A study on maximizing the lifetime of wireless sensor networks

In the other issue, we focus on maximizing the lifetime of dual-radiowireless sensor network by using a backbone structure that is used to trans-mit aggregated data to sink.. 17 3 On Max

國 立 高 雄 應 用 科 技 大 學

電 子 工 程 系 博 士 班

博 士 論 文

最大化無線感測網路壽命之研究

A Study on Maximizing the Lifetime of

Wireless Sensor Networks

研 究 生：范文忠 (Van-Trung Pham)

指導教授：劉炳宏 (Bing-Hong Liu)

中 華 民 國 一百零四 年 七 月

A Study on Maximizing the Lifetime of

Wireless Sensor Networks

研 究 生：范文忠 (Van-Trung Pham) 指導教授：劉炳宏 (Bing-Hong Liu)

國 立 高 雄 應 用 科 技 大 學

電 子 工 程 系 博 士 班

博 士 論 文

A Thesis submitted to Institute of Electronic Engineering National Kaohsiung University of Applied Sciences

in Partial Fulfillment of the Requirements

for the Degree of PhD of Engineering

in Electronic Engineering

July 2015 Kaohsiung, Taiwan, Republic of China

中 華 民 國 一百零四 年 七 月

I, Van-Trung Pham, declare that this thesis titled, ‘A Study on Maximizing the Lifetime of Wireless Sensor Networks’ and the work presented in it are my own I confirm that:

clear exactly what was done by others and what I have contributed myself.

Signed:

Date:

Van-Trung

i

本論文主要著重在研究最大化無線感測網路壽命方法，為了能在這個問題上取得進展，本論文針對兩個議題提出研究。在第一個議題中，我們研究在不同時間下使用不同的資料匯集樹以延長網路壽命，在此網路中，資料可以透過資料匯集函式來匯集，如 MAX， MIN， COUNT，

方法，此方法為根據距離匯集點 k-hop 的區域資訊來建造適用於不同時間的資料匯集樹，在建造資料匯集樹時，在匯集點 k-hop 內的樹路徑可以根據網路壽命和感測器所收集到的資料進行重建。

在另一個議題中，我們著重在使用骨幹結構來傳送資料到匯集點以最大化雙頻無線感測網路壽命，在雙頻無線感測網路中，每個感測器可以使用雙頻無線電傳送資料，即小範圍無線電傳送及大範圍無線電傳送。我們在此提出應用於雙頻無線感測網路的骨幹網路建構演算法，在骨幹網路中的節點使用小範圍無線電或大範圍無線電傳送來維護骨幹網路，其餘非骨

路排程方法，可以讓不同的骨幹網路交替使用以延長網路壽命。

關鍵詞: 無線感測網路, 虛擬骨幹, 網路壽命, 資料匯集, NP困難

Author: Van-Trung Pham Supervisor： Dr Bing-Hong Liu

Institute of Electronic Engineering National Kaohsiung University of Applied Sciences

ABSTRACT

Wireless sensor networks are composed of many wireless sensors deployed

in a wide range of areas to collect, process, and store environmental mation Because many environments or objects that need to be monitoredare difficult to approach, such as disaster areas, volcanos, and battle fields,charging batteries of sensors or redeploying sensors is costly Therefore,prolonging the lifetime of wireless sensor networks is an important issue.This thesis focuses on the proposal of solutions for maximizing the life-time of wireless sensor networks For making progress on these issues, thereare two novel issues proposed in this thesis In the first issue, we study theproblem of constructing data aggregation trees for different time to maxi-mize the network lifetime In which, the data is aggregated to the nodes

infor-in a tree by some aggregated data functions, such as MAX, MIN, COUNT,and SUM In addition, the scheduling of data aggregation trees for maxi-mizing the lifetime of wireless sensor networks is proposed In which, thek-hop local information to the sink is used to construct data aggregationtrees for different time While constructing a data aggregation tree, the treetopology within the k-hop of the sink is reconstructed by the lifetime andthe collected data of sensors

In the other issue, we focus on maximizing the lifetime of dual-radiowireless sensor network by using a backbone structure that is used to trans-mit aggregated data to sink In the dual-radio wireless sensor network, everysensor is assumed to have dual radios, that is, small-range radio and large-range radio We then propose an algorithm to construct a backbone in dual-radio wireless sensor networks, where the backbone nodes use small-rangeradio or large-range radio to maintain the backbone, and the rest nodes usesmall-range radio to connect to the backbone In addition, a schedulingfor constructing virtual backbones is proposed such that the constructedbackbones can work sequentially to prolong the network lifetime.

Keywords: Wireless sensor network, virtual backbone, network lifetime,data aggregation, NP-Hard

First at all, I wish to express my sincere thanks to my advisor, Dr Bing-Hong Liu, Associate Professor, National Kaohsiung University of Applied Sciences, Tai- wan, for his earnest guidance and support throughout the course of this research.

