evenly distribute the burden of high energy required cluster heads task among all
nodes, and have an equal lifetime for all nodes. What follows would give a brief description
of existing wireless sensor network clustering algorithms in line with cluster head selection
and rotation and their overall performance is given.
12
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
30/165
2.2.1 LEACH: Low Energy Adaptive Clustering Hierarchy
LEACH [19]-[20] is a distributed cluster based data communication algorithm. It uses the
hybrid CDMA/TDMA technique as the multiple access control mechanism. Each cluster has
its own Spread Spectrum code so that the interference between clusters is minimized. Forintra cluster communications, TDMA slots are assigned for each member to minimize media
contention. The LEACH algorithm periodically rotates the role of cluster heads among all
nodes to evenly distribute the energy dissipation rate. A pre-determined percentage of
sensor nodes become cluster heads in LEACH. The probability of a node to become a
cluster head is self determined in a manner, a sensor node, which has not become a cluster
head recently is more likely to be a cluster head. The nodes that are selected to become
cluster heads first advertise their candidacy to the rest of the sensor nodes. Hearing the
advertisements, each sensor node chooses the closest cluster head and registers itself as a
cluster member. The cluster head prepares the TDMA schedule, which assigns a time slot
for each member to periodically communicate with the cluster head and broadcast it among
its cluster members. Eventually, clusters are formed and periodic data gathering from nodes
start. During the data gathering phase, the sensor nodes periodically wake up, sense and
update the results in the -bit length data packets to the cluster head in the allotted time
slot before going back to sleep. Subsequently, each cluster head combines all-bit data into
a single -bit message and sends it directly to the base station.
LEACH has compared with Minimum Transmission Energy, Static Clustering and a
centralized dynamic clustering algorithm similar to LEACH named LEACH-C where cluster
heads are selected using the knowledge of all node locations and positions at the central
base station in [19]. According to the comparison based on the simulation results, LEACH
is far superior to Minimum Transmission Energy and Static Clustering. The performance of
LEACH is only 40% below LEACH-C. Based on these results LEACH has shown that it is
an attractive dynamic, distributed and randomized clustering algorithm for ad-hoc deployedwireless sensor networks. However, LEACH algorithm suffers with following drawbacks.
1. LEACH algorithm does not perform well in heterogeneous energy sensor networks [53].
These networks have nodes with different amounts of initial residual energy. This is a
direct result of the LEACH algorithm assuming a homogeneous energy sensor network
at the time of initiating the algorithm.
2. The algorithm does not produce well distributed cluster heads [54]. LEACH can
produce two or more adjacent cluster heads. This is illustrated in Figure 2.1.
13
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
31/165
Figure 2.1: Non uniform cluster formation in LEACH
3. Actual number of cluster heads produced by LEACH algorithm does not concentrate
in a small range around the expected number. Further, there may a be chance that
the entire sensor network being served with only one cluster head. This adverse effect
is a result of self election of a node as a cluster head using random number. This
effect increases the total energy consumption and reduces the effective lifetime of the
sensor network. [46]
4. Cluster head broadcasting message has to cover the entire wireless sensor network.
Hence this would require significant energy compared to covering a local neighborhood.
5. The cluster head rotation is carried out after a predetermined constant number of
normal operation rounds, i.e. in every T time units. If the number of data gathering
rounds is too small then there will be a large cluster setup overhead. On the other
hand if cluster setup happens after a large number of regular data gathering rounds,
then the existing cluster heads would not have enough energy to function as regular
sensor nodes after they relinquish the role of cluster head. Therefore, proper selection
of the number of data rounds before cluster head role rotation is crucial for the network
lifetime. Typically, this value is found at design stage and not changed in run time.
Hence, this method is not flexible in accommodating the unforeseen events, which
would affect the energy dissipation rate or the battery energy content of a cluster
head. As a result, this method is more vulnerable to such unforeseen events.
