無線感測器自動化網路佈署協定設計與實作

國 立 交 通 大 學

電信工程學系

碩 士 論 文

無線感測器自動化網路佈署協定設計與實作

Coverage-aware Sensor Deployment Schemes and

Implementation of an Automated Home Monitoring Network

研究生：劉葳庭

指導教授：林亭佑 博士

無線感測器自動化網路佈署協定設計與實作

Coverage-aware Sensor Deployment Schemes and Implementation

of an Automated Home Monitoring Network

研 究 生：劉葳庭 Student：Wei-Ting Liu

指導教授：林亭佑 Advisor：Ting-Yu Lin

國 立 交 通 大 學

電信工程學系

碩 士 論 文

A Thesis

Submitted to Department of Communication Engineering College of Electrical Engineering and Computer Science

National Chiao Tung University in partial Fulfillment of the Requirements

for the Degree of Master

in

Communication Engineering

June 2010

Hsinchu, Taiwan, Republic of China

中華民國九十九年一月

無線感測器自動化網路佈署協定設計與實作

學生：劉葳庭

指導教授

：

林亭佑

博士

國立交通大學電信工程學系﹙研究所﹚碩士班

摘

要

近年來無線感測網路的技術蓬勃發展，其應用也推陳出新，事實上，對於無線感測 網路而言，如何提供有效的感測覆蓋率是決定網路運作效率的重要因素。在這份三年的 計畫中，我們致力於設計居家智慧型無線感測網路，我們在感測器上配置行動裝置使其 具有行動能力，並針對居家環境設計一自動化感知傳測器佈署協定(Coverage-Aware Sensor Automation protocol，以下簡稱 CASA)，實現此居家型高智慧網路，藉由動態感 測器的自動佈署，以提供使用者所需的感測覆蓋率。此外，有別於其他先前的研究，我 們所設計的 CASA 協定允許網路中同時使用齊性或非齊性感測器，也就是說，CASA 協 定亦適用於感測範圍(sensing range)不同的感測器，在使用上具有較大的彈性。事實上， CASA 協定主要由 EVFA-B、CFPP、SSOA 這三個機制構成。EVFA-B 會針對我們設計

的距離門檻值

d

_th使感測器彼此之間運作引力或斥力，其合力結果會將感測器逐漸推向 合適的位置，以強化網路中的感測覆蓋率。為了達到高品質的感測覆蓋率，我們研究出 EVFA-B 中所使用的環境參數與網路拓墣有相當大的關係，例如:監控面積大小與網路中 的感測器數量，我們期望 EVFA-B 能提供有效的自動化佈署。此外，我們發現當感測器 重新佈署時，在移動的過程中可能會有碰撞問題發生，因此我們規劃 CFPP 演算法，針 對每一台感測器的移動路徑，事先偵測潛在的碰撞發生地點，並重新調整感測器的移動 時程，藉此避免碰撞發生。除此之外，當網路中有某些感測器發生故障或電力不足的情 形，我們設計 SSOA 演算法進行局部的修復行動，也就是說，當有感測破洞發生(sensing void)時，SSOA 會選擇此破洞周圍某些合適的感測器去修補它，而不是使用 EVFA-B 重 新佈署整個網路，藉此有效節省電力消耗。除此之外，我們發現如何選擇合適的救援感 測器，事實上屬於 Maximum-Weight Clique Problem(以下簡稱 MWCP)，此問題被公認為 NP-hard，我們將 MWCP 簡化(reduce)為選擇救援感測器的問題，發現我們的 weight 值 可能有正有負，然而目前能解決 MWCP 的演算法只考慮 weight 值恆正的情況。因此， 最終我們定義救援感測器選擇的問題時，只考慮 weight 值恆正的情況，如此一來才存在 有效率的 polynomial-time 演算法，而 weight 值為負的情況就使用 EVFA-B 代為解決， 藉由此合作機制，CASA 協定可以達到有效率的覆蓋率要求。在真實的環境中使用嵌入 式系統運行 CASA 協定，藉此設計一個可以容許感測器故障，藉由自動化佈署以延長網 路使用壽命的居家智慧型監控網路(MoNet)。此外我們會藉由觀察覆蓋率達成率、網路 自我修復能力、移動所耗費的電力，並實地模擬當緊急災害的發生時，MoNet 的事件回 報率，藉此估測 CASA 協定的效率。

Coverage-aware Sensor Deployment Schemes and Implementation

of an Automated Home Monitoring Network

Student：Wei-Ting Liu

Advisor：Dr.

Ting-Yu Lin

Department﹙Institute﹚of Communication Engineering

National Chiao Tung University

ABSTRACT

For the wireless sensor network (WSN) to operate successfully, a critical issue is to provide sufficient sensing coverage. In this thesis, we target on the home environment and deal with both the homogeneous (having identical sensing radius) and heterogeneous sensors (having different sensing ranges) equipped with locomotion facilities to assist in the sensor self-deployment. A coverage-aware sensor automation (CASA) protocol is proposed to realize an automated home monitoring network. Three centralized algorithms are included in the CASA protocol suite: EVFA-B, CFPP, and SSOA. Unlike most previous works that tackle the deployment problem only partially, we intend to address the sensor deployment-related problems in a holistic manner. The enhanced virtual forces algorithm with boundary forces (EVFA-B) exerts weighted attractive and repulsive forces on each sensor based on predefined distance thresholds. The resultant forces then guide the sensors to their suitable positions with the objective of enhancing the sensing coverage (after a possibly random placement of sensors). To achieve high coverage ratio, we prove that good choices for the associated weight constants greatly depend on sensor population and monitored area size, while independent of sensing radius. When sensors move around to self-deploy, the collision-free path planning (CFPP) algorithm, based on geometric formulations, comes into play by carefully scheduling the moving paths to avoid sensors colliding each other. Furthermore, in the presence of sensor power depletions and/or unexpected failures, our sensor self-organizing algorithm (SSOA) is activated to perform local repair by repositioning sensors around the sensing void (uncovered area). This capability of local recovery is advantageous in terms of saving the communication and moving energies. Selection of local rescue sensors with mixed negative and positive weights is NP-hard, and can be reduced from the

maximum-weight clique problem. We resolve this selection problem by considering

only positive weights (leaving the negative weights to be handled by EVFA-B), so that efficient polynomial-time computation can be utilized. In the case that local repairing is unable to provide required sensing coverage, SSOA invokes EVFA-B to globally redeploy sensors. As a result, adequate sensing coverage can be maintained even in the face of sensor node failures, effectively extending network functioning time. Performance of the proposed sensor deployment strategies is evaluated in terms of surveillance coverage, network self-healing competence, and moving energy consumption. We also implement our CASA protocol suite in a real-life home monitoring network (MoNet) to demonstrate the protocol feasibility and validate the

