NETWORK ISSUES FOR 3D WIRELESS SENSOR NETWORKS
By
Fernando J. Cintr´n
o

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Computer Science — Doctor of Philosophy
2013

ABSTRACT
NETWORK ISSUES FOR 3D WIRELESS SENSOR NETWORKS
By
Fernando J. Cintr´n
o
Wireless sensor networks (WSN) give the opportunity to monitor the environment by
performing sensing tasks in places that are difficult to reach or dangerous for humans. Nevertheless, topographical characteristics of such places and the sensor node’s limitations introduce new issues in WSN performance. Additionally, in scenarios where sensors are moving
or in rugged terrain, there is a high chance for them to be out of communication range,
causing network connectivity problems. Hence, solutions have to take into consideration the
aspect of the topography consisting of its three dimensional characteristics, namely, type of
terrain, terrain unevenness, and obstacles.
This dissertation discusses several research topics addressing issues relevant to WSN
connectivity and area coverage problems. First, changes in sensor communication range
are studied by varying sensors’ heights relative to the surface. A novel communication
technique that relies on the jumping capabilities of sensors is proposed. While the jumping
sensor robots are airborne, the change in elevation enhances their ability for a short time
to successfully communicate with other sensors that are out of communication range at the
ground level. Field experiments were conducted and results show a considerable improvement
in wireless communication ranges.
Second, the impact of network connectivity and area coverage in a jumping sensor network is further studied. A Hopping Sensor Network Model is defined to increase sensing
area coverage along with the enhancement of network connectivity. A Hopping Sensor Rout-

ing Protocol is designed from the model that balances the energy consumption on active
jumping sensor nodes. Results from simulations show the increase in area coverage obtained
from jumping sensor networks, and the effectiveness of the routing protocol to optimize
communication paths while balancing energy depletion in the network.
Third, a distributed wireless sensor network organization to establish a functional network, without requiring initial topology information, is presented. Two decentralized algorithms that use the jumping capabilities of sensors are designed for the discovery of isolated
sensors. Simulation results show the success of the algorithms to enhance base station
reachability. Additionally, cluster to cluster (C-to-C) packet forwarding schemes relying on
boundary jumping sensor gateways are defined and analyzed, showing remarkable savings in
network energy consumption.
Fourth, in order to have a functional network, it is important to address connectivity
issues in an application oriented manner. This work presents an efficient node redeploymentdecision process to produce a functionally heterogeneous (jumping and non-jumping sensors)
WSN with a performance guarantee. Network performance is defined as a network fitness
formula considering the network Quality of Connectivity (QoC). Decision making algorithms
for node relocation and topology defragmentation are presented, along with a discussion of
their performance.
Fifth, a multi-step procedure to produce a direction oriented jumping sensor network is
presented. A jumping sensor robot approach is introduced for collecting and processing signal
strength data into relative geographical orientation information. A directional-orientation
decision algorithm is defined to process the orientation information. Furthermore, an error
identification and correction procedure is established. This has proven to accurately fix the
true orientation of the nodes by using only a pair of location aware beacon nodes.

Copyright by
´
FERNANDO J. CINTRON
2013

To my family.

v

ACKNOWLEDGMENTS

Pursuing a Ph.D. degree requires a high level of commitment and perseverance. Completing
this dissertation has been possible thanks to the guidance and support of many people. I
would like to thank everyone who has contributed directly or indirectly toward this accomplishment, marking one big step in my professional development.
I would like to express my sincere gratitude to Dr. Matt W. Mutka for his advices,
financial support, and incessant patience. As a professor, he introduced me to a new area
of research in wireless sensor networks. His experience and guidance helped me to become a
better researcher and to successfully complete this dissertation.
I would like to thank Dr. Li Xiao, Dr. Abdol-Hossein Esfahanian, and Dr. Ning Xi
for their valuable comments in this dissertation and for taking the time to serve on my
Ph.D. guidance committee. I am also thankful to the Department of Computer Science and
Engineering and the Graduate School of Michigan State University for providing me teaching
assistantships, and awarding me a fellowship to support this research.
I would like to express my thanks and deepest appreciation to Dr. Percy Pierre, for his
generosity, support, and his role with the Sloan program, giving me the opportunity to visit
and meet so many great people at Michigan State University before making my decision
to study in this institution. Without any doubt, the Alfred P. Sloan foundation program
played an important role from the beginning, deciding on an academic institution to pursue
the Ph.D. degree; while on board, giving me the support and encouraging me to stay focused;
to the end, becoming part of such a great network of people with diverse backgrounds.
I would like to thank my colleagues in the ELANS laboratory. We spent time having
discussions that resulted in great research collaborations.
vi

Last but not least, I would like to thank my family, who encouraged me to continue on
this journey and to always keep growing as a person. I would like to express my deepest
thanks to my wife Maria Teresa, for her unconditional support, patience and love.

