Base Station Positioning, Nodes’ Localization and
Clustering Algorithms for Wireless Sensor Networks
A Thesis submitted
in Partial Fulfilment of the Requirements
for the Degree of
DOCTOR OF PHILOSOPHY
by
RAJIV KUMAR TRIPATHI
(Y7104097)
DEPARTMENT OF ELECTRICAL ENGINEERING
INDIAN INSTITUTE OF TECHNOLOGY KANPUR
October, 2012
Base Station Positioning, Nodes’ Localization and
Clustering Algorithms for Wireless Sensor Networks
A Thesis Submitted
in Partial Fulfilment of the Requirements
for the Degree of
DOCTOR OF PHILOSOPHY
by
RAJIV KUMAR TRIPATHI
(Y7104097)
to the
DEPARTMENT OF ELECTRICAL ENGINEERING
INDIAN INSTITUTE OF TECHNOLOGY KANPUR
October, 2012
Synopsis
Name of the Student : Rajiv Kumar Tripathi
Roll Number : Y7104097
Degree for which submitted : Ph.D.
Department : Electrical Engineering
Thesis Title : Base Station Positioning, Nodes’ Localizationand Clustering Algorithms forWireless Sensor Network
Thesis Supervisors : Dr. Yatindra Nath SinghDr. Nishchal Kumar Verma
Month and year of submission : October, 2012
Wireless sensor networks have found hundreds of applications to simplify the man-
agement of complex problems. Energy conservation in wireless sensor nodes is prime
concern to engineers in most of its applications. This becomes important as increase in
the network life time depends mainly on minimizing the energy consumption in sensor
nodes. Thus conserving and balancing the energy consumption is of utmost importance.
Main challenge in a wireless sensor network is to design algorithms with minimum en-
ergy expenditure. LEACH [1, 2] was one of the clustering/routing protocols for two
tiered wireless sensor networks which minimizes the energy consumption. This thesis
has expanded on the LEACH protocol to improve the wireless sensor network perfor-
mance.
SYNOPSIS iii
The main objective of the thesis is to investigate the mechanisms to conserve and
balance the energy consumption in two tiered wireless sensor networks. In this thesis,
the problems related to above aspect have been investigated.
The first issue is where to place base station in a wireless sensor network. The
base station position have been geometrically optimized to maximize the life of sensor
network in [3-6]. The energy minimization also depend on path loss along with the
position of deployed sensor nodes. In this thesis, we have considered path loss as well
as geometrical parameters for defining new optimal base station location. An algorithm
to estimate the sub-optimal base station position has also been proposed.
The second problem is anchor free localization in a single hop wireless sensor
network. Anchor free localization is important for localizing all the wireless nodes
without using GPS [7-11]. Average of distances is used for the first time to localize the
reference nodes in a wireless sensor network in this thesis.
Clustering techniques are important to save energy in a wireless sensor network
[12-22]. Two different clustering techniques have been proposed to minimize as well as
to balance the energy consumption among the nodes. Also the problem of identifying
bottleneck nodes have been investigated to conserve the energy in a WSN.
The thesis has been organized in the following six chapters.
Chapter 1 is the introduction to the wireless sensor networks. This chapter dis-
cusses the architecture of the wireless sensor networks and basics like sensors, base
station, radio model and the energy consumption model. This thesis is built around
the LEACH routing protocol. LEACH and its derivative Fixed-LEACH routing algo-
rithm have been discussed in detail as the background of work reported in this thesis.
SYNOPSIS iv
Chapter 2 discusses optimal location of base station in two tired wireless sensor
network. The geometry as well as path loss between the sensor nodes and the base
station have been considered. Weighted centroid method to find sub-optimal position
of base station has been proposed This method has been used in many localization
algorithms in literature. We have proposed new weights for computing the sub-optimal
location.
In chapter 3, anchor free localization algorithm for single hop wireless sensor net-
work has been proposed using internode distances. Merits and demerits of existing
anchor free localization techniques have also been discussed. Simulation results have
been presented to show that our method is better than the existing algorithms.
In chapter 4, We have proposed two methods for balancing energy consumption in
a WSN. In first one, new cluster head selection algorithm have been proposed based on
LEACH. The proposed algorithm achieved more balanced energy consumption among
sensor nodes. As this algorithm is based on number of supported node in previous
round, it is named as N-LEACH. Simulation results show significant improvement in
lifetime of sensor nodes compared to LEACH clustering.
In second method, we have proposed use of variable duty cycle of sensor nodes for
balancing energy consumption depending on distance of a sensor node and sink. The
nodes which are located near the sink consume lower energy compared to the farther
ones due to the smaller transmission distance. With same information transfer rates,
there will be uneven energy consumption in sensor nodes and farther nodes will die
earlier than the nodes closer to the sink. We propose transmission with reduced duty
cycle for farther nodes compare to nodes nearer to sink. We have presented a mechanism
to enforce equal energy consumption using variable duty cycle for all nodes participating
SYNOPSIS v
in a wireless sensor network. We have evaluated performance of proposed algorithm and
compared it with existing algorithms and found it better in terms of lifetime, average
energy consumption and balanced energy consumption for all the nodes.
Chapter 5 discusses the localized detection of bottleneck nodes in a wireless sensor
network. Criticality of bottleneck nodes has also been defined using partitioning in the
network. Different category bottleneck nodes can be found in a wireless sensor network.
Here the network is assumed to have four sink in each direction at the corners of network
as a root nodes, which use flooding for bottleneck nodes detection.
In chapter 6, the major conclusions of the work reported in the thesis have been
presented, and problems for further investigations have been suggested.
Dedicated to
Bhavya, Aadya
Ashutosh, Ajitesh
Acknowledgements
The wisest mind has something yet to learn.
— George Santayana
I feel immense pleasure in expressing my profound sense of gratitude to my thesis
supervisors Dr. Yatindra Nath Singh and Dr. Nishchal Kumar Verma under whose
supervision and inspiring guidance, I had privilege to carry out my research work. I
am indebted to them for their constant and ungrudging encouragement, valuable sug-
gestions and ingenious ideas. Words won’t be sufficient to quantify there immense
knowledge and understanding of the subject. They showed me different ways to ap-
proach a research problem and the need to be persistent to accomplish any goal. I am
also grateful to all the other faculty members for their support and encouragement.
I immensely express my heartiest thanks to my friends Rajiv, Anurag, Ashutosh,
Sandeep, Sanjeev, Amit, Arun and Ruchir for their support and help. I have spent some
of the craziest and most memorable moments with them. I would like to acknowledge
my lab coordinator Mr Vijay who helped me many times for software related problems.
Acknowledgements viii
I would also like to acknowledge my friend Rajiv Srivastava whose valuable sug-
gestions and discussions, helped me in my work. I wish to express my deep gratitude
to all of my the teachers whose consistent help and encouragement boosted my moral
and confidence.
I would also like to thanks to my friends Deepak Sharma, Pankaj Dumka, Amiya
Sahoo, Rishika Chauhan from Jaypee University of Engineering and Technology for
their support during my PhD.
Though it is beyond the scope of any acknowledgement for all that I have received
from my parents Mr. Bhanu Shanker Tripathi and Mrs. Maya Rani Tripathi, by the way
of inspiration, patience and encouragement at all times but most conspicuously during
this period, yet I make an effort to express my heartfelt and affectionate gratitude to
them. May God guide me to the wishes of my parents so that they feel the joy of having
lived a contained life in my conduct to them. I also want to express my deep gratitude
to my brothers Sanjeev Tripathi and Vikas Tripathi and my bhabhi Sarita Tripathi
who helped me by giving time to complete my thesis. Finally I want to thank my wife
Dr. Aparna Tripathi, who supported me in every step of my life and encouragement to
pursue PhD from this reputed institute, IIT Kanpur.
(Rajiv Kumar Tripathi )
Acronyms and Abbreviations
ABC Assumption Based CoordinatesACK AcknowledgementAFL Anchor Free LocalizationAoA Angle of ArrivalAPTEEN Adaptive Threshold sensitive Energy Efficient Sensor Network ProtocolBS Base StationCH Cluster HeadGPS Global Positioning SystemGER Global Energy RatioKPS Knowledge-based Positioning SystemLEACH Low-Energy Adaptive Clustering HierarchyMDS Multi Dimensional ScalingMEMS Micro Electro Mechanical SystemsMSPA Matrix transform-based Self Positioning AlgorithmNLOS Non Line of Sightnon-CH Non Cluster HeadPDM Proximity Distance MappingPEGASIS Power Efficient Gathering in Sensor Information SystemsQoS Quality of ServiceRSSI Received Signal Strength IndicatorRx ReceiverRF Radio FrequencySPA Self Positioning AlgorithmTEEN Threshold sensitive Energy Efficient Sensor Network ProtocolTDoA Time Difference of ArrivalTDMA Time Division Multiple AccessToA Time-of-ArrivalTx TransmiterVDC Variable Duty CycleWSN Wireless Sensor Network
List of Symbols
M Length of side of square wireless sensor network topologyL Total number of transmitted bitsF Location where sum of distance power 4 is minimized from all the nodesP Location of proposed pointC Location of centroid of distributed nodesd Distance between two nodesEelec Energy used for transmission and reception of electronicsEamp(L, d) Energy used by amplifier to transmit L bits at distance dETx(L, d) Energy used to transmit L bits at distance dERx Energy used to receive L bitsEDA Energy used for data aggregationECH Energy of a Cluster HeadEnon−CH Energy of a normal nodeECluster Energy dissipated in a clusterERound Energy dissipated by all the nodes during one RoundETotal Energy dissipated by all the nodes during one Epochεfs Free space loss constant measured in J/bit/m2
εmp Multi-path loss constant measured in J/bit/m4
n Total number of nodes in WSN topologyk Number of clusters in WSN topologynk
Average number of nodes per cluster in WSN topologykopt Optimum number of cluster in WSN topologyd0 Constant threshold distance for swapping amplification modelα Path loss constantDTR Distance between transmitter and receiverVs Velocity of acoustic signalVRF Velocity of radio frequency signalTn Constant threshold for cluster head selectionG Eligibility constant for cluster head selectionr Current round during simulation
List of Symbols xi
Dave Average distance among all nodes
Dc Normalized distance of all the nodes from the centroid of distributed nodesDiC Distance of ith node from the centroid of distributed nodes
ˆDhave Normalized average distance among all nodes when hop count is usedfor distance measurement
Ebs Energy for transmitting data to base stationEC Total energy of all the nodes if base station is placed at centroidEO Total energy of all the nodes if base station is placed at optimal pointEd2 Amplification energy consumption when all nodes suffer only free space lossEd4 Amplification energy consumption when all nodes suffer only multipath lossEd2,4 Amplification energy consumption when some nodes suffer free space and
remaining nodes multipath lossEMEC Total energy of all the nodes if base station is placed at center of minimum
enclosing circle
Contents
List of Figures xv
List of Tables xx
1 Introduction 1
1.1 Wireless Sensor Network . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Sensor Nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Base Station . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.3 Radio Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.4 Energy Consumption Model . . . . . . . . . . . . . . . . . . . . . 6
1.2 Research Issues and Literature Survey . . . . . . . . . . . . . . . . . . . 8
1.2.1 Deployment Strategy . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.2 Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.3 Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2 Placement of Base Station in a Wireless Sensor Network 28
CONTENTS xiii
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4 Algorithm for Base Station Location . . . . . . . . . . . . . . . . . . . . 34
2.4.1 Optimal location of base station - An example . . . . . . . . . . . 37
2.5 Simulation And Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3 Anchor Free Localization Technique for Singlehop Wireless Sensor
Network 49
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3 Challenges with Localization Techniques . . . . . . . . . . . . . . . . . . 52
3.3.1 Localization using Hop Count . . . . . . . . . . . . . . . . . . . . 52
3.3.2 Accumulation of Localization Error . . . . . . . . . . . . . . . . . 55
3.4 Metric for Proposed Algorithm . . . . . . . . . . . . . . . . . . . . . . . 56
3.5 Algorithm for Singlehop Localization . . . . . . . . . . . . . . . . . . . . 57
3.6 Simulation & Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4 Balancing Energy Consumption in a Wireless Sensor Network 64
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
CONTENTS xiv
4.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.3 Method 1: N-LEACH, a Balanced Cost Cluster-Heads Selection Algorithm 68
4.4 Proposed Cluster Head Selection Algorithm . . . . . . . . . . . . . . . . 70
4.5 Method 2: Balancing Energy Consumption using Variable Duty Cycle
(VDC) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.6 VDC Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.7 Simulation and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5 Localized Detection of Bottleneck Nodes and Quantification of Criti-
cality in a Wireless Sensor Networks 81
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.3 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6 Conclusions and Future Works 89
6.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.2 Suggestions for future works . . . . . . . . . . . . . . . . . . . . . . . . . 91
Bibliography 92
List of Publications 106
Appendix A 108
Appendix B 110
Appendix C 117
List of Figures
1.1 Wireless Sensor Network as Subset of Wireless Network. . . . . . . . . . 1
1.2 A Wireless Sensor Network . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Block diagram of Sensor Node . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 A Sensor Node . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 A Base Station Node . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.6 Radio Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.7 Clustering in WSN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.8 RSSI vs Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.9 TDoA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.10 Hyperbolic Tri-lateration . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.11 Triangulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.12 Tn for One Epoch Time in Different Rounds . . . . . . . . . . . . . . . . 22
1.13 Frequency of Cluster Heads in LEACH . . . . . . . . . . . . . . . . . . . 23
1.14 Distribution of Cluster Heads in LEACH . . . . . . . . . . . . . . . . . . 23
1.15 Frequency of Cluster Heads in Fixed-LEACH . . . . . . . . . . . . . . . 24
LIST OF FIGURES xvii
1.16 Distribution of Cluster Heads in Fixed-LAECH . . . . . . . . . . . . . . 24
2.1 A Wireless Sensor Network topology . . . . . . . . . . . . . . . . . . . . 33
2.2 Weights calculation for Algorithm . . . . . . . . . . . . . . . . . . . . . . 35
2.3 Distances from Centroid of distributed nodes . . . . . . . . . . . . . . . . 38
2.4 Distances from Theoretical Optimal Position of base station . . . . . . . 38
2.5 Distances from Proposed Point as base station . . . . . . . . . . . . . . . 39
2.6 Distances from center of minimum enclosing circle as base station . . . . 39
2.7 Average percentage reduction in amplifier energy for proposed point com-
pared to centroid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.8 Average percentage reduction in amplifier energy when negative values
are clipped to zero . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.9 Average percentage reduction in amplifier energy for proposed point com-
pared to centroid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.10 Average percentage reduction in amplifier energy when negative values
are clipped to zero . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.11 Average percentage reduction in amplifier energy for minimum enclosing
circle compared to centroid . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.12 Average percentage reduction in amplifier energy for minimum enclosing
circle compared to centroid . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1 Localization of a node in AFL . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 Nodes are localized integral multiples of maximum radio range R . . . . . 53
3.3 Localization using Hop Count . . . . . . . . . . . . . . . . . . . . . . . . 54
LIST OF FIGURES xviii
3.4 Relation between Dc and ˆDave . . . . . . . . . . . . . . . . . . . . . . . . 57
3.5 Estimation of Reference node 1 . . . . . . . . . . . . . . . . . . . . . . . 59
3.6 Estimation of Reference node 2 . . . . . . . . . . . . . . . . . . . . . . . 59
3.7 Estimation of Reference node 3 . . . . . . . . . . . . . . . . . . . . . . . 59
3.8 Estimation of Reference node 4 . . . . . . . . . . . . . . . . . . . . . . . 59
3.9 Estimation of Reference node 5 . . . . . . . . . . . . . . . . . . . . . . . 59
3.10 Localization for node j . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.11 Localization Problem for non rectangular sensor field . . . . . . . . . . . 60
3.12 GER vs size of topology for constant number of nodes . . . . . . . . . . . 61
3.13 GER vs number of nodes for constant size of topology . . . . . . . . . . . 61
3.14 Average localization error vs number of nodes for constant size of topology 62
3.15 Average localization error vs size of topology for constant number of nodes 62
4.1 Balanced Energy Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2 LEACH Algorithm Flowchart . . . . . . . . . . . . . . . . . . . . . . . . 71
4.3 Proposed Algorithm Flowchart . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4 A Wireless Sensor Network topology . . . . . . . . . . . . . . . . . . . . 74
4.5 Number of Alive Nodes vs Time (in number of rounds) . . . . . . . . . . 76
4.6 Standard Deviation of residual energy for all the nodes at each epoch . . 77
4.7 Cumulative Distribution Function of mean energy lost in one epoch . . . 79
4.8 Probability Density Function of mean energy lost in one epoch . . . . . . 79
LIST OF FIGURES xix
5.1 Example of Bottleneck Node . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.2 Undesirable Energy Histogram . . . . . . . . . . . . . . . . . . . . . . . . 83
5.3 Desirable Energy Histogram . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.4 Root Nodes Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.5 Bottleneck Nodes Detection . . . . . . . . . . . . . . . . . . . . . . . . . 87
B-1 Geometric Median (x, y) . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
B-2 Geometric medians for two points . . . . . . . . . . . . . . . . . . . . . . 112
B-3 Geometric medians for three points . . . . . . . . . . . . . . . . . . . . . 113
B-4 Geometric medians for four points . . . . . . . . . . . . . . . . . . . . . . 113
B-5 Geometric medians for five points . . . . . . . . . . . . . . . . . . . . . . 114
B-6 Geometric medians for Six points case 1 . . . . . . . . . . . . . . . . . . 115
B-7 Geometric medians for Six points case 2 . . . . . . . . . . . . . . . . . . 115
C-1 Point A (x,y) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
C-2 Point A when only two point are in field i.e. n=2. . . . . . . . . . . . . 118
C-3 Point A when only three point are in field i.e. n=3. . . . . . . . . . . . 118
List of Tables
1.1 Radio Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Different deployment technique . . . . . . . . . . . . . . . . . . . . . . . 12
1.3 Different localization technique . . . . . . . . . . . . . . . . . . . . . . . 20
1.4 Different Routing Techniques . . . . . . . . . . . . . . . . . . . . . . . . 26
2.1 Excess energy used by different methods compared to the optimal location 40
2.2 Nodes’ ratio when centroid as center . . . . . . . . . . . . . . . . . . . . 46
2.3 Nodes’ ratio when proposed point as center . . . . . . . . . . . . . . . . . 46
2.4 Distance in meter between centroid and point P . . . . . . . . . . . . . . 47
3.1 Average Actual Distance vs Average Distance due to number of hops . . 55
4.1 Average Lifetime of nodes in number of rounds . . . . . . . . . . . . . . . 77
5.1 % of Bottleneck Nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
B-1 Geometric medians for different cases . . . . . . . . . . . . . . . . . . . . 116
Chapter 1
Introduction
1.1 Wireless Sensor Network
A Wireless Sensor Networks (WSNs) form a subset of Ad-hoc networks as shown in
Fig.1.1. Wireless sensor networks have many restrictions compared to Ad-Hoc net-
works in terms of its sensor nodes’ capability of memory storage, processing and the
available energy source. Wireless sensor networks are generally assumed to be energy
restrained because sensor nodes operate with small capacity DC source or may be placed
such that replacement of its energy source is not possible. Even though sensor networks
are a subset of ad hoc networks, the protocols designed for ad hoc networks cannot be
used as it is due to the reasons cited in [23].
Figure 1.1: Wireless Sensor Network as Subset of Wireless Network.
1.1 Wireless Sensor Network 2
1) The number of nodes in a sensor network is very large compared to ad-hoc networks.
Thus sensor networks require different and more scalable solutions.
2) In comparison to ad-hoc networks, sensor nodes have limited power supply and
recharging of power is impractical considering the large number of nodes and the en-
vironment in which they are deployed. Therefore energy consumption in WSN is an
important metric to be considered.
A WSN (see Fig.1.2) consists of spatially distributed autonomous sensors nodes to
cooperatively monitor physical or environmental conditions. The nodes communicate
wirelessly and often self-organize after being deployed in an ad-hoc fashion. A WSN
may consist of hundreds or even thousands of nodes. Source nodes transmit their data
to destination nodes through intermediate nodes. This destination node is connected to
a central gateway, also known as base station. Central gateway provides a connection
to the wired world where the data can be collected, processed, and analyzed.