誌謝

研究所生涯終於告一段落，這一切都要感謝許多人對我的提攜與

鼓勵。首先要感謝我的指導教授－林亭佑博士，在我碩士生涯的過程

中，老師悉心的教導並指點我研究的方向，使得我在這段日子中獲益

匪淺，尤其是在準備論文與口試期間，多虧有老師的協助與指導，使

得本論文能夠更臻完整與嚴謹。

再來，要感謝冠勳學長在硬體實作上給我的協助，感謝學長教導

我許多實作相關的知識與技術，也要感謝冠樺學弟、宗欽學弟、修銘

學弟與柏齊學弟在各式嵌入式競賽中的付出與努力；另外要感謝

Charka，對於此論文的諸多貢獻與付出，也要謝謝實驗室同學們 – 境

宜、昆儒、承穎，在生活中給我的幫助與鼓勵；此外，也感謝其他所

有 Bun Lab 的成員，有你們的陪伴讓我的碩士生活增添不少色彩。

最後，要感謝我的父母親，從小對我的悉心栽培與諄諄教誨，給

予我無限的鼓勵與協助，讓我能無後顧之憂專心致志於研究上，並且

達成目標，謹以此文獻給我摯愛的雙親。

誌於 2010.01 新竹 交大

葳庭

Contents

Abstract-Chinese i Abstract-English ii Appreciation iii Contents iv List of Tables vi

List of Figures vii

1 Introduction 1

2 Prior work 4

3 Coverage-aware Sensor Automation (CASA) Protocol 8

4 Enhanced Virtual Forces Algorithm with Boundary Forces (EVFA-B) 10

4.1 Distance Threshold . . . 12

4.2 Weight Constants . . . 14

4.3 Verification of Parameter Settings . . . 17

5 Collision-Free Path Planning (CFPP) 19 5.1 Path Planning Strategy . . . 20 5.2 CFPP Algorithm Summary . . . 26

6 Sensor Self-Organizing Algorithm (SSOA) 29

6.1 Local Selection of Rescue Sensors . . . 30 6.2 Physical Movements Performed by Selected Rescue Sensors . . . 35 6.3 SSOA Algorithm Summary . . . 36

7 Performance Evaluation 37

7.1 Improved Surveillance Coverage . . . 38 7.2 Network Self-healing Capability . . . 40 7.3 Energy Conservation on Physical Movements . . . 41 8 Implementation of an Automated Home Monitoring Network (MoNet) 43

9 Conclusions 46

List of Tables

4.1 Summary of notations used in our EVFA-B . . . 18 5.1 Summary of notations used in the CFPP algorithm . . . 26

List of Figures

1.1 Illustration of an automated home monitoring network, and the importance of (movement-assisted) network self-healing capability to tolerate sensor

faults (no need to deploy new sensors). . . 2

4.1 Concept of attractive, repulsive, boundary forces, and virtual movement exerted on a sensor node. . . 10

4.2 Distance threshold (dij_th) settings for two arbitrary sensors si and sj under

four different environmental conditions. . . 13

4.3 Extreme node configuration used to derive the proper wr

wa ratio setting. . . 15

4.4 Impact of wa, wr (= wb) parameter settings on the coverage ratio of

mon-itored 200 × 200 area (HSR with ea ≥ 1). . . 16

4.5 Performance justification of proper choices for dij_th, wa, wr(wb) values in our

EVFA-B algorithm. . . 17 5.1 Possible intersection (collision) cases generated by moving paths of any two

sensors si and sj, where pi (pj) denotes the original position of si (sj) and

5.2 Every sensor si in the potential moving set Mmin order should be analyzed

by identifying its intersection (collision) relationship with each member

in Ci, in which intersection cases D-II, D-IV, S-I, S-II, S-III, S-IV, and

S-V require further consideration/processing, before including si into the

moving set (allowed to move in the current round). . . 25 5.3 Example illustrating the operations of our proposed CFPP algorithm for

sensor physical movements. . . 27

6.1 Experiments on possible selections of rescue sensors set Rdead to locally

recover the sensing void caused by faulty sensor sdead. . . 31

6.2 Construction of graph Gr for our rescue sensors selection problem (RSSP). 32

6.3 1st_{-tier and 2}nd_{-tier physical movements applied on selected rescue sensors}

and their affected immediate neighbors, respectively. . . 35 7.1 Sensor deployment status after 50 rounds (virtual movements) using Zou,

Zou-B, and our proposed CASA strategies, respectively (m = 200, n =

200, k = 80, HSR with ea ≥ 1). . . 38

7.2 Coverage performance accomplished by Zou, Zou-B, and our CASA de-ployment strategies under various amounts of sensor nodes in a monitored 200 × 200 area. Note that the results are obtained after the first redeploy-ment (no faulty sensor occurs yet). . . 39 7.3 Network self-healing performance comparison in a monitored 120 × 120

environment with 70 sensors where some faulty sensor occurs every unit

time (number of sensors reduced to only 32 at the 38th _{time unit). . . 40}

7.4 Physical movement energy consumption comparison after the first redeploy-ment is respectively completed by Zou, Zou-B, and CASA under various amounts of sensor nodes in a monitored 200 × 200 area. . . 41

8.1 Validation of the proposed CASA protocol suite by implementing a real-world home monitoring network (MoNet) via commodity hardware com-ponents. . . 43 8.2 Performance results obtained from our home MoNet prototype. . . 44

Chapter 1

Introduction

Advances of micro-electromechanical system (MEMS), sensing technology, and wire-less communication have significantly encouraged the development of wirewire-less sensor net-works (WSNs) in the past decade. A WSN is widely used for habitat and environmental surveillance, medical application (with the purpose of improving quality of health care), agricultural assistance, and as solutions to military problems [5, 11, 15, 16]. Several ex-perimental testbeds are also implemented to investigate various aspects of WSN-related performance issues [10,19,21,23]. Since different environments usually guide WSN studies to distinct research directions and design considerations, it is necessary to firstly define the target environment under investigation. In this thesis, we focus on the indoor home environment, as depicted in Fig. 1.1. To furnish the home with monitoring capability, one possibility could be embedding a secret compartment under the roof, and deploying smart sensors inside the double-deck structure on the ceiling. For a successful home surveillance, providing sufficient sensing coverage is essential. Manual placement of static sensors in-volves labor effort (reaching the ceiling to perform the planned deployment) and lacks network self-healing competence (when faulty sensors occur). Thanks to the availabil-ity of motion facilities, we consider smart sensors with mobilavailabil-ity capabilavailabil-ity to accomplish self-deployment after an initial random placement of sensors. Furthermore, since sensing devices are inherently unreliable, faulty sensors due to power depletions or unexpected

sensor sensing range

faulty sensor leaving a monitoring void

Full home monitoring restored on completion of network self-healing process Network self-healing

process automatically activated

Coverage-preserving sensor deployment scheme performed to form a full home monitoring network

A typical home environment Sensors randomly placed on the ceiling (home area not fully monitored)

sensor sensing range

Figure 1.1: Illustration of an automated home monitoring network, and the importance of (movement-assisted) network self-healing capability to tolerate sensor faults (no need to deploy new sensors).

errors may occur over time, leaving monitoring voids (uncovered sensing holes). With the movement ability, instead of replacing faulty sensors with new ones, those smart sensors reposition themselves to restore the sensing coverage, as illustrated in Fig. 1.1. According to the above descriptions, several deployment-related issues need be addressed. First, a coverage-aware sensor deployment scheme should be developed to ensure suf-ficient sensing coverage. Second, when sensors reposition themselves, a collision-free

route scheduling strategy is required to prevent sensors colliding each other. Third,

in the face of sensing node failures, a sensor self-organizing mechanism need be devised to efficiently recover the sensing void and restore the required sensing coverage. In this work, we do not intend to study the energy-conserving sensor communication behavior (though we try to reduce the moving energy by keeping sensors from moving far away when performing self-deployment), nor the issue of required amount of sen-sors to achieve certain degree of sensing coverage. Rather, given any number of sensen-sors, we investigate the deployment-related problems and propose a coverage-aware sensor

au-tomation (CASA, which means ”home” in Spanish) protocol including the aforementioned three deployment-related designs, with the objective of providing/maintaining high sens-ing coverage. Our ultimate goal is to realize an automated home monitorsens-ing network, so that detection applications of various emergence events can be practically implemented.