vii

TABLE OF CONTENTS

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xii

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii

LIST OF ALGORITHMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi

Chapter 1 INTRODUCTION . . . . . . . . . . . . . .
1.1 WSN Challenges . . . . . . . . . . . . . . . . . . .
1.1.1 Connectivity and Area coverage . . . . . . .
1.1.2 Wireless Communication Challenges . . . .
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . .
1.2.1 Communication Range Increase in WSN . .
1.2.2 Jumping Sensor Networks and Connectivity
1.2.3 Network Fitness . . . . . . . . . . . . . . . .
1.2.4 Relative Orientation . . . . . . . . . . . . .
1.3 Structure of the Content . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

1
2
2
3
4
4
5
6
7
8

Chapter 2 BACKGROUND SURVEY . . . . . . . . .
2.1 Wireless Sensor Design . . . . . . . . . . . . . . . . .
2.2 Network Energy Management . . . . . . . . . . . . .
2.3 Connectivity and Area Coverage . . . . . . . . . . . .
2.4 Topology Control . . . . . . . . . . . . . . . . . . . .
2.4.1 Degree-based Mobility Model . . . . . . . . .
2.4.2 Transmit Power Adjustment . . . . . . . . . .
2.4.3 Local Minimum Spanning Tree (LMST) . . .
2.5 Mobile Robots . . . . . . . . . . . . . . . . . . . . . .
2.6 State of the Art Sensors . . . . . . . . . . . . . . . .
2.7 Wireless Node Localization and Relative Orientation

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

9
9
11
12
13
14
16
18
21
22
23

Chapter 3 WIRELESS COMMUNICATION RANGE INCREASE
3.1 Preliminaries and Motivations . . . . . . . . . . . . . . . . . . . . . .
3.1.1 Chapter Organization . . . . . . . . . . . . . . . . . . . . . . .
3.2 Received Signal Strength . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.1 Impact of communication distance on RSS . . . . . . . . . . .
3.2.2 Impact of sensor node elevation on RSS . . . . . . . . . . . . .
3.2.3 Impact of transmission power on RSS . . . . . . . . . . . . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

26
26
27
27
27
30
32

viii

3.3

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

32
33
35
36
37
39
40
42
44

Chapter 4 HOPPING SENSOR NETWORK . . . . . . . .
4.1 Preliminaries and Motivations . . . . . . . . . . . . . . . .
4.1.1 Chapter Organization . . . . . . . . . . . . . . . . .
4.2 Network Connectivity and Area Coverage . . . . . . . . . .
4.2.1 Connectivity and Area Coverage Evaluation . . . .
4.3 Hopping Sensor Network Model . . . . . . . . . . . . . . .
4.3.1 Vertical Displacement . . . . . . . . . . . . . . . .
4.3.2 Horizontal Displacement . . . . . . . . . . . . . . .
4.3.3 Hop Energy Vs. Transmission Energy Consumption
4.3.4 Application Parameter . . . . . . . . . . . . . . . .
4.4 Hopping Sensor Routing Protocol . . . . . . . . . . . . . .
4.4.1 HSRP Evaluation . . . . . . . . . . . . . . . . . . .
4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

45
45
45
46
47
52
52
52
53
54
54
57
60

3.4

3.5

Airborne Communication . . . . . . . . .
3.3.1 Time of Flight . . . . . . . . . . .
3.3.2 Jumping Algorithm . . . . . . . .
3.3.2.1 Jump Data Partitioning
3.3.3 Airborne Communication Process
Experimental Evaluation . . . . . . . . .
3.4.1 Airborne RSS . . . . . . . . . . .
3.4.2 Airborne Throughput . . . . . . .
Summary . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

Chapter 5 WIRELESS SENSOR NETWORK CONNECTIVITY INCREASE
WITH JUMPING SENSORS . . . . . . . . . . . . . . . . . . . . . 61
5.1 Preliminaries and Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.1.1 Chapter Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.2 Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.3 Boundary Nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.4 Discovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.4.1 Boundary Node Aggressive Neighbor Discovery . . . . . . . . . . . . 66
5.4.1.1 Pre-hop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.4.1.2 Hopping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.4.1.3 Post-hopping . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.4.2 Boundary Node Smart Neighbor Discovery . . . . . . . . . . . . . . . 69
5.5 Simulation Results and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.5.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.5.2 Discovery Algorithm Performance . . . . . . . . . . . . . . . . . . . . 72
5.5.3 Network Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