Figure 1.2: A Wireless Sensor Network
The WSN consist of two main components:
1. Sensor Nodes, and
2. Base Station (Central Gateway).
1.1 Wireless Sensor Network 3
1.1.1 Sensor Nodes
Sensors nodes are typically built of few sensors and a mote unit as shown in Fig.1.3.
A Sensor is a device which senses the information and pass it on to mote. Sensors
are typically used to measure the changes in physical environmental parameters like
temperature, pressure, humidity, sound, vibration and changes in the health parameter
of person e.g. blood pressure and heartbeat. MEMS based sensor have found good
use in sensor nodes. A mote consists of processor, memory, battery, A/D converter
for connecting to a sensor and a radio transceiver for forming an ad hoc network. A
mote and sensor together form a Sensor Node. A sensor network is a wireless ad-hoc
network of sensor nodes. Each sensor node can support a multi-hop routing algorithm
and function as forwarder for relaying data packets to a base station.
Figure 1.3: Block diagram of Sensor Node
1.1.2 Base Station
A base station links the sensor network to another network. It consists of a processor,
radio board, antenna and USB interface board. It is preprogrammed with low-power
mesh networking software for communication with wireless sensor nodes. Deployment
of the base station in a wireless sensor network is very important as all the sensor nodes
handover their data to the base station for processing and decision making. Energy
1.1 Wireless Sensor Network 4
conservation, coverage of sensor nodes and reliability issues are taken care of during
deployment of base station in sensor network. Generally base stations are assumed
static in nature but in some scenarios they are assumed to be mobile to collect the
data from sensor nodes. The Crossbow sensor node and base station [24] are shown in
Fig.1.4 and Fig.1.5 respectively.
Figure 1.4: A Sensor Node Figure 1.5: A Base Station Node
1.1.3 Radio Model
We have assumed the same radio model which has been used in earlier works [2, 25].
For the radio hardware, the transmitter dissipates energy to run the transmitter radio
electronics and power amplifier, and the receiver dissipates energy to run the receive
radio electronics as shown in Fig.1.6.
For the scenarios described in this thesis, both the free space (d2 power loss) and the
multi path fading (d4 power loss) channel models were used depending on the distance
between the transmitter and the receiver. If the distance is less than a threshold, the
free space (fs) model is used; otherwise, the multi path (mp) model is used.
1.1 Wireless Sensor Network 5
Figure 1.6: Radio Model
In the radio model of Tx amplifier, we use α = 2 for free space and α = 4 for
multipath model. Thus if a node transmits L number of bits, the energy used in
transmission will be
ETX(L, d) =Eelec.L+ Eamp(L, d), (1.1.1)
ETX(L, d) ={L.Eelec+L.εfs.d
2 if d<d0L.Eelec+L.εmp.d4 if d≥d0 . (1.1.2)
Here threshold
d0 =
√εfsεmp
. (1.1.3)
To receive L message bits, the radio spends energy
ERX(L) =Eelec.L. (1.1.4)
1.1 Wireless Sensor Network 6
The parameters shown in Table 1.1 are used to evaluate the energy consumption in
a wireless sensor network.
Table 1.1: Radio ParametersDescription Symbol ValueEnergy consumed by the amplifier totransmit at a shorter distance
εfs 10pJ/bit/m2
Energy consumed by the amplifier totransmit at a longer distance
εmp 0.0013pJ/bit/m4
Energy consumed in electronic circuitto transmit or receive the signal
Eelec 50nJ/bit
Data aggregation energy EDA 5nJ/bit
1.1.4 Energy Consumption Model
Before discussing the energy consumption model, understanding of cluster is important.
Clustering (see Fig.1.7) is a process of grouping nodes using an algorithm to perform
certain tasks efficiently as per the requirements. Clustering can also be used to divide
the topology into sub-regions based on certain criteria e.g. whole area should be covered,
minimum energy consumption, maximum lifetime, etc..
Figure 1.7: Clustering in WSN
1.1 Wireless Sensor Network 7
In a wireless sensor network, cluster heads are selected from all the nodes. Each node
chooses the nearest cluster head for forwarding packets to Base station. All the nodes
attached to cluster head, form the cluster. Let there be n nodes uniformly distributed
in an M ×M area and k clusters in topology. There will be on an average nk
nodes per
cluster. Out of these, there will be one cluster head node and remaining (nk− 1) non
cluster head nodes. Energy consumption for a cluster is derived further.
The energy consumption Enon−CH for a single non-cluster head node is only for
transmission of L bits to cluster head.
Enon−CH =L.Eelec + Eamp(L, d). (1.1.5)
ECH for a particular Cluster Head node is due to reception of data from all non-Cluster
Head node of that cluster, data aggregation and transmission of aggregated data to
base station.
ECH =(n
k− 1).L.Eelec +
n
k.L.EDA + ETX(L, d). (1.1.6)
Where EDA is energy used by Cluster Head for data aggregation. Now the energy
consumption in a cluster is given by
ECluster =ECH + (n
k− 1)Enon−CH . (1.1.7)
More detailed energy consumption model will depend on specific problem in a wireless
sensor network. We shall be looking at these details later in thesis.
1.2 Research Issues and Literature Survey 8
1.2 Research Issues and Literature Survey
Wireless sensor networks are quite different from general wireless networks due to var-
ious constraints and highly application specific nature of WSNs. Consequently, WSNs
pose different research challenges. In wireless communication system, the models for
signal strength drop over a distance are well developed. Effects of signal reflection, scat-
tering and fading are well understood. In an actual WSN, cost and other application
specific issues affect the communication properties of the system. For example, radio
communication in WSN is of low power and short range compared to any other wireless
communication network. The system performance characteristics vary considerably in
WSN even though the same basic principles of wireless communication network are
used in WSN. The size, power, cost and their tradeoffs are fundamental constraints
in WSNs. Considering the basic differences with the wireless communication systems,
many issues have been identified and investigated. Major issues affecting the design
and performance of a wireless sensor network are the following:
1) Deployment strategy
2) Localization
3) Clustering for hierarchal routing
4) Coverage efficacy
5) Efficient medium access control
6) Efficient database centric design
7) Quality of service implementation
8) Acceptable security
We have restrained ourselves to the study of first three issues. This thesis concentrate
mainly on deployment strategy, localization and clustering for hierarchal routing.
1.2 Research Issues and Literature Survey 9
1.2.1 Deployment Strategy
The deployment strategy depend mainly on the type of sensors and the application.
The following deployments strategies are generally used in WSNs.
Random deployment: Random deployment is the most practical way of placing the
sensor nodes. For a dynamic sensor network, where there is no a-priori knowledge of
optimal placement, random deployment is a natural option.
Incremental deployment: The incremental placement strategy is a centralized, one-
at-a-time approach to place the sensors. The implementation makes use of information
gathered through the previously deployed nodes to determine the ideal deployment
location of the next sensor node. This can be calculated at base station.
Computational geometry approach: The computational geometry approach is the
simplest method for sensor deployment. In this approach, the target sensing region is
constructed by a set of grids or polygons thus deciding the node placement.
Movement-assisted deployment: In certain scenarios, deployment scheme is prone
to deviate from the plan, because the actual landing positions cannot be controlled
due to the existence of wind and obstacles. For industrial use, where the detection
of some part of a machine is required, or in a wildlife sanctuary, one may use mobile
sensor nodes, which can move to the desired places to provide the required coverage. In
movement-assisted deployment, initially random deployment is performed. Thereafter
re-deployment of the mobile sensor nodes is performed with incremental optimization.
The WSN consists of sensor nodes and a base station. So researchers can solve two
different problems of deployment. In this thesis, the literature survey for deployment is
divided in two parts. In the first part, sensor nodes’ deployment is discussed. Thereafter
the literature survey on base station deployment have been done.
1.2 Research Issues and Literature Survey 10
Sensor node deployment problems have been studied in terms of energy conservation
by Howard et al. [26] in 2002 and Heo et al. [27] in 2003. Dhillon et al. [28] optimized
the number of sensors in 2004. Also in the same year, Zou et al. [29] proposed virtual
force algorithm (VFA) as a sensor deployment strategy to enhance the coverage after
an initial random placement of sensors. One of the successful application of the com-
putational geometry in WSNs deployment problem is the Voronoi diagram approach
investigated by Wang et al. [30] in 2004. In 2005, Lin et al. [31] proposed near-optimal
sensor placement algorithm to achieve complete coverage. Liu et al. [32] proposed
power aware sensor node deployment in 2006. Poe et al. [33] studied deployment of
sensor nodes while considering area coverage, energy consumption, and worst-case de-
lay of the WSNs in 2009. Park et al. [34] achieved optimum sensor nodes deployment
using Fuzzy C-means algorithm in 2011. The research in last decade for sensor node
deployment have indicated it to be a application specific problem. Most of the research
approaches are geometrical and also optimize the number of sensor nodes in general.
Some of the famous research projects are as follows: Redwoods project [35], which
sought to observe the microclimate variations surrounding Redwood trees in a coastal
forest. Volcano monitoring using acoustic sensors to detect the seismic activity and
record high-frequency measurements [36], and ZebraNet project to gather position data
of zebras in their natural environment in Kenya [37].
Presently main issues in sensor node deployment are-
1) Sensor nodes deployment to maximize the coverage in the network. Detection of
coverage holes, repositioning of sensor nodes to repair coverage holes,
2) Deployment to achieve minimal energy consumption configuration,
3) Deployment of sensor nodes for a large scale WSN to minimize the event sensing
delays.
1.2 Research Issues and Literature Survey 11
As in this thesis, investigation in the base station deployment problem has been
presented so the literature survey on base station deployment is important. There
are several proposed algorithms available in the literature for optimal location of base
station in a WSN. Though the Fermat point is very old problem for optimal facility
location but when the optimal location depends on several parameters other than only
distance, new methods have been evolved. Pan et al. (2003, 2005) [3, 4] provided
algorithms for locating a single base station using an upper bound and a lower bound
initially. Bogdanov et al. (2004) [38] optimized the positions of base stations in a data
collecting sensor network to minimize the power consumed by the sensors. Efrat et al.
(2004) [39] used approximation schemes for optimally locating a base station. Wong
et al. (2006) [40] proposed novel binary integer programming formulation of the base
station placement problem. Akkaya et al. (2007) [5] showed that dynamic positioning
of the BS is an effective means for boosting the network dependability. Basheer et
al. [41] proposed delaunay refinement based sub optimal receiver placement technique
which reduces the dilution of localization accuracy. This method solve two problems
of wireless sensor network simultaneously, one is base station positioning and other,
minimization of localization error. Recently Paul et al. (2010) [6] also proposed a new
optimal location of base station using geometrical approach for maximum lifetime of
the sensor network. Most of the algorithms use power minimization between nodes and
base station. In this thesis also, power consumption minimization has been considered
along with the geometry as well as the path loss exponent of the medium. The issues
for investigation in base station deployment are the following.
1) Deployment of base station to maximize network lifetime,
2) Dynamic and multiple base station positioning,
3) Base station deployment for minimal energy consumption.
Literature survey summary is shown in Table 1.2.
1.2 Research Issues and Literature Survey 12
Table 1.2: Different deployment technique
Technique Algorithm Deployment Method
Sensor node Deployment Howard et al. [26] Deployment of a mobile nodes inan unknown environment.
Heo et al. [27] Deployment of mobile sensornodes in the region of interest.
Dhillon et al. [28] Probabilistic optimization ofnumber of sensors.
Zou et al. [29] Virtual force algorithm as a sen-sor deployment strategy.
Wang et al. [30] Deployment problem using theVoronoi diagram approach .
Lin et al. [31] Near-optimal deployment toachieve complete coverage.
Liu et al. [32] Power-aware sensor node deploy-ment.
Poe et al. [33] Considering coverage, energy con-sumption, and delay.
Park et al. [34] Nodes deployment using Fuzzy C-means algorithm.
Base Station Deployment Pan et al. [3, 4] Minimum Enclosing Circle withlower and upper bound.
Bogdanov et al. [38] Minimizing power consumptionof sensor nodes.
Efrat et al. [39] Approximation schemes for locat-ing a base station.
Wong et al. [40] Binary integer programming forbase station placement.
Akkaya et al. [5] Dynamic positioning of base sta-tion to increase network lifetime.
Basheer et al. [41] Receiver placement using delau-nay refinement by reducing local-ization error.
Paul et al. [6] Using geometrical approach formaximum lifetime.
1.2 Research Issues and Literature Survey 13
1.2.2 Localization
Localization of sensor nodes in WSNs, has been and still is quite a vital area of research.
Localization is a method to determine the accurate physical position of sensor nodes.
Wireless sensor networks are used in the different environments to perform various
monitoring tasks such as search, rescue, disaster relief, target tracking. In many such
tasks, node localization is important to the application. Existing localization algorithms
basically consists of two basic phases-
(1) Distance (or angle) estimation, and
(2) Distance (or angle) combining.
1.2.2.1 Distance (or angle) estimation
We determine distance or angle between nodes in phase 1 using following methods.
Received Signal Strength Indicator (RSSI): This techniques measure the received
power of the signal at the receiver. If the transmit power is known, effective propagation
loss can be calculated. Theoretical and empirical models can be used to translate this
loss into a distance estimate. This method has been used mainly for RF signals. Using
popular Friis’ equation [42], the detected signal strength PRx is given by
PRx =PTx .GTx .GRx .(λ
4πd)2. (1.2.8)
It shall be noted that PRx decreases quadratically with the distance d from the sender.
The sensor nodes convert the received signal strength to received signal strength indi-
cator (RSSI). It is ratio of the received power to the reference power (PRef ).
RSSI =10.log(PRxPRef
). (1.2.9)
1.2 Research Issues and Literature Survey 14
Figure 1.8: RSSI vs Distance
Typically, the reference power represents an absolute value of PRef = 1mW and
RSSI is represented in dBm. When 2.4 GHz signal is transmitted with PTx = 50W ,
RSSI (in dBm) as a function of the distance is shown in Fig.1.8. RSSI is the prime
candidate for range measurements due to its simplicity. Given a known transmission
power and a good model of the wireless channel, the distance between transmitter and
receiver can be estimated based on the received power. Unfortunately, the accuracy
of these RSSI range measurements is highly sensitive to multi-path fading, non-line of
sight (NLOS) scenarios, and other sources of interference, which may result in large
errors. These errors can propagate through all subsequent triangulation computations,
leading to useless information.
Angle-of-Arrival (AoA): Nodes estimate the angle at which signals are received and
use simple geometric relationships to calculate node positions. Localizations in AoA can
be solved using triangulation. AoA localization is susceptible to measurement noise and
few additional problems. This technique require knowledge of reference direction. Thus
AoA approaches require costly antenna arrays on each node. Usually such antennas are
not affordable in cheap sensor nodes.
1.2 Research Issues and Literature Survey 15
Time Difference of Arrival (TDoA): The propagation time can be directly
translated into distance, based on the known signal propagation speed. These methods
can be applied to many different signals, such as RF, acoustic, infrared and ultrasound.
The short transmit ranges of 1 to 10 m result in unacceptably high synchronization
demands of 3 psec per cm resolution, when TDoA techniques are employed.
For example, RF and acoustic signal have different speed, and if transmitted simul-
taneously by transmitter at known time, will be received by receiver at time T1 and
T2 respectively, due to the difference in their speed as shown in Fig.1.9. Thus time
difference is delay ∆T = (T1 − T2). Let the distance between transmitter and receiver
be DTR, and velocity of acoustic and RF signal be Vs and VRF respectively. DTR is
given by
DTR =∆T × Vs(1− Vs
VRF). (1.2.10)
As Vs << VRF ,
DTR =∆T × Vs. (1.2.11)
Figure 1.9: TDoA
1.2 Research Issues and Literature Survey 16
1.2.2.2 Distance (or angle) combining
In phase 2, phase 1 data is used to determine (X,Y) coordinates of all the nodes. The
most common methods for phase 2 are the following.
Hyperbolic Tri-lateration: This process uses distances or absolute measurements
of time-of-flight from three or more sites. As shown in Fig.1.10.
Figure 1.10: Hyperbolic Tri-lateration
Equation of Spheres are
a2=x2+y2+z2,b2=(x−d)2+y2+z2, and
c2=(x−i)2+(y−j)2+z2 .
(1.2.12)
Solving, we get
x =a2 − b2 + d2
2d, (1.2.13)
y =a2 − c2 + i2 + j2
2j− i
j.x, and (1.2.14)
z =±√a2 − x2 − y2. (1.2.15)
1.2 Research Issues and Literature Survey 17
Thus using proper assumptions Hyperbolic Tri-lateration can be used to solve localiza-
tion problems.
Triangulation: This process uses at least one distance estimate and two measured
angles. The angle measurement may be done with receiver antenna diversity or phase
comparison. As shown in Fig.1.11, Sine theorem, i.e.
Figure 1.11: Triangulation
a
sinA=
b
sinB=
c
sinC(1.2.16)
and Cosine theorem, i.e.
c2=a2+b2−2ab.CosC
b2=a2+c2−2bc.CosB
a2=b2+c2−2bc.CosA
(1.2.17)
can be used. But these methods require at least two references for localization of
unknown nodes. Geometry of the references nodes and configuration of network nodes,
decide accuracy in localization through triangulation. Localization also depend on
accuracy of the range measurements in phase 1.
Initially Harter et. al developed a small device called Bat for localization of various
objects in 1999. Bats consist of a radio transceiver, controlling logic and an ultrasonic
transducer. Objects to be localized were tagged with bats. The location of these bats
were tracked by the system to build a location database of these objects [43].
1.2 Research Issues and Literature Survey 18
The literature survey for localization can be divided into two parts. In first part, the
anchor based localization is discussed and in second part, anchor free localization.
Anchor based localization: Priyantha et al. [44] at MIT developed a hardware plat-
form known as Cricket. The sensor nodes localization is done using inter node ranges
in that hardware. Doherty et al. [45] in 2001, developed semidefinite program based on
connectivity and pairwise angles between nodes for localization. In 2002, Savvides et al.
[46] derive the Cramer-Rao lower bound (CRLB) for network localization, expressing
the expected error characteristics for an ideal algorithm, and compared it to the actual
error in an algorithm based on multilateration. In a well known work APS (Ad Hoc
Positioning System), Niculescu et al. [47] proposed two localization algorithms using
angle of arrival in 2003.In same year, He et al. [48] proposed APIT (Approximate PIT)
localization using neighbor information. Meertens et al. [49] (2004) have shown that
how sensor nodes in a WSN can construct, align and reconcile a spatial map of its own
position and those of other, nearby nodes based on inter-node distance information,
thus providing a global coordinate system. Peng et al. [50] (2006) focused on localiza-
tion techniques based on angle of arrival information between neighboring nodes and
proposed a new localization and orientation scheme that considers beacon information
multiple hops away. Cheng et al. [51] (2007) proposed an algorithm which uses two
different localization techniques, multidimensional scaling and proximity distance map-
ping (PDM), in a phased approach. The phased approach has comparable complexity
to PDM but less than MDS. Huang et al. [52] (2008) used Monte Carlo sampling steps
in the context of the localization using a single beacon for various types of observations
such as ranging, Angle of Arrival (AoA) and connectivity. Basheer et al. [53] (2011)
proposed localization technique using constrained simulated annealing with tunnelling
transformation to solve non-tractable posterior distribution.
1.2 Research Issues and Literature Survey 19
Anchor free localization: Initially, Bulusu et al. [7] (2000) gave the anchor free
localization algorithm for the wireless sensor networks. In 2001 Capkun et al. [8] (2001)
proposed the self positioning algorithm (SPA). They used distances between the nodes
to build a relative coordinate system. In the same year Savarese et al. [9] proposed
the Assumption Based Coordinates (ABC) algorithm. It determines the locations of
the unknown nodes, one at a time. Priyantha et al. [10] (2003) proposed Anchor
Free Localization (AFL) using hop count. The key idea in AFL is fold-freedom, where
nodes first configure into a topology that resembles a scaled and unfolded version of the
true configuration, and then run a force-based relaxation procedure. Moses et al. [11]
(2003) used a calibration signal for measuring time-of-arrival and direction-of-arrival
of sensor nodes in anchor free localization. Moore et al. [54] (2004) localized nodes
in a sensor network with noisy distance measurements, using no beacons or anchors.