He give me the moral support, persistent encouragement and perpetual ideas His depth of knowledge and enthusiasm for research has inspired me a lot during 4 years in KUAS.

I would like to thank all members of Wireless Networking and Distributed Computing Lab for their helping to finish this work More generally, the opportu- nity to work with the members of Wireless Networking and Distributed Computing Lab has helped me increase my knowledge of the Wireless Sensor Network field.

I would like to thank National Kaohsiung University of Applied Sciences and Institute of Electronic Engineering, for providing me with a precious scholarship and resources for me to concentrate on pursuing a PhD In additon, I would thank all the professors of Department of Electronic Engineering for taking a course with them and gain knowledge and gaining a style of teaching from them during the period of my researching in KUAS.

I would like to thank Pham Van Dong University and Faculty of Information Technology, for believing in me and giving me the support to come to KUAS to finish my doctor degree.

In addition, I would like to thank my friends, who are studying in KUAS, for their encouragement and help me quite a lot in sharing the problems occurred in

my life during four years in KUAS.

Finally, I would like to thank my parents and the members in my family, for everything that they have given me too much Especially, I would like to thank

my wife for understanding, sacrifices and take care our babies while I studied in KUAS.

v

Declaration of Authorship i

1.1 Literature Survey and Motivation 3

1.2 Contributions of the thesis 7

1.2.1 Constructing Virtual Data Aggregation Trees Scheduling 7

1.2.2 Construct Virtual Backbone Scheduling 8

1.3 Organization of the thesis 8

2 Background 10 2.1 An overview of Wireless Sensor network 10

2.2 Specific Issues in Wireless Sensor Network Systems 14

2.2.1 The problem of Duty cycle scheduling 14

2.2.2 The problem of Data aggregation 15

2.2.3 Maximum lifetime problem in WSNs 16

2.2.4 Coverage problem in WSNs 16

2.3 Summary 17

3 On Maximizing the Lifetime for Data Aggregation in Wireless Sensor Network Using Virtual Data Aggregation Trees 18 3.1 Problem Definition and Its Hardness 20

3.1.1 Network Model 21

vi

3.1.2 Data Aggregation Tree 22

3.1.3 The Maximum Lifetime Data Aggregation Tree Scheduling Problem and Its Hardness 25

3.2 Local-Tree-Reconstruction-Based Scheduling Algorithm 31

3.2.1 Distributed Construction of a Shortest Path Tree 32

3.2.2 Local-Tree-Reconstruction Algorithm 33

3.2.3 The Proposed Scheduling Algorithm 39

3.3 The Correctness and the Time Complexity of the LTRA 40

3.4 Performance Evaluation 43

3.4.1 Uniform and Non-uniform Energy 44

3.4.2 Aggregation Ratio 45

3.4.3 Number of Relay Nodes 46

3.4.4 The Number of Units of Raw Data Generated by Sources 46

3.4.5 Field Size 47

3.5 Summary 48

4 An Efficient Algorithm of Constructing Virtual Backbone Schedul-ing for MaximizSchedul-ing the Lifetime of Dual-Radio Wireless Sensor Networks 50 4.1 Network model and problem definition 51

4.1.1 Network model 51

4.1.2 Problem Definition and Its Hardness 55

4.2 The Dominating-Set-Based Algorithm (DSBA) 56

4.2.1 Construction of Dominating Set 57

4.2.2 Establishment of Backbone 60

4.2.3 Refinement of Backbone 63

4.3 Analysis of the DSBA 65

4.4 Performance Evaluation 69

4.4.1 Network Lifetime 70

4.4.2 Size of Virtual Backbone 72

4.4.3 Residual Energy 73

4.5 Summary 75

5 Conclusions 77 5.1 Contributions 77

5.2 Future directions 79

1.1 Example of a Wireless sensor network 2

2.1 Four basic components of the sensor node 12

3.1 Example of the network lifetime while using virtual data tion trees (a) and (b) show the data aggregation trees T1 and T2, respectively 19

aggrega-3.2 Example of a connected weighted graph and data aggregation trees (a) shows a connected graph G = (V G , E G , w G , ρ G ), where the left number and the right number in parentheses represent the corre- sponding energy power and number of units of generated raw data, respectively (b) and (c) show two data aggregation trees T 1 and

T2, respectively 20

3.3 Example of virtual data aggregation trees in the network (a) shows

a data aggregation tree T1 with lifetime equal to 4 (b) shows the residual energy power of nodes after four working rounds (c) shows another data aggregation tree T2 with lifetime equal to 1 (d) shows the residual energy power of nodes after two working rounds 26

3.4 Example of the reduction from the X3C problem to the MLDATS problem (a) shows the corresponding instance of the MLDATS problem while given the instance of the X3C problem, including

X = {1, 2, 3, 4, 5, 6} and C = {C1, C2, C3}, where C1 = {1, 3, 5},

C2 = {2, 4, 6}, and C3 = {1, 2, 3} (b) shows a data aggregation tree T in the graph shown in (a) 28