ix

Chapter 6 WIRELESS SENSOR NETWORKS AND TOPOLOGY REDEPLOYMENT . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.1 Preliminaries and Motivations . . . . . . . . . . . . . . . . . . . . . . . . . .
6.1.1 Chapter Organization . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.2.1 Connectivity and Sensor Clusters . . . . . . . . . . . . . . . . . . . .
6.3 Network Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3.1 Jumping Airborne Communication . . . . . . . . . . . . . . . . . . .
6.3.2 Gateway Airborne Communication . . . . . . . . . . . . . . . . . . .
6.4 Quality of Connectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.4.1 QoC Enhancement . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.5 Cluster Merging Decision Algorithms . . . . . . . . . . . . . . . . . . . . . .
6.5.1 Centralized Approaches . . . . . . . . . . . . . . . . . . . . . . . . .
6.5.1.1 BMN and BMN-BS . . . . . . . . . . . . . . . . . . . . . .
6.5.1.2 LIMN and LIMN-BS . . . . . . . . . . . . . . . . . . . . . .
6.5.1.3 RMN and RMN-BS . . . . . . . . . . . . . . . . . . . . . .
6.5.2 Distributed Approaches . . . . . . . . . . . . . . . . . . . . . . . . .
6.5.2.1 DBMN, DBMN-BS, and S-DBMN . . . . . . . . . . . . . .
6.5.2.2 DLIMN . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.5.2.3 DRMN . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.6 Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.6.1 Simulation Environment . . . . . . . . . . . . . . . . . . . . . . . . .
6.6.2 Evaluation and Results Analysis . . . . . . . . . . . . . . . . . . . . .
6.6.2.1 BMN, BMN-BS, and DBMN-BS . . . . . . . . . . . . . . .
6.6.2.2 LIMN, LIMN-BS, and DLIMN . . . . . . . . . . . . . . . .
6.6.2.3 S-DBMN and DBMN . . . . . . . . . . . . . . . . . . . . .
6.6.2.4 RMN, DRMN, and RMN-BS . . . . . . . . . . . . . . . . .
6.6.3 Performance Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

80
80
82
82
83
84
84
89
91
94
98
100
100
101
101
101
102
102
103
103
103
104
105
105
106
107
107
114

Chapter 7 LEVERAGING HEIGHT IN JUMPING SENSORS
TO OBTAIN RELATIVE ORIENTATION . . . . . . .
7.1 Preliminaries and Motivations . . . . . . . . . . . . . . . . . . . .
7.1.1 Organization . . . . . . . . . . . . . . . . . . . . . . . . .
7.2 Sensor Orientation . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.1 Oriented Direction . . . . . . . . . . . . . . . . . . . . . .
7.2.2 Experimentation . . . . . . . . . . . . . . . . . . . . . . .
7.2.3 Multiple Minima . . . . . . . . . . . . . . . . . . . . . . .
7.3 Orientation Algorithm . . . . . . . . . . . . . . . . . . . . . . . .
7.3.1 OD Selection . . . . . . . . . . . . . . . . . . . . . . . . .
7.4 Mirror Free Topology . . . . . . . . . . . . . . . . . . . . . . . . .
7.4.1 OD Paths and Mirrored Topology Identification . . . . . .
7.4.2 Special Cases . . . . . . . . . . . . . . . . . . . . . . . . .
7.5 Additional Proofs for Special Cases . . . . . . . . . . . . . . . . .
7.5.1 Proof for Odd Crossings . . . . . . . . . . . . . . . . . . .

115
115
118
119
120
121
124
126
128
132
133
135
140
141

x

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

7.5.1.1 OD-paths with cx > 1
Proof for Even Crossings . . . .
7.5.2.1 OD-path with cx = 2
7.5.2.2 OD-paths with cx > 2
Summary . . . . . . . . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

141
144
144
149
151

Chapter 8 SUMMARY AND FUTURE WORK . . . . .
8.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . .
8.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . .
8.2.1 Heterogeneous Wireless Sensor Network Test-bed
8.2.2 Neighbor Discovery . . . . . . . . . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

152
152
155
155
156

7.5.2

7.6

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

xi

LIST OF TABLES

Table 2.1

Wireless Measurements Systems Specifications . . . . . . . . . . . .

23

Table 5.1

Discovery Algorithm Performance . . . . . . . . . . . . . . . . . . .

72

Table 6.1

Cluster’s Sizes and Distribution by Connectivity Level of Network
Topology in Fig. 6.3 . . . . . . . . . . . . . . . . . . . . . . . . . . .

95

Table 6.2

Algorithms’ Characteristic Overview . . . . . . . . . . . . . . . . . . 100

Table 7.1

Antenna True Yaw Orientation for IRIS mote . . . . . . . . . . . . . 122

xii

LIST OF FIGURES

Figure 1.1

Active Airborne Jumping Sensor, Zhao et al. (2009). For interpretation of the references to color in this and all other figures, the reader
is referred to the electronic version of this dissertation. . . . . . . . .

5

Figure 3.1

RSSI on Concrete . . . . . . . . . . . . . . . . . . . . . . . . . . . .

28

Figure 3.2

RSSI on Grass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

29

Figure 3.3

Communication Range with Varying Heights . . . . . . . . . . . . .

31

Figure 3.4

Height Needed Above Threshold Height VS. Communication Time .

34

Figure 3.5

Time of Flight for 1 Meter Jump Height . . . . . . . . . . . . . . . .

35

Figure 3.6

Jump Data Partitioning Flow Diagram . . . . . . . . . . . . . . . .

37

Figure 3.7

Airborne Communication Process . . . . . . . . . . . . . . . . . . .

38

Figure 3.8

RSS Airborne Transmitted Packets (20msec rate)

. . . . . . . . . .

40

Figure 3.9

RSS Airborne Transmitted Packets (50msec rate)

. . . . . . . . . .

41

Figure 3.10

RSS Airborne Transmitted Packets at 25 meters (50msec rate) . . .

41

Figure 3.11

Throughput Airborne Data Communication (10msec) . . . . . . . .

42

Figure 3.12

Throughput Airborne Data Communication (20ms) . . . . . . . . .