Fang et al. [55] (2005) had proposed Knowledge based Positioning System (KPS). The
nodes are deployed as one group at a time centered around a group coordinates. The
nodes in a group gets distributed around center as per source distribution function. In
this scheme, each sensor first finds out the number of its neighbors from each group.
Using node degree in a group, nodes estimate a location based on the principle that
the estimated location should maximize the probability of the observation. Basheer et
al. [56] (2009) introduced a new parameter R-factor, which enhances the localization
accuracy.Wang et al. [57] (2010) developed a standardized clustering-based approach
for the local coordinate system formation. Wang et al. [58] (2010) considered residual
energy, node degree and principles of triangle inequality to heuristically build clusters,
then fuse the clusters into one cluster. Basheer et al. [59] (2011) used cross correlation
of shadow fading noise and copulas to improve localization accuracy. The related work
for localization is summarized in the Table 1.3.
1.2 Research Issues and Literature Survey 20
Table 1.3: Different localization technique
Technique Algorithm Localization Method
Anchor Based Localization Priyantha et al. [44] Localization using time differenceof arrival (TDoA).
Doherty et al. [45] Semidefinite program based local-ization.
Savvides et al. [46] Localization using collaborativemultilateration.
APS [47] Localization using angle of arrival.APIT[48] Approximate PIT Test using neigh-
bor information.Meertens et al. [49] Localization using inter node dis-
tance.Peng et al. [50] Localization using angle of arrival.Cheng et al. [51] Localization using MDS and prox-
imity distance mapping.Huang et al. [52] Localization using Monte Carlo
method.Basheer et al. [53] Localization using stochastic tun-
nelling.Anchor Free Localization Bulusu et al. [7] GPS-less low cost outdoor localiza-
tion.ABC [9] Assumption Based Coordi-
nates (Solved geometrically).SPA [8] Localization using distances be-
tween the nodes.AFL [10] Anchor Free Localization using hop
count.Moses et al.[11] Localization using time-of-arrival
and direction-of-arrival.Moore et al. [54] Localization using robust quadri-
laterals.KPS [55] Knowledge-based Positioning Sys-
tem.Basheer et al. [56] Localization using receive error fac-
tor and diversity.MSPA [57] Matrix transform-based Self Posi-
tioning Algorithm.Wang et al. [58] Cluster based localization using
residual energy and node degree.Basheer et al. [59] Localization using cross correlation
of shadow fading noise and copulas.
1.2 Research Issues and Literature Survey 21
1.2.3 Routing
Routing in the WSNs is challenging due to the inherent characteristics that distinguish
these networks from the other wireless networks like ad hoc networks or cellular net-
works. It is not possible to use a global addressing scheme for the deployment of a
relatively large number of sensor nodes. Thus, traditional IP-based protocols may not
be applicable to WSNs. Sensor nodes are tightly constrained in terms of energy, pro-
cessing, and storage capacities. Thus, they require careful resource management. In
most of the application scenarios, nodes in WSNs are generally stationary after deploy-
ment except for a few mobile nodes. Nodes in other traditional wireless networks are
free to move, which results in unpredictable and frequent topological changes. Routing
protocols in the WSNs mainly depend on the application and the network architecture
of the sensor networks.
Routing in the WSNs can be classified into two broad categories:
(1) Network structure based routing, and
(2) Protocol operation based routing.
Network structure based routing: Flat addressing based routing, hierarchical based
routing, and location based routing can be used depending on the network structure.
In flat addressing based routing, all nodes are typically assigned equal roles or function-
ality. In hierarchical routing, nodes will play different roles defining the hierarchy in
the network. In location-based routing, sensor nodes’ positions are exploited to route
the data in the network.
Protocol operation based routing: Protocol operation based routing can be clas-
sified into multipath based, query based, negotiation based, QoS based, or coherent
routing technique.
1.2 Research Issues and Literature Survey 22
Hierarchical routing is based on network structure. It is also known as cluster-
based routing, and is a well-known technique with special advantages related to energy
efficient communication. As this thesis is based on LEACH routing, the literature
survey focusses only different hierarchical routing protocols.
Low Energy Adaptive Clustering Hierarchy (LEACH): In LEACH [1] topol-
ogy is divided in to clusters. Few nodes are selected as cluster heads and other nodes
use these cluster heads as routers to the sink. All the data processing are performed at
cluster head. Cluster heads change randomly over time in order to balance the energy
consumption in nodes. A node becomes a cluster head for the current round if the
chosen random number is less than the following threshold,
Tn =
{ p
1−p.(rmod 1p )
if n in G
0 otherwise (1.2.18)
Here p is the desired percentage of cluster heads (In [2], p is 5% of number of nodes
for maximum lifetime), r is current round, and G is a number, which decides eligibility
to become cluster head and form a set of nodes that have not been cluster heads in
current epoch. Fig.1.12 shows the value of Tn for p = 0.1 and 10 rounds in one epoch.
Figure 1.12: Tn for One Epoch Time in Different Rounds
1.2 Research Issues and Literature Survey 23
LEACH was the first protocol which tried achieving the balance in energy consump-
tion among nodes. There were also with some shortcomings in LEACH,
(1) Residual energy of nodes was not considered during cluster head selection.
(2) Probability function for becoming a cluster head was based on the assumption that
all nodes have same initial energy, which is not true in case of heterogeneous networks.
(3) Number of cluster head nodes are not fixed in LEACH due to stochastic selection
depending on probability rule.
Every round in LEACH will have the different cluster numbers as shown in Fig.1.13.
We have chosen p = 0.05 in 100 nodes topology, 5 nodes become cluster head with
highest frequency. After every epoch, every node again becomes eligible to be a cluster
head as all the nodes have become cluster head once in earlier epoch. Fig.1.14 shows
the number of cluster heads in various rounds in different epochs.
Figure 1.13: Frequency of Cluster Heads inLEACH
Figure 1.14: Distribution of Cluster Headsin LEACH
In real situations, sensor nodes drain out non uniform amount of energy due to
different distances between sensor nodes and sink and different transmission rates. In
heterogeneous network there may be two or more groups of nodes which have different
initial energy. If nodes have different amount of energy, then the nodes with more
energy should be cluster heads more often than the nodes with less energy, to ensure
1.2 Research Issues and Literature Survey 24
that all nodes die approximately at the same time.
Fixed-LEACH is one of the solutions to this problem. The number of cluster heads
are fixed in Fixed-LEACH as shown in Fig.1.15 and the frequency remain fixed as shown
in Fig.1.16. But as the nodes choose its nearest cluster head, the number of supported
nodes may be different for each cluster head. This leads the uneven energy dissipation
among nodes. In this thesis, we have proposed a better solution compared to Fixed-
LEACH, which provide more balanced energy consumption in a WSN.
Figure 1.15: Frequency of Cluster Heads inFixed-LEACH
Figure 1.16: Distribution of Cluster Headsin Fixed-LAECH
We can also reduce the energy consumption in LEACH algorithm using compression
scheme for data aggregation in wireless sensor network. Such a scheme was proposed
by Kasirajan et al. [60] (2010). More than 50% of energy was saved with low overhead
and less than 5% of distortion.
Other Hierarchical Protocols: After LEACH, several algorithms have been de-
rived to form the energy efficient wireless sensor network while balancing remaining
energy level among the nodes. LEACH inspired many hierarchical protocols such as
Power-Efficient Gathering in Sensor Information System (PEGASIS) [14], Threshold
sensitive Energy Efficient sensor Network protocol (TEEN) [12] and Hybrid Energy
Efficient Distributed clustering (HEED) [15] etc..
1.2 Research Issues and Literature Survey 25
Manjeshwar et al. [12, 13] (2001,2002) had proposed two hierarchical routing proto-
cols, TEEN and APTEEN; these protocols are designed for time-critical applications
using TDMA schedule. Lindsey et al. [14, 17] (2002) proposed PEGASIS, an enhanced
protocol over LEACH to extend network lifetime by using only one node in a chain to
transmit to the sink instead of multiple nodes. Younis et al. [15, 16] (2004) proposed
a Hybrid Energy-Efficient Distributed Clustering (HEED). They assumed availability
of multiple power levels in sensor nodes. Ye et al. [18] (2005) proposed the algorithm
EECS, which elects cluster heads with more residual energy through local radio com-
munication while achieving good cluster head distribution. Hansen et al. [19] (2006)
have shown that using a minimum separation distance between cluster heads improves
energy efficiency, measured by the number of messages received at the base station
per unit energy consumed. Junping et al. [20] (2008) improved LEACH and proposed
TB-LEACH. In this algorithm, competition for cluster heads no longer depends on a
random number as in LEACH. Nodes which have the shortest time interval for each
node’s random time interval will win the competition and become cluster heads en-
sure that the partition of cluster is balance and uniform. Kumar et al. [21] (2009)
proposed energy efficient heterogeneous clustered scheme for wireless sensor networks.
They improved lifetime and performance of the network using weighted probability of
the election of cluster heads. Otgonchimeg et al. [22] (2009) proposed Energy Efficient
Clustering algorithm for Event-Driven (EECED) wireless sensor networks. This algo-
rithm aimed at prolonging the lifetime of a sensor network by balancing energy usage
of the nodes. EECED makes higher chances to be selected as cluster head for the nodes
with more residual energy. Said et al. [61] (2010) proposed Improved and Balanced
LEACH (IB-LEACH) to decrease probability of failure of nodes and to prolong the time
interval before the death of the first node. The related work for hierarchical routing is
summarized in the Table 1.4.
1.2 Research Issues and Literature Survey 26
Table 1.4: Different Routing Techniques
Algorithm Clustering Rule
LEACH [1] Random probabilistic clustering.LEACH-C [2] Centralized clustering algorithm to produce better clusters.LEACH-F [2] Clustering with fixed number of clusters.TEEN [12] Total number of transmissions are reduced by allowing the
nodes to transmit only when the sensed value less than athreshold value.
APTEEN [13] New TDMA schedule is introduced to avoid collisions of close-by nodes which fall in the same cluster. As these nodes sensesimilar data and try to send their data simultaneously.
PEGASIS[14, 17]
Each node communicates only with a close neighbor and takesturns in transmitting to the base station, thus reducing theamount of energy spent per round.
HEED [15, 16] Cluster heads are selected periodically according to a hybrid ofthe node residual energy and node proximity to its neighborsor node degree.
EECS [18] Clustering based on residual energy through local radio com-munication while achieving better cluster head distribution.
Hansen et al.[19]
Clustering for large sensor networks using a minimum sepa-ration distance.
TB-LEACH[20] Nodes which have the shortest time intervals from randomlyselected time interval by nodes will win the competition andbecome cluster heads to ensure that the partition of cluster isbalanced and uniform.
EEHC [21] Cluster head selection based on weighted election probabilitiesaccording to the residual energy in each node. Heterogeneoustopology is assumed.
EECED [22] Nodes with more residual energy have more chances to beselected as cluster head.
IB-LEACH [61] Clustering using characteristic parameters of heterogeneity.MMCR [62] Multi-interface multi-channel routing for enhancing capacity
of wireless sensor network.
1.3 Thesis Organization 27
1.3 Thesis Organization
Objective of the work done in this thesis is to improve the conservation of energy and
balancing the energy consumption of sensor nodes, which leads to improvement in the
lifetime of a sensor network. The anchor free localization of sensor nodes in wireless
sensor networks have also been investigated.
This thesis is organized in the following structure. Chapter 2, discuses the base sta-
tion positioning based on multipath loss and a mathematical model have been presented
to analyze the base station positioning performance in terms of total energy used from
different points. In chapter 3, anchor free localization technique is discussed. In chapter
4, modifications have been proposed in the LEACH algorithm, to improve its perfor-
mance in terms of lifetime and balancing the energy consumption in nodes along with
variable duty cycle algorithm for balancing energy consumption among sensor nodes. In
chapter 5, bottleneck nodes have been defined and detected along with their criticality.
Chapter 6, concludes the thesis.
Chapter 2
Placement of Base Station in aWireless Sensor Network
2.1 Introduction
Base station location is an important factor in a wireless sensor network as it affects
its lifetime. In this chapter, the question of “Where should the base station be placed
in two tiered wireless sensor network field ?”1 has been investigated. The objective is
to minimize the overall energy consumption in a wireless sensor network. A heuristic
algorithm has been proposed to find such a base station location. Considering some
nodes to be far enough to use a different path loss model for their signals to the base
station, our proposed algorithm considers two categories of nodes and hence two differ-
ent path loss models based on their distance from the base station. The results show
that the new algorithm provides a better base station location than earlier methods.
The overall energy consumption is quite close to the optimal solution.
1Rajiv Kr. Tripathi, Yatindra Nath Singh and Nishchal Kr. Verma, “Two tiered wireless sensornetworks - Base station optimal positioning case study,” Published in IET Journal Wireless SensorSystem, Vol.2, Issue.4 December 2012, pp. 351-360..
2.2 Related Work 29
2.2 Related Work
In the literature so far, many heuristic algorithms have been proposed to find sub-
optimal solutions [39] of optimum base station positioning in two tiered wireless sensor
network. Although these heuristics are shown to be effective, their algorithms depend on
the topology and are based on structural metrics. Akkaya et al. [5] discussed a different
method for finding the optimal position of a base station. Pan et al. [3] considered only
free space loss and found that minimum enclosing circle gives the maximum lifetime for
a sensor network.
The optimal location of a base station can be analyzed with respect to minimum
energy expenditure or maximum lifetime of a sensor network. Even though both seems
to have the same objectives but this is not true. Pan proved in his paper that the
center of a minimum enclosing circle, is the optimal location for the n-of-n lifetime.
n-of-n lifetime means the time after which the first node dies or time upto when n out
of n nodes remain alive. The minimum enclosing circle is equivalent to minimizing the
maximum distance between the base station and any sensor node in the network, i.e.,
(x0, y0) =arg min(x,y)
(max∀i∈N
√(x− xi)2 + (y − yi)2
). (2.2.1)
Here (x0, y0) is the center of the minimum enclosing circle. This approach is also
known as the minmax algorithm for the optimum base station location [63], which
provides maximum lifetime for a static single base station in a two tiered WSN. Recently
Paul et al. [6] proposed an optimal location of base station by combining multi-hop
inter-cluster routing with single-hop intra-cluster routing where the death of the first
sensor node indicates the end of network lifetime. Paul used the center of minimum
enclosing circle as the optimal location of the base station.
2.2 Related Work 30
We prove that the overall energy consumption is minimized at the centroid of the
nodes in Appendix A, when path loss exponent α is 2. In this method, average energy
expenditure is minimized but first nodes may die earlier compared to minmax approach,
as energy consumption in the sensor nodes, is rather uneven. For example, it may
happen that most of the nodes are close to the base station, while a few nodes are
far from it. Therefore, these far off sensors use more energy than nearer ones and
hence deplete their energy at a faster rate and die sooner. This approach in known as
minimum average or minavg [63].
Vass et al. [63] has also shown that maximum relative energy is a better approach as
compared to minimum average or minmax approach. This approach uses current status
of the sensor nodes, thus being more effective for a short period of time. But in the long
run, minavg approach is a better choice. Lin et al. [31] minimized the total distance
from sensor nodes to the base station to reduce the number of data relays, and placed
the base station in an area with a high density of sensor nodes. This is the point where∑ni=1wi.di is minimized. Here di is the distance between sensor i and base station and
wi is sensor’s density in the vicinity of sensor i.
Son et al. [64] proposed that Fermat point is the best point for positioning the base
station. He used hexagonal topology for showing lifetime optimization. Power aware
base station positioning was proposed in [38]. It was proved that the choice of position
of base station depends on the data rate or equivalently, the power efficiency of the
network.
We have used a weighted centroid method for finding the optimal base station loca-
tion. This method has also been recently used for localization algorithms in the wireless
sensor networks [65-70].
2.3 Problem Formulation 31
2.3 Problem Formulation
The energy consumption in network for one round is the sum of energy dissipated by
all the clusters.
ERound =k∑j=1
ECluster(j). (2.3.2)
Now using equation (1.1.7) from energy consumption model,
ERound =L.(2n− k).Eelec + n.L.EDA +k∑i=1
Eamp(L, dbs(i)) +n∑i=1
Eamp(L, dch(i)).
(2.3.3)
As signals from all the nodes are assumed to suffer free space loss in reaching to respec-
tive cluster head, and following [2],
n∑i=1
Eamp(L, dch(i)) =L.(n− k).εfs.M2
2πk. (2.3.4)
Here, M is the length of the square field used for WSN deployment. There are three
different cases depending on distance between cluster head node and the base station.
Case 1: When all the nodes in a sensor network are near to the base station, such
that there is free space loss for the transmission from all nodes to base station. Then
ERound is given by
ERound =L
[(2n− k).Eelec + n.EDA + (n− k).εfs.
M2
2πk+ εfs
k∑j=1
d2j
]. (2.3.5)
Here dj is the distance between cluster head and the base station. After nk
rounds when
every node has become cluster head once, total energy spent is given by
Etotal =L.n
k
[(2n− k).Eelec + n.EDA + (n− k).εfs.
M2
2πk
]+ L.εfs
n∑j=1
d2j . (2.3.6)
2.3 Problem Formulation 32
Case 2: When base station is far away from all the nodes, that is multipath loss exists
for transmission from nodes to base station. Then Etotal spent in nk
rounds is given by
Etotal =E1 + L.εmp
n∑j=1
d4j , (2.3.7)
where
E1 =L.n
k
[(2n− k).Eelec + n.EDA + (n− k).εfs.
M2
2πk
]. (2.3.8)
Case 3: When some nodes are near and some nodes are far away from the base station
then Etotal is given by
Etotal =E1 + L.εfs
p∑i=1
(d2i ) + L.εmp
q∑j=1
d4j . (2.3.9)
Here node i and node j are from different sets, and p number of nodes are nearer to base
station and q number of nodes are farther from base station (di<d0 ⇒ nearer nodes
and dj ≥ d0⇒ farther nodes). Since E1 is same in all the three cases, we use second
and third term in equation (2.3.9) for further investigation.
For Case 1, Energy for transmitting data to base station Ebs in each nk
rounds is
Ebs =L.εfs
n∑j=1
d2j . (2.3.10)
For Case 2,
Ebs =L.εmp
n∑j=1
d4j . (2.3.11)
For Case 3,
Ebs =L.εfs
p∑i=1
d2i + L.εmp
q∑j=1
d4j . (2.3.12)
We need to find optimal location for the base station, which minimizes Ebs.
2.3 Problem Formulation 33
Figure 2.1: A Wireless Sensor Network topology
Let n sensor nodes be uniformly distributed in a rectangular field at (x1, y1), (x2, y2),
..., (xn, yn) respectively, and base station be deployed at (x, y) as shown in Fig.2.1.
The Euclidean distance between the base station and the nodes are d1, d2, d3, ..., dn
respectively. Here di is
di =√
(x− xi)2 + (y − yi)2. (2.3.13)
Case 1 (di < d0 ;∀ i): Let d2 be the sum of squares of all the Euclidean distances. As
all the nodes will have only free space loss, energy consumption in amplification, Ed2 in
the sensor network will be sum of amplifier energy consumption of the all the nodes.
Ed2 =L.εfs.(d21 + d2
2 + d23 + ....+ d2
n) (2.3.14)
Sum of square of all the Euclidean distances, d2 will be minimum at the centroid of
nodes [93, 95, 96]. Thus amplification energy consumption Ed2 will also be minimum if
base station is placed at the centroid.
Case 2 (di ≥ d0 ;∀ i): When all the nodes are far away from the base station and will
have only multi-path loss. Then we need to minimize
Ed4 =L.εmp.(d41 + d4
2 + d43 + ....+ d4
n), (2.3.15)
for optimal base station location.
2.4 Algorithm for Base Station Location 34
Ed4 will be minimum at a point, say F ; we can determine point F only by heuristic
search methods. This point F converge to the centroid of nodes when all the nodes are
placed symmetrically.
Case 3 (Some nodes with di < d0 and remaining nodes with di ≥ d0): When
few node are near and other remaining nodes are far away from the base station then
we have to minimize the following equation.
Ed2,4 =L.εfs.(d21 + ..+ d2
p) + L.εmp.(d41 + ..+ d4
q). (2.3.16)
Where p+ q = n and p, q ≥ 1.