14
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
32/165
2.2.2 LEACH-D: Low Energy Adaptive Clustering Hierarchy with De-
terministic Cluster Head Selection
LEACH-D [22] extends LEACHs stochastic cluster head selection algorithm by a determin-
istic component of individual nodes residual energy compared to its initial energy at thetime of the deployment of sensor network. This has resulted in 25% lifetime improvement
of LEACH-D over LEACH as given in the simulation results [22].
However this deterministic component has failed to identify the relative energy level
of a node compared to its neighbors. Hence, LEACH-D has failed to get the service of
high energy nodes in favor of low energy ones as the energy demanding cluster heads in
all occasions. Further LEACH-D also suffers with the draw backs that have identified with
LEACH as its an extension of LEACH.
2.2.3 SEP: A Stable Election Protocol for Clustered Heterogeneous Wire-
less Sensor Networks
SEP [24] is also an extension to the LEACH algorithm. It proposes the use of a small
percentage of advanced nodes along with normal nodes. Then, it uses a technique to
allocate these advanced nodes as cluster heads more often when compared to normal nodes,
and thus try to prolong the overall network lifetime. The randomized rotation of cluster
heads is weighted by the proportion of extra initial battery energy of the nodes. This
weighting is such that SEP selects advanced nodes (1 + ) times more often than a normal
node, where is the extra energy content incorporated into the advanced nodes. Rest of
the SEP algorithm is identical to LEACH.
According to the simulation results presented in the [24], SEP has taken full advantage
of heterogeneity (i.e. extra energy of advance nodes), thus, the life time of the wireless
sensor network has increased by 26% compared to LEACH.
SEP-E [55] is an extension of SEP by introducing another node type named intermediate
nodes, which serves as a bridge between the advanced nodes and the normal nodes as
described above. Intermediate nodes are equipped with energy content between normal
nodes and advanced nodes. Number of intermediate nodes is higher than advanced nodes
and much lower than normal nodes. Rest of the SEP-E algorithm is identical to SEP.
Simulation results given in [55] indicate that there is a slight increase in the stability of
SEP-E over SEP and a significant reduction of the instability of SEP-E. This is due to
the introduction of the intermediate nodes to SEP-E, which acts as a bridge between the
advanced nodes and the normal nodes in SEP-E, thus lowering the instability region.
15
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
33/165
While SEP and its derivatives have a wireless sensor network lifetime increase compared
to LEACH, still they inherit many drawbacks identified with LEACH such as, random head
election that cannot guarantee the desired number of cluster heads be elected or the elected
heads be evenly positioned, cluster head candidacy needs to cover entire wireless sensor
network and fixed time based cluster head role rotation. Class of SEP algorithms cannot
be used in a true random heterogeneous sensor networks. (Note : The SEP and SEP-E
algorithms assume two and three types of energy nodes respectively. Hence, it cannot be
considered a strict homogeneous network. On the other hand it also cannot be considered
a heterogeneous network since random energies are not assigned to nodes.)
2.2.4 HEED: Hybrid Energy Efficient Distributed Clustering
HEED [27] periodically selects cluster heads according to a hybrid of their residual energy
and a secondary parameter such as node proximity to its neighbors or the node degree
(number of members assigned to a cluster head). HEED is also a distributed clustering
algorithm. HEED has eliminated the non uniform cluster forming problem that was ob-
served in LEACH and its derivatives. HEED requires cluster head announcement to cover
only a local neighborhood. Furthermore, HEED algorithm has the ability to perform in a
heterogeneous energy networks as it considers node residual energy in cluster head election.
HEED uses a complex weight based cluster setup procedure, where cluster head isselected with many round of iterations. This has adversely resulted in the communication
and coordination energy overhead during cluster setup. Hence, HEED itself admits in [27]
that LEACH protocol expends less energy in clustering and produces longer lifetime than
HEED. HEED too rotates cluster heads after a constant predetermined number of data
gathering rounds. Hence, same vulnerabilities faced by LEACH on fixed time based cluster
head rotation are applicable to HEED.