43

Figure 4.1

% Connectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

48

Figure 4.2

% Area Covered . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

49

Figure 4.3

Topology Area Coverage for 30cm Jump Height

. . . . . . . . . . .

50

Figure 4.4

Topology Area Coverage for 40cm Jump Height

. . . . . . . . . . .

51

xiii

Figure 4.5

Hopping Sensor Network Model . . . . . . . . . . . . . . . . . . . .

52

Figure 4.6

Hopping Sensor Energy Consumption . . . . . . . . . . . . . . . . .

53

Figure 4.7

Network Energy Consumption . . . . . . . . . . . . . . . . . . . . .

58

Figure 5.1

WSN Topology at ground level (31% connectivity) . . . . . . . . . .

62

Figure 5.2

WSN Topology with gateway jumping sensors (100% connectivity) .

62

Figure 5.3

Discovered Area Gain of BNs Subset Selection for Discovery

. . . .

69

Figure 5.4

Remaining Network Energy for Different Communication Loads . . .

74

Figure 5.5

Remaining Network Connectivity for Different Communication Loads

74

Figure 5.6

Mean Packet Delay Time for Different Communication Loads . . . .

75

Figure 5.7

Network Energy Consumption for 20% Nodes Generating Packets

.

76

Figure 5.8

Network Energy Consumption for 10% Nodes Generating Packets

.

77

Figure 5.9

Network Energy Consumption for 5% Nodes Generating Packets . .

77

Figure 5.10

Network Energy Consumption for 2% Nodes Generating Packets] . .

78

Figure 5.11

Network Energy Consumption for 1% Nodes Generating Packets . .

78

Figure 6.1

Jumping Robot Jump Cycle and Time of Flight for 73.6 cm Jump
Height . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

86

Figure 6.2

Single base station clustered topology depicting connectivity levels .

92

Figure 6.3

∗
Network Topology Example for QoC2 Contribution Evaluation . . .

96

Figure 7.1

Orientation with GPS . . . . . . . . . . . . . . . . . . . . . . . . . . 116

Figure 7.2

Relative Orientation with Jumping Sensors . . . . . . . . . . . . . . 117

Figure 7.3

Yaw, Pitch, and Roll . . . . . . . . . . . . . . . . . . . . . . . . . . 119

xiv

Figure 7.4

Node A’s Antenna Yaw RSSI (w/ Roll −45◦ ) Measured at Node B

123

Figure 7.5

Node A’s Antenna Yaw RSSI (w/ Roll −90◦ ) Measured at Node B

124

Figure 7.6

ODs Choice Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 127

Figure 7.7

OD-paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

Figure 7.8

OD-path with One Edge Crossing . . . . . . . . . . . . . . . . . . . 136

Figure 7.9

OD-path with 3 Edges Crossing . . . . . . . . . . . . . . . . . . . . 141

Figure 7.10

True OD-path with 2 Edges Crossing . . . . . . . . . . . . . . . . . 144

Figure 7.11

Mirrored OD-path with 2 Edges Crossing . . . . . . . . . . . . . . . 147

Figure 7.12

True OD-path with 4 Edges Crossing . . . . . . . . . . . . . . . . . 149

xv

LIST OF ALGORITHMS

4.1

Hopping Sensor Routing Protocol . . . . . . . . . . . . . . . . . . . . . . . .

56

7.1

Estimate OD from a set with n directions . . . . . . . . . . . . . . . . . . . 125

7.2

Neighbor OD Selector

7.2

Neighbor OD Selector (cont’d) . . . . . . . . . . . . . . . . . . . . . . . . . . 130

7.3

OD Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

xvi

Chapter 1
INTRODUCTION
Wireless Sensor Networks (WSNs) serve a broad range of applications, for e.g., environmental monitoring that includes sensing of temperature, sound, movement, earth tectonic plate
vibrations, pressure, and toxic condition among others. WSNs offer the ability to monitor
large scale areas where wired networks may be unviable or infeasible to establish. Furthermore, they present the opportunity to monitor the environment by performing sensing tasks
in places that are difficult to reach or dangerous for humans. Nevertheless, topographical characteristics of such places and the sensor node’s limitations introduce new issues in
WSN performance. Additionally, in scenarios where sensors are moving or in a rugged terrain, there is a high chance for them to be out of communication range, causing network
connectivity problems.
This dissertation presents solutions to improve WSN connectivity and network deployment efficiency. It considers sensor nodes’ limitations in combination with three-dimensional
characteristics of the topography, namely, type of terrain, terrain unevenness, and obstacles.
With this goal present, novel techniques optimized for sensor nodes with limited capabilities are presented. In this chapter, challenges of today’s WSN are introduced, along with
motivations for connectivity efficient and enduring networks.

1

1.1

WSN Challenges

Although WSNs are self-organizing and provide an ease of deployment, the sensor nodes have
inherent shortcomings in form of limited battery, processing capabilities, and communication
range.