The remainder of this thesis is organized as follows. Chapter 2 reviews several prior research efforts and summarizes our unique contributions. In Chapter 3, we introduce the coverage-aware sensor automation (CASA) protocol and provide the environmen-tal assumptions made by the protocol. The proposed CASA protocol consists of three closely-related algorithms to address the sensor deployment scheme (EVFA-B), collision-free route planning (CFPP), and sensor self-organizing mechanism (SSOA), respectively. Chapter 4, Chapter 5, and Chapter 6 elaborate on the detailed operations of EVFA-B, CFPP, and SSOA separately. Chapter 7 presents the performance and comparison re-sults, while Chapter 8 reports our prototype of a home monitoring network (MoNet) and demonstrates the detection capability of CASA-enabled MoNet. Finally, we draw our concluding remarks in Chapter 9.

Chapter 2

Prior work

Depending on the target applications, earlier studies in WSNs generally focus on either outdoor large-scale environments, where planned sensor deployment is difficult, or indoor small-scale monitoring zones, where sensor deployment mechanism is feasible and beneficial. For large-scale WSNs, several works have been proposed to address the energy conservation issue [14, 22, 26, 29, 30]. Given sufficient number of sensors randomly deployed (scattered) over the monitoring field to ensure a certain degree of redundancy in sensing coverage, those proposals design node working schedules such that sensors can rotate between active and sleep modes. The objective of those proposed working schedules (node-scheduling protocols) is to achieve power conservation (prolonging system lifetime), while preserving reasonable sensing coverage and network connectivity.

For the monitoring environments where planned sensor deployment is possible, various static deployment strategies have been introduced to enhance the surveillance coverage [7, 8, 12, 25, 27]. In this kind of research studies, one commonly considered metric is to minimize the number of sensors required to achieve a certain sensing coverage. Due to different sensor capabilities (e.g., distinct attainable sensing/detection ranges) and manufacturing expenses, this metric is sometimes transformed into minimizing/optimizing the required total device cost for those deployed sensors, making this research subject

more interesting yet challenging [7,25]. However, such static deployment involves manual sensor placement/installation, and is incapable of dynamically repairing sensing voids (uncovered areas) in the presence of unexpected sensor failures.

Consequently, a number of research efforts have explored the movement-assisted sensor deployment techniques by utilizing mobile sensors to enhance the sensing coverage after an initial random placement of sensors [24,28,31]. With the motion facilities equipped at the sensing devices, sensors can move around to self-deploy. Given any number of ran-domly placed sensors, in [31], the authors present a centralized force-guided algorithm, inspired by the disk packing theory and virtual force field concept from robotics, to es-tablish motion paths for sensors. Assuming there exists a powerful clusterhead, capable of communicating with all sensors and obtaining sensor locations, the proposed algorithm evaluates all attractive and repulsive forces and obtains the resultant force exerted on each sensor. The computed resultant force then directs the sensor to move to a desired position. Also utilizing mobile sensors, the authors in [24] introduce a distributed sensor self-deployment scheme. They suggest to firstly identify the coverage holes (sensing voids) based on Voronoi diagram, and then propose three algorithms (choices) to guide sensor movements toward the detected holes. However, accurate Voronoi polygon constructions are not always possible to achieve, due to unevenly distributed sensors with limited com-munication distances. Therefore some optimization heuristic is needed to prevent sensors from moving too far and keep a reasonable number of total movements, further complicat-ing the deployment computations. Furthermore, since the termination condition for the Voronoi-based deployment strategy is coverage, for a monitoring environment with sensor number much larger than necessary, unbalanced sensor distribution (some areas are much more highly populated than other areas, even with an overall sensing coverage required) is likely to occur. As a result, the authors in [28] develop a scan-based movement-assisted sensor deployment (SMART) method to address the unbalanced problem. Instead of

tack-ling the deployment problem directly, SMART focuses on sensor load balancing by using 2D scanning and dimension exchanges to achieve a balanced network state. As claimed by the authors, SMART can operate on top of existing sensor deployment schemes, and pro-duces good performance especially for unevenly distributed WSNs. The aforementioned movement-assisted sensor deployment techniques all consider homogeneous sensors (with equal sensing/detection radius), and no specific route planning strategies are available to perform collision-free movements between sensors.

We observe that most previous works explore the sensor deployment problem only par-tially, leaving issues such as heterogeneous sensors (with different sensing ranges), sensor moving path scheduling, and locally recovering sensing holes (caused by sensor failures) unaddressed. However, in practice, those closely-related deployment isues should be re-solved as a complete protocol set to achieve an operative WSN with high detection ca-pability. In light of this, we investigate the movement-assisted sensor deployment subject by considering those deployment-related problems in a holistic manner. A coverage-aware sensor automation (CASA) protocol suite is proposed to address the global sensor deploy-ment scheme (EVFA-B), the sensor moving path planning (CFPP), and sensing coverage recovery in the presence of sensor failures (SSOA). We summarize our unique contribu-tions as follows. First, we develop the enhanced virtual forces algorithm with

boundary forces (EVFA-B) based on the concept of potential field and disk packing theory. Though sharing similar idea of virtual forces with [31], our EVFA-B

deals with both the homogeneous and heterogeneous sensors, while [31] only discusses the case of homogeneous sensors, where a global distance threshold value is adopted in

determining whether an attractive (with weight constant wa) or repulsive (with weight

constant wr) force should be applied on a sensor. However, in realistic settings, where

varying sensing distances are common, the distance threshold (determining the desirable

globally. In addition, since the observed environment is usually in a bounded area, our EVFA-B incorporates the boundary force (with weight constant wb) as a kind of

repul-sive force from the boundaries to keep sensors staying inside the monitoring area. Since

the boundary force is considered as a type of repulsive force, we use the same value for

wr and wb. In [31], no boundary force is modeled, and no specific design guidelines are

available for determining suitable wa and wr (=wb) weight constants. The authors only

suggest to select wr >> wa. However, we discover that arbitrary settings (even satisfying

wr>> wa) do not always yield desirable sensing coverage. Motivated by the observations,

we investigate and prove that good choices for wa and wr (=wb) greatly depend on sensor

population and monitored area dimensions, while independent of sensing radius. Second, we propose a collision-free path planning (CFPP) strategy, based on geo-metric formulations, to avoid sensors colliding each other when performing self-deployment. This route scheduling is necessary in order to achieve effective

sen-sor deployment in real environments. Third, the sensen-sor self-organizing algorithm