3.5 The lifetime of networks whose number of sensors ranging from 100

to 1000 The initial energies of the sensors are uniform in (a) and non-uniform in (b) 45

3.6 The lifetime of networks that have aggregation ratio α ranging from

1 to 5 46

3.7 The lifetime of networks that have relay nodes ranging from 50 to

250 47

3.8 The lifetime of networks that have β ranging from 10 to 50 47

3.9 The lifetime of networks whose field size ranges from 10 × 10 to

50 × 50 48

viii

4.1 Example of dual-radio wireless sensor networks, where the number close to a node denotes the node’s energy (a) The network topology

G 1 (V G 1 , E G 1 , w G 1 ) formed by the nodes using small-range radios (b) The network topology G2(VG2, EG2, wG2) formed by the nodes each using large-range or small-range radio 52

4.2 An example of reducing the energy consumption of backbone nodes (a) Nodes u, v, x, and y use a large-range radio in the backbone (b) Node v uses a small-range radio, and the others use a large-range radio in the backbone 64

4.3 An example of pruning redundant backbone nodes (a) The network has a redundant backbone node v (b) Node v is pruned from the backbone shown in (a) 64

4.4 The network lifetime versus the number of network nodes ranging from 50 to 500 The initial energies of nodes in the networks are

400 in (a) and randomly chosen from the interval [0, 400] in (b), respectively 71

4.5 The network lifetime versus β, where the backbone nodes using a large-range radio (or small-range radio) consume β (or 1) energy power for one working cycle, the network has 500 nodes and the initial energy of nodes in the networks is 400 73

4.6 The average size of virtual backbones versus the number of network nodes ranging from 50 to 500 The initial energies of nodes in the networks are 400 in (a) and randomly chosen from the interval [0, 400] in (b), respectively 74

4.7 The average residual energy of all nodes in the networks versus the number of network nodes ranging from 50 to 500 The initial energies of nodes in the networks are 400 in (a) and randomly chosen from the interval [0, 400] in (b), respectively 74

DBA Distributed Backbone Algorithm

x

Wireless sensor networks are composed of many wireless sensors deployed in

a wide range of areas, where each sensor is able to communicate with othersthrough inter-sensor wireless communication Recently, many applications

of wireless sensor networks have been developed, such as, environmentalmonitoring, animal tracking, surveillance, endangered species protection,fire detection and seismic monitoring [1 3] The most importance operation

in these applications in WSNs is data aggregation, to collect sensing datafrom the sensor nodes and periodically report to a specific node, called

a sink Fig 1.1 shows the an illustration of a Wireless sensor networkincluding some sensor nodes and one sink node

Because of the requirement of applications, wireless sensor network can

be constructed in two different manners, that is, the static wireless sensornetwork and mobile wireless sensor network In the type of static wirelesssensor network, the sensor nodes are deployed at fixed position in the fieldarea network and can form a steady routing structure to transmit theirsense data to sink Because of static of network topology, the cooperationmechanism of sensor nodes in the network should be simple, implemented by

1

sensor node sink node radio range network edge

Figure 1.1: Example of a Wireless sensor networklow complexity algorithm requiring little space for data storage, thus savingenergy resources However, the static network structure is not suitable forsome applications For example, in situations where the sensor nodes need to

be attached directly to moving objects, such as, animal tracking, cell-phones,vehicles in urban areas In the type of mobile wireless sensor network, whichcomposes the sensor nodes can move in the field area network In thismodel network, the communication between the nodes in the network isquite unreliable because of the moving of sensor nodes Hence, the routingstructure also is changed

The most important component to construct a WSNs is the sensor nodeswhich are capable of performing some processing, gathering sensory infor-mation and communicating with other connected nodes The sensor nodesare generally equipped with a radio transceiver, a micro controller, a mem-ory unit, and a set of transducers using which it can acquire and processdata from the deployed regions [4] Because the energy power of sensors is

limited, energy efficiency in wireless sensor networks is the major challenge

in the design of the network to enhance the network lifetime

In this section, we present a brief survey of literature on the topics of terest to the thesis The scope of survey is divided into the following areas

in-in brin-ingin-ing out the motivation of the thesis work: data gatherin-ing problem

in wireless sensor networks, clustering structure for data aggregation, treestructure for data aggregation, backbone structure in wireless sensor net-work and problem of scheduling the backbones in wireless sensor networks.This survey provides the motivation of the problems that have been worked

on in the thesis

The data gathering is a well-known mechanism for collect sensingdata from sensors and reports to a specific node, called a sink Because theenergy powers of sensors are limited, efficiently gathering data is a majorchallenge in WSNs Recently, researchers have studied efficiently gather-ing data when multiple data are allowed to be aggregated into one packet[5, 6] In [7], Madden et al propose methods for improving the reliabilityand performance of retrieving data via a tree when basic database aggre-gates, including min, max, sum, average, and count, are used with group-ing In [8], Kalpakis et al propose scheduling methods based on admissionflow networks to maximize the network lifetime In [9], Krishnamachari et