1.1.1

Connectivity and Area coverage

Sensor nodes used in the network have the task to monitor their surrounding environment.
The area that they cover depends on two factors: (1) the sensing range of the sensors, and
(2) the connectivity of the sensors. The sensing range is a property dependent on the type of
sensing, which is defined by the application’s demands. In other words, it can be considered
as an application-logistic problem. For the scope of this dissertation, connectivity is defined
as follows: a single node is said to have connectivity if it has a direct communication path
to the base station (one communication hop) or an indirect communication path to the
base station via its neighbor nodes (multi-hop routing communication). In contrast to the
sensing range, the connectivity presents a dynamic problem for mobile sensor networks and
topography scenarios requiring less precise deployment techniques.
Network Connectivity is an important factor because when a sensor is not able to reach
the base station (i.e., it does not have connectivity), the data gathered by the sensor cannot
be transferred to the base station. Depending on the application’s tolerance to communication delay and unpredictability of events for the disconnected sensor to regain connectivity,
the collected data can become outdated. Therefore, connectivity is a crucial aspect for the
network to operate.

2

1.1.2

Wireless Communication Challenges

Wireless communication is affected by many factors: environment, distance, antenna type
and height, etc. These factors can cause signal fading, resulting in reduced communication
range, which can negatively affect the transmission of network packets.
The distance between sensor nodes is one of the critical factors that dictates the received
signal strength (RSS). The farther away the sensors are from each other, the weaker the
received signal strength will be, and the tendency for packet loss increases. Ideally, all sensors would be placed within the wireless communication range of each neighbor. However,
as previously stated, this becomes infeasible in scenarios where sensors are moving, or in
difficult-to-reach terrains where more aggressive deployment techniques (e.g., air-borne dispersion) are required. In such cases, the deployment of sensors becomes unevenly spread
over the target field. Furthermore, as more aggressive the deployment technique becomes,
the impact on the design of the sensor, e.g., usage of short length antennas, creates a further
reduction of communication range.
Beside the limitation of sensor nodes, sensors’ specification often advertise communication characteristics under ideal conditions, not reflecting the adversity of the conditions
presented to today’s use of sensor networks. For instance, sensor networks are deployed in
rough terrains where ground unevenness and obstacles greatly interfere with wireless communication. Moreover, sensor nodes tend to be small in size and height, making heavy foliage
a potential line of sight obstacle in wireless communication.
Given the nature of sensor network deployment and the constraining factors, it is almost
certain for a WSN to experience connectivity related problems.

3

1.2
1.2.1

Motivation
Communication Range Increase in WSN

It is known that a change in antenna elevation affects wireless communication range. The
communication range of a sensor node increases with its elevation with respect to ground
level, within practical bounds.1 A temporary change in sensor elevation is seen in jumping
sensor robots. While the sensor is jumping, the change in elevation enhances, for a short
time, its ability to communicate with other sensors that were not reachable at ground level.
The jumping sensor (also known as hopping sensor for its movement resemblance to
grasshoppers) concept is inspired by natural wild life behaviors to perform sensing tasks in
environments where it becomes difficult for other types of sensor robots, e.g., wheeled sensors.
The jumping motion is not as accurate as the wheeled motion, but is adequate for rough
terrains where wheeled sensors cannot perform as desired. Figure 1.1 shows three jumping
sensors2 at ground level, while one jumping sensor is in the act of jumping (airborne). The
picture was enhanced to display a close view of the jumping sensor. Dotted red lines highlight
the airborne sensor and the embedded magnified view.
Jumping sensors provide an alternative to address common WSN’s problems. The use
of jumping sensors requires the study of three-dimensional aspect of network topologies.
However, WSN are usually modeled in a two-dimensional space, leaving out the effects of
topology characteristics issues, and dynamic changes in sensor nodes’ height with respect to
others.
1 Communication impact from sensor node elevation is presented in Chapter 2.
2 Jumping sensor prototypes presented in Fig. 1.1 are developed by Zhao et al. (2009).

4

Figure 1.1: Active Airborne Jumping Sensor, Zhao et al. (2009). For interpretation of the
references to color in this and all other figures, the reader is referred to the electronic version
of this dissertation.

1.2.2

Jumping Sensor Networks and Connectivity

When wireless sensor nodes are randomly deployed in a field, there may be areas that lack
sensor nodes. These areas tend to create communication holes, and in the worst case, break
the topology into disconnected clusters of nodes, i.e., disconnected groups of interconnected
nodes.
The problem of communication holes in WSN can be addressed using different approaches.
An intuitive method is to increase the node density, i.e., amount of nodes per unit area. This
approach provides a feasible solution for tasks, which can be performed with relatively cheap
technology. However, leading edge technology demanding applications are expensive, given
that the price per node tends to be higher. On the other hand, node relocation provides the
5

opportunity to use those nodes that are redundant in some areas to cover communication
holes. However, relocation is not possible in difficult to navigate terrains such as river stream,
lakes, or harmful areas. As an alternative to reach farther, signal transmission power can be
amplified. Nevertheless, many topology control protocols for WSN use maximum transmission
power of the nodes at their initial phases, and still experience connectivity problems, due to
the presence of communication holes.
Complete and persistent network connectivity within a random wireless sensor network
topology is difficult to guarantee. Instead, jumping sensors can be used to temporarily
increase connectivity in wireless sensor networks. However, a topology of jumping sensor
nodes configured to enhance network connectivity by means of jumping can significantly
reduce the network life span, if not well managed. With the help of selected jumping sensors
on disconnected clusters, enhanced connectivity can be achieved without fully compromising
network energy.