2.2.5 ANTCLUST based Energy-Efficient Clustering Method for Data
Gathering in Sensor Networks
ANTCLUST [26],[56]-[57] is an algorithm that considers the ant model of colonial closure
to solve ad-hoc deployed wireless sensor network distributed clustering problem. It regards
a sensor node (the object) as an ant and a cluster as a nest. In ANTCLUST, it is assumed
that two randomly chosen objects meet. Based on their similarity threshold values, they
create, merge, or delete clusters. By repeatedly conducting random meetings, clusters are
appropriately organized, so that objects in the same cluster become more similar with one
16
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
34/165
another than those in different clusters.
ANTCLUST algorithm elects a node with highest residual energy in a given neigh-
borhood as the cluster head. Further, the algorithm guarantees no two nodes in a given
neighborhood are cluster heads. Once cluster heads are elected, a set of randomly chosen
non cluster head nodes referred as social sensors in the algorithm, broadcast their status
information. The nodes which hear those broadcasts have the ability to get updated infor-
mation about the neighbors. This is partly analogous to ant meetings. In ant meetings,
both ants have the ability to get updated about the environment from the information ex-
change, whereas, in the sensor network, those sensors that hear the broadcast message get
a chance to receive environment information from the other. Non cluster head nodes select
their clusters based on the residual energy of the neighboring cluster heads, its distance to
the neighboring cluster heads, and an estimation of the cluster size based on the information
gathered from local meetings. Eventually, energy efficient clusters are formed, that result
in an extension of the lifetime of the sensor network.
This algorithm does not have non uniform cluster forming issue as with LEACH and
its derivatives such as SEP. Further, this algorithm can be applied to heterogeneous energy
networks. According to the simulation results presented in [26], the number of rounds in
which more than 80% of the sensor nodes keep alive when ANTCLUST is used is 25% to
55% higher than that when LEACH is used.However, the ANTCLUST based clustering algorithm has the following limitations.
1. The algorithm requires prior knowledge of the location information of all sensor nodes.
This assumption is not realistic considering the ad-hoc nature of the deployment of
sensors and the limited availability of battery energy. (Use of GPS would increase
hardware cost of motes and energy overhead in using GPS. Similarly, use of non GPS
based methods such as triangulation would have adverse energy overhead.)
2. This algorithm too changes cluster heads after a predetermined period of time in
normal operation similar to LEACH and HEED. However, this algorithm too does
not mention how to identify optimal point in doing so.
2.2.6 EDAC: Energy Driven Adaptive Clustering Data Collection Proto-
col in Wireless Sensor Networks
Having a proper mechanism is essential for heterogeneous energy networks to conserve
energy of nodes with less battery energy and to extract the advantage of nodes with moreenergy to prolong the lifetime of the network. EDAC [35] tries to achieve this objective by
17
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
35/165
an energy based cluster head selection and rotation mechanisms. A selected cluster head
will function until its residual energy fall below a thresholdand then cluster head rotation
take place. Then a node with the highest residual energy in this cluster heads member
base will take over the new cluster head role. Hence there will not be change of cluster
boundaries.
EDAC uses the approach outlined in the LEACH algorithm to determining the first set of
cluster heads, i.e. a fixed proportion of nodes randomly declare themselves as cluster heads.
This can lead to the creation of non-uniform clusters, especially since two or more close by
nodes may now become cluster heads similar to the situation that occurred in the LEACH
algorithm. Further, there can be situations where the number of clusters produced in the
first round is far apart from what is expected as mentioned in the drawbacks of LEACH. If
the initial cluster setup phase has these problems, it can propagate to subsequent rounds
with non-uniform clusters and/or non optimal number of clusters. Additionally, EDAC also
expects nodes to know their position. When the nodes are ad-hoc deployed, this information
can only be retrieved using GPS or triangulation technique. Both of these methods consume
a significant amount of energy.
Simulation results presented in [35] shows that EDAC has about 10% better lifetime
performance compared to LEACH in a homogeneous energy wireless sensor network. At the
same time, in a heterogeneous energy wireless sensor network, overall lifetime performance ofEDAC is 100% to 200% better than LEACH. This performance improvement in EDAC is a
direct result of energy based cluster head role rotation. Whereas in LEACH, performance is
low due to the predetermined fixed time duration of the cluster head role rotation. Further,
the simulation experiments of EDAC have shown that EDAC is good for varying data traffic
conditions (heavy and light traffic with variable packet sizes) when compared to LEACH,
as it concerns the residual energy of the nodes. Finally, [35] concludes that the energy
based cluster head rotation is robust compared to fixed time based cluster head rotation.