Now minimizing Ed2,4 will depend on which type of nodes are dominating in energy
expenditure, nearby nodes or farther nodes. If nearby nodes are dominating i.e. p >> q,
then centroid of p nodes will be the optimal position. If q ≥ p, then farther nodes will
decide the optimal location for the base station and then centroid of q nodes (if q
nodes are symmetrically placed) will be the optimal position. When both type of nodes
are equally dominating, the optimal location of the base station can be found by the
proposed algorithm.
2.4 Algorithm for Base Station Location
Problem: To find the location of base station where Ed2,4 is minimized.
Step 1: Find centroid (Cx, Cy) of the nodes distributed in the field. This is the point,
where Ed2 is minimized, and is given by
Cx =
∑ni=1 xin
, (2.4.17)
and
Cy =
∑ni=1 yin
. (2.4.18)
2.4 Algorithm for Base Station Location 35
Step 2: Find the nodes which are at less than d0 distance from the centroid.
Step 3: Weights are calculated using centroid for all the nodes as
wi =
{1 if diC<d0diC
2
d02 if diC≥d0
(2.4.19)
Here diC is the distance between ith node and the Centroid. The justification for above
mentioned weights is as follows.
Calculation of weights for proposed algorithm: We take weight w as 1 for the
nodes which are at less than d0 distance from the centroid, and something else for other
nodes which are at equal or higher distance than d0 from the centroid. As the proposed
point will be weighted average of all points, the weights w for all the node will be 1,
when all the nodes are closer than d0 from centroid. Thus proposed point P is same as
centroid. The proposed point will be different only when some nodes are farther than
d0 distance from centroid.
Let us have a square field with length of side ‘M ’. We assume that M = d0/√
2.
In this case wherever we place our base station, only free space loss will be suffered by
transmission from all the nodes inside the field (see Fig.2.2).
Figure 2.2: Weights calculation for Algorithm
2.4 Algorithm for Base Station Location 36
We have taken three nodes at (x1, y1), (x2, y2) and (x3, y3) in the field, then the base
station location P (Px, Py) is given by proposed point will be P (Px, Py) as all the nodes
can suffer only free space loss
Px =x1 + x2 + x3
3, (2.4.20)
and
Py =y1 + y2 + y3
3. (2.4.21)
But when side M > d0/√
2 and let the base station is at centroid of nodes. Then two
nodes will suffer free space loss and third node will suffer multipath loss. If weight to
(x3, y3) is w and for other nodes it is 1, then proposed point P (Px, Py) is given by
Px =x1 + x2 + w.x3
2 + w, (2.4.22)
and
Py =y1 + y2 + w.y3
2 + w. (2.4.23)
In equation 2.4.22 and 2.4.23, assume w is very large say ∞, then Px will become x3
and Py will be y3. If w is very small say 0, then Px will become x1+x2
2and Py will be
y1+y22
. Thus, we need to make these weights more than 1 so that proposed point shifts
towards nodes, which are having multipath loss.
Weights must increase with distance from the centroid for shifting proposed point
towards nodes which are having multipath loss. Using w as diC2
d02 provide us good results
as w is always greater than 1 and proportional to the distance and thus with amplifier
energy also.
w =εmp.d
4iC
εfs.d2iC
=εmp.d
2iC
εfs=
(diC)2
(d0)2. (2.4.24)
2.4 Algorithm for Base Station Location 37
Step 4: Determine the weighted average of nodes’ positions as proposed optimal po-
sition (xp, yp) for the base station.
xp =
∑ni=1wi.xi∑ni=1wi
, (2.4.25)
and
yp =
∑ni=1wi.yi∑ni=1wi
. (2.4.26)
This is a heuristic algorithm which provides an approximate solution. For more accurate
solution (theoretical optimal point), we can place our base station in such a way that
the distance from all the nodes which are at less than d0 from the centroid, may increase
up to the maximum of d0 while reducing the distance from the remaining nodes which
are at greater than d0 distance from centroid. We have cited an example which give
more insight into the algorithm and thus justify it.
2.4.1 Optimal location of base station - An example
Let us consider three points in a 150×150 square area plane, at positions (0, 150),(0, 0)
and (150, 150). The centroid for these points turns out to be at (50, 100). The distances
from the centroid to points 1, 2 and 3 (see Fig.2.3) are d1C , d2C & d3C respectively. Here,
d1C = 70.7107 and d2C = d3C = 111.8034.
Thus amplifier energy in J/bit-node for three nodes at centroid is
EC =(εfs.d21C + εmp.d
42C + εmp.d
43C)/3 = 1.5208× 10−7 J/bit− node (2.4.27)
We have assumed the values of εmp=1.3×10−15J/bit/m4 as multipath loss constant and
εfs=10−11J/bit/m2 as free space loss constant in this example. This gives us d0 = 87.7m.
2.4 Algorithm for Base Station Location 38
Figure 2.3: Distances from Centroid ofdistributed nodes
Figure 2.4: Distances from TheoreticalOptimal Position of base station
Now we calculate the theoretical optimal position (see Fig.2.4), by placing base
station at d0 from point 1 along the perpendicular bisector of line joining points 2 and
3. This point is ( d0√2,150- d0√
2). It will have d1O, d2O and d3O distances from points 1,
2 and 3 respectively. Now d1O = 87.7000 and d2O = d3O = 107.6444. The amplifier
energy in J/bit-node for three nodes at the optimal point O is
EO =(εmp.d41O + εmp.d
42O + εmp.d
43O)/3 = 1.4200× 10−7 J/bit− node (2.4.28)
The proposed base station location P using weighted average is given by
xp =0 + 0 +
d23Cd20.150
1 +d22Cd20
+d23Cd20
=57.35, (2.4.29)
and
yp =150 + 0 +
d23Cd20.150
1 +d22Cd20
+d23Cd20
=92.65. (2.4.30)
This point (57.35, 92.65) will have d1p, d2p & d3p as distances from points 1, 2 and 3
(see Fig.2.5). Now d1p = 81.1051 and d2p = d3p = 108.9635. Thus amplifier energy in
J/bit-node for the three nodes with base station at point P , is
EP =(εfs.d21p + εmp.d
42p + εmp.d
43p)/3 = 1.4410× 10−7 J/bit− node (2.4.31)
2.4 Algorithm for Base Station Location 39
Figure 2.5: Distances from ProposedPoint as base station
Figure 2.6: Distances from center of min-imum enclosing circle as base station
We also calculated the amplifier energy by considering center of minimum enclosing
circle as base station location (see Fig.2.6). The center for minimum enclosing circle
comes out to be (75, 75). This will have d1M , d2M and d3M distances from points 1, 2
and 3 respectively. These distances are d1M = d2M = d3M = 106.066. The amplifier
energy EMEC in J/bit-node for three nodes at this point is
EMEC =(εmp.d41M + εmp.d
42M + εmp.d
43M)/3 = 1.6453× 10−7 J/bit− node. (2.4.32)
From these results, we see that the proposed point leads to,
EP − EOEO
× 100 =1.47%, (2.4.33)
more energy consumption compared to optimal location of base station.
While using centroid energy consumption exceeds by
EC − EOEO
× 100 =7.10% (2.4.34)
with respect to optimal point.
2.4 Algorithm for Base Station Location 40
If we consider center of minimum enclosing circle as base station location then dif-
ference in amplifier energy between optimal location and center of minimum enclosing
circle is
EMEC − EOEO
× 100 =15.86%. (2.4.35)
Using Lin’s weighted method [31], the base station deployment comes at (0, 150) and
the distance from node 1 and node 3 is now 150 and from node 2 is zero. so the amplifier
energy in J/bit-node for the three nodes at this point will be EL.
EL =(εmp.d41 + 0 + εmp.d
43)/3 = 4.38× 10−7 J/bit− node. (2.4.36)
where d1 = d3 = 150 and d2 = 0 and compare to optimal position the amount of excess
energy is given by
EL − EOEO
× 100 =208.45%. (2.4.37)
Comparison of existing algorithms with our algorithm is shown in Table 2.1. This shows
that how much excess energy is used in these methods [31, 3, 63] and proposed method
compared to optimal location.
Table 2.1: Excess energy used by different methods compared to the optimal locationMethod Excess Energy Used
Weighted position by Lin [31] 208.45%Minimum enclosing circle or minmax [3] 15.87%
Centroid or minavg [63] 07.10%Proposed weighted position 01.47%
2.5 Simulation And Results 41
2.5 Simulation And Results
Example in previous section shows a comparison of different techniques to find optimal
location of base station. Merely citing an example do not provide the novelty of work.
Thus in this section, generic results are shown to prove the validity of our hypothesis.
We have calculated the average amplification energy in J/bit-node, Ec, of all the
nodes, for the base station at centroid is
EC =Ek∑n
i=1 dαiC
n. (2.5.38)
Average amplification energy in J/bit-node, Ep, for base station at the proposed point
is given by
EP =Ek∑n
i=1 dαiP
n. (2.5.39)
Average amplification energy in J/bit-node, EMEC , for base station at the center of
minimum enclosing circle is given by
EMEC =Ek∑n
i=1 dαiM
n. (2.5.40)
Here E2 and E4 are εfs and εmp respectively. For nodes having diC or diP or diM< d0,
α = 2, and for all others, α = 4.
The percentage reduction in average amplification energy with base station at pro-
posed point (say P) compared to base station at centroid is given by
Percentage Reduction =EC − EPEC
.100%. (2.5.41)
The percentage reduction in average amplification energy with base station at center of
minimum enclosing circle compared to base station at centroid is given by
Percentage Reduction =EC − EMEC
EC.100%. (2.5.42)
2.5 Simulation And Results 42
Simulation was performed for 1000 different random sensor placements. Corresponding
Ec, Ep and percentage reduction have been evaluated. Average of this percentage
reduction is taken and plotted for different number of nodes (5 to 200 in step of 10)
distributed in different size field (square region with side length 30 meter to 500 meter
in steps of 10 meter).
In Fig.2.7, we have shown average percentage reduction by placing base station at
proposed point compared to centroid of distributed nodes. This has been done for
different number of nodes and length of side of square field. We see that for the area
with side less than 150 meter almost all the nodes suffer free space loss (as the size of
topology is small, farthest node from the centroid of nodes will have distance less than
d0) so point P converges to centroid and the percentage reduction become zero in this
area (as all the weights become 1). The reduction peaks when number of nodes ranges
from 5 to 50 and length of square field between 160 meter to 270 meter (this is region
where both types of nodes are existing, nodes which suffer free space loss as well as the
multipath loss. As some nodes will have distance less than d0 and rest will have greater
than d0 from the centroid of distributed nodes). It can be observed that the percentage
reduction is significantly large in this region and varies from 0 to 4.85%. For side length
of more than 300 meter and for 5 number of nodes the average percentage reduction
becomes negative as the weights are high in magnitude comparison to the side less than
300 meter of topology weights. But for the higher number of nodes because of average
process the higher weights are not dominating. Thus the average percentage reduction
is near to zero in this case.
2.5 Simulation And Results 43
Figure 2.7: Average percentage reduction in amplifier energy for proposed point com-pared to centroid
Figure 2.8: Average percentage reduction in amplifier energy when negative values areclipped to zero
Fig.2.8 shows the results when only the positive average percentage reduction has
been considered. The results for negative reduction are clipped to zero to emphasize
positive reduction results in the figure. This gives the region where our algorithm will
be best fit. The ignored values corresponds to large field with few nodes or small field
with large number of nodes, which is not of much interest.
2.5 Simulation And Results 44
Figure 2.9: Average percentage reduction in amplifier energy for proposed point com-pared to centroid
Figure 2.10: Average percentage reduction in amplifier energy when negative values areclipped to zero
2.5 Simulation And Results 45
Fig.2.11 shows the improvement w.r.t. centroid when center of the minimum enclos-
ing circle is considered as the base station location. This shows that it is never better
than centroid for given distribution. Fig.2.9, Fig.2.10 and Fig.2.11 present the three
dimensional figures for the results shown in Fig.2.7, Fig.2.8 and Fig.2.12 respectively.
Figure 2.11: Average percentage reduction in amplifier energy for minimum enclosingcircle compared to centroid
Figure 2.12: Average percentage reduction in amplifier energy for minimum enclosingcircle compared to centroid
2.5 Simulation And Results 46
We have used Table 2.2 and 2.3 to validate the results shown in Fig.2.7 and Fig.2.8.
We have calculated ratio of number of nodes (Nd2) lying inside the circle of radius d0 to
the ones outside the circle (Nd4) with center being the centroid and then the proposed
point respectively in Table 2.2 and 2.3. This node ratio is given by
Node Ratio =Nd2
Nd4
. (2.5.43)
As we increase number of nodes for a constant length of square side usually this ratio
decreases. This shows that number of nodes who suffer multipath loss increases with
node density. Table 2.2 and 2.3 verify the same because as we increase the number of
node for a constant size of topology and vice versa, gain in amplifier energy at proposed
point reduces. This is because the point F is dominated by number of nodes with
multipath loss.
Table 2.2: Nodes’ ratio when centroid as centerLength of square side Number of Nodes
⇓ 30 50 70 90 110180m 3.90 3.49 3.27 3.22 3.14200m 2.01 1.79 1.70 1.66 1.64220m 1.22 1.14 1.07 1.06 1.05240m 0.89 0.80 0.77 0.76 0.76260m 0.66 0.61 0.59 0.58 0.57
Table 2.3: Nodes’ ratio when proposed point as centerLength of square side Number of Nodes
⇓ 30 50 70 90 110180m 3.75 3.42 3.23 3.17 3.11200m 1.85 1.69 1.62 1.60 1.59220m 1.09 1.07 1.03 1.02 1.02240m 0.78 0.75 0.73 0.73 0.73260m 0.59 0.57 0.56 0.55 0.55
Results in Table 2.2 and 2.3 verify the results of Fig.2.7, Fig.2.8, Fig.2.9 and Fig.2.10.
2.6 Conclusions 47
Table 2.4 shows the distance between the centroid and the proposed point. In this
table, for a fixed field size, as we increase the number of nodes this distance decreases.
Table 2.4 also verify our results that as we increase the node density for a smaller
constant size of topology the proposed point and centroid of nodes comes closer to each
other. Also the gain in amplifier energy at proposed point reduces as the centroid of
nodes dominates. For smaller size of topology, proposed point merge with centroid so
the distance become zero for smaller topology. As we increase the size of topology the
distance between the proposed point and centroid increase this shows that centroid is
less dominating as we increase the size of topology.
Table 2.4: Distance in meter between centroid and point PLength of square side Number of Nodes
⇓ 10 50 90 130 17030m 0 0 0 0 0180m 3.8 2.3 1.8 1.6 1.4330m 21.3 11.4 8.7 7.2 6.7480m 34.6 17.6 13.0 11.6 10.3
2.6 Conclusions
Lifetime maximization and overall energy consumption minimization are two different
problems in a wireless sensor network. The optimal point for the maximum lifetime of
a sensor network is the center of the minimum enclosing circle of all nodes. If all the
nodes have free space path loss (α = 2) then the energy consumption is minimized when
the base station is placed at centroid. If all the nodes suffer only multipath loss (α = 4)
then the optimal base station location will be point F . For the case when both type
of nodes (some with free space path loss and some with multipath loss) are there in a
sensor network, we have proposed a new location of the base station for minimization of
2.6 Conclusions 48
energy consumption. This point is centroid of weighted position of the deployed nodes.
These weights are dependent on the path loss. Our proposed point provides a sub-
optimal solution. Results show that the proposed point provides a better performance
compared to centroid as well as center of minimum enclosing circle for sensor nodes.
These results can be used for deployment of a base station in two tiered wireless sensor
networks.
Chapter 3
Anchor Free Localization Techniquefor Singlehop Wireless SensorNetwork
3.1 Introduction
Most of the applications of wireless sensor network are associated with position infor-
mation. Generally GPS devices are not used in large scale sensor network due to the
cost and poor availability of GPS signals due the environmental problems. It is very
important to acquire the nodes position without GPS. As in most of the applications
sensor nodes are directly connected to the base station. We have proposed an anchor
free localization algorithm for singlehop wireless sensor network. The proposed algo-
rithm is based on internode distances. Average of the internode distances is used as
main metric to localize the reference node in the proposed algorithm.
3.2 Related Work 50
3.2 Related Work
Capkun [8] proposed first Self Positioning Algorithm (SPA) for Ad-Hoc wireless net-
works. After that, Anchor Free Localization (AFL) [10], Assumption Based Coordinates
(ABC) [9], Knowledge-based Positioning System (KPS) [55], GPS-less low cost outdoor
localization [7], Matrix transform-based Self Positioning Algorithm (MSPA) [57], Mobile
Geographic Distributed Localization (MGDL) [71], Ad Hoc Positioning System (APS)
[47] were proposed.
In SPA [8] algorithm, every node establishes its local coordinate system by setting
itself as the origin. Two other nodes are randomly chosen such that the three nodes
do not lie on the same line to minimize triangulation error and can communicate with
each other. If from a node the distance to each of the three nodes can be estimated
then it can be localized.
Anchor Free Localization (AFL) [10] algorithm has two phases. First phase is a
heuristic that produces a layout which looks similar to the original layout. In second
phase, nodes locally correct and balance the optimal solution by adopting the opti-
mization algorithm to ensure that the energy in new location is less than that in the
original location. In AFL, localization is dependent on communication range of nodes
and gives good result only when average connectivity of network is more than seven.
Localization error may increase due to unavailability of direct path between nodes and
results into an estimate which is multiple of communication radius R, depending on
number of hops.
3.2 Related Work 51
Assumption Based Coordinates (ABC) [9] uses cooperative ranging approach for
localization. This algorithm localizes one unknown node at a time by making assump-
tions, compensating the errors through corrections and use of redundant calculations as
more and more information becomes available. In ABC accumulation of errors increases
with number of beacon-free node.
Knowledge-based Positioning System (KPS) [55] algorithm is based on pre-configured
knowledge of probability distribution of configuration. This algorithm uses two main
concepts which are configuration point and the probability distribution function. Con-
figuration point is the location of the determined point which is in the node group, when
deploying the nodes. The probability distribution function is obeyed by each group’s
nodes after the sensor nodes’ configuration.
Bulusu [7] in his paper on GPS-less low cost outdoor localization, used overlapping
region of nodes to localizes new nodes considering some reference nodes which are
already localized. Nodes localized themselves to the centroid of their reference points
and localization error depend on separation distance between two adjacent reference
nodes and the transmission range of these reference points.
Wang [57] developed a clustering-based approach for the local coordinate system for-
mation wherein a multiplication factor is introduced to regulate the number of master
and slave nodes and the degree of connectivity among master nodes. Using homoge-
neous coordinates, Wang derived a transformation matrix between the two Cartesian
coordinate systems to efficiently merge them into a global coordinate system.
APS [47] and Peng et al. [50] used Angle Of Arrival (AoA) capability of the nodes
to derive position information. Moses et al. [11] used time of arrival and direction of
arrival to localize nodes for anchor free localization.
3.3 Challenges with Localization Techniques 52
3.3 Challenges with Localization Techniques
In SPA algorithm, communication cost and convergence time increases exponentially
with the number of nodes. Thus, SPA algorithm is not useful for energy constrained
WSN. ABC algorithm can be realized easily, but the localization accuracy is poor.
KPS require deployment knowledge which is generally not available with sensor nodes.
GPS-less low cost outdoor localization also depend on communication range and sepa-
ration of reference nodes. AFL algorithm is fully distributed, only requires a relatively
small number of communications and simple calculations, and has good scalability and
robustness. Apart from these, following problems are generally faced by this algorithm.
3.3.1 Localization using Hop Count
There are two problems with use of hop count as a measurement of distance [72].
1. Distance measurements are always integral multiples of maximum radio
range R : In AFL localization, five reference nodes n0, n1, n2, n3 and n4 are found
in phase 1 as shown in Fig.3.1. Node n0 lies in the center of topology. For a node nj,
hop-counts h0,j, h1,j, h2,j, h3,j and h4,j from n0, n1, n2, n3 and n4 respectively, are used
for localization. Localization is expressed in polar coordinates as (ρj, θj)
ρj =h0,j ×R; (3.3.1)
here R is maximum radio range.