1.2.3

Network Fitness

In order to have a functional network it is important to address connectivity issues in an application oriented manner. However, efficient node redeployment is a key aspect for enduring
WSN.
WSN applications require a certain level of network performance guarantee to properly
function. Some applications can withstand an intermittent connectivity (delay tolerant),
meaning that communications between node pairs do not require long lasting wireless communication links. Instead, the communication occurs for short periods of time during the
network lifespan. In such cases, network cost for redeployment could far exceed the gains
provided by jumping sensors. It is clear that connectivity alone is not representative of
6

potential network performance. Therefore, an evaluation metric representing the fitness of
a WSN in terms of the quality of its connectivity is required.
Part of this research is dedicated to providing a Quality of Connectivity (QoC) performance metric for WSN. Furthermore, the contribution of having such metric is twofold.
First, it provides a non-heuristic performance metric for the network topology, allowing almost instant topology evaluation after deployment. Second, it serves as an optimization
metric for node redeployment.

1.2.4

Relative Orientation

Relative orientation is used by humans on a daily basis, as a set of directions to reach from
one place to another, without knowledge of detailed location information. Many times, the
orientation is obtained in the simplest form as place reference, guiding ourselves in a reference
relaying manner, e.g., “Let’s meet in the coffee shop north of the campus library.”. While
the orientation concept seems relatively simple in nature, it is still a challenge to purely
obtain orientation information without relying on precise localization systems, sometimes
impractical for a highly resource constrained (CPU, power, form-factor, etc.) WSN.
It is clear that as newer generations of jumping sensors become available, opportunities
open up to solve many existing Wireless Sensor Network (WSN) problems, which include a
node’s relative orientation. We define a node’s relative orientation as the ability of a node to
identify the direction of a target (e.g., neighbor node) from its current location relative to a
reference direction. Obtaining a reference direction, such as the magnetic north, has become
simple with the usage of inexpensive magnetic compass sensors. However, the challenge
remains to determine the relative direction of a neighbor node.
Jumping sensors display an opportunity to utilize the resulting airborne rotation, ob7

tained from the jumping action, to collaboratively identify neighboring nodes’ relative position. This can be achieved with the usage of signal strength readings at different orientation
angles. Such a method, presents a series of challenges including signal strength analysis,
prediction and error correction methodologies.

1.3

Structure of the Content

This dissertation is organized as follow. Chapter 2 gives a background survey. Chapter 3
discusses wireless signal propagation from a WSN perspective. Fundamental factors responsible for decrease in successful wireless communications are presented and evaluated.
Furthermore, it introduces an approach to decrease packet loss and increase communication
range in a wireless jumping sensor network. Chapter 4 shows the impact of jumping sensors
on connectivity and area covered, and presents a Hopping Sensor Network Model to increase
the sensing area coverage. A hopping sensor network topology, and the introduction of two
algorithms for node discovery and connectivity increase with the use of jumping sensors as
gateways are discussed in Chapter 5. Chapter 6 presents an efficient node redeploymentdecision process to produce a functionally heterogeneous (jumping and non-jumping sensors)
WSN with a performance guarantee. Network performance is defined as a network fitness
formula considering the network Quality of Connectivity (QoC). Chapter 7 a multi-step
procedure to produce a direction oriented jumping sensor network topology is presented.
Finally, Chapter 8 concludes this dissertation with an overall summary, and highlight areas
for future research work.

8

Chapter 2
BACKGROUND SURVEY
This chapter provides a survey of related work to wireless sensor networks. It first presents
existing work in the area of wireless sensor design. Second, since energy is at premium in
sensor nodes, related work to efficient network energy management protocols is discussed.
Third, connectivity and area coverage are topics directly addressed in this dissertation,
therefore, work relevant to these topics is covered. Forth, topology control methodology is
reviewed. Fifth, one decade of research progress in the area of jumping robots prototypes
is presented. Sixth, a survey of state of the art sensors is provided. Last, related work to
sensor localization and its challenges is discussed.

2.1

Wireless Sensor Design

The design of a wireless sensor network is a non trivial process. R¨mer and Mattern (2004),
o
provided insight to different dimensions of the design space of a wireless sensor network.
Some of the design aspect discussed are:
• Deployment: The deployment of sensor nodes into a target field is the initial phase
to build a sensor network. However, it may be a continuous process during the life
spam of the network, e.g., failed node replacement. Nodes may be installed at precise
locations or deployed randomly. The deployment technique affects network properties
such as node density, node locations, coverage, and connectivity.
9