al use the data aggregation tree to model data-centric routing to yieldenergy-efficient dissemination In [10], Wu et al study the construction of

a data-gathering tree to maximize the network lifetime In addition, Many

researchers have studied efficiently gathering data in WSNs when a fixednumber of data are allowed to be aggregated into one packet [11–13] In[11], Liu and Cao propose solutions to design a fault-tolerant and energy-efficient distributed monitoring system in WSNs In [12], Kuo and Tsaipropose methods of constructing data aggregation trees such that the totalenergy cost for gathering data is minimized In [13], Liu and Jhang proposenovel data aggregation and routing structures for gathering data With thestructures, a distributed data scheduling algorithm is proposed to scheduledata to nodes such that the energy cost for gathering data is minimized,when all data are aggregated at most once However, minimizing the totalenergy cost for gathering data does not imply that the network lifetime ismaximized.

Clustering structure for data aggregation is a well-known structurefor data aggregation in WSNs In which, each cluster head collects data fromthe corresponding cluster members The data aggregated in the clusterhead is then sent to the sink Recently, many data aggregation methodsare based on the clustering structure [14–16] H Chen et al propose amethod to select the cluster heads to cover whole sensor network so thattotal information transmitted through the sensor network is minimized [14]

In [15], Y Liang et al propose a novel clustering algorithm which better suitthe periodical data gathering applications to effectively reduce redundantdata transmission and the whole energy consumed in the network In [16],

K Dasgupta et al propose an efficient clustering-based heuristic to solvethe data-gathering problem with aggregation in sensor networks such thatthe system lifetime is maximized

The tree for data aggregation is a common structure that leads the

data generated in WSNs to the sink By using the tree structure, each node

in the tree is responsible for forwarding its generated data and the datareceived from its child nodes to its parent node In [17], Lee and Wongpropose an overlay tree structure to prolong the lifetime of sensors in thenetwork In the tree, the nodes that have higher residual energy powerare selected as the parents to facilitate data collection In addition, whensensors are no longer functional or network links are broken, the tree can

be reconstructed by their proposed method In [18], Luo et al study theproblem of selecting a maximum lifetime tree from a set of shortest pathtrees In [19], Dasgupta et al propose an approximation method that usesintelligent selection of trees to solve the maximum lifetime data collectionproblem in sensor networks In [20], Wu et al propose an approximationalgorithm of constructing a spanning tree to prolong the network lifetimewhen a single sink exists In these studies, the data is directly forwarded viathe tree without minimizing the size of the aggregated data, and therefore,

it spends lots of energy for data gathering

The backbone structure is a well-known technique for various aspects

of wireless sensor networks, such as routing, multicasting, and broadcasting[21, 22] In the backbone technique, some nodes in the backbone alwaysturn on their radio to maintain the activation of the network, and the rest

of the nodes turn off their radio to save energy Recently, many researchstudies are mostly designed to minimize the size of the backbone to, there-fore, reduce the energy consumption in the network In [23], Zhang et al.proposed an algorithm to construct a connected backbone in the wirelesssensor network such that the network achieves much higher throughput anddelivery ratio, and much lower end-to-end delay and routing distance In

[22], an algorithm was proposed to construct a backbone in the wireless sor network such that the cost of routing would be reduced In [24], Chuang

sen-et al proposed a novel heuristic-based backbone algorithm, called Bone In SmartBone, two steps are required to form a backbone to improveenergy saving in the network The first step is to find backbone nodes bythe proposed flow-bottleneck preprocessing method, and the second step is

Smart-to reduce the number of redundant nodes by the proposed dynamic sity cutback method Recently, some researchers focused on prolonging thenetwork lifetime while using the backbone technique The network lifetime

den-is often determined by the time span from when the network den-is active towhen one node in the backbone runs out of energy In [25], Torkestani in-troduced an extended version of the connected dominating set problem formodeling the energy-efficient backbone formation problem in wireless sensornetworks In addition, an algorithm of constructing the network backbonewas proposed to maximize the lifetime

Scheduling the backbones on WSNs which allows multiple bones work for different time periods Because the nodes in the backboneare always active, their energy is depleted quickly so that the backbone iseasily broken Therefore, to prolong the network lifetime, some researchersstudied scheduling nodes to become backbone nodes during some periods

back-of time and non-backbone nodes during other periods In [26], Zhao et

al studied the problem of finding multiple backbones and scheduling thebackbones to work for different time periods such that the network lifetimewould be maximized Hereafter, the multiple backbones used for differentperiods were called virtual backbones The problem is shown to be NP-hard

in [26] In addition, centralized algorithms are proposed to find and schedule

virtual backbones to maximize the network lifetime However, centralizedalgorithms are often not feasible in a wide range of wireless sensor networks[27, 28].