(SSOA) is devised to provide network self-healing (automated fault recovery) capability, which most previous sensor deployment protocols do not handle. Fourth, we

observe that most existing works do not have a real-life testbed to demonstrate their proposed protocols/algorithms. In this work, we implement a home monitoring

network (MoNet), based on embedded platforms, sensing components, communication

Chapter 3

Coverage-aware Sensor Automation

(CASA) Protocol

Three deployment-related mechanisms are incorporated in our CASA protocol set: EVFA-B, CFPP, and SSOA. The detailed operations of respective mechanism, with the objective of enhancing/preserving/recovering the sensing coverage for a home environ-ment, are elaborated in Chapter 4, Chapter 5, and Chapter 6, respectively. Below we summarize the environmental assumptions made in this work.

(A1) There exists a powerful clusterhead responsible for performing centralized compu-tations. All sensors are able to communicate with the clusterhead via single-hop or multi-hop wireless transmissions.

(A2) Sensors have the isotropic sensing shape and the binary sensing/detection behavior, in which an event is detected (not detected) by a sensor with complete certainty if this event occurs inside (outside) its sensing radius. Both the homogeneous (having identical sensing range) and heterogeneous (having varying sensing ranges) sensors are allowed in our model. Information of respective sensing ranges is provided by all sensors and made available at the clusterhead for deployment-related computations.

(A3) We adopt the discrete coordination system, in which the monitoring area (sensing field) is represented by a 2D grid network. Locations of all sensors are obtained via the pre-deployed RFID platform or some existing localization technique, and constantly updated to the clusterhead. Neighboring nodes under the adopted

co-ordination system are defined as sensors within the sensing range (rs), which is

normally much smaller than the radio communication distance (rc). Without loss

of generality, we assume that rc > 2rs in our model. According to the derivations

in [14, 30], if the radio communication range (rc) is at least twice the sensing radius

(rs), complete coverage of a convex area implies connectivity among the working set

of sensor nodes. Consequently, in this work, we only deal with the sensing coverage, and network connectivity follows accordingly.

Chapter 4

Enhanced Virtual Forces Algorithm

with Boundary Forces (EVFA-B)

The concept of virtual forces is inspired by the combined idea of potential field and disk packing theory [9,13]. Each sensor behaves as a source giving a force to others. This force can be either positive (attractive) or negative (repulsive). If two sensors are too close, they exert repulsive forces to separate each other, otherwise they exert attractive forces to draw

each other. We quantify the definition of ”closeness” by using the distance threshold dij_th

for any two sensors si and sj with respective sensing radius ri and rj (design guidelines on

dij_th are provided in Chapter 4.1). Given k sensors (denoted as s1, s2, . . . , sk with sensing

j s 2 x ib F o ik F o ij F o 0 il F o i F o next position of sensor si 1 2 3 4 5 6 7 8 i F o 1 x ib F o 2 y ib F o 1 y ib F o i s k s l s y axis x axis (0, 0) (n, 0) (0, m) i T ib d boundary 1 ib b ib F w d o § · ˜¨ ¸ © ¹

Figure 4.1: Concept of attractive, repulsive, boundary forces, and virtual movement ex-erted on a sensor node.

radius r1, r2, . . . , rk, respectively) deployed in the monitoring area, for any two sensors si

and sj located at coordinates (xi, yi) and (xj, yj), we adopt the Euclidean distance dij to

indicate how far the two sensors are spaced, where dij =

p

(xi− xj)2+ (yi− yj)2. As a

result, if dij > dij_th, then attractive force is applied. On the other hand, repulsive force is

generated if dij < dij_th. Define

− →

Fij as the directed virtual force acting on si from sj, now

we have − → Fij =            (wa(dij − dij_th), θij) for dij > dij_th (0, 0) for dij_ij = dth (wr(dijth− dij), θij+ π) otherwise            , (4.1)

where θij = tan−1 (y_(xii−y_−xj_j)₎ and wa(wr) represents the weight measurement for the attractive

(repulsive) force (detailed design guidelines on the two weight constants are elaborated

in Chapter 4.2). Take si in Fig. 4.1 for example, attractive force

− →

Fij from sj (to draw

si closer) and repulsive force

− →

Fik from sk (to repel si) are acting simultaneously on si.

In the case of setting distance threshold as the summation of two sensing ranges, the

virtual force −→Fil from sl equals zero (no force imposed on si by sl). In addition, we

incorporate the boundary force −→Fib to quantify the virtual force acting on si from the

monitored boundaries. By boundary forces, we can significantly reduce the unwanted

coverage outside the sensing field. As depicted in Fig. 4.1, the magnitude of −→Fib should

be inversely proportional to the perpendicular distance between si and the boundary, and

is formulated as |−→Fib| = wb(_d1_ib), where wb represents the weight measurement for the

boundary force. In this work, we regard the boundary force as a type of repulsive force,

and use the same value for wr and wb. In a rectangular area, boundary forces could be

from the four boundaries surrounding the monitoring region. Thus −→Fib is actually the

combined force from all boundaries, where −→Fib =

− → Fx1 ib + − → Fx2 ib + − → Fy1 ib + − → Fy2 ib. In Fig. 4.1,

to a zero −→Fib. Considering all attractive, repulsive, and boundary forces, we have the

resultant force −→Fi exerted on sensor si being defined as

−→ Fi = k X j=1,j6=i − → Fij + − → Fib. (4.2)

The determined resultant force −→Fi then guides si to virtually move to its next position.

Since we adopt the discrete coordination system, the next position for si is defined as the

closest possible grid point. As illustrated in Fig. 4.1, given the resultant moving angle

θi, with respect to the positive x axis in counterclockwise direction, we obtain the actual

motion angle θ0_i by approximating θ0_i = π

4round(

θi

π/4). Consequently, sensor si moves to

grid point 4, shown in Fig. 4.1, as its next position.

Our EVFA-B mechanism terminates when either the required sensing coverage

thresh-old (cth) is achieved or the maximum allowable virtual movements performed by each

sensor (Maxloops) is reached.