θj =arctan
(h1,j − h2,j
h3,j − h4,j
). (3.3.2)
3.3 Challenges with Localization Techniques 53
Figure 3.1: Localization of a node in AFL
As we see in Fig.3.2, node A and B lie within one hop transmission from center node,
thus in AFL localization ρA = ρB = R as h0,A = h0,B = 1. Similarly for node C and
node D, ρC = ρD = 2R as h0,C = h0,D = 2. Thus nodes are localized with resolution of
maximum radio range R. This inaccuracy corresponds to a total error of about 0.5R
per measurement, which can be too high for some applications.
Figure 3.2: Nodes are localized integral multiples of maximum radio range R
Nagpal [73] had shown that even better hop count distance estimates can be com-
puted by averaging distances with neighbors, which can reduce hop count error down
to 0.2R. But for multihop scenario, it can accumulate to high localization error. As
average connectivity increases, the probability of nodes lying along the straight-line
path increases rapidly this happens till average connectivity of 15 nodes, there after
any further increase in sensor density has diminishing returns. Thus algorithms using
hop count techniques work properly with average connectivity in the range of 7 to 15.
3.3 Challenges with Localization Techniques 54
Thus many different nodes will be assigned same ρj in AFL localization. If R is
increased then more number of nodes will have same ρj and if decreased then problem
of unavailability of direct edges increases, This is discussed in next subsection.
2. Problem of unavailability of direct edges : As we see in Fig.3.3. Node A
and B are not in direct communication range as actual distance between A and B, i.e.
dAB >R. Similarly node A and node C are not in direct communication range due to
obstacle in between these nodes. Here hA,B = 3 and hA,C = 4, thus algorithm using
hop count localization will localize node B at 3R distance and node C at 4R distance
away from the node A respectively.
Figure 3.3: Localization using Hop Count
Such cases will be very frequently observed in a sensor network with low density
of sensor nodes. Thus unavailability of direct edges may lead to substantially high
localization errors. Only in case of node D, the hop distance measurement will be
correct as hA,D = 4 as the node D is approximately 4R away from node A.
For different value of communication range, the connectivity varies and thus average
distance due to number of hops. Actual average distance between nodes remains con-
stant as topology is fixed. Error between actual average distance and hopcount average
distance is
%Error =(Dhave −Dave).100
Dave
, (3.3.3)
3.3 Challenges with Localization Techniques 55
We have calculated actual average distance, Dave, between 40 nodes randomly dis-
tributed in a 100× 100 m2 field and then average distance in terms of number of hops,
Dhave, between nodes by choosing different communication range from 30m to 50m
with step of 5m. Results are shown in Table 3.1.
Table 3.1: Average Actual Distance vs Average Distance due to number of hopsMax. RadioRange(m)
AverageConnectivity
Dave(m) Dhave (m) % Error
30 9.40 946.6 1395.7 47.4435 12.20 946.6 1383.3 46.1340 14.75 946.6 1429.0 50.9545 17.75 946.6 1431.0 51.1650 21.20 946.6 1445.0 52.64
The average of this percentage error comes out to be 49.66%. Thus one can conclude
that merely using hop counts will give an error of approximately 50% for localization
in any algorithm irrespective of average connectivity for the nodes.
3.3.2 Accumulation of Localization Error
In the present techniques which are used for anchor free localization, first the reference
nodes are detected, and then all other nodes are localized with the help of these reference
nodes. For multihop scenario, the nodes are localized in tiered fashion. First tier nodes
which are nearest to reference nodes, are localized first with some localization error.
Then second tier of nodes are localized using the first tier nodes as a reference nodes.
So localization error of first tier nodes is directly added to second tier of nodes. Thus
localization error accumulates as we localize higher tier nodes. This is the serious
problem with ABC algorithm [9], where the average estimated position error reaches as
much as 60%.
3.4 Metric for Proposed Algorithm 56
3.4 Metric for Proposed Algorithm
Let n sensor nodes be randomly distributed in a topology at (x1, y1), (x2, y2), (x3, y3),...
and (xn, yn). Let distances between node i and other nodes be di1, di2, ....din. Here dij
is defined as
dij =√
(xi − xj)2 + (yi − yj)2 (3.4.4)
The average distance of a node (Dave) among distributed nodes is having nearly similar
estimation to the distance from that node to the centroid of nodes (Dc). As we do have
position information to calculate the centroid, we use Dave for localization. This Dave
for ith node among n distributed nodes is defined as [93, 95, 96].
(Dave)i =
∑nj=1 dij
n− 1for ∀ j; j 6= i (3.4.5)
Normalized Dave can be obtained by dividing maximum Dave among all nodes to Dave
to particular node. ˆDave is given by:
ˆDave =Dave
(Dave)max(3.4.6)
We have plotted ˆDave and Dc (Normalized distance of all the nodes from the centroid
of distributed nodes) for 20 distributed nodes in 100 × 100 square meter topology as
shown in Fig.3.4. This shows that as we increase the distance from the centroid of a
node the Dave increase similarly. As we proposed anchor free localization so we have to
localize reference nodes first then other nodes. We used this Dave to localize reference
nodes. Thus using only internode distance we can localize reference nodes.
3.5 Algorithm for Singlehop Localization 57
Figure 3.4: Relation between Dc and ˆDave
3.5 Algorithm for Singlehop Localization
Let n sensor nodes are randomly distributed in a topology.
Step 1 : Every node broadcast a hello message with its id using TDMA scheduling.
Each node calculates (n-1) distances (dij) using RSSI from all other nodes. Further
every node calculates average of all the distance estimates. Thus, Dave for ith node is
calculated using equation (3.4.5).
Step 2 : Every node will broadcast this Dave and the node having highest Dave is
the farthest node from the centroid and will lie at the corner of topology as shown in
Fig.3.5. This node will be first reference node for the system.
Step 3 : Farthest node from the first reference node will be second reference node
as shown in Fig.3.6. This node will always lie in opposite corner of node 1.
3.5 Algorithm for Singlehop Localization 58
Step 4 : Now node 2 will send data of all distances d2j to node 1. Node 1 finds
(d1j + d2j) ∀ j; j 6= 1, 2. The node which has highest value of this sum will be the node
farthest from 1 & 2 nodes. This node will also lie in the corner of topology, shown as
node 3 in Fig.3.7. This node will be third reference node for the system.
Step 5 : Now node 3 find the farthest node from it and this will be fourth refer-
ence node. This node will lie in opposite corner of node 3 as shown in Fig.3.8.
Step 6: Node having minimum Dave will have minimum distance from the centroid.
This node will be the fifth reference node for our protocol. This node will lie at the
center of topology as shown in Fig.3.9 and act as origin for localization.
Thus, we have calculated five reference positions. Now we localize rest of nodes as
follows. Origin is assigned to reference node 5. With respect this node, we calculate
angle of different nodes for localization. Reference node 1 is considered on positive
x axis, and reference node 2 considered near to negative x axis. Reference node 3 is
considered near to positive y axis, and reference node 4 is considered near to negative
y axis.
Thus node j can be localized in the topology by estimating θ as shown in Fig.3.10.
cos(θ) =d2
51 + d25j − d2
1j
2.d51.d5j
(3.5.7)
Thus node j is localized in topology depending on in which quadrant the node is lying.
For example if the node j is lying in 1st quadrant then coordinate for node j is given by
x =d5j.cosθ, (3.5.8)
and
y =d5j.sinθ. (3.5.9)
3.5 Algorithm for Singlehop Localization 59
Figure 3.5: Estimation of Refer-ence node 1
Figure 3.6: Estimation of Refer-ence node 2
Figure 3.7: Estimation of Refer-ence node 3
Figure 3.8: Estimation of Refer-ence node 4
Figure 3.9: Estimation of Refer-ence node 5
Figure 3.10: Localization fornode j
3.6 Simulation & Results 60
Figure 3.11: Localization Problem for non rectangular sensor field
This method will give incorrect results when the field in which sensor nodes are
distributed is non rectangular e.g. as shown in Fig.3.11. Here the reference nodes are
not diagonally away so will give high triangulation error which lead to wrong estimation
of position of sensor nodes in this sensor field. One need to build additional algorithms
to split such fields into multiple rectangular fields and use this algorithm in them.
3.6 Simulation & Results
We have used the average percentage error of the calculated distances compared to the
true distance between neighbors, as well as Global Energy Ratio (GER) which capture
the true behavior of auto-localization algorithms [10]. The global energy ratio is the
root-mean-square normalized error value of the node-to-node distances and is given by
GER =
√∑i, j: i<j e
2ij
N(N−1)2
(3.6.10)
Normalized error eij is given by
eij =dij − dijdij
(3.6.11)
Here the error eij is the difference between the true distance dij and the estimated
distance from the algorithms, dij.
3.6 Simulation & Results 61
This measure captures both the edge length errors and the structural error of the
graph, because it has contributions from both nodes that are neighbors as well as nodes
that are not. GER metric also captures the errors in the final configuration due to
erroneous range estimates.
Fig.3.12 shows the graph between GER and number of nodes for different Size of
topology. In the proposed algorithm simulation, we have taken square fields of lengths
100m to 400m with step of 100 and number of nodes varied from 100 to 250 with step
of 50.
Fig.3.13 shows the graph between GER and Size of topology for different communica-
tion range of nodes. GER remains nearly constant. This shows that GER is independent
of size of topology in a wireless sensor network.
Figure 3.12: GER vs size of topology forconstant number of nodes
Figure 3.13: GER vs number of nodes forconstant size of topology
3.6 Simulation & Results 62
Figure 3.14: Average localization errorvs number of nodes for constant size oftopology
Figure 3.15: Average localization error vssize of topology for constant number ofnodes
We also have used the same metric as used in [72, 45], for localization error estimation
for proposed algorithm.
Error =1
n
n∑i=1
||ei − li||. (3.6.12)
where n is the number of sensors, li denotes the real location of the ith node and ei
denotes the location estimate of the ith sensor where i = 1, ..., n.
We have used same parameters as in GER calculations. Average localization error
for different number of nodes in proposed algorithm is shown in Fig.3.14. Average
localization error decreases with number of nodes.
Fig.3.15 shows the graph between Average localization error and Size of topology
for different communication range of sensor nodes. Average localization error increases
with size of topology. As GER is independent of size of topology average localization
error gives better performance for localization error.
3.7 Conclusions 63
3.7 Conclusions
In this chapter we proposed anchor free localization techniques for singlehop wireless
sensor network using internode distance only. Proposed algorithm is very simple and
effective in terms of average localization error. GER results shows that localization
error decreases with increases in number of nodes. Here we are able to localize the
sensor nodes with average localization error less than 6% error in all topologies. The
proposed new metric Dave is able to perform localization in anchor free scenario.
Chapter 4
Balancing Energy Consumption in aWireless Sensor Network
4.1 Introduction
Balancing energy consumption is an important issue. This leads to maximum lifetime
of a sensor network. Energy consumption in a node depends on many factors such
as distance from the sink (base station), number of nodes supported by a node when
it is a cluster head and duty cycle of transmission by the nodes. Depending on these
parameters, some sensor nodes spend higher amount of energy compared to other nodes.
This leads to the uneven energy consumption, which reduces the time after which the
first node dies in the sensor network. By adjusting these parameters, we can force the
balanced energy consumption among sensor nodes thereby increasing its lifetime. This
thesis pertains to wireless sensor network which is a highly distributed system with very
limited computing capacity and only locally observed information at each node. The
clustering methods which have been given mostly in literature refer to algorithms and
methods when all information is known to the executor of algorithm and he then forms
the clusters based on various parameters. We have not looked into these algorithm
4.2 Related Work 65
as these are impractical for wireless sensor networks. In WSN, we will usually prefer
extremely simple algorithms which can be executed at the sensor nodes considering only
the local observations. In such a scenario, no node have complete picture of the cluster
formation. LEACH was first logical mechanism and later only variation of this basic
mechanisms were invented by various researchers for improving the performance. We
have presented the comparison with most of such algorithms. In this chapter, we have
proposed two methods for balancing energy consumption in a wireless sensor network.
4.2 Related Work
There are two main category of sensor network depending on routing of data to sink.
One is multi-hop and another is single-hop. In a multi-hop wireless sensor network
energy hole is a common problem. Some nodes are used frequently to hand over the
data to base station as all the forwarded data come through these nodes only. We
need to take care of all such nodes which are closer to base station to avoid energy hole
problem. This problem is solved by different researcher with different methods. Haenggi
(2003) [74] proposed four methods to solve the energy hole problem. In first method,
he proposed that distance between nodes must be decreased as the nodes are moving
towards base station. This ensure large data can be transmitted to a lower distance with
equal amount of energy. But this assumes that nodes are spread in an ordered fashion
in the sensor field. In the second method, he proposed balanced data compression to
achieve effectively one packet handing by each node. In third method, he showed data
can be directly handed over to base station rather than forwarding to next node if base
station is within its communication range. In his last method, he used equalization of
the end-to-end reliability method to balance energy consumption in a wireless sensor
4.2 Related Work 66
network. Olariu et al. (2006) [75] developed the theoretical model of the uneven energy
depletion phenomenon in sink based wireless sensor networks. Powell et al. (2007) [76]
computed the optimal parameters for probabilistic data propagation in wireless sensor
network and proved that these parameters maximize the networks lifetime in centralized
manner. Bouabdallah et al. (2009) [77] showed that significant amount of energy can
be saved by sending data through multiple paths, instead of a single path. Zhang H. et
al. (2009) [78] proposed energy-balanced data gathering (EBDG) protocol for solving
energy hole problem. Zhang X. et al. (2011) [79] solved this problem by deploying
nodes non uniformly in a wireless sensor network. The density of nodes is increased
towards the direction of base station.
But in most of the applications only one tiered or two tiered sensor network is used.
Single hop or two tiered scenario is different compared to multi-hop scenario, thus energy
consumption also will be different. Two tiered architecture of sensor network itself is a
solution to energy hole problem in a wireless sensor network. For example, in LEACH
if a node is selected as cluster head again and again then its energy will deplete first.
Such a scenario we need different protocols to balance the energy consumption. Many
researchers have tried to balance energy consumption in two tiered sensor network. In
LEACH, cluster head selection among sensor nodes is done randomly and two tiered
sensor network topology is used. Although LEACH algorithm avoids energy hole prob-
lem but cluster heads deplete their energy much faster compared to other nodes. This
is main cause of unbalanced energy consumption in LEACH. Hybrid Energy-Efficient
Distributed Clustering (HEED) proposed by Younis et al. (2004) [15, 16] is another
well-known clustering based routing algorithm in two tiered wireless sensor network. In
this work relationship between residual energy and reference energy is used for cluster
head selection. Zytoune et al. (2009) [80] worked on balancing energy consumption in
4.2 Related Work 67
two tiered wireless sensor network. Their probabilistic cluster head selection depends
on the position of a node. They divided sensor field in sub-regions (N1, N2, N3, and
N4 as shown in Fig.4.1) and depending on distance (d1, d2, d3, and d4) between these
sub-regions and the base station probability of all nodes (p1, p2, p3, and p4) are decided.
Figure 4.1: Balanced Energy Algorithm
Zytoune et al. used stochastic cluster head selection algorithm by a modifying the
probability of each node to become cluster head based on its required energy to transmit
to the base station. Zytoune et al. also solved this problem for two tiered wireless sensor
networks.
Thus due to relevance of two tiered wireless sensor networks, We have discussed
problem of balancing energy in two tired wireless sensor network in this thesis. We
have proposed two methods for energy balancing. The first one is for small area sensor
network, where the eligibility of cluster head is decided by number of nodes supported
in last round. In second one, we have varied duty cycle of a node depending on the
distance from a node to base station. This work is similar to Zytoune et al. balanced
energy consumption algorithm.
4.3 Method 1: N-LEACH, a Balanced Cost Cluster-Heads Selection Algorithm 68
4.3 Method 1: N-LEACH, a Balanced Cost Cluster-
Heads Selection Algorithm
In LEACH algorithm [1], load on individual cluster head nodes increases (decreases)
depending on more (less) than optimum number of nodes supported by it. As only the
number of cluster heads are fixed and the remaining nodes choose the nearest cluster-
head node for data transmission, the number of supported nodes for each cluster head
vary quite a bit. This leads to uneven load distribution among nodes in a Wireless
Sensor Network (WSN). We propose a new cluster head selection method, N-LEACH,
based on number of supported nodes in earlier rounds.1
Problem Statement: Any node become cluster head once in a epoch and rest round
it behave like normal nodes. When a node i becomes a cluster head and it supports Ni
nodes for data transmission, the energy consumption is
ECH =L.Ni.(ERX + EDA) + ETX−amp(L, ditobs) (4.3.1)
As there are nk
nodes in a cluster. So node i will behave as a normal node for (nk− 1)
rounds in each epoch. Thus energy spent by node i as non cluster-head node is
Enon−CH =(n
k− 1).ETX−amp(L, ditoch) (4.3.2)
Energy consumption for node i in one epoch is sum of equations (4.3.1) and (4.3.2).
Ei =(n
k− 1).ETX−amp(L, ditoch) + L.Ni.(ERX + EDA) + ETX−amp(L, ditobs) (4.3.3)
Similarly for jth node energy consumption in one epoch is
Ej =(n
k− 1).ETX−amp(L, djtoch) + L.Nj.(ERX + EDA) + ETX−amp(L, djtobs) (4.3.4)
1Rajiv Kr. Tripathi, Yatindra Nath Singh and Nischal Kr. Verma, “N-LEACH, a balancedcost cluster-heads selection algorithm for Wireless Sensor Network,” Proceeding of IEEE NationalConference on Communications, NCC 2012 , Vol.1, No.1, IIT Kharagpur, India, Feb. 2012, pp.1-5.
4.3 Method 1: N-LEACH, a Balanced Cost Cluster-Heads Selection Algorithm 69
In a small area network energy consumed in transmission of data from non cluster head
nodes to cluster heads is nearly same as energy consumed in transmission from cluster
head nodes to base station. It is with the assumption that base station is located at
the centroid of the deployed sensor nodes. So ETX−amp is approximately same for both
the nodes either for transmission from normal node to cluster head or for transmission
from cluster head node to base station.
Thus for a small area network
ETX−amp(L, ditoch) ≈ ETX−amp(L, djtoch), (4.3.5)
and
ETX−amp(L, ditobs) ≈ ETX−amp(L, djtobs). (4.3.6)
If Ni is much larger than Nj then energy consumed in receiving and data aggregation
play more important role compared to transmission of data.
Ei − Ej ∼=L. (Ni −Nj) . (ERX + EDA). (4.3.7)
So here main difference between nodes’ energy consumption is due to reception and
data aggregation only.
If the optimum number of clusters are k then the average number of supported nodes
by a cluster head are nk
for a particular round. The optimum number of cluster head
for a two tiered wireless sensor network is defined by Heinzelman et al. [2]
k =
√n√2π·√εfsεmp· M
d2toBS
(4.3.8)
Where n is the total number of nodes, M is length of side of square topology and dtoBS
is the average distance of different cluster heads to base station.
4.4 Proposed Cluster Head Selection Algorithm 70
The cluster number in some round may become less than k or in some round greater
than k. If the cluster head nodes support more than nk
nodes, then it spends more
energy which it should have been spending in nk
number of rounds. If cluster heads are
more than k, then cluster head nodes support less than nk
nodes. Thus cluster heads
spend less energy compared to others nodes in nk
number of rounds.
The different cluster numbers [84] in WSNs along with random choice of cluster
heads will make the node numbers in every cluster different and leading to uneven
energy dissipation for all nodes in each round. Therefore, we propose a new technique
which leads to dissipation of approximately 1 epoch energy on an average in each round.
4.4 Proposed Cluster Head Selection Algorithm
Step 1: In the first round of data transmission G is set to be −1 for all nodes.
Step 2: Operation for the epoch is performed after every nk
rounds. The value of G is
reduced by one for all the nodes which are having G ≥ 0 at the end of each epoch.
Step 3: All the nodes which are having G < 0 are eligible to become cluster head.
Step 4: The nodes will become cluster-head by randomly choosing a threshold Tn
between 0 and 1 using equation (1.2.18).
Step 5: If a node becomes a cluster head, it supports N number of nodes. If this N is
greater than Nave = nk, then this node is going to loose higher amount of energy, and if
N is less than Nave, then this node is going to save some energy compared to other nodes
which have already become cluster heads or will become cluster head within remaining
rounds in same epoch.