Schedul-ing

The first main contribution of this thesis is the design of Virtual Data gregation Trees Scheduling for maximizing the lifetime of Wireless SensorNetwork In this work, we study the problem of constructing data aggre-gation trees for different time to maximizing the network lifetime In thismanner, we consider to the bottleneck nodes, which run out their energyfirstly in WSNs, to construct a reconstruction data aggregation tree algo-rithm to reduce the collected data at source node as bottleneck nodes toprolong its lifetime and then to prolong the lifetime of the data aggregationtree We then present a maximum lifetime data aggregation tree schedulingproblem to increase the number of working rounds for the data aggrega-tion in the network The lifetime of schedule for the data aggregation tree

Ag-is denoted the lifetime of network Based on the simulation, we illustratethat our proposed solutions achieve the significantly prolong the networklifetime

1.2.2 Construct Virtual Backbone Scheduling

In the second contribution of this thesis, we design the Virtual backbonescheduling for maximizing the lifetime of Wireless Sensor Network In thiswork, to maximize the network lifetime, we study the problem of schedulingvirtual backbones for a dual-radio wireless sensor network such that the net-work lifetime is maximized This Maximum Lifetime Backbone Schedulingfor Dual-Radio Wireless Sensor Network problem occurs when each node

in the network is equipped with two radio interfaces, including small-rangeradio and large-range radio In addition, the nodes belong the backbone canuse small-range radio or large-range radio to maintain the backbone, the rest

of nodes use small-range radio to connect the backbone Simulation resultsshow that the proposed algorithm outperforms some existing algorithms

In addition to this introduction chapter, the thesis concludes 4 chapters and

is organized as following In chapter 2, the fundamental background on thewireless sensor network and the energy constraint issues is provided Inaddition, a review of related works with the energy constraint problem ofthe sensor networks is presented Through the discussion of related works,the research’s objective, methodology and the originality of the schemesinvestigated in this thesis are proposed

In chapter 3, we study the problem of constructing data aggregationtrees for different time to maximizing the network lifetime, termed maxi-mum lifetime data aggregation tree scheduling problem An algorithm that

is based on the k-hop local information to the sink to construct data gation trees for different time is proposed.

aggre-In chapter 4, we study the problem of constructing virtual backbones

in dual-radio wireless sensor networks to maximize the network lifetime, adistributed algorithm is then proposed for a wide range of wireless sensornetworks to find a backbone when a new one is required The last chapter ofthis thesis, we conclude this thesis and proposes directions for future works

In chapter 5, we summary the results of this research work for the specificproblems considered in this thesis and the future directions

This chapter provides an overview of important topics to understand thework reported in this thesis This overview aims to briefly present the mainconcepts and ideas concerning them, highlighting those that have a relation

to the content of the work in this thesis, to ease the understanding of itscontribution This chapter includes as following: An overview of WirelessSensor network, and Specific Issues in Wireless Sensor Networks

Wireless sensor network (WSN) is a network system which is designed forsensing and processing signals from the environment, such as temperature,humidity and sound and transmitting the sensed data through wireless chan-nels Wireless sensor networks consist of many small compact devices thatform a wireless network Each sensor node in the network is equipped withsensors to collects information from its surroundings, and sends the sensingdata to a base station (sink) directly or via the immediate neighbor nodes.The sink node in WSNs can be static or mobile Depending on the WSN

10

application and the multiplicity of the requesting users, multiple sinks may

be present in the network The purpose of the sink nodes is to provide aninterface between the WSN and another type of network from which theend-user will be able to access the data acquired by the sensor nodes Thefirst wireless network, namely the Sound Surveillance System (SOSUS), isdeveloped by the United States Military in the 1950s to detect and trackSoviet submarines [29] This network used submerged acoustic sensors –hydrophones – distributed in the Atlantic and Pacific oceans Nowadays,this sensing technology is still in service and develop with more peacefulfunctions of monitoring undersea wildlife and volcanic activity

The sensor node consists four basic components such as sensing unit,transceiver unit, processing unit and power unit, which is showed as Fig

2.1 The sensing unit uses to sense data from its surroundings and convertthe sensed data to digital signals, and then fed into the processing unit.The processing unit is generally associated with a small storage unit and itcan manage the procedures that make the sensor node collaborate with theother nodes to carry out the assigned sensing tasks The transceiver unitcan connect the other node to transmit the sensed data together within thefixed radio range In addition, the power unit is one of the most importantcomponents of a sensor node This unit can be supported by a power scav-enging unit and used to provide the energy to other units Thereto, due

to each application, the sensor node is plug other subunits such as GlobalPositioning System (GPS), camera, etc

Nowadays, WSN is an important and exciting new technology with greatpotential for improving almost applications such as medicine, transporta-tion, agriculture, industrial process control, and the military In addition,

Sensor Converter Processor Storage Transmitter

it is necessary to consider some basic characteristics of WSNs as following:

• Flexibility: WSNs should be scalable, and they should be able todynamically adapt to changes in node density and topology For ex-ample, in case the sensor node will exit from the networks because ofthe battery exhaustion and other failures; or in some applications forobject tracking where the sensors always change their position in thenetwork These will bring about changes in the topology of network,

so the WSN topology must have the function of the reconfiguration,dynamic and self-adjustment

• Coverage: Coverage is a fundamental issue in a WSN, which mines how well a phenomenon of interest is monitored or tracked bysensors Each sensor node is able to sense the phenomenon in a finitesensing area which is normally assumed to be a disk with the sensorlocated at the center The radius of the disk is called the sensing range

deter-of the sensor Any point in the sensing area deter-of a sensor is said to becovered by the sensor

• Connectivity: Connectivity is an important issue in WSNs whichconcerns with delivering the sensed data from the source sensor tothe destination via radio transmissions Each sensor node has onlylimited communication range compared with the size of the monitoredarea Two sensors are called neighbors if they are within each other’scommunication range The sensor nodes and the communication linksbetween each pair of neighbors build the network topology, which isrequired to be connected by the connectivity requirement.

• Multi-hop communication: Each sensor node in WSNs only cancommunicate with direct neighbors within its radio range In case onenode need communicate with the nodes, which is beyond the coverage

of the node’s radio range, it must be use multi-hop route transmitdata through the intermediate nodes Multi-hop communications arenecessary when the source node cannot reach the destination nodedirectly

• Data collection: In WSNs, all sensed data of nodes will be send

to sink via single-hop or multi-hop by the data collection mechanism.These cause non-uniform power consumption patterns that may over-burden forwarding nodes This is particularly harsh on nodes providingend links to sink, which may end up relaying traffic coming from allother nodes, thus forming a critical bottleneck for throughput of thenetwork Therefore, the data gathering mechanism must be carefullydesigned to save the energy consumption of sensors and to reduce thetraffic data of the nodes in WSNs

• Storage, search and retrieval: Since the sensed data in WSNs

is continuous and needs to be processed in real time, the traditional

database are not suitable In addition, the sensor nodes have storageconstraints, processing and memory constraints Therefore, it is nec-essary to require significant modifications to existing techniques of thetraditional database.

• Network lifetime: Network lifetime is extremely critical for mostapplications, and its primary limiting factor is the energy consumption

of the nodes, which need to be self-powering The network lifetime isoften determined by the time span from when the network is active towhen the first node in the network runs out of energy

• Latency: Latency is the delay time from when a source node sends

a packet data until this packet data is successfully received by thedestination node In general, a low latency network connection spendssmall delay times, while a high latency connection spends long delaytimes

Systems

In WSNs, it is necessary for nodes to relay and send their sensed data

to sink However, since sensor nodes have a limited power energy, it isunavoidable that duty cycle scheduling for nodes be exploited to conserveenergy when the application has long lifetime requirements The use of dutycycle scheduling will keep the nodes in a low duty cycle to decrease the powerconsumption In this manner, when working in a low duty cycle, nodes will

be in the inactive state for most of the time A node in the inactive statewill turn off their radio transceivers and other onboard equipments to saveenergy Therefore, when a fixed duty cycle scheduling scheme is used in theWSNs, the sensor nodes are divided two groups, that is, the active group andinactive group In the active group, the sensors always turn on their radio

to maintain the activation of the network Otherwise, the sensor nodes inthe inactive group turn off their radio to save energy In this thesis, the dutycycle scheduling problem for the wireless sensor network will be examinedbased on the backbone structure and data aggregation tree structure of thenetwork

The most importance operation in these applications in WSNs is data gregation, to collect sensing data from the sensor nodes and report to asink, at each time unit The collection data process is repeated until thedata packet of all nodes reaches to sink Data aggregation is a well-knowtechnique for data collection to reduce the energy consumption of reportingdata in WSNs In which, the collected data is generated by some aggrega-tion functions [5, 6], such as, max, min, sum, etc., and would be aggregatedinto a data packets with a fix constant size before they are transmitted Inthis thesis, the data aggregation problem for the wireless sensor network will

ag-be examined based on the data aggregation tree structure of the network

2.2.3 Maximum lifetime problem in WSNs

Maximizing network lifetime is an important issue in WSNs which defineshow long the deployed WSNs can function well Because the sensor nodesare equipped with battery which cannot be replenished and has a life span-ning from the point of deployment till its battery exhaust Therefore, thenetwork lifetime often indicates the time elapsed until the first node runs outits energy which is responsible to die down the network Thus, improvingthe network lifetime is a challenging issue for design the system in wirelesssensor networks that can conserve energy resource In this thesis, the max-imum lifetime problem will be examined based on the optimization dataaggregation and scheduling virtual backbones for wireless sensor network

With the rapid technological development of sensors, many applicationshave been designed to use wireless sensor networks to monitor a certain areaand provide quality-of-service guarantees Recently, the coverage problem

is a well-know technique for some issues, such as, constructing a minimumsize wireless sensor network to fully cover critical squares in a sensor field

to prolong the network lifetime, constructing a wireless sensor network bydeploying sensors with minimum cost to fully cover a sensor field, etc