4.1

Distance Threshold

The distance threshold effectively defines the desired overlapping degree of two sen-sors. For homogeneous sensors, the distance threshold can be made as a global constant. However, for heterogeneous sensors, the value of distance threshold should be designed on per node-pair basis to obtain a similar degree of overlapping under different sensing distances. Specifically, for two sensors with small sensing ranges, the distance threshold should be made smaller than that of two sensors with large sensing distances, in order to keep reasonably similar overlapping level for the two sensor pairs (couples). Besides sens-ing ranges, the design of distance threshold also depends on the sensor density. Suppose

the monitoring area has size A, and the maximum area size covered by all sensors is As,

where As = π

P_k

2r 2r 2r

ISR with < 1 a ISR with a t 1 HSR with < 1a HSR with a t 1

Case I Case II Case III Case IV

i r j r ) (i j ij th r r d D  k s i s j s cos 6 r § ·_{¨ ¸}S © ¹ r i j rr i s j s k s

Figure 4.2: Distance threshold (dij_th) settings for two arbitrary sensors si and sj under

four different environmental conditions.

ratio ea < 1 implies the total number of sensors is insufficient to fully cover the monitoring

area. In this case, we cannot afford overlapping between sensors. On the other hand,

coverage ratio ea ≥ 1 indicates the sensor population is capable of fully covering the whole

area, in which case a certain degree of overlapping is desirable to minimize the sensing holes (uncovered zones). Based on the above principles, we propose to separately design

the distance threshold dij_thfor any two sensors si and sj under four environmental settings.

For homogeneous sensors, we use the abbreviation ISR to reflect the fact of having Iden-tical Sensing Radius. For heterogeneous sensors, we use HSR to represent the condition of possessing Heterogeneous Sensing Ranges. As illustrated in Fig. 4.2, Case I and Case

III deal with insufficient sensor population (reflected by ea < 1) for homogeneous and

heterogeneous sensors respectively, where overlapping is not desirable. As a result, the distance threshold is simply designed as the sum of two sensing ranges. In Case II and

Case IV, where sensor population is sufficient to allow overlapping (due to ea ≥ 1), the

design of distance threshold should try to minimize the sensing holes. In Case II, it is

easy to obtain the perfect (minimum) overlapping by setting dij_th = 2r cos(π/6), while in

Case IV, we set dij_th = α(ri + rj) by introducing a system tunable factor α to control

the desired overlapping degree, where 0 < α < 1. Consequently, we have the distance

dij_th =                 

2r for ISR with ea < 1

2r cos(π

6) for ISR with ea ≥ 1

ri+ rj for HSR with ea < 1 α(ri+ rj) for HSR with ea ≥ 1                  . (4.3)

4.2

Weight Constants

For the self-deployment algorithm based on virtual forces to perform effectively in achieving high sensing coverage in a bounded m × n area, the design of weight constants

waand wrassociated with the attractive and repulsive forces is a critical issue. Intuitively,

wrshould be set much larger than wa(as suggested in [31]), considering the relatively small

number of neighboring sensors (exerting repulsive forces) compared to the large number of non-neighboring nodes out there (exerting attractive forces). However, experimental

experiences reveal that arbitrary settings of a large wr and a small wa do not produce

effective sensing coverage in many cases. In this section, we attempt to characterize the

relationship between wr and wa by deriving a better formulated equation for setting the

two weight constants than simply suggesting to use wr >> wa (with arbitrary settings).

Consider an extreme node configuration shown in Fig. 4.3, where all the sensors (except

for si and sj) are located in one corner of the m×n sensing field. For sensor si, the virtual

forces it receives include the repulsive force from sj and attractive forces from all the other

(k − 2) nodes. The magnitude of repulsive force from sj is denoted as |

− →

FR

i |. Based on the

definition of repulsive force provided in Eq. (4.1), we have |−→FR

i | = wr|dij_th− dij| = wr∆,

where ∆ is a small value that represents the tolerable overlapping between si and sj. On

the other hand, since the average distance between si and all the other (k − 2) nodes is

approximately (√m2_{+ n}2₋√_2r

i−

√

2rk), the magnitude of total attractive forces acting

on si is given by | − → FA i | = (k − 2)wa( √ m2 _{+ n}2₋√_2(r

2 2 n m 

m

ij d k s 2r_k 2r_i

n

j s i s

Figure 4.3: Extreme node configuration used to derive the proper wr

wa ratio setting.

values of ri and rk compared to the area dimensions (m and n), we neglect the term

√

2(ri+ rk). Moreover, by approximating (k − 2) ≈ k, we have |

− → FA i | = wak √ m2_{+ n}2_. − → FR i and − → FA

i are two forces that drive sensor si toward the opposite directions. To keep si

in a balanced state without being drawn toward the center or pushed outside the sensing

field, we adopt the equality of the two forces by making |−→FR

i | = | − → FA i |. Consequently, we have wr wa = k √ m2_{+ n}2 ∆ , (4.4)

where m, n, and k are environmental constants, while ∆ (= |dij_th − dij|) varies with the

tolerable overlapping degree of respective sensor pair (related to the sensing ranges and

resultant dij_th). Based on the above derivations, proper choices for the weight constants

can be made by setting wr= k

√

m2_{+ n}2 _{and w}

a= ∆.

Next, we intend to further relax wafrom the dependency on sensing radius by

consid-ering setting wa inversely proportional to the sensor population k as another alternative

to the positive (attractive) weight value. In the case of having a large sensor population (with large k), the weight associated with the positive force should be made small to avoid exerting too much total attractive force on a sensor, and vice versa. To maintain a balanced force interaction, it is reasonable to relate the attractive weight measurement to the actual sensor population (parameter k). As a result, we propose another alternative

arbitrary setting (wa 1,wr 1000) 2 2 , a r b w ' w w k m n wa 1,wr wb k m2 n2 k  73.63% covered 75.66% covered 84.94% covered 95.53% covered 97.91% covered 63.57% covered 50 k 50 k k 50 70 k k 70 k 70

Figure 4.4: Impact of wa, wr (= wb) parameter settings on the coverage ratio of monitored

200 × 200 area (HSR with ea ≥ 1).

to proper weight choices by setting wr = k

√

m2_{+ n}2 _{and w}

a = 1_k.

In addition, since the monitored home environment is usually in a bounded area, we

also incorporate the boundary forces (with weight constant wb) in our EVFA-B

mecha-nism. We use the same value for wr and wb, considering the boundary force is also a kind

of repulsive force. In Fig. 4.4, we perform EVFA-B (with Maxloops = 100, cth = 0.95,

α = 0.9) and experiment on two sensor populations (k = 50 and k = 70) under three

different settings of wr and wa as discussed earlier. As we can see from the figure,

ar-bitrary setting (though wr >> wa) without boundary forces performs poorly, while the

third alternative by making wa inversely proportional to k performs the best with the

highest coverage ratio achieved. Interestingly, by setting wa = _k1 (independent of

sens-ing radius), we actually obtain a better senssens-ing coverage than that by settsens-ing wa = ∆

de-pend on the sensor population (parameter k) and monitoring dimensions (m and n), and can be made independent of sensing radius. This implication greatly simplifies the

de-sign of weight constants when dealing with heterogeneous sensors (having varying sensing

ranges). Therefore we adopt the third alternative by setting wr = k

√

m2_{+ n}2 _{and wa =} 1

k

in our EVFA-B mechanism thereafter.