4.4 Proposed Cluster Head Selection Algorithm 71
We add(kn
)*N to G when a node become cluster head. Thus value of G become
proportional to N, if a node support large number of nodes then this will loose its
eligibility criterion for next few nk
rounds as this will not become eligible unless G ≤ 0.
If it supports lesser than Nave nodes then it remains eligible to become cluster head
again. Thus a node can spend only Nave energy in every nk
rounds. The LEACH and the
proposed algorithm are shown in the Fig.4.2 and Fig.4.3 respectively as the flowcharts.
Figure 4.2: LEACH Algorithm Flowchart
4.4 Proposed Cluster Head Selection Algorithm 72
Figure 4.3: Proposed Algorithm Flowchart
4.5 Method 2: Balancing Energy Consumption using Variable Duty Cycle (VDC) 73
4.5 Method 2: Balancing Energy Consumption us-
ing Variable Duty Cycle (VDC)
A sensor node consumes energy for event sensing, coding, modulation, transmission,
reception and aggregation of data. Data transmission has highest share in total energy
consumption [81, 82]. The required transmission power of a wireless radio is propor-
tional to square or an even higher order exponent of distance in the presence of obstacles.
Thus the distance is main factor responsible for energy consumption in a wireless sensor
network. As some nodes are nearer to sink they loose lesser energy compared to distant
nodes. This leads to an unbalanced energy consumption among sensor nodes. In this
subsection, we have proposed the use of variable duty cycle of sensor nodes for balancing
the energy consumption depending on distance of a sensor node from the base station.
4.6 VDC Algorithm
Zytoune et al. [83],[25] analyzed that sensor nodes drain out non-uniform energy due
to different distances between sensor nodes and base station. Thus nodes will have
different amount of residual energy depending on distance from the base station. One
of the solutions is that the nodes with more energy should be cluster heads more often
than the nodes with less energy. This ensures that all nodes die approximately at the
same time. Zytoune et al. used probability of cluster head selection as a function of
distance of the node from the base station to solve this problem.
Problem Statement: Let there be two nodes, node i and node j, and Ei and Ej are
the energy required to send one bit from respective nodes to base station as shown in
Fig.4.4. Node i will consume less energy compare to node j which is farther compared
4.6 VDC Algorithm 74
to node i, from the base station.
Figure 4.4: A Wireless Sensor Network topology
If we fix duty cycle for node i, Di, duty cycle for node j for equal average energy
consumption per cycle will be
Dj =Ei.Di
Ej. (4.6.9)
Here the duty cycle means, periodicity of transmission of data for a node. If transmission
cycle period is Ti then Di = 1Ti
. Thus EiTi
=EjTj
.
Every node can communicate with base station and vice versa. Base station will
inform all nodes how far it is from the nearest node and each node will estimate its own
distance from base station (Thus energy which need to be transmitted per bit to base
station). Thus all nodes can set their duty cycle Dj using equation (4.6.9). Let di be the
distance of the nearest node from the base station. Total number of bits transmitted
by node i in one cycle be Ni. It will consume least energy for transmitting data packets
to base station and has highest probability to die in the last. We take this nearest node
as reference; other nodes who want to have same energy consumption must transmit
lesser number of bits compared to this reference node.
We can achieve this adjustment by reducing duty cycle of the nodes. Here assumption
is that each node transmits same number of bits per cycle. Thus Nj=Ni,∀ j. Thus nodes
4.7 Simulation and Results 75
differ in periodicity of transmission (duty cycle).
Energy used to transmit Ni bits from reference node to base station is
ETX(Ni, di) ={Ni.Eelec+Ni.εfs.d
2i if di<d0
Ni.Eelec+Ni.εmp.d4i if di≥d0
. (4.6.10)
Now for any node j, the energy used for transmission is
ETX(Nj, dj) ={Nj .Eelec+Nj .εfs.d
2j if dj<d0
Nj .Eelec+Nj .εmp.d4j if dj≥d0
. (4.6.11)
We can determine the duty cycle for node j using equation (4.6.9), (4.6.10) and (4.6.11)
with assumption that nearest node to base station will have free space loss:
Dj =
ETX+εfs.di
2
ETX+εfs.dj2 .Di if dj<d0
ETX+εfs.d2i
ETX+εmp.d4j
.Di if dj≥d0. (4.6.12)
Thus we can achieve the balanced energy consumption in the sensor network by adjust-
ing the duty cycle of all sensor nodes using equation (4.6.12).
4.7 Simulation and Results
We have used same simulation parameters for both the methods. These parameters are
as follows: 100 nodes were randomly deployed in 100× 100 m2 area. The initial power
of all nodes is considered to be 0.5 J. Data packet size is assumed L = 20000bits/packet.
Base station is assumed to be at centroid of nodes. We carry out a comparison among
the LEACH, Fixed-LEACH and the proposed methods in this simulation.
The simulation is performed for 50 different random topologies in all the algorithms.
The averaged results are shown and analyzed for their effectiveness. Duty cycle for
4.7 Simulation and Results 76
nodes in LEACH, Fixed-LEACH and in node nearest to base station in the proposed
method 2 (VDC) has been fixed to 50%. Other nodes in proposed algorithm will have
duty cycle less that 50% depending on the distance from the base station.
Fig.4.5 shows percentage of node alive in the topology with time. LEACH and Fixed-
LEACH nodes die earlier compared to proposed method 1 and 2. Thus showing that
N-LEACH and VDC performs better than LEACH and Fixed-LEACH. In method 1
algorithm curve is above than that for LEACH until around 50% nodes are dead and As
N-LEACH utilizes the nodes energy evenly. This balancing results in sudden reduction
of alive nodes in the end. In method 2, curve is much above than LEACH and Fixed-
LEACH for all the time as VDC algorithm utilizes less amount of energy compared to
LEACH and Fixed-LEACH. It may be noted that channel capacity has been sacrificed
for far off nodes.
Figure 4.5: Number of Alive Nodes vs Time (in number of rounds)
4.7 Simulation and Results 77
Table 4.1 shows the number of rounds after which 1st and 50% node are dead in
the network for 50 different topologies for all the protocols. Lifetime for N-LEACH
and VDC is more than 12.21% and 12.3% higher compared to LEACH and 5.1% and
6.1% higher compared to Fixed-LEACH for first node death respectively. Lifetime for
VDC is more than 19% higher compared to LEACH and 18.2% higher compared to
Fixed-LEACH for 50% node death.
Table 4.1: Average Lifetime of nodes in number of rounds% of dead node Algorithm
⇓ LEACH Fixed-LEACH Method 1 Method 21 156 167 175 17850 241 246 243 301
We have calculated average energy consumption of all the nodes after every epoch.
Using this we have calculated residual energy of all the nodes at the end of each epoch.
Then standard deviation of residual energy after every epoch is plotted in Fig.4.6, The
method of calculating standard deviation of residual energy was earlier used by Shin
[85]. Results show that the variation in balance energy after every epoch in the sensor
network is always less for the proposed algorithms as compared to LEACH as well as
Fixed-LEACH protocol.
Figure 4.6: Standard Deviation of residual energy for all the nodes at each epoch
4.7 Simulation and Results 78
We have also estimated cumulative distribution function and probability density
function of nodes for their mean energy lost in one epoch. We have used Gaussian kernel
density estimator [99] defined for given N independent realizations χN ≡ {X1, ..., XN}.
This method is used to estimate probability density function f of an unknown contin-
uous probability density function f on X. f is defined as
f(x; t) =1
N
N∑i=1
φ(x,Xi; t), x ∈ <, (4.7.13)
where
φ(x,Xi; t) =1√2πt
e−(x−Xi)
2
2t(4.7.14)
is a Gaussian p.d.f. (kernel) with location Xi and scale√t. The scale is usually referred
to as the bandwidth.
Fig.4.7 is the cumulative distribution function of mean energy lost in one epoch for
all the nodes for LEACH, Fixed-LEACH, N-LEACH and VDC respectively. This figure
shows that the energy spend by sensor nodes in N-LEACH and VDC is having lower
spread compared to LEACH and Fixed-LEACH. Mean of energy lost in N-LEACH is
centered near around LEACH and Fixed-LEACH, but in VDC mean of energy lost is
much lesser compared to all other algorithms.
Fig.4.8 is the estimated probability density function of mean energy lost in one epoch
for all the nodes for LEACH, Fixed-LEACH and N-LEACH respectively. This figure
also shows that in N-LEACH and VDC expenditure of node is having lower spread and
closer to the mean energy of all nodes. It implies more balanced energy consumption
in proposed protocols compared to LEACH and Fixed-LEACH.
4.7 Simulation and Results 79
Figure 4.7: Cumulative Distribution Function of mean energy lost in one epoch
Figure 4.8: Probability Density Function of mean energy lost in one epoch
4.8 Conclusions 80
4.8 Conclusions
In method 1 the lifetime of the network increases significantly without introducing new
parameters or assumptions regarding sensor nodes capabilities and data overheads for
routing. We have shown results for different topologies. This shows that proposed
cluster head selection method is robust against topology changes also. This algorithm
can also be used for dynamic wireless sensor networks.
In method 2 the nodes which are located near the sink consume lower energy com-
pared to the farther ones, due to the smaller transmission distance. With same infor-
mation transfer rates, there will be asymmetric energy consumption in sensor nodes
and farther nodes will die earlier than the nodes closer to the sink. If farther nodes
transmit with reduced duty cycle compared to nodes nearer to sink a more balanced
energy consumption can be achieved. We have presented a mechanism to enforce equal
energy consumption using variable duty cycle for all nodes participating in a wireless
sensor network. We have evaluated performance of proposed algorithm and compared
it with existing algorithms and found it to be better as increases lifetime almost by
10%, reduces average energy consumption and balacne the energy consumption in all
the nodes.
Chapter 5
Localized Detection of BottleneckNodes and Quantification ofCriticality in a Wireless SensorNetworks
5.1 Introduction
In a WSN, sensor nodes work without knowledge of their position in the network. If
we look at multihop scenario, after random deployment of nodes, some of the nodes
are in critical positions as they connect one group of sensor nodes to another. If these
nodes are removed from the network it may lead to partitioning of network. Most of the
routing protocols identify a node as a bottleneck node depending upon residual energy
of the node. The nodes would have consumed considerable amount of energy before
they could have been identified as bottleneck nodes. We propose an alternative method
to detect the bottleneck nodes in the network just after their deployment.1
1Rajiv Kr Tripathi, and Shakya, R. and Verma, N.K. and Singh, Y.N., “Localized detection ofbottleneck nodes and quantification of criticality in a wireless sensor networks,” Engineering (SIBIR-CON), 2010 IEEE Region 8 International Conference on , vol.1, no.1, pp.326-328, 11-15 July 2010.
5.1 Introduction 82
Energy consumption is not uniform for all the nodes hence nodes must conserve more
energy if it is more critical in the network. A node is more critical, if its removal leads
to the large increase in probability of partitioning of the network. As the nodes need
to be low cost, they are assumed to be not having the GPS capability for localization.
The consequence is that the nodes are not aware of their exact position in the network.
Thus we can not use localization methods to identify the bottleneck nodes in a sensor
network. Here we propose a simple method based on local information for detection of
bottleneck nodes with their criticality.
Figure 5.1: Example of Bottleneck Node
Bottleneck nodes (as shown in Fig.5.1) are such nodes that have to be shared by
multiple paths because of low density in its vicinity. These nodes are located at such
positions in the network, that they are used quite frequently to forward the packets
from one part of network to other, consuming more energy compared to other nodes
in the network. Removal of these nodes will cause partitioning of network most of the
time.
The objective of this chapter to study the algorithm using local information to iden-
tify bottleneck nodes as well as their criticality rating. The criticality level thus should
affect the nodes contribution in the forwarding packets. Conventionally, when we iden-
tify criticality based on residual energy, we assumed that probability of network parti-
tioning due to removal of a node is same for all the nodes.
5.1 Introduction 83
The researcher working on energy efficient routing algorithms, maximization of life of
sensor network and efficient distributed topology, have designed routing protocols and
topology distributions, which reduces traffic load on critical nodes for extending the
lifetime of a sensor network. Ritter and Winter [86] have discussed partition detection
for a mobile ad-hoc network by identifying critical nodes.
Figure 5.2: Undesirable Energy Histogram Figure 5.3: Desirable Energy Histogram
Xiao and Dai [88] had tried to balance the nodes’ energy consumption for increasing
the life of a sensor network and shown that the nodes which lie in the central part
of topology consumes more energy compared to other nodes. The energy used by all
the nodes in the a sensor network with and without criticality of nodes is compared by
Schurgers and Srivastava [87] (Fig.5.2 and 5.3). It is evident from these, that there exist
such bottleneck nodes. Fig.5.2 presents a typical energy consumption histogram at a
certain point in time. Some nodes have rarely been used as they have less neighbors
and hence less packet for forwarding; while others have almost completely drained their
energy as they had large number of neighbors and were being used frequently to forward
data packets. Our aim is to find these nodes who will potentially completely drain their
energy much sooner compared to other nodes in a topology. The histogram in Fig.5.3 is
more desirable than the one in Fig.5.2, although the total energy consumption is same
in both, during observation period.
5.2 Algorithm 84
Jorgic [89] introduced several localized definitions of critical nodes and critical links.
They are based only on topological and positional information. In our study, the nodes
are not mobile and the topology of the network is static. There are no base stations in
the network. So any node can transmit data to any node as it expected to happen in
an self configured distributed sensor network.
5.2 Algorithm
Nodes are divided into two types- root nodes which initiates the flooding in the network,
and the others who forward root nodes’ unique signal throughout the network. Root
nodes have to be chosen carefully to ensure that most of the area in the network is
monitored. Their key property is that they have a relatively small amount of neighbors
compared to the nodes which are not root nodes. The reason for this kind of selection
is that with a relatively low neighbor count, root nodes are most likely at the border
of the network topology. The number of neighbors, a node has, allows us to select the
most appropriate root nodes. In other words, it allows us to identify border nodes. So
after deploying nodes, we start with the metric- neighbor count within certain radius
of transmission of all the nodes. The critical node detection in a topology is performed
in two steps.
Step 1: Finding root nodes: There is a reset timer in each node. If timer expires,
the node level is made∞ to identify the node, which has not got any flood packet, as an
isolated node. Each node also has Tx timer; after every transmission, it is reset. Upon
expiry of Tx, of the flooding transmission is done. The Tx timer should have value less
than the reset timers. Just after deployment, each nodes will send a PING broadcast
message. All nodes which are inside communication range of radius R will receive it
5.2 Algorithm 85
and respond with the ACK message. Each node will also store transmitting node’s
entry as neighbor in their database. Neighbors count will be the criterion to decide
if a node is a root node or not. Each node sends the neighbors’ count to each of its
neighbors. Thus each node knows its neighbors and their neighbors count. Any node
which has only one neighbor identify itself as a root node. The root node if identified
send flooding packet to all the neighbors which in turn further propagate it. If all the
nodes are having more than one neighbors, then there will not be any root node so far.
If time out happens and no flooding packets are seen, each node will reset its Tx timer.
Then nodes increase the threshold number of neighbors for deciding if it is a root node
or not. One important thing which must be taken care of that is the communication
range must be chosen at its largest possible value for estimating the neighbors count.
By doing this, the chances to select a root node at border will be maximum. For a high
density sensor network, these nodes will surely lie at the border of the topology. One
point must be taken care of, that the two root nodes must not lie in the each other
communication range directly to avoid false detection of bottleneck node.
Step 2: Detection of critical nodes : Let n nodes decide to become root nodes
throughout the topology. These nodes start flooding with its own level set at zero. The
neighboring nodes who receive flooding request from zero level will assign level one to
themselves and further propagate the flooding packet. Nodes give next higher level to
their neighbors and so on. All root nodes will generate different flooding packets. Each
node rebroadcast the received flooding packet only if it is makes the current node a
higher level node. If it is making it of same level or lower, then it is discarded. Once
a node accepts a request for level change while it has already been assigned a higher
level from a different root node, it is considered critical node. If a critical node receives
flooding packet at same level from third root node, then it is more critical node as this
5.3 Simulation Setup 86
node connects three root nodes. Thus criticalness of a node will depend on number of
different flood packets received from different root nodes. Thus the each node knows
its criticalness before the actual transmission of data. For this step the communication
radius must be optimum to save energy and no isolated nodes should be there in the
topology.
5.3 Simulation Setup
In the simulation, sensor nodes were randomly distributed within an area of 1000 ×
1000 m2. Each sensor is given an initial level 0 and a unique node ID. We have taken
160 nodes with communication radius 185m in our simulation. We have chosen root
nodes at the corners of the rectangular area. In a topology, where most of the time,
the traffic moves across the network, the nodes can use this information to detect their
criticality and so enhance the life time of sensor network by taking appropriate routing
decisions.
Each root node sets its level to zero. Root nodes start flooding with unique source
id. The neighbors of root nodes are assigned level one with unique ID of corresponding
root nodes. When two different root nodes’ signals come at a node it will consider itself
as bottleneck node of category-1. If any of these category-1 bottleneck nodes gets a
flood signal from the third root node this will be category-2 bottleneck node. As there
are n root nodes so at most bottleneck nodes will be of category-(n−1) criticality which
are connected to all n root nodes. These will be most important nodes as they may be
used to forward packet to whole topology of the network.
5.3 Simulation Setup 87
Figure 5.4: Root Nodes Selection Figure 5.5: Bottleneck Nodes Detection
Here simulation is performed for 150 different topologies. The following table shows
the percentage of bottleneck nodes in a topology. These bottleneck nodes must be
treated differently than the other ordinary nodes for different applications. As we have
taken 4 root nodes so there are 3 category of bottleneck nodes. Among them, category-3
are the most critical bottleneck nodes.
In Fig.5.4 the red color circle nodes shows the random deployment of the nodes in
the topology and root node selection is shown, the blue circles are the root nodes. In
Fig.5.5 the blue star nodes are category-1 bottleneck nodes and magenta color nodes
are category-2 and most critical bottleneck node i.e. category-3 is shown by green color.
Table 5.1: % of Bottleneck NodesCategory Category (1) Category (2) Category (3)
% of Bottleneck Nodes 19.308% 3.533% 1.512%
Table 5.1 shows the percentage of bottleneck nodes in the WSN. Bottleneck node
of Category-1 comes out be nearly 20% of the total sensor nodes. So nearly 25% total
nodes are bottleneck of all categories in a WSN. So rest 75% nodes can be used more
compared to these bottleneck nodes for data transmission in a WSN.
5.4 Conclusions 88
5.4 Conclusions
There exists nearly 25% bottleneck node in a sensor network. Detection of these nodes
may be performed based on residual energy or localized detection method. Localized
detection of bottleneck nodes is apriori that means we can detect bottleneck nodes in a
sensor network just after deployments of nodes. But if we use residual energy method
which is posteriori method that means after using sensor network for a sufficient time,
we check residual energy of nodes. Then after we can detect the bottleneck nodes in a
sensor network. Thus Localized detection is better as the criticality of nodes is defined
before energy consumption start. Thus the load distribution on sensor nodes can be
managed depending of its criticality.
Chapter 6
Conclusions and Future Works
6.1 Conclusions
The conclusions of the work presented in the thesis are as follows.
The Optimal location of base station can be obtained such that either the transmis-
sion energy expenditure is minimum or lifetime of the sensor network is maximum. If
all nodes suffer only free space transmission loss to base station then centroid of nodes
is the optimal position for base station deployment. If all nodes suffer only multipath
transmission loss to base station then point F (as discussed in thesis, point F is the
point, which minimizes∑n
i=1 d4i , where di is the distance from the point F to node i)
is the optimal position for base station deployment. If some nodes suffer free space
loss and remaining nodes, multipath loss to base station, then point O (as discussed
in thesis) is the optimal position for base station deployment. This location can not
be calculated by any method. It can be guessed only for certain situations. We have
suggested a method to find base station location quite closer to the optimal point. Our
method can be used in any topology in general.
1. Conclusions 90
For the case when lifetime of the network is to be maximized, placing the base station
at the center of minimum enclosing circle is the choice. This is with the assumption
that lifetime is defined as time after which first node die due to energy depletion.