• Mobility: Sensor nodes may have automotive capabilities, enabling to relocate them
to target destinations. However, mobility may occur in a passive (incidental) manner,
as the result from environmental influences such as wind, water streams, or carried by
a mobile entity.
• Resources and Energy: The target application and the field environment dictate the
type of sensors that the wireless network will be composed of, e.g., hundreds of small
sensors vs. few sophisticated sensors. As sensor nodes become more sophisticated, the
cost of a single unit becomes higher, which can be suitable for networks where few
nodes are needed, but economically impractical for large-scale networks. Furthermore,
the size of a sensor node can constraint available energy, computing, storage, and
communication resources.
• Heterogeneity: Sensor networks may be composed of different sensor nodes and devices.
Some nodes with special capabilities, such as, more computing power, greater storage
resource, or longer communication range may be positioned on key locations to collect,
process and route data from other sensors.
• Infrastructure: Wireless networks are built as infrastructure-based networks, ad-hoc
networks, or a combination of both. In an infrastructure-based network, a set of base
stations devices are deployed, to be used by sensor nodes to communicate. Base stations
serves as intermediary relays on any communication. On the other hand, in an ad-hoc
network, nodes directly communicate with each other in a collaborative manner, i.e.,
nodes can serve as routers for those nodes that cannot directly communicate, without
the use of an intermediary base station.
• Network Topology: The maximum number of communication hops between any two
10

nodes in a network is known as the network diameter. This property is important since
it affects network latency, and tolerance to node failure.

2.2

Network Energy Management

Among all the design aspects, energy and communication constraints have the highest influence on the node’s performance. Wireless sensor life is usually limited by a battery source,
hence it is important to reduce the energy consumption to maximize sensor life. There
has been a significant amount of work carried out to reduce the impact of energy usage for
wireless sensor operation.
Shah and Rabaey (2002), proposed an energy aware routing scheme that uses sub-optimal
paths occasionally to provide substantial gains. Their approach is based on the premise that
always using lowest energy paths may not be optimal from the point of view of network
lifetime and long-term connectivity. An uncontrolled usage of lowest energy paths may
deplete the energy on nodes located on critical routes, creating a disruption in network
connectivity.
Data-gathering is an important issue in wireless sensor network since energy is usually a
scarce resource on the sensor nodes. The problem of energy efficient data-gathering escalates
when nodes are capable of mobility. Liu et al. (2004); Liu and Lee (2004) provided clusteringbased and time-driven protocols to minimize the energy consumption for data-gathering in
wireless mobile sensor networks.
Carbunar et al. (2006), focused on the detection and elimination of redundant sensors
without affecting network coverage, improving energy efficiency on the network. Similarly,
an approach was introduced by Cheng and Yen (2006), to obtain an optimal active sensor

11

selection while retaining system coverage. Likewise, Ding et al. (2007), presented a connectivity based Partition Approach (CPA) to partition sensors into groups. Groups follow a
sleep scheduling for sensor nodes, preserving network communication quality and reducing
the energy consumption on the sensor network. CPA differs from other approaches since it
is based on measured connectivity, and not on nodes’ locations.

2.3

Connectivity and Area Coverage

In wireless communication signal strength fades over distance. Low signal strength negatively
contributes to high packet loss, increasing the possibility of communication links breakage.
Aguayo et al. (2004), analyzed the importance of network planning over packet loss and the
implications for MAC and routing protocol design. In order to better understand wireless
network performance, signal strength estimation for indoor communications was performed
by Biaz et al. (2005). Similarly, Kavak et al. (1999) studied the effects of base station antenna
height and mobile node movement over wireless communication.
Maintaining system connectivity in a wireless sensor network is a key issue for maximizing
area coverage. Wang et al. (2003), studied the relation between wireless sensor network
coverage and connectivity. In fact, they observed that coverage infers connectivity if the
radio transmission range is at least twice the sensing range, i.e. Rt ≥ 2Rs . The relation
allowed them to treat coverage and connectivity as one problem. Similarly, Zhang and Hou
(2005), obtained the same coverage and connectivity relation (Rt ≥ 2Rs ), proving that when
the relation is satisfied, complete coverage of a convex area implies connectivity among the
set of nodes in that area.
Clustering techniques have been discussed to achieve higher network connectivity. Younis

12

et al. (2006) studied two clustering methods, for wireless sensor networks. In the first method,
multiple gateways are deployed, and sensor nodes associate with them to form clusters. The
second method is a multi-tier architecture, in where selected sensors are designated as agents
to connect unreachable nodes to a single gateway. Multi-tier clustering showed superior node
reachability, but requires more sensors involved in network management.
In mobile wireless sensors, area coverage is often achieved by relocating sensors to needed
areas. Howard et al. (2002), used virtual forces to redistribute sensors. Sensors are repelled
from obstacles objects or from others nodes in order to maximized the covered area. Similarly, Zou and Chakrabarty (2003) proposed a Virtual Force Algorithm (VFA) to enhance
sensor field coverage. However, even with groundbreaking relocation techniques there will be
situations where the relocation of the sensors to needed areas is not possible due to obstacles
or terrain types (eg., river streams, lakes, etc.). This become a major problem when sensors
are sparsely distributed over the field (e.g., when using air deployment), because there is
a higher possibility for sensors to be left without connectivity, consequently affecting area
coverage.