4.3

Verification of Parameter Settings

We conduct more EVFA-B experiments (Maxloops = 100, cth = 0.95) in this section

to observe the combined impact of dij_th, wa, wrsettings on the attainable coverage ratio. In

Fig. 4.5, two dij_th designs are experimented (where ¯r = 1

k

P_k

i=1ri, representing the average

sensing radius), both with three different wa, wr settings. As depicted in the figure, by

setting wa = 1_k and wr = k

√

m2_{+ n}2_{, we obtain the highest coverage under both d}ij

th

val-ues. Moreover, even higher coverage ratio is attainable if we make the distance threshold

on per node-pair basis by setting dij_th= α(ri+ rj). The results indicate the importance of

proper parameter settings on the distance threshold (dij_th) and weight constants (wa, wr,

wb), further validating our parameter designs proposed in Chapter 4.1 and Chapter 4.2.

2 ij th d r  a 1, r 1000 w w ( ) with 0.9 ij th i j d D r r D  a 1, r 1000 w w (k)   1 2 2 , a r b w w w k m n k   2 2 a r b w w w k m n ' _  2 2 a r b w w w k m n ' _   1 2 2 , a r b w w w k m n k 

Figure 4.5: Performance justification of proper choices for dij_th, wa, wr(wb) values in our

4.4

EVFA-B Algorithm Summary

Table 4.1: Summary of notations used in our EVFA-B

Notation Description

m Length of the monitored field

n Width (breadth) of the monitored field

k Total number of sensor nodes (denoted as s1, s2, · · · , sk with radius

r1, r2, · · · , rk)

(xi, yi) Coordinate (position) of sensor si

dij_th Distance threshold for two arbitrary sensors s_i and s_j (j 6= i)

w_a Tunable measure weight for the attractive force

wr(wb) Tunable measure weight for the repulsive force (boundary force)

− →

Fi Resultant force exerted on sensor si(attractive, repulsive, boundary

forces considered)

M axloops Maximum number of virtual movements performed by each sensor

c_th Desired coverage ratio threshold

Algorithm 1 Enhanced Virtual Forces Algorithm with Boundary Forces (EVFA-B)

1: set loops = 0;

2: set cnow = cinit; // initial coverage ratio

3: while (loops < M axloops) && (cnow < cth) do

4: for each sensor si ∈ {s1, s2, ..., sk} do

5: compute −→F_i=Pk_j6=i,j=1−→F_ij+−→F_ib;

6: end for

7: perform virtual movements; // all sensors virtually move to their next positions

8: update coverage ratio c_now;

9: set loops = loops + 1;

10: end while

Table 4.1 summarizes the notations used in the EVFA-B mechanism, and Algorithm 1 provides the pseudocode for EVFA-B operations. Note that in the end of each loop, every sensor performs virtual movement without physically moving to the new position.

Physical movements are conducted once the EVFA-B process terminates (either cth or

Maxloops has been reached), and this is when our collision-free path planning (CFPP)

Chapter 5

Collision-Free Path Planning

(CFPP)

In practical deployment, a collision-free moving path scheduling is essential, so that mobile sensors can reach their destinations without colliding with each other. However, the scheduling strategy is non-trivial for various collision cases need be systematically classified and handled/resolved in different ways. In this work, we assume the sensor volume is neglected and regarded as a moving point on a 2D plane, while every moving path (performed by a sensor) regarded as a line. Suppose no two moving paths share the same line (i.e., no path lies in the sub-path of another). We identify the collision cases based on the following geometric theorem.

Theorem 1. With respect to the line ax + by + c = 0 on a 2D plane, points Q1(x1, y1)

and Q2(x2, y2) fall in the same side if (ax1+ by1+ c)(ax2+ by2+ c) > 0, in different sides

if (ax1 + by1 + c)(ax2+ by2 + c) < 0, while one or both reside(s) exactly on the line if

(ax1+ by1+ c)(ax2+ by2+ c) = 0.

For an arbitrary sensor si departing from point pi (with coordinate (xi, yi)) to point

ij p ( i, ij) d p p d p p( j_, ij) ' ' ( ) ( ) 0 & & ( ) ( ) 0 j i j i i j i j L p L p L p L p   ' ( ) 0 & & ( ) ( ) 0 i j j i j i L p L p L p  ' ' ( ) 0 & & ( ) ( ) 0 i j j i j i L p L p L p  ' ( ) 0 & & ( ) ( ) 0 j i i j i j L p L p L p  ' ' ( ) 0 & & ( ) ( ) 0 j i i j i j L p L p L p  j ij p p i p ' i p ' j p i p ' i p ' j p j p ' j ij p p i p ' i p j p j p i ij p p ' j p ' i p j p ' i ij p p ' j p i p

Figure 5.1: Possible intersection (collision) cases generated by moving paths of any two

sensors si and sj, where pi (pj) denotes the original position of si (sj) and p

0

i (p

0

j) indicates

the physical movement destination for sensor si (sj).

Li. Similarly, the moving path of another sensor sj is given as Lj. Define pij as the

intersection point of lines Li and Lj, which can be easily obtained by solving the two

line equations. According to Theorem 1, we can now classify five possible intersection

(collision) cases for any two sensors si and sj, as illustrated in Fig. 5.1, where d(pi, pij)

and d(pj, pij) represent the Euclidean distances from pi to pij and from pj to pij. Case

I shows the case in which points pi and p

0

i fall in different sides of line Lj, while points

pj and p

0

j fall in different sides of line Li as well. In Case II, the departure point pj of

sensor sj gets in the way of the moving path of si, while in Case IV, on the contrary,

the departure point pi of sensor si blocks the moving path of sj. Case III draws the

condition in which the destination point p0_j of sensor sj lies on the moving path of si,

while Case V, on the opposite side, displays the condition that destination point p0_i of

sensor si falls on the moving path of sj.

5.1

Path Planning Strategy

Given the five potential collision (intersection) cases caused by any two moving paths,

we establish colliding set Ci, which includes all sensors whose moving paths intersect

with that of si, for each sensor. Instead of performing one-time physical movements, we

at the expenses of increased moving latency. Define orderi as the cardinality of set Ci

(orderi = |Ci|) for sensor si, indicating its moving order. We start from performing

movements for sensors with the least order value. All sensors with the currently least

(smallest) order value are contained in set Mmin order. Intuitively, sensors with order value

of zero can move simultaneously since no other sensors pose potential colliding sources

to them. For any sensor si with non-zero orderi value, potential colliding conditions

(on per node-pair basis) caused by all members in its Ci set should be analyzed and

handled case by case. Specifically, all sensors are divided into moving groups (batches) based on their order values and processed round by round (batch by batch). Sensors in

set Mmin order are evaluated in the same round. The evaluation and processing details

will be provided later in this section. After the evaluations, a subset of Mmin order (or

probably the whole Mmin order set) is determined and all sensors included in the subset

are allowed to move simultaneously in the current round. For sensor si that has been

evaluated and permitted to move, the tf lagi is set true, indicating its moving intention.

Once the physical movement has been successfully performed by sensor si, moving flag

mf lagi is set true and si is removed from the consideration list. All order values for the

remaining sensors (physical movements not performed yet) should be refreshed, and the batched scheduling procedure starts over accordingly.