2.3 Summary

By analysing the some basic characteristics of WSNs and specific Issues, it

is possible to notice that a main concern in this research area is how to vide suitable solutions that prolong the lifetime of WSNs This observationmotivates the study of solutions that can provide these features which areproposed in Chapter 3 and 4 in this thesis

pro-On Maximizing the Lifetime for Data Aggregation in Wireless

Sensor Network Using Virtual

Data Aggregation Trees

In WSNs, some sensors, termed source nodes, are responsible for sensingthe environment and generating sensing data, and some sensors, termedrelay nodes, are allowed to forward received data to others When anysource node runs out of the energy, the end of the lifetime for the WSN isreached because the network cannot provide adequate quality of service forapplications To efficiently prolong the network lifetime, our idea is to use

a set of data aggregation trees that work in turn for different time periodssuch that the network lifetime is maximized Hereafter, the data aggregationtrees used in different time periods are called virtual data aggregation trees.Take Fig 3.1, for example In Fig 3.1, node s is the sink, and the numberclose to a node denotes the node’s remaining energy For simplicity, weassume that the energy consumption for any non-leaf node is 2 per round

18

c e

a d

b

s

8 8

7

(a)

c e

a d

b

s

5 5

1

(b)

relay node source node network edge tree edge

Figure 3.1: Example of the network lifetime while using virtual data tion trees (a) and (b) show the data aggregation trees T1 and T2, respectively.

aggrega-to receive and forward data In addition, we also assume that the energyconsumption for the node that is a source node and leaf in the tree is 1 perround to transmit data If we use the trees T1, as shown in Fig 3.1(a),and T2, as shown in Fig 3.1(b), in turn to be the data aggregation trees inthe network, T1 can survive 3 rounds, and T2 can survive 2 rounds Notethat node b in Fig 3.1(b)does not consume any energy because it is a relaynode, and has no data to receive or transmit We have that the network cansurvive 5 rounds, which is better than that using only one data aggregationtree This motivates us to study the problem of scheduling virtual dataaggregation trees to maximize the network lifetime when a fixed number ofdata are allowed to be aggregated into one packet, termed the MaximumLifetime Data Aggregation Tree Scheduling (MLDATS) problem

The remaining sections of this chapter are organized as follows The inition and the hardness of the MLDATS are formally illustrated in Section

def-3.1 In addition, a local-tree-reconstruction-based scheduling algorithm isproposed in Section 3.2 to find virtual data aggregation trees for the WSN.The analysis is provided in Section 3.3 The performance of our proposed

(65, 0)

k j

i h

i h

i h

relay node source node network edge tree edge

Figure 3.2: Example of a connected weighted graph and data aggregation trees (a) shows a connected graph G = (V G , E G , w G , ρ G ), where the left number and the right number in parentheses represent the corresponding energy power and number of units of generated raw data, respectively (b) and (c) show two

data aggregation trees T 1 and T 2 , respectively.

method is evaluated in Section 3.4 Finally, the chapter is concluded inSection 3.5

In this section, we first describe the network model for WSN that can berepresented by a connected graph in Section 3.1.1 Based on the networkmodel, we present the structure of a data aggregation tree used for data

aggregation in WSNs and its lifetime in Section3.1.2 Finally, the MaximumLifetime Data Aggregation Tree Scheduling (MLDATS) problem and itshardness are developed in Section 3.1.3.

In this paper, the communication model in the wireless sensor network isassumed to be a unit disk graph model [31] In the model, a sensor u canreceive messages from sensor v if u is within the transmission range of v.Hereafter, u is said to be v’s neighboring node in the network if u can receivemessages from v When all sensors have the same transmission range, theWSN can be represented as a connected weighted graph G(VG, EG, wG, ρG),where node v ∈ VG denotes a sensor in the networks, edge (u, v) ∈ EGrepresents that u and v can communicate with each other, wG(v) denotesthe energy power of v, and ρG(v) is the number of units of raw data generated

by v to report to sink s ∈ VG per unit time, where the sink s is a specialnode in the network that is responsible for data collecting, processing, andanalysis The data generated in a unit time have to be reported to thesink in the same unit time, which is called a working round in this paper

In addition, the nodes v with ρG(v) > 0 are called source nodes; and thenodes v with ρG(v) = 0 are called relay nodes hereafter The relay nodecan receive data from other nodes and forward the data to the next nodefor reporting data to the sink The source node can generate its own rawdata for each working round and works like a relay node to help relay data.Fig 3.2(a) shows an example of the connected graph representing a WSN,where the number of units of raw data generated by a node is shown as the

right number in parentheses Note that node s is the sink, three nodes a, eand g are relay nodes, and the rest of the nodes are source nodes.