However, here, they do not provide a method to find a suitable value for the cluster head
rotation triggeringthreshold.
2.2.7 MEDIC: Medium-Contention Based Energy-Efficient Distributed
Clustering Algorithm
MEDIC [48] is designed to replace the cluster formation occurring at the beginning of
each round in LEACH. It follows Dutch auction principle for its time efficiency. There
is no global broadcast in MEDIC. Each node first broadcasts its vital information such
18
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
36/165
as residual energy at the maximum radio power level so that the knowledge is spread as
widely as possible. Then, each node counts its neighbors and broadcasts the number of its
neighbors at an adjusted power level corresponding to the desired cluster size. If a node
has the potential to qualify as a cluster head compared to its neighbors, it will try to claim
the cluster head role by broadcasting locally, which can be viewed as placing a bid for the
cluster headship.
The bidders will contend with each other until a node with satisfactory potential wins.
By doing so, the head-to-be expels other possible heads in its neighborhood, and in conse-
quence, the clusters with desired size are formed. The headship potential is an important
parameter, which replaces the self-electing probability used in native LEACH. The nodes
energy is important to determine its potential because the headship can be rotated among
nodes by assigning more potential to the nodes with higher energy. Further, MEDIC take
the number of neighbors into consideration as it is energy-efficient to let the node with more
neighbors wins the headship.
Once a node successfully sends out the headship claim, its neighbors must join it by
sending Request to join. Since these requests can be eavesdropped by the neighbors, they
can correspondingly correct their numbers of un-clustered neighbors. If a node finds all
its neighbors are clustered, it can elect to be a cluster head by sending out a headship
claim. Those nodes outside the neighborhood of existing cluster heads cannot join anyclusters. When the public channel is idle again, which indicates there is no node in its
neighborhood trying to join existing clusters, another round of auction will begin until all
nodes are clustered.
MEDIC is a complex algorithm which has managed to overcome undesired features of
LEACH such as bad cluster head distribution. According to the simulation results MEDIC
has a 25% better lifetime than LEACH.
2.3 Anatomy of a Wireless Sensor Node
A brief description of wireless sensor node anatomy is given in this section. The key features
of the hardware parts of a node that would be useful in proposing a wireless sensor network
self organization and communication algorithm will be discussed herein.
A wireless sensor node consists of ultra low power micro controller (C), relevant sen-
sor(s), very low power RF transceiver unit and battery with optional environment energy
harvesting circuitry as shown in Figure 2.2.
19
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
37/165
Battary
+
OptionalEnvironmental Energy
Harvesting Device
Sensor(s)
Low Power
RF Tranceiver
c
Figure 2.2: Anatomy of a wireless sensor node
2.3.1 Ultra Low Power Micro Controllers
TI MSP430 family, Microchip ultra low power PIC family, low power 8051 based imple-
mentations and Atmel AVR family micro controllers are widely used in experimental sensor
nodes. Typically these micro controllers operational efficiency is around 165 A/MIPS
in active mode. Further, these micro controllers have multiple power modes. The lowest
power mode consumes 0.1 A in RAM retention and 0.7 A in RTC mode [58]. It says
that MSP430 requires less than 0.6 s to wake-up from low-power modes to active mode.
These micro controllers are equipped with ADC, Comparators, Op-amps, Timers, RTC,
Supply Voltage Supervisors, Watch Dogs and Temperature Sensors. Hence, they inherently
support most of low power sensor interfacing and other requirements of a wireless sensor
node.
2.3.2 Low Power Wireless Transceivers
There are many single chip ultra low power RF transceivers ideal for wireless sensor nodes
in the market. Some of them are TI CC low power RF family [49], Atmel low power RFfamily [59], and Semtech SX low power family [60]. Most of these families have members
which operate in sub 1 GHz bands and 2.4 GHz bands. All these family transceivers require
only very few external passive components (one crystal, few capacitors, few inductors and
an antenna) to complete the RF design. Some of the key features of these transceivers that
are considered important for this study are discussed below. Most of those facts are derived
from the TI CC1100 family data sheet [49].