Now we detail on the evaluation procedures for determining a set of movable sensors in a single round (batch). Based on the idea of batched movements, we regard all sensors with the currently minimum order value as a potential moving batch and include them

in set Mmin order. We then analyze all members in set Mmin order one by one to determine

their moving possibilities. In our design, we start the evaluation from sensor with the

smallest ID, say s1, and identify all possible collision cases caused by members in its

colliding set C1. For any two sensors si and sj with moving orders orderi and orderj,

relationship of orderi and orderj. Suppose si ∈ Mmin order, sj ∈ Ci, and orderi = orderj,

we term the five collision cases as Case S-I, Case S-II, Case S-III, Case S-IV, and

Case S-V, where ’S’ indicates that sensors si and sj are potentially scheduled to move

in the ”same” round due to equal order value. On the other hand, if orderi < orderj

(note that orderi > orderj is not possible since si ∈ Mmin order), we define another five

collision cases as Case D-I, Case D-II, Case D-III, Case D-IV, and Case D-V,

where ’D’ means si and sj are potentially scheduled to move in ”different” rounds due

to their unequal order values. In each potential collision case, on detecting a colliding

possibility, si tries to resolve the collision by adjusting/prolonging the waiting time Tj

or increasing the moving speed Vj of sensor sj. Originally all waiting times are set to

zero, and moving speeds all set at a constant velocity V . If the adjustment (on either waiting time or moving speed) is successful, the colliding possibility is eliminated and

si moves on to evaluate collision cases with other members in Ci. To avoid repeated

adjustments on a single sensor, in our design, each sensor is allowed to be adjusted (either

on waiting time or moving velocity) once. In addition, si itself cannot be adjusted by

other sensors in set Mmin order that are evaluated after it, if si is indeed scheduled to

move in the current round. We keep track of the adjustment possibility for sensor si

by the dirtyi bit, implying adjustable if set f alse and not adjustable if set true. When

si intends to resolve a collision by adjusting another sensor with dirty bit set true, the

adjustment is prohibited and si is not allowed to move in the current round (tf lagi set to

f alse), since the collision remains. Only when all members in Ci with various colliding

possibilities are all resolved can sensor si be included into the movable set and perform

physical movement. Upon receiving the moving instruction from the clusterhead, si waits

for Ti (possibly adjusted) and then moves with speed Vi (possibly adjusted). In our route

scheduling strategy, we try to include as many sensors as possible to move simultaneously in the same round (batch).

For each of the ten collision cases identified, we define corresponding actions (Action D-I, Action D-II, · · · , Action S-I, Action S-II, · · · ) to evaluate respective case and perform necessary adjustments. If colliding possibility remains due to unsuccessful

ad-justment, physical movement by sensor si is not allowed and should be deferred. Thus we

additionally define Action Deferred to perform corresponding operations. Note that

in Case D-I, Case D-III, and Case D-IV, no action is needed since si and sj are

scheduled in different rounds (no collision is likely to happen in the three cases despite intersection exists between the two moving paths). For the rest of seven cases, we describe the evaluation principles exercised by respective action as follows (detailed operations are available in Algorithm 2, Chapter 5.2).

Action D-II In this case, since sj gets in the way of si’s moving path, the clusterhead

instructs sj to slightly adjust its location along line

−−→ pjp

0

j to avoid collision. Assume the

location adjustment is small enough to have no effect on other moving paths.

Action D-V Sensor si is not allowed to move, for its destination point p

0

i will block

the moving path of sj in a later round. In this case, the moving order of si should be set

to be larger than that of sj (orderi = orderj+1) to postpone si’s physical movement after

sj. In addition, a f ix orderi flag should be set true, indicating no updates on orderi will

be performed in later rounds to ensure the delayed movement after sj, and then Action

Deferred is invoked for si.

Action S-I Define the traveling time from pi to the intersection point pij as tpi→pij

(obtained from available d(pi, pij) and Vi), the clusterhead evaluates if Ti + tpi→pij =

Tj + tpj→pij, where Ti and Tj are the waiting times of si and sj as defined earlier. If

equality holds, a collision at the intersection is expected, and the waiting time Tj of sj

should be increased by a small amount of ∆t to avoid the collision. However, in case sj

has already been processed with dirtyj set true, the adjustment is prohibited and si is

(orderi = orderi+ 1) and Action Deferred is invoked for si.

Action S-II If si reaches the intersection point pij no later than sj’s departure time,

the clusterhead should instruct sj to slightly adjust its location along line

−−→ pjp

0

j to avoid

collision.

Action S-III If sj reaches the intersection point pij no later than si, the destination

point p0_j of sj will block the moving path of si. In this case, the clusterhead should instruct

sj to increase its waiting time Tj by setting Tj = Ti + (tpi→pij − tpj→pij) + ∆t to ensure

the delayed arrival of sj at pij (p

0

j). If the adjustment of Tj is not successful due to a true

flag of dirtyj, then sj is not allowed to move in the current round. Consequently, moving

order of sj is increased (orderj = orderj + 1) and Action Deferred is invoked for sj.

Action S-IV If sj reaches the intersection point pij no later than si’s departure time,

the clusterhead should increase the waiting time of sj by setting Tj = Ti− tpj→pij+ ∆t. In

case the adjustment is not allowed due to a true value of dirtyj, the clusterhead instructs

si to slightly adjust its location along line

−→ pip

0

i to avoid collision.

Action S-V If si reaches the intersection point pij no later than sj, the destination

point p0_i of si will block the moving path of sj. In this case, the clusterhead should

instruct sj to increase its moving speed Vj by setting Vj = _d(pi,pVi_ij·d(p_)+Vj,p_i(Tij_i)_−Tj) + ∆v, where

∆v is a small amount of speed increment to ensure sj’s earlier arrival at pij (p

0

i) than si.

However, if the adjusted Vj is larger than the maximum possible moving speed Vmax or

the adjustment of Vj is prohibited due to a true value of dirtyj, then si is not allowed

to move in the current round. Moving order of si is increased (orderi = orderi+ 1) and

Action Deferred is invoked for si.

Action Deferred Since si (sj) is not allowed to move in the current round, tf lagi

(tf lagj) is set f alse. In addition, the clusterhead should confirm if this not-moving

decision leads to moving path blocking of any sensor in Mmin order set that is already

1 p ' 2 p 3 p ' 1 p 2( 23) p p ' 3( 34) p p 4( 24) p p ' 4 p i s 1 s 2 s 3 s 4 s i C 2(S I), (S III), (D I)3 4 s  s  s 

1(S I), (S IV), (D II)3 4

s  s  s 

1(S V), (S II), (D V)2 4

s  s  s 

1(D I), (D IV), (D III)2 3

s  s  s  ˜˜˜ i order 3 3 3 5 1 2 Moving set = { , }s s min_ order 1 2 3 M = { , , }s s s _ i

fix order dirtyi tflagimflagi

0 0 0 0 0 0 0 0 0 0 i s 1 s 2 s 3 s 4 s i C 4(D V) s  3(D III) s  ˜˜˜ i order 3 6 i

dirty tflagi mflagi

1 1 0 0 0 0 1 1 1 12 p 1 0 1 1 _ i fix order ' 2 p 3 p ' 1 p ' 3( 13 34) p p p 4( ) p adjusted ' 4 p 4( ) p original 0 0 0

(a) Before moving

(b) After moving

Figure 5.2: Every sensor si in the potential moving set Mmin order should be analyzed

by identifying its intersection (collision) relationship with each member in Ci, in which

intersection cases D-II, D-IV, S-I, S-II, S-III, S-IV, and S-V require further

considera-tion/processing, before including si into the moving set (allowed to move in the current

round).

location adjustment to resolve the blocking.