A data aggregation tree constructed for the network G(VG, EG, wG, ρG) is aspanning tree T = (VT, ET), where VT = VG, ET ⊆ EG A data aggregationtree has the following characteristics:

• A data aggregation tree T is a connected graph rooted at the sinkwithout cycle Each node in T forwards its generated/received data toits parent node in T

• The data generated by source nodes in G are forwarded to and collected

in the sink via T in each working round

Figs 3.2(b)-3.2(c) show two data aggregation trees that span all nodes inthe WSN shown in Fig 3.2(a)

Because each node u in a data aggregation tree T has to forward itsgenerated/received data to its parent node in T , u has to help forward alldata generated in the subtree rooted at u Let u.tot denote the total number

of units of raw data generated and received by u Also let uT.CH denotethe set of u’s child nodes in T Then u.tot can be defined in the followingequation:

units of raw data allowed to be aggregated into one unit-size packet [11–13].The number of unit-size packets that are forwarded by each node u in T ,denoted by u.δ, is defined with the following equation:

u.δ =  u.tot

α



(3.2)

Take Fig 3.2(b)-3.2(c), for example Assume that the aggregation ratio

α = 2 In Fig 3.2(b), it is clear that d.tot = 10 because 2, 1, 3, 2, 2 units

of raw data are generated by nodes d, h, i, l, m, respectively, which are thenodes in the subtree rooted at d In addition, a.tot = 10 because d.tot = 10and no raw data are generated by node a Then a.δ = d.δ = 102  = 5 InFig 3.2(c), because d.tot = 5 and a.tot = 5, d.δ = a.δ =52 = 3

Let etx (or, erx) denote the energy consumption of the sensor node totransmit (or, receive) one unit-size packet Because each node u in T has toreceive packets from its child nodes in T and forward the aggregated data

to its parent node in T , therefore, the energy consumption for node u in

T to receive and forward data within each working round is defined as thefollowing equation:

In Fig 3.2(c), because d.δ = a.δ = 3, the energy consumption of a for eachworking round is therefore 1 × 3 + 1 × 3 = 6

In WSN, each node u has limited energy power wG(u) to maintain itsactivity The lifetime of u, denoted by u.`, is therefore defined as the maxi-mum number of working rounds for u to sustain [10, 23, 26], as defined bythe following equation:

u.` = wG(u)

u.eng



(3.4)Take Figs 3.2(b)-3.2(c), for example Assume that etx = erx = 1 and theaggregation ratio α = 2 In Fig 3.2(b), it is clear that a.` = 6510 = 6 Inaddition, because e.δ = 0, e.` = ∞ In Fig 3.2(c), it is clear that a.` =

T2, that is, T2.`, is equal to 4 because node e has the minimum lifetime 4among all nodes in T2

Because we can find a set of data aggregation trees, T1, T2, , Tp,that work in turn for time periods t1, t2, , tp, respectively, to prolongthe network lifetime Hereafter, the data aggregation tree working for acertain time period is called a virtual data aggregation tree The network

lifetime is therefore defined as the sum of the number of working rounds ofall virtual data aggregation trees, that is,

no energy power on forwarding data because e is a leaf in T2 and generates

no raw data required to be reported to the sink Because T2.` = 1, T2 cansustain for at most 1 working round Fig 3.3(d) shows the status of allnodes in the network after 1 working round It is clear that when T1 and T2are used for the network, the network lifetime is 4 + 1 = 5

Schedul-ing Problem and Its Hardness

While we are given a WSN, our problem is to find the scheduling of tual data aggregation trees to maximize the network lifetime, termed theMaximum Lifetime Data Aggregation Tree Scheduling (MLDATS) problemhereafter The MLDATS is formally illustrated as follows:

vir-INSTANCE: Given a graph G = (VG, EG, wG, ρG), etx ∈ R+, erx ∈ R+,

an aggregation ratio α ∈ Z+, and a constant k ∈ Z+

QUESTION: Does there exist a schedule of virtual data aggregation trees

in the network, that is, a list of 2-tuples {(T1, R1), (T2, R2), , (Tp, Rp)},

g

(65, 0)

k j

i h

i h

i h

(c)

s

(33, 0)

k j

i h

(d)

relay node source node network edge tree edge

Figure 3.3: Example of virtual data aggregation trees in the network (a) shows a data aggregation tree T 1 with lifetime equal to 4 (b) shows the resid- ual energy power of nodes after four working rounds (c) shows another data aggregation tree T2with lifetime equal to 1 (d) shows the residual energy power

of nodes after two working rounds.

such that the lifetime of the network is not less than k, where each tuple(Ti, Ri) represents that the data aggregation tree Ti has to start to work atworking round Ri?

Take Fig 3.3, for example Assume that etx = erx = 1 and the tion ratio α = 2 Let T1 and T2 be the virtual data aggregation trees shown

aggrega-in Figs 3.3(a) and 3.3(c), respectively We can set {(T1, 0), (T2, 4)} to bethe scheduling of virtual data agggregation trees in the network