Low power transceivers have configurable baud rates typically from 1.2 kBaud to 500
kBaud. Most of these transceivers support multiple modulation modes such as 2-FSK,
20
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
38/165
GFSK, MSK, OOK and ASK. They also have very low sleep mode current consumption
(250 nA). Further, these transceivers have fast startup time (240 s) from sleep to Rx or
Tx mode. The transceivers support Rx signal input sensitivity as low as -112 dBm with
1% packet error rate. The Rx mode current of them varies between 14 mA to 17 mA
depending on the operating frequency, baud rate, input signal strength and temperature.
Typically, these transceivers can be programmed to output -30 dBm to 12 dBm RF Tx
power. Tx mode current consumption depends on the Tx output power level. Figure 2.3
gives an interpolated piecewise linear curve of currant consumption for different Tx output
power levels. According to this, 10 dBm Tx power requires about 33 mA. Whereas -6 dBm
requires only 16 mA. Hence, Rx mode power consumption is comparable to Tx mode power
consumption. This results in idle listening having a negative effect on battery life. Therefore,
these transceivers have automatic sleep-wake-sleep mode by which they generate interrupt
to wake the micro controllers if there are any incoming packets. This feature is referred
to as Wake-on-Radio. Wake-on-Radio support different duty cycles for this automatic Rx
polling. These different duty cycles consume different amounts of power. According to the
CC1100 family data sheet, one second automatic Rx poling would draw 10 A. Conversely,
when it is set to poll every 15th second, the transceiver would draw 1.5 A. Further, these
transceivers support multi channel selection. Hence, they can be used in frequency hopping
spread spectrum (FHSS) or a multichannel protocol as the frequency diversity makes thesystem more robust with respect to interference from other systems operating in the same
frequency band. This feature can be used to eliminate interference caused by neighboring
cluster communication in wireless sensor network. These transceivers also have Receiver
Signal Strength Indicators (RSSI) which is used to determine the relative distance of the
sender as well as calculating the minimum required power to reach back the sender. Most of
these RF transceivers consist of Programmable Carrier Sense Indicator. In addition, they
have direct hardware support for Clear Channel Assessment which is used to indicate if
the current channel is free or busy. This is very useful in CSMA/CA systems. Most of
these transceiver chips have hardware Link Quality Indicators. This gives a metric of the
current quality of the received signal. Further, there is a Programmable Preamble Quality
Indicator for detecting preambles and improved protection against sync word detection in
random noise.
It should be noted that there are single package SoC with both micro controller and
RF transceiver requiring very few additional components. Texas Instrument CC430 RF
SoC Series provides both MSP430 micro controller core and RF module in one package.
21
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
39/165
30 20 10 0 10 2010
15
20
25
30
35
Output TX Power (dBm)
CurrentConsumption
(mA)
Figure 2.3: Current consumption of CC1101 transceiver for different Tx power output levels
Further Texas Instrument CC1110 is a combination of CC1100 RF transceiver and 8051
micro controller core in one package.
2.3.3 Battery and Optional Energy Harvesting Techniques
Proper selection of batteries and any supplementary energy harvesting mechanisms are
crucial for long life of a wireless sensor node [61]. In theory, a 1000 mAh battery could
support a processor consuming 10 mA for 100 hours. In practice, this is not always true.
Voltage and current levels of the battery vary depending on how the energy is extracted
from it because of battery chemistry. At the same time, as batteries discharge, their voltage
drops. If the system is not tolerant to a decrease in voltage, it may not be possible to use the
full rated capacity of the battery. For example, a 1.5 V Alkaline battery is not considered
empty by the manufacturer until its voltage is only 0.8 V.There are three common battery technologies that are applicable to wireless sensor nodes
- Alkaline, Lithium, and Nickel Metal Hydride (NiMH). An AA Alkaline battery is rated
as 2850 mAh at 1.5 V, but during operation it ranges from 1.65 to 0.8 V. With a volume
of just 8.5 cm3, it has an energy density of approx 1500 J/cm3. Even though Alkaline
batteries provide cheap and high capacity energy source, their large physical size and wide
voltage range can be identified as drawbacks. Additionally, lifetimes beyond 5 years cannot
be achieved in Alkaline batteries because of battery self-discharge.