Fig. 5.2 illustrates a snapshot of the CFPP operations. Note that s4 has more

inter-sections with other sensors, which are not shown in the figure (omitted for brevity). In

the current round, potential moving set Mmin order includes s1, s2, and s3, all having the

currently smallest order value of 3. For s1, colliding conditions caused by all members in

C1 are analyzed and handled case by case. In this example, since s1 and s2 are evaluated

to reach intersection p12 simultaneously, the clusterhead adjusts the waiting time of s2 by

setting T2 = T2+ ∆t to resolve the collision. Next, since s3 is found to reach intersection

p13 earlier than s1, blocking s1’s moving path, the clusterhead instructs s3 to increase its

waiting time by setting T3 = T1 + (tp1→p13− tp3→p13) + ∆t. As to s4 (scheduled to move

in a later round), no action is required since no collision is likely to happen between s1

apply to s2. In our example, s2 has no colliding possibilities with s1 and s3. However,

since the departure location p4 of s4 blocks s2’s moving path, the clusterhead instructs s4

to slightly move from p4 (original) to p4 (adjusted), as shown in Fig. 5.2 (b). As a result,

s2 is also included into the moving set. For s3, in our example, both s1 and s2 do not

pose colliding sources to s3. Unfortunately, since the destination point p

0

3 of s3 will block

the moving path of s4 in a future round, s3 is not allowed to move before s4 (not included

into the moving set), and order3 should be updated to 6 (order4+ 1) with f ix order3 set

true. After the evaluations, sensors included in the moving set (i.e., s1 and s2) perform

physical movements simultaneously, and order4 and set C4 are updated accordingly.

5.2

CFPP Algorithm Summary

Table 5.1: Summary of notations used in the CFPP algorithm

Notation Description

Ci Set of potential colliding sensors against si

orderi Moving order of si, where orderi= |Ci|

f ix orderi Indicates the order value of si is henceforth fixed

dirtyi Indicates whether si has been processed in the current round

tf lagi Indicates whether si is allowed to move in the current round

mf lagi Indicates whether si has moved from pi to p

0

i

Mmin order Set of sensors with the minimum order value in the current round

Table 5.1 summarizes the notations used in CFPP, and Algorithm 2 provides the pseudocode for CFPP operations. In addition, a running example illustrating the CFPP route scheduling procedures is available in Fig. 5.3. Note that in Round 3 of this example,

s9 is excluded from the moving set due to a unsuccessful adjustment of s11’s waiting time

(since T11 has been adjusted by the clusterhead to resolve collision with s7 and can only

be adjusted once according to the scheduling principles adopted by CFPP). After the

clusterhead decides that s9 is not allowed to move in the current round, s9 no longer

i s 1 s 2 s 3 s 4 s 5 s 6 s 7 s 8 s 9 s 10 s 11 s i C 2,4, ,5 6 s s s s 1,10 s s 1,5 s s 1,4 s s 9,11 s s 9 s 7, ,8 11 s s s 2 s 7,9 s s 4 2 0 2 2 2 1 3 1 11' 10' 9' 8' 7' 6' 5' 4' 3' 2' 1' 1 2 3 4 5 6 9 7 8 10 11 3 8 10 Moving set = {s , s , s } 11' 10' 9' 8' 7' 6' 5' 4' 3' 2' 1' 1 2 4 5 6 9 7 11 2 Moving set = {s } 11' 10' 9' 8' 7' 6' 5' 4' 3' 2' 1' 1 4 5 6 9 7 11 4 5 7 11 Moving set = {s , s , s , s } 11' 10' 9' 8' 7' 6' 5' 4' 3' 2' 1' 1 6 7 1 9 Moving set = {s , s } 11' 10' 9' 8' 7' 6' 5' 4' 3' 2' 1' 6 2 1 1 s i dirty 0 0 0 0 0 0 0 6 Moving set = {s } 11' 10' 9' 8' 7' 6' 5' 4' 3' 2' 1' min_ order 3 6 8 10 M = { , , , s s s s } Mmin_ order= { }s2 min_ order 4 5 7 9 11 M = { , , , , s s s s s } 9 min_ order 1 9 M = { ,s s} min_ order 6 M = { }s i mflag 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 i tflag 0 0 0 0 0 0 0 0 1 1 1 i

order fix order_ i

0 0 0 0 0 0 0 0 0 0 0 i s 1 s 2 s 3 s 4 s 5 s 6 s 7 s 8 s 9 s 10 s 11 s i

dirty tflagimflagi i

order fix order_ i

0 0 0 0 0 0 0 2,4, ,5 6 s s s s 1 s 1,5 s s 1,4 s s 9,11 s s 7,11 s s 7,9 s s 1 s i C 4 2 2 5 2 1 2 2 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 1 1 i s 1 s 2 s 3 s 4 s 5 s 6 s 7 s 8 s 9 s 10 s 11 s i

dirty tflagi mflagi i

order fix order_ i

0 0 0 0 0 0 1 4, ,5 6 s s s 1,5 s s 1,4 s s 9,11 s s 7,9 s s 7,11 s s 1 s i C 2 2 2 2 3 2 5 0 0 0 0 0 0 0 1 1 1 1 0 1 1 1 1 1 0 1 1 1 0 1 0 0 i s 1 s 2 s 3 s 4 s 5 s 6 s 7 s 8 s 9 s 10 s 11 s i

dirty tflagimflagi i

order fix order_ i

0 0 1 6 s 1 s i C 0 1 1 1 1 1 1 1 0 0 1 1 0 0 1 1 1 5 1 0 i s 1 s 2 s 3 s 4 s 5 s 6 s 7 s 8 s 9 s 10 s 11 s i

dirty tflagimflagi i

order fix order_ i

1 i C 1 1 1 1 1 1 1 0 0 1 0 5 i s 1 s 2 s 3 s 4 s 5 s 6 s 7 s 8 s 9 s 10 s 11 s i

dirty tflagimflagi i

order fix order_ i i C 1 1 1 1 1 1 1 1 1 1 1 1 1

Figure 5.3: Example illustrating the operations of our proposed CFPP algorithm for sensor physical movements.

by the clusterhead.

In the CFPP strategy, we propose batched movements to successfully resolve moving collisions between sensors at the cost of global deployment latency. While most existing self-deployment works do not handle this collision problem, our proposed CFPP strategy is essential in practical deployment, and we believe the disadvantage of increased deployment time can be effectively reduced by the local recovery capability provided by our SSOA mechanism (detailed in Chapter 6), which leads to infrequent global redeployments.