The second issue investigated by us is Anchor free localization. It can be performed
by either using average distance between nodes or using hop count as distance mea-
surement. Using average distance is very simple and in single hop localization, approx-
imately 6% localization error occurs for the rectangular sensor field.
We also observe that localization error of approximately 50% is introduced when
using hop counts as distance measurement in any localization algorithm. It may be
noted that nodes are localized in integral multiples of maximum communication radio
range R, in hop count based metric.
We have also looked at how to balance the energy consumption and thereby increase
the lifetime of the sensor network. Energy consumption is mainly due to reception and
data aggregation for a smaller size sensor network.Thus energy consumption mainly
depend on number of nodes supported by the cluster head node. Consequently, number
of supported nodes is an important factor for balancing energy consumption in WSN.
We have exposed the mechanism to decide candidacy of a node to become cluster head
node depending on number of nodes supported in the past.
Energy consumption among sensor nodes in a WSN can also be balanced using
variable duty cycle transmission. This is a viable technique when energy consumption is
mainly due to data transmission over longer distance in sensor networks. This technique
is for large sized sensor networks. In this method, length of transmission duty cycle
of sensor nodes is proportional to distance between sensor nodes and base station to
achieve balanced energy consumption.
6.2 Suggestions for future works 91
In a wireless sensor network some nodes need to work more as they are used frequently
to forword the data for all the nodes. These are bootleneck neck nodes.In a WSN, nearly
25% such nodes exits. Bottleneck nodes consumes more energy compared to other nodes
in the network. Thus identification and conserving energy specifically in these nodes is
more important.
6.2 Suggestions for future works
This thesis mainly concentrate on the conserving and balancing energy in a wireless
sensor network. We have investigated only few of the problems. Some of the problems
can be further explored are as follows. Algorithms for placing more than one base station
can be explored depending on communication parameters. More accurate methods can
be developed for anchor free localization in WSN. Models for identification of bottleneck
nodes and resulting life time enhancement techniques. Further, we fed at the end of this
study that overheads needed to implement any method should be as low as possible and
low enough to made it viable. Further if random access MAC (medium access control)
is used, there will be random delay and energy overhead which gets added. The impact
of these need more investigations along with new methods to minimize this delay and
to improve throughput per unit energy consumed.
Bibliography
[1] Heinzelman, W. R., Chandrakasan, A. and Balakrishnan, H. : ‘Energy-Efficient
Communication Protocol for Wireless Microsensor Networks’, Proceedings of the
International Conference on System Sciences-2000, pp. 3005-3014.
[2] Heinzelman, W. R., Chandrakasan, A. and Balakrishnan, H. : ‘An application spe-
cific protocol architecture for wireless microsensor networks”, IEEE Transactions
on Wireless Communications Vol.1, No.4, 2002, pp. 660-670.
[3] Pan, J., Cai, L., Hou, T., Shi, Y. and Shen, S. X. : ‘Optimal Base-Station Lo-
cations in Two-Tiered Wireless Sensor Networks’, IEEE Transactions on Mobile
Computing, Vol.4, No.5, Sept. 2005, pp. 458–473.
[4] Pan, J., Cai, L., Hou, y., Shi, Y. and Shen, S. X. : ‘Topology control for wireless
sensor networks’, Proceedings of the 9th ACM International Conference on Mobile
Computing and Networking, San Diego, USA, Sep. 2003, pp. 286-299.
[5] Akkaya K., and Youssef, W. : ‘Positioning of Base Stations in Wireless Sensor
Networks’, IEEE Communications Magazine, April 2007, pp.96-102.
[6] Paul, B. and Matin, M. A. : ‘Optimal geometrical sink location estimation for
two-tiered wireless sensor networks ’, IET-Wireless Sensor Systems, Vol.1, Iss.2,
Feb. 2011, pp. 74-84.
BIBLIOGRAPHY 93
[7] Bulusu N., Heidemann J. and Estrin D. : ‘GPS-less low-cost outdoor localization
for very small devices’, Journal of Personal Communications, IEEE, Vol.7 Iss.5,
Oct. 2000 , pp. 28-34.
[8] Capkun S., Hamdi M., and Hubaux J. P. : ‘GPS free positioning in mobile ad
hoc networks’, Proceedings of the 34th Hawaii International Conference on System
Sciences, Maui, HI, USA, Jan. 2001; pp. 3481-3490.
[9] Savarese C., Rabaey J. M. and Beutel J. : ‘Locationing in distributed ad-hoc wire-
less sensor networks’, Proceedings of IEEE International Conference on Acoustics,
Speech, and Signal Processing, Vol.4, 2001, pp. 2037-2040.
[10] Priyantha N. B., Balakrishnan H., Demaine E. and Teller S. : ‘Anchor-free dis-
tributed localization in sensor networks’, Technical Report 892, MIT Lab. for Com-
puter Science, Apr. 2003.
[11] Moses R. L., Krishnamurthy D. and Patterson R. M. :‘A Self-Localization Method
for Wireless Sensor Networks’, EURASIP Journal on Applied Signal Processing,
Vol.4, 2002, pp. 348–358.
[12] Manjeshwar , A. and Agrawall, D.P.: ‘TEEN : A Protocol for Enhanced Efficiency
in Wireless Sensor Networks’, Proceedings of the 1st International Workshop on
Parallel and Distributed Computing Issues in Wireless Networks and Mobile Com-
puting, San Francisco, CA, Apr. 2001.
[13] Manjeshwar , A. and Agrawall, D. P.: ‘APTEEN: A Hybrid Protocol for Efficient
Routing and Comprehensive Information Retrieval in Wireless Sensor Networks’,
Proceedings of the 2nd International Workshop on Parallel and Distributed Com-
puting Issues in Wireless Networks and Mobile computing, Ft. Lauderdale, FL,
Apr. 2002.
BIBLIOGRAPHY 94
[14] Lindsey, S. and Raghavendra, C. S.: ‘PEGASIS: Power-Efficient Gathering in Sen-
sor Information Systems’, Proceedings of IEEE Aerospace Conference, Vol.3, Mar.
2002.
[15] Younis, O. and Fahmy, S.: ‘HEED: A Hybrid, Energy-Efficient, Distributed Clus-
tering Approach for Ad-hoc Sensor Networks’, IEEE Transactions on Mobile Com-
puting, Vol.3 , Iss.4, Oct. 2004, pp. 366–379.
[16] Younis, O. and Fahmy, S.: ‘Distributed clustering in ad-hoc sensor networks: A hy-
brid, energy-efficient approach’, Proceedings of the IEEE Conference on Computer
Communications, Hong Kong, 2004.
[17] Lindsey, S., Raghavendra, C. S. and Sivalingam, K.: ‘Data gathering in sensor
networks using the energy delay metric’, Proceedings of the IPDPS Workshop on
Issues in Wireless Networks and Mobile Computing, Apr. 2001.
[18] Ye, M., Li, C., Chen, G. and Wu J.: ‘EECS: An energy efficient clustering scheme in
wireless sensor networks’, Proceedings of the 24th IEEE International Performance,
Computing, and Communications Conference, 2005. IPCCC 2005, Apr. 2005, pp.
535–540.
[19] Hansen, E., Neander, J., Nolin, M. and ,Bjrkman M.: ‘Energy-Efficient Cluster
Formation for Large Sensor Networks using a Minimum Separation Distance’,
Mlardalen Real-Time Research Centre, Mlardalen University, Sweden, 2006.
[20] Junping, H., Yuhui, J. and Liang, D.: ‘A Time-based Cluster-Head Selection Algo-
rithm for LEACH’, IEEE Symposium on Computers and Communications, 2008.
ISCC 2008, pp. 1172–1176.
BIBLIOGRAPHY 95
[21] Dilip, K., Trilok, C. A. and Patel, R. B.: ‘EEHC: Energy efficient heterogeneous
clustered scheme for wireless sensor networks’, Journal of Computer Communica-
tions, Vol.32, 2009, pp. 662-667.
[22] Otgonchimeg, B. and Kwon, Y.: ‘EECED: Energy Efficient Clustering Algorithm
for Event-Driven Wireless Sensor Networks’, Proceedings of the 5th International
Joint Conference on INC, IMS and IDC, 2009, pp. 1758-1763.
[23] Gowrishankar, S., Basavaraju, T. G., Manjaiah, D. H., and Sarkar, S. K., : ‘Issues
in Wireless Sensor Networks’, Proceedings of the World Congress on Engineering,
Jul. 2008, London, U.K.
[24] ‘Crossbow Wireless 2007 Catalog’ http://www.investigacion.frc.utn.edu.ar/sensores
/Equipamiento/Wireless/Crossbow Wireless 2007 Catalog.pdf, Accessed in May.
2010.
[25] Zytoune, O., Fakhri, Y. and Aboutajdine, D. : ‘A Balanced Cost Cluster-Heads
Selection Algorithm for Wireless Sensor Networks’, International Journal of Com-
puter Science Vol.4, Iss.1, 2009, pp. 21-24.
[26] Howard, A., Mataric, M. J. and Sukhatme, G. S. : ‘Mobile Sensor Network De-
ployment using Potential Fields: A Distributed, Scalable Solution to the Area
Coverage Problem’ Proceedings of the 6th International Symposium on Distributed
Autonomous Robotics Systems (DARS02) Fukuoka, Japan, Jun. 2002.
[27] Heo, N. and Varshney, P. K. : ‘Energy-Efficient Deployment of Intelligent Mobile
Sensor Networks’ IEEE Transactions on Systems, Man and Cybernetics-part A:
System and Humans, Vol.35, No.1, Jan. 2005, pp. 4548-4553.
BIBLIOGRAPHY 96
[28] Dhillon, S. S. and Chakrabarty K. : ‘Sensor Placement for Effective Coverage and
Surveillance in Distributed Sensor Networks’ Proceedings of the IEEE Wireless
Communications and Networking Conference (WCNC03), Mar. 2003.
[29] Zou, Y. and Chakrabarty K. : ‘Sensor Deployment and Target Localization Based
on Virtual Forces’ Proceedings of the 22nd Annual Joint Conference of the IEEE
Computer and Communications Societies, Vol.2, Apr. 2003, pp. 1293-1303.
[30] Wang, G., Cao G., and Porta, T. L. : ‘Movement-Assisted Sensor Deployment’
Proceedings of the IEEE INFOCOM 2004, Vol.4, pp 2469-2479.
[31] Lin, M., Wu, Q., Cai, X. and Rao, N. S. V. : ‘Optimizing Base Station Deploy-
ment in Wireless Sensor Networks Under One-hop and Multi-hop Communication
Models’, Proc. Int. Conf. on Parallel and Distributed Systems, 2009, pp. 96-103.
[32] Liu, Y., Ngan H., and Lionel, M. N. : ‘Power-aware Node Deployment in Wireless
Sensor Networks’ Proceedings of the IEEE International Conference on Sensor
Networks, Ubiquitous, and Trustworthy Computing.
[33] Poe,, W. Y. and Schmitt, J. B. : ‘Node Deployment in Large Wireless Sensor
Networks: Coverage, Energy Consumption, and Worst-Case Delay’ Proceedings of
the AINTEC’09, Nov. 2009.
[34] Park, Y. K., Lee, M. G., Jung, K. K., Yoo, J. J. and Lee, S. H. : ‘Optimum Sensor
Nodes Deployment using Fuzzy C-means Algorithm’ International Symposium on
Computer Science and Society, 2011.
[35] Tolle, G., Polastre, J., Szewczyk, D., Culler, D., Turner, N., Burgess, S., Daw-
son, T., Buonadonna, P., Gay, D. and Hong, W. : ‘A macroscope in the redwoods’
Proceedings of the Sensys, Nov. 2005.
BIBLIOGRAPHY 97
[36] Werner-Allen, G., Johnson, J., Ruiz, M., Lees, J., and Welsh, M. : ‘Monitoring
Volcanic Eruptions with a Wireless Sensor Network’ Proceedings of the EWSN,
Jan. 2005.
[37] Liu, T., Sadler, C. M., Zhang, P. and Martonosi, M., : ‘Implementing Software
on Resource-Constrained Mobile Sensors: Experiences with Impala and ZebraNet’
Mobisys, Jun. 2004.
[38] Bogdanov, A., Maneva, E. and Riesenfeld, S.: ‘Power-aware Base Station Position-
ing for Sensor Networks’, Proceedings of the IEEE INFOCOM 2004, pp. 261–274.
[39] Efrat, A., Har-Peled, S. and Mitchell, W. : ‘Approximation Algorithms for Two
Optimal Location Problems in Sensor Networks’, Proceedings of the International
Conference on Broadband Communication, Networks and System, 2005., pp. 1–14.
[40] Wong, J. K. L., Mason, A. J., Neve, M. J. and Sowerby, K. W. : ‘Base sta-
tion placement in indoor wireless systems using binary integer programming’, IEE
Proc.-Communication, Vol.3, No.5, October 2006.
[41] Basheer, M. R.; Jagannathan, S. : ‘A New Receiver Placement Scheme Using
Delaunay Refinement-Based Triangulation,’ Proceedings of the Wireless Commu-
nications and Networking Conference (WCNC), pp.1-6, 18-21 April 2010.
[42] Rappaport, T. : ‘Wireless Communications: Principles Practice’, Englewood Cliffs,
NJ, Prentice-Hall, 1996.
[43] Harter, A., Hppper, A., Steggles, P., Ward, A., and Webster, P. : ‘The Anatomy
of a Context-Aware Application’ Proceedings of the 5th ACM MOBICOM, 1999.
BIBLIOGRAPHY 98
[44] Priyantha N. B., Chakraborty A. and Balakrishnan H.: ‘The Cricket Location-
Support System’, Proceedings of 6th ACM International Conference on Mobile
Computing and Networking, Boston, MA, August 2000.
[45] Doherty L., Pister K. S. J. and Ghaoui L. E. : ‘Convex Position Estimation in
Wireless Sensor Networks’, Proceedings of 20th Annual Joint Conference of the
IEEE Computer and Communications Societies, Vol.3, 2001, pp. 1655-1663.
[46] Savvides A., Han C. and Strivastava M. B. : ‘Dynamic Fine Grained Localization
in AdHoc Networks of Sensors’, Proceedings of ACM SIGMOBILE 2001, Rome,
Italy, pp. 166-179.
[47] Niculescu D. and Nath B. : ‘Ad Hoc Positioning System (APS) Using AOA’,
Proceedings of Twenty-Second Annual Joint Conference of the IEEE Computer
and Communications, INFOCOM 2003, Vol.3, pp. 1734-1743.
[48] He T., Huang C., Blum B., Stankovic J., and Abdelzaher T.: ‘Range-free local-
ization schemes in large scale sensor network’, Proceedings of the International
Conference ACM MOBICOM’03, pp. 81-95.
[49] Meertens L. and Fitzpatrick S.: ‘The Distributed Construction of a Global Co-
ordinate System in a Network of Static Computational Nodes from Inter-Node
Distances’, Kestrel Institute Technical Report, Kestrel Institute, Palo Alto, 2004.
[50] Peng R. and Sichitiu M. L. :‘Angle of Arrival Localization for Wireless Sensor
Networks’, Sensor and Ad Hoc Communications and Networks, 2006, pp. 374-382.
[51] Cheng K. Y., Lui K. S., and Tam V.: ‘Localization in sensor networks with lim-
ited number of anchors and clustered placement’, Proceedings of the International
Conference IEEE WCNC, Hong Kong, Mar. 2007; pp. 4425-4429.
BIBLIOGRAPHY 99
[52] Huang R. and Zaruba G. V.: ‘Monte Carlo localization of wireless sensor networks
with a single mobile beacon’, Wireless Networks, Vol.15, No.8, 2009, pp. 978-990.
[53] Basheer, M. R.; Jagannathan, S. : ‘Localization of objects using stochastic tun-
nelling,’ Proceedings of IEEE Wireless Communications and Networking Confer-
ence (WCNC), pp.587-592, 28-31 March 2011.
[54] Moore D., Leonard J., Rus D., and Teller S.: ‘Robust distributed network localiza-
tion with noisy range measurements’, Proceedings of the International Conference
Embedded Networked Sensor Systems, Nov. 2004, pp. 50-61.
[55] Fang L., Du W. and Ning P. : ‘A beacon-less location discovery scheme for wire-
less sensor networks’, Proceedings of 24th Annual Joint Conference of the IEEE
Computer and Communications Societies, Vol.1, pp. 161-171.
[56] Basheer, M. R.; Jagannathan, S. : ‘R-Factor: A New Parameter to Enhance Loca-
tion Accuracy in RSSI Based Real-time Location Systems,’ Proceedings of the 6th
Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc
Communications and Networks, 2009. SECON ’09., pp.1-9, 22-26 June 2009.
[57] Wang L., and Xu Q. : ‘GPS-Free Localization Algorithm for Wireless Sensor Net-
works’, Sensors 2010, Vol.10, pp. 5899-5926.
[58] Wang, X., Cao, R., Huang, Q., Ke, Z., and Liu, Z. : ‘A novel anchor-free position
algorithm based on cluster technique’, Proceedings of the International Conference
on Software Technology and Engineering(ICSTE), Hefei 230027, China, 2010.
[59] Basheer, M. R.; Jagannathan, S. : ‘Localization of Objects Using Cross-Correlation
of Shadow Fading Noise and Copulas,’Proceedings of IEEE Global Telecommuni-
cations Conference (GLOBECOM 2011), pp.1-5, 5-9 Dec. 2011.
BIBLIOGRAPHY 100
[60] Kasirajan, P.; Larsen, C.; Jagannathan, S. : ‘A New Adaptive Compression Scheme
for Data Aggregation in Wireless Sensor Networks,’Proceedings of IEEE Wireless
Communications and Networking Conference (WCNC), pp.1-6, 18-21 April 2010.
[61] Said, B. A., Abdellah, E., HSSANE, A. B. and HASNAOUI, M. L.: ‘Improved
and Balanced LEACH for heterogeneous wireless sensor networks’, International
Journal on Computer Science and Engineering, Vol.2, No.8, 2010, pp. 2633-2640.
[62] Anguswamy, R.; Zawodniok, M.; Jagannathan, S. : ‘A Multi-Interface Multi-
Channel Routing (MMCR) Protocol for Wireless Ad Hoc Networks,’Proceedings of
IEEE Wireless Communications and Networking Conference, 2009. WCNC 2009,
pp.1-6, 5-8 April 2009.
[63] Vass, D. and Vidacs, A.: ‘Positioning Mobile Base Station to Prolong Wireless
Sensor Network Lifetime’, Proceedings of the International Conference on Emerging
network experiment and technology, 2005, pp. 300–301.
[64] Son, J., Pak, J. and Han, K. : ‘Determination of Aggregation Point Using Fermat’s
Point in Wireless Sensor Networks’, APWeb Workshops 2006, Springer-Verlag
Berlin Heidelberg, 2006, pp. 257-261.
[65] Wang, J., Urriza, P., Han, Y. and Cabric, D.: ‘Weighted Centroid Localization
Algorithm: Theoretical Analysis and Distributed Implementation’, IEEE Trans-
actions on Wireless Communications, Vol.10, No.10, 2011, pp. 3403–3413.
[66] Li, R. Z., Luo, X. L. and Lin, J. R.: ‘Weighted Centroid Correction Localization in
Cellular Systems’, American Journal of Engineering and Applied Sciences, Vol.4,
No.1, 2011, pp. 37–41.
BIBLIOGRAPHY 101
[67] Chen, Y., Pan, Q., Liang, Y. and Hu, Z.: ‘AWCL: Adaptive Weighted Centroid
Target Localization Algorithm Based on RSSI in WSN’, Proceedings of the Inter-
national Conference on Computer Science and Information Technology (ICCSIT),
2010, Vol.9 pp. 331–336.
[68] Blumenthal, J., Grossmann, R., Golatowski, F., and Timmermann, D.: ‘Weighted
Centroid Localization in Zigbee-based Sensor Networks’, IEEE International Sym-
posium on Intelligent Signal Processing, 2007, pp. 1–6.
[69] Xing, H., Zhou, J., and Dong, L.: ‘The Study of Localization Algorithm Based
on RSSI’, Proceedings of the International Conference on Information Science and
Technology, Mar. 2011, pp. 766–769.