Lithium batteries provide an incredibly compact power source. The smallest versions
22
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
40/165
are just a few millimeters across. Additionally, they provide a constant voltage supply that
decays little as the battery is drained. Devices that operate with Lithium batteries do not
have to be as tolerant to voltage changes as devices that operate with Alkaline batteries.
Additionally, unlike Alkaline batteries, Lithium batteries are able to operate at temperatures
down to -400 C. Typical Lithium batteries has energy density of 2400 J/cm3. One of the
drawbacks of Lithium batteries is that they often have very low nominal discharge currents.
NiMH is the third battery type. They have the benefit of being easily rechargeable. The
downside to rechargeable batteries is a significant decrease in energy density. An AA size
NiMH battery has approximately half the energy density of an Alkaline battery. Further,
it should be noted that NiMH battery cells are 1.2 V.
One other factor that should be noted for a long life of any of these batteries, is that the
batteries should be used in duty cycles to allow recovery. In other words, the total lifetime
of a battery being continuously used at a recommended current is far less than using it in
a duty cycle, allowing the battery to recover during unused times.
Optional environmental energy harvesting mechanisms, along with the batteries, are
proposed in literature for wireless sensor nodes [62]-[63]. It is possible to use Solar, Seismic
and Vibration to harvest energy from the operating environment. However, the cost and
size of such a setup increases with the required level of power. Hence, even if an application
decided to equip energy harvesting mechanisms, the power generation rate and usage ratemay not be the same [63]. Therefore, the energy generated by such techniques has to be
stored in rechargeable batteries or super capacitors. Environment energy harvesting is a
complementing technique with clustering, where cluster head role is rotated to match the
energy dissipation and generation rates.
2.4 Sensor Network Model
This section presents a wireless sensor network model to be used in the rest of this research.
This model has derived from the existing related literature. The assumptions that have
been considered with respect to this model are presented first. Then, this section presents
the energy consumption model of a wireless sensor node. Finally, a discussion on different
wireless sensor network lifetime measurement matrices is carried out.
2.4.1 Assumptions
Practical wireless sensor networks are complex. Hence, researchers have looked at differentaspects of wireless sensor networks with appropriate assumptions [1]. For the purpose
23
8/13/2019 Sankalpa PhD Thesis Final 2010-10-28
41/165
of this research, following preliminary assumptions are made to make the sensor network
model mathematically tractable. These assumptions are in line with the previous literature
[20],[24],[26] and [27].
1. All sensor nodes are identical: It is assumed that all sensor nodes are equipped withidentical micro controllers, radio communication equipment and same capacity bat-
teries.
2. The Base Station has the ability to guide the existing cluster head rotation operation:
This means that, whenever existing cluster head identifies that it can no longer con-
tinue as a cluster head, it request the base station help to inform this among other
cluster heads to start a new global cluster head selection phase. In other words, base
station can reach any cluster head asynchronously and has the ability to commandthem. Cluster head to base station communication is contention based MAC. Further,
the research assumes base station does not have any energy limitations. Moreover,
the research assumes 100% reliability and availability of the base station due to many
to one communication, where base station is the data sink. Hence, central point of
failure is not applicable.
3. TDMA based data transmission is used for intra cluster communication: This implies
that non cluster head sensor nodes periodically wake up and update their sensed datato the cluster head in their allocated TDMA time slot and goes back to sleep mode
to preserve energy. TDMA is appropriate due to its simplicity, low overhead, short
communication duty cycle, and zero packet collisions. In literature, it is shown that
the effectiveness of TDMA is only applicable when the number of transmitting nodes
is relatively stable over time [64]. This is applicable in periodic data gath