[70] Schuhmann, S., Herrmann, K., Rothermel, K., Blumenthal, J., and Timmer-
mann, D.: ‘Improved Weighted Centroid Localization in Smart Ubiquitous Envi-
ronments’, Proceedings of the International Conference on Ubiquitous Intelligence
and Computing, 2008, pp. 20–34.
[71] Xu Y., Ouyang Y., Le Z.,Ford J., and Makedon F. : ‘Mobile Anchor-free Localiza-
tion for Wireless Sensor Networks ’, Journal of Distributed Computing in Sensor
Systems, 2007.
[72] Bachrach, J. and Taylor, C. : ‘Localization in Sensor Networks’, JRB Project,
Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute
of Technology, Cambridge, pp. 1-38.
[73] Nagpal, R., Shrobe, H. and Bachrach, J. : ‘Organizing a Global Coordinate System
from Local Information on an Ad Hoc Sensor Network’, A.I. Memo 1666, MIT A.I.
Laboratory, August 1999.
BIBLIOGRAPHY 102
[74] Haenggi, M., : ‘Energy-Balancing Strategies for Wireless Sensor Networks’, IEEE
International Symposium on Circuits and Systems (ISCAS’03), (Bangkok, Thai-
land), May 2003.
[75] Olariu, S., and Stojmenovic, I.: ‘Design guidelines for maximizing lifetime and
avoiding energy holes in sensor networks with uniform distribution and uniform
reporting’, Proceedings of IEEE Conference on Computer Communications, IN-
FOCOM, 2006, pp. 1-12.
[76] Powell, O., Leone, P., and Rolim, J.: ‘Energy optimal data propagation in wireless
sensor networks’, Journal of Parallel and Distributed Computing 67, 2007, pp.
302-317.
[77] Bouabdallah, F., Bouabdallah, N., and Boutaba, R.: ‘On Balancing Energy Con-
sumption in Wireless Sensor Networks’, IEEE Transactions on Vehicular Tech-
nology, Vol.58, No.6, July 2009, pp.2909-2924.
[78] Zhang, H. and Shen, H. : ‘Balancing Energy Consumption to Maximize Network
Lifetime in Data-Gathering Sensor Networks’, Parallel and Distributed Systems,
IEEE Transactions on , Vol.20, No.10, Oct. 2009, pp.1526–1539.
[79] Zhang, X. and Wu, Z. : ‘The balance of routing energy consumption in wireless
sensor networks’, Journal of parallel and distributed computing,Vol.71, No.7, 2011,
pp. 1024–1033
[80] Zytoune, O., Fakhri, Y., and Aboutajdine, D. : ‘A Balanced Cost Cluster-Heads
Selection Algorithm for Wireless Sensor Networks’, Proceeding of World Academy
of Science, Engineering and Technology, 2009, pp. 767–770.
BIBLIOGRAPHY 103
[81] Bharathidasan, A., Anand, V. and Ponduru, S. : ‘Sensor Networks: An Overview’,
Technical report, Dept. of Computer Science, University of California at Davis,
2002.
[82] Akyildiz, I. F., Su, W., Sankarasubramaniam, Y.,and Cayirci, E. : ‘Wireless sensor
networks: a survey’, Computer Networks: The International Journal of Computer
and Telecommunications Networking, vol.38, no.4, pp. 393–422.
[83] Zytoune, O., Aroussi, M. E., Rziza, M. and Aboutajdine, D.: ‘Stochastic Low
Energy Adaptive Clustering Hierarchy’, Journal of ICGST- CNIR, Vol.8, No.1,
2008, pp. 47-51.
[84] Ci, S., Guizani, M. and Sharif, H. : ‘Adaptive clustering in wireless sensor networks
by mining sensor energy data’, Elsevier: Computer Communications, 2007.
[85] Shin, I., Kim, M., Mutka, M. W., Choo, H. and Lee, T. J. : ‘MCBT: Multi-Hop
Cluster Based Stable Backbone Trees for Data Collection and Dissemination in
WSNs’, Sensors 2009, Vol.9, pp. 6028-6045.
[86] Ritter, H., Winter, R. and Schiller, J. : ‘A Partition Detection System for Mo-
bile Ad-hoc networks’, Sensor and Ad Hoc Communications and Networks (IEEE-
SECON), 2004, pp. 489-497.
[87] Schurgers, C. and Srivastava, M. B. : ‘Energy Efficient Routing in Wireless Sen-
sor Networks’, Military Communications Conference (IEEE-MILCOM), 2001, pp.
357-361.
[88] Du, X., Xiao, Y. and Dai, F. : ‘Increasing network lifetime by balancing node
energy consumption in heterogeneous sensor networks’, Wireless Communications
and Mobile Computing, Wiley Interscince Jounral, Vol.8, 2008, pp. 125-136.
BIBLIOGRAPHY 104
[89] Jorgic, M., Stojmenovic, I., Hauspie, M. and SimplotRyl, D. : ‘Localized Algo-
rithms for Detection of Critical Nodes and Links for Connectivity in Ad Hoc Net-
works’, 20th International Conference on Advanced Information Networking and
Applications, Vol.2, 2006, pp. 336-340.
[90] Apostol, R. and Mnatsakanian, M. A. : ‘Finding Centroids the Easy Way’, The
Mathematical Association of America, Math Horizons, Vol.8, No.1, Sep. 2000),
pp. 7-12.
[91] Hoffman, D. : ‘Contemporary Calculus’, Online Lecture notes on Calculus at
http://scidiv.bellevuecollege.edu/dh/ccal/CC5.4.pdf.
[92] Bajaj, C. : ‘The algebraic degree of geometric optimization problems’, Journal of
Discrete and Computational Geometry, 1988.
[93] Chen, R.: ‘Location problem with costs being sums of powers of Euclidean dis-
tances’, The Journal of Computer and Operation Research Vol.11, No.3, Oct. 1984,
pp. 285-294.
[94] Drezner, Z. : ‘Bounds on the Optimal Location to the Weber Problem under
Conditions of Uncertainty’, Journal of the Operational Research Society, Vol.30,
No.10, Oct. 1979, pp. 923-931.
[95] Drezner, Z. : ‘Facility Location: A Survey of Applications and Methods’, Springer
Series in Operations Research. Springer-Verlag, New York, 1995.
[96] Apostol, R. and Mnatsakanian, M. A. : ‘Sums of Squares of Distances in m-Space’,
The Mathematical Association of America, Jul. 2003, pp. 516-526.
BIBLIOGRAPHY 105
[97] Bose, P., Maheshwari, A. and Morin, P. : ‘Fast approximations for sums of dis-
tances, clustering and the fermat–weber problem’, Comput. Geom. Theory Appl.,
Vol.24, No.3, 2003, pp. 135–146.
[98] Jiawang, N., Parrilo, P. A. and Bernd , S. : ‘Semidefinite representation of the
k-ellipse’, Algorithms in Algebraic Geometry. IMA Volumes in Mathematics and
its Applications. 146. Springer-Verlag, 2008, pp. 117–132.
[99] Botev, Z. I., Grotowski, J. F. and Kroese, D. P. : ‘Kernel density estimation via
diffusion’, Annals of Statistics, Vol. 38, No.5, 2010, pp 2916-2957.
[100] http://en.wikipedia.org/wiki/Geometric median, accessed September, 2011.
List of Publications
Published Papers:
1. Rajiv Kr Tripathi, Y. N. Singh and Nischal K. Verma, “Clustering Algorithm
for Non Uniformly Distributed Nodes in a Wireless Sensor Network,”In IET Jour-
nal of Electronics Letter, Vol.49, No.4, February 2013, pp. 299-300.
2. Rajiv Kr Tripathi, Y. N. Singh and Nischal K. Verma, “Two tiered wireless
sensor networks - Base station optimal positioning case study,” In IET Journal
of Wireless Sensor System, Vol.2, No.4, December 2012, pp. 351-360.
3. Rajiv Kr Tripathi, Y. N. Singh and Nischal K. Verma, “N-LEACH, a Balanced
Cost Cluster-Heads Selection Algorithm for Wireless Sensor Network,”Proceeding
of IEEE National Conference on Communications, NCC 2012 , Vol.1, No.1, IIT
Kharagpur, India, Feb. 2012, pp.1-5.
4. Rajiv Kr Tripathi, Rajeev Shakya, Nischal K. Verma and Y. N. Singh, “Local-
ized Detection of Bottleneck Nodes and Quantification of Criticality in a Wire-
less Sensor Networks,” Proceeding of IEEE Region 8 SIBIRCON-2010, Irkutsk
Listvyanka, Russia, pp. 326–328.
LIST OF PUBLICATIONS 107
Communicated Papers:
1. Rajiv Kr Tripathi, SatishKrishna Dulli, Y. N. Singh and Nischal K. Verma,
“Analysis of Weights for optimal positioning of base station in a wireless sensor
network,” In IET Journal Electronics Letter, February 2013.
Appendix A 108
Appendix A - Optimal location of base station when all nodes suffer only
free space loss
Let d2 be the sum of square of all the Euclidean distances as all the nodes have only
free space loss.
d2 =d21 + d2
2 + d23 + ....+ d2
n. (A-1)
From equations (2.3.13) and (A-1),
d2 =n∑i=1
[(x− xi)2 + (y − yi)2
]. (A-2)
d2 =n∑i=1
[x2 + x2
i − 2xxi + y2 + y2i − 2yyi
]. (A-3)
Expanding the summation.
d2 =n(x2 + y2) +n∑i=1
(x2i + y2
i )− 2xn∑i=1
xi − 2yn∑i=1
yi. (A-4)
now let
C1 =n∑i=1
xi, (A-5)
C2 =n∑i=1
yi, (A-6)
and
C3 =n∑i=1
(x2i + y2
i ). (A-7)
Now equation (A-4) becomes,
d2 =n(x2 + y2)− 2xC1 − 2yC2 + C3. (A-8)
Taking ∂(d2)∂x
and ∂(d2)∂y
for finding maxima and minima.
∂(d2)
∂x=2nx− 2C1 = 0. (A-9)
Appendix A 109
⇒ x =C1
n. (A-10)
∂(d2)
∂y=2ny − 2C2 = 0. (A-11)
⇒ y =C2
n. (A-12)
As ∂2d2
∂x2 and ∂2d2
∂y2both are positive for all values x and y, equation (A-10) and (A-12)
provide minima of equation (A-1).
∂2d2
∂x2=∂2d2
∂y2=2n. (A-13)
Thus sum of square of distances from n points is minimized at the centroid of given
points.
Appendix B 110
Appendix B -Contribution in 2D Geometry-Revealing Line of symmetry
The Center of mass, which converges in to Centroid for uniformly distributed mass
of some objects lies in the line of symmetry [90, 91]. It has been proved [92] that it is
impossible to find the exact location of geometric medians other than the centroid. But
after observing many different cases, we have concluded that other geometric medians
also lies on same line. We consider the cases where many points are distributed in a
two dimension plane. Within this if symmetry exists, we will have at least one line of
symmetry. We will investigate the positioning of various geometric medians in context
of line of symmetry.
Figure B-1: Geometric Median (x, y)
The geometric median of a discrete set of sample points in a Euclidean space is the
point minimizing the sum of distances to the sample points [100]. For n points (x1, y1),
(x2, y2), (x3, y3),...(xn, yn) (as shown in Fig.B-1) distributed in 2-Dimensional plane the
kth geometric median Mk is the point (x, y) which minimizes.
n∑j=1
(dj)k (B-1)
Here dj is given by
dj =√
(x− xj)2 + (y − yj)2. (B-2)
Appendix B 111
For k = 1, this become optimal facility location, also known as Fermat-Weber point or
1-median or M1. It is the (x, y) such that M1 is minimized.
M1 =(d1 + d2 + d3 + ...+ dn−1 + dn) (B-3)
Computing the optimal facility location is a challenge as it has been shown that no
explicit formula exist for finding the optimal facility location. Only numerical approxi-
mations to the solution of this problem are possible.
For k = 2, this is known as centroid (Cx, Cy) or center of mass, defined similarly to
the geometric median as minimizing the sum M2 of the squares of the distances to each
sample [80-83].
M2 =d21 + d2
2 + d23 + ...+ d2
n−1 + d2n. (B-4)
This location can be find by
Cx =
∑ni=1 xin
(B-5)
Cy =
∑ni=1 yin
(B-6)
Similarly we can define higher order geometric medians (kth median) as point which
minimizes
Mk =dk1 + dk3 + dk3 + ...+ dkn−1 + dkn. (B-7)
Similar to 1-median, there is no explicit formula to determine the location of these
higher order medians. Pierre de Fermat posed Fermat Problem in the early 1600.
But it was not studied or applied substantially until the twentieth century and still
it need to further work for more results. In the latest development in 2003, Bose
et al. [97] described more sophisticated geometric optimization procedures for finding
Appendix B 112
approximately optimal solutions to this problem and in 2008, Nie, Parrilo and Sturmfels
[98] showed that the problem can also be represented as a semidefinite program. Here
we have shown that these medians lie on the line of symmetry with some interesting
observations.
Examples: We have calculated the geometric medians using MATLAB in the fol-
lowing examples.
Example 1: There are two points P1 and P2 having line of symmetry at center of
both point and perpendicular to the plane of the points (see Fig.B-2). All medians M1,
M2,...Mn coincide each other and lie on point of intersection of line of symmetry and
the line joining the points.
Figure B-2: Geometric medians for two points
Example 2: Three points P1, P2 and P3 lie on the vertices of right angle isosceles
triangle. All medians M1, M2,...Mn lie on the line from right angle vertices to center
of line joining to other two points. This is the line of symmetry for three points. Mn
median for large value of n, will lie on the center of line joining to other two points.
Fermat point i.e. point M1 for right angle isosceles triangle will lie at 0.1726× L from
the right angle vertices on the line of symmetry (see Fig.B-3). Where L is the equal
sides of the triangle.
Example 3: Four points P1, P2, P3 and P4 (see Fig.B-4) lie on the vertices of square
of side L. All medians M1, M2,...Mn coincide to each other where two lines of symmetry
cut each other i.e. at the center of the square.
Appendix B 113
Figure B-3: Geometric medians for three points
Figure B-4: Geometric medians for four points
Appendix B 114
Example 4: Out of Five points, four points P1, P2, P3 and P4 lie on the vertices of
square of side L and one point P5 on the line of symmetry (see Fig.B-5). All medians
M1, M2,...Mn distributed on the line joining point P5 to center of square. Mn median
for large value of n, will lie on the center of the square.
Figure B-5: Geometric medians for five points
Example 5: For six point we have discussed two example where by keeping M2
constant by moving two points inwards for case 1 and outwards for case 2. In case
1 the geometric medians contract from both side towards M2 that M1 and Mn both
approaches to M2. In case 2 the geometric medians diverge from both side from M2
that M1 and Mn both move away from M2.
Case 1: Six points out of these, four points P1, P2, P3 and P4 lie on the vertices of
square of side L and points P5 and P6 lie on the one side of square and moving towards
center of points P1 and P3 (see Fig.B-6). All medians M1, M2,...Mn distributed from
center line of P1 and P3 to center of P1, P2, P3 and P4. Mn median for large value of n,
will lie on the center of P1, P2, P3 and P4.
Appendix B 115
Figure B-6: Geometric medians for Six points case 1
Figure B-7: Geometric medians for Six points case 2
Appendix B 116
Case 2: Six points out of these, four points P1, P2, P3 and P4 lie on the vertices of
square of side L and points P5 and P6 lie on the one side of square and moving away
from points P1 and P3 (see Fig.B-7). All medians M1, M2,...Mn distributed from center
line of P1 and P3 to center of P1, P2, P3 and P4. Mn median for large value of n, will
lie on the center of line joining points P1 and P3.
Table B-1: Geometric medians for different casesM1 M2 M3 M100
Two Points (75.0, 0.0) (75.0, 0.0) (75.0, 0.0) (75.0, 0.0)Three Points (31.7, 118.2) (50.0, 100.0) (55.3, 94.7) (72.8, 77.2)Four Points (75.0, 75.0) (75.0, 75.0) (75.0, 75.0) (75.0, 75.0)Five Points (28.1, 75.0) (45.0, 75.0) (50.4, 75.0) (50.3, 75.0)
Six Points Case 1 (14.2, 75.0) (50.0, 75.0) (62.4, 75.0) (75.0, 75.0)Six Points Case 2 (60.9, 75.0) (50.0, 75.0) (31.3, 75.0) (0.0, 75.0)
Observations: Results shown in Table B-1 verifies the examples discussed above
and strengthen the observations. Though there is no explicit proof of geometric median
except median-2, through analysis and observation we have found the following facts
which are useful in designing optimal location facility.
(1) For the kth median, when value of k is very large, standard deviation of array
D = [d1, d2, d3, ...dn−1, dn] is minimized at the geometric median. We have discussed
this problem in Appendix C. Although literature regarding this could not be found.
(2) Fermat point i.e. point M1 for right angle isosceles triangle will lie at 0.1726× L
from the right angle vertices on the line of symmetry (see Fig.B-3). Where L is the
length of sides of the triangle.
Future Work: Though these observations lead to some interesting results for symmet-
rically placed points in a plane but further research is recommended for those interested
on geometric medians localization for randomly distributed n points when axis of sym-
metry may or may not exit.
Appendix C 117
Appendix C -Open Problem
Let n points be randomly selected inside finite bounded area at (x1, y1), (x2, y2), (x3, y3)
.....(xn, yn) and a point A (x, y) be inside that area (As shown in Fig.C-1).
Figure C-1: Point A (x,y)
Distances between point A and nodes are d1, d2, d3, ....dn. The di is defined in equa-
tion (2.3.13). Let D be array of distances from point A to all the points.
D = [d1 d2 d3 ... dn−1 dn]
Problem statement: What value of (x, y) inside finite bounded area minimizes the
standard deviation of array D.
Example 1: See in Fig.C-2, where we have chosen only two point i.e. n=2. Here
choice of point A is trivial that is midpoint of two point.
Standard deviation σ for n = 2 is given by
σ =
√(d1 − d1+d2
2
)2+(d2 − d1+d2
2
)22
. (C-1)
Appendix C 118
Figure C-2: Point A when only two point are in field i.e. n=2.
So for d1 =d2 = (d1+d2)2⇒ σ = 0.
Thus standard deviation is minimized at midpoint of these two point. And Point A is(x1+x2
2, y1+y2
2
).
Example 2: Now we see for n = 3, The best choice is circumcenter (see Fig.C-3) of
three point.
σ =
√√√√(d1 −∑3i=1 di3
)2
+(d2 −
∑3i=1 di3
)2
+(d3 −
∑3i=1 di3
)2
3. (C-2)
Where d1 =d2 =d3⇒ σ = 0
Figure C-3: Point A when only three point are in field i.e. n=3.
Now for n = 4 and more points as no circumcenter is possible in general, so now
standard deviation will not be zero. Because all the nodes can’t have equal distances
Appendix C 119
from any point. Standard deviation will be minimized at some point with some nonzero
value and given by
σ =
√√√√(d1 −∑4i=1 di4
)2
+(d2 −
∑4i=1 di4
)2
+ ...+(d4 −
∑4i=1 di4
)2
4. (C-3)
which can be simplified as,
σ =
√(4.d1 −
∑4i=1 di
)2+(4.d2 −
∑4i=1 di
)2+ ...+
(4.d4 −
∑4i=1 di
)264
(C-4)
σ =1
4√
4
√√√√16.d21 +
(4∑i=1
di
)2
− 8.d1.4∑i=1
di + ...+ 16.d24 +
(4∑i=1
di
)2
− 8.d4.4∑i=1
di.
(C-5)
Adding similar terms,
σ =
√√√√1
4.
4∑i=1
di2 − 1
16.
(4∑i=1
di
)2
.. (C-6)
Similarly for n = 5, this becomes
σ =
√√√√1
5.
5∑i=1
di2 − 1
25.
(5∑i=1
di
)2
, (C-7)
and in general for n nodes, this becomes
σ =
√√√√ 1
n.
n∑i=1
di2 − 1
n2.
(n∑i=1
di
)2
. (C-8)
From the equation (C-8) if we check for n = 2 and n = 3 in both cases σ will become
zero. The problem is to find the point A(x, y) such that standard deviation of array D
(equation (C-8)) is minimized.