2.4

Topology Control

The topology of an ad hoc network is defined as the set of communication links between
nodes pairs used explicitly or implicitly by a routing mechanism, Ramanathan and RosalesHain (2000). Different factors can affect a topology. Some of them can be classified as
uncontrollable such as node unpredicted movement, noise interference, and obstacles. Others
can be controllable such as node transmission power, node relocation, and by limiting the
amounts of neighbors that each node keeps track.

13

Topology control is important since wrong topology control will result in degraded performance and even poor connectivity on the network. For instance, a too dense topology
will limit the spatial reuse, hence network capacity is reduced. On the contrary, if the
topology is too sparse, connectivity can be degraded. Moreover, a good power control is
necessary to extend battery life of nodes, and to avoid extremes power levels. In fact, too
high transmit power level result in excessive interference among nodes, creating high MAClevel contentions. On the other hand, too low transmit power level results in a disconnected
network.
The following methods address the problem of controlling the network topology by adjusting the above mentioned controllable factors.

2.4.1

Degree-based Mobility Model

Mobility of nodes in wireless network introduces more complexity in the system models.
There are some studies in literature that propose mobility models in order to decrease mobility effect on network performance and permit to test algorithms for realistic scenarios.
Mobility models can represent mobile nodes whose movements are independent of each other,
or mobile nodes whose movements are dependent on each other Camp et al. (2002). Some
well-known mobility models are: random walk, random waypoint, random direction, GaussMarkov and parallel path, they differ from each other about the choice of speed and direction
of the nodes.
• Random walk: a simple mobility model based on random directions and speeds.
• Random waypoint: a model that includes pause times between changes in destination
and speed.
14

• Random direction: a model that forces mobility nodes to travel to the edge of the
simulation area before changing direction and speed.
• Gauss-Markov: a model that uses one tuning parameter to vary the degree of randomness in the mobility pattern.
• Parallel path: a model where a mobile node picks a random speed and move across
the geographical area following a direction parallel to one of the boundary lines.
In Santi (2005), a different approach of mobility model is proposed: mobility is used to
achieve a physical topology with better performance in connectivity, instead of decrease its
impact. At this aim, mobility is created in an artificial way: a static wireless network is
overlaid by a mobile wireless network. The static network is assumed as a wireless sensor
network or a wireless mesh network (where nodes take data and send it to the sink or to
the backbone) and the mobile network is constituted by nodes that follow a certain mobility
pattern. Direction of the mobile nodes is determinate by a measure of importance of static
neighbors. In particular, a degree-based mobility model is defined, where the probability
distribution for the direction depends on the degree of the static nodes.
In its work, topology is generated in the following manner: a sink node is placed in the
central geographic area and static nodes are placed around it in a growing manner, where
new static nodes are placed around previously placed static nodes, to avoid isolated nodes
and to have a fully-connected network. Every static node is ensured to be in the coverage
area of at least one and at most k neighbors, where k is a chosen parameter accordingly to
the desired coverage. A multi-hop connectivity is considered between nodes and the sink,
when data is sent to the sink.

15

The speed of mobile nodes is uniformly distributed between [0, Vm ] and the probability
distribution of direction is proportional to the importance of the neighboring nodes. Hence,
mobile nodes move through more important static nodes. In particular, in Yanmaz (2008),
the distribution of the direction of a mobile node depends on the degree of its neighbors,
where degree of a node is defined as the number of connections of that node with nodes that
are within its transmission range.
This mobility model increases the connectivity of a sparsely-connected grid topology and
the random topology. Moreover, using mobile nodes overlaying a static network decreases
the average static node-to-sink number of hops, creating shortcuts between nodes of the
static network.

2.4.2

Transmit Power Adjustment

Topology control is considered in Ramanathan and Rosales-Hain (2000), as a constrained
optimization problem of practical importance; particularly as minimizing transmit power
subject to two constraints: network connected or network biconnected. Therefore the problem in this case is adjusting the transmit powers of nodes to obtain a desired topology.
There are two centralized algorithms in order to minimize the maximum power used per
node for use in static networks: CONNECT and BICONN-AUGMENT. For mobile networks
two distributed heuristics are introduced, Local Information Non Topology (LINT) and
Local Information Link-State Topology (LILT). They adapt node transmit powers related
to topological changes and aim to have a connected topology using minimum power.
CONNECT is a simple algorithm where every component is merged iteratively until one
component is left. In the BICONN-AUGMENT a connected network is increased to obtain
a biconnected network, which is based on CONNECT. In both cases a post processing can
16

be used to delete redundant connections and to ensure per-node minimally.
In LINT, each node has three parameters: a desired node degree, a high and a low
threshold on the node degree. Periodically, each node changes its power level with respect
to the number of its active neighbors, keeping the node degree between the two thresholds.
LILT is an improvement of LINT, it overrides the high threshold when there is a topology
change from the routing protocol update results (i.e. link-state protocols). It is composed
of two main parts:
• Neighbor Reduction Protocol (NRP): In charge of maintaining the node connectivity
degree around a certain configured value.
• Neighbor Addition Protocol (NAP): Employed to increase the node transmit power if
the topology changes indicated by the routing update is in undesirable connectivity.
Initially all nodes start with the maximum possible power to obtain the maximum connected network, enabling successful propagation of updates and the initialization of a network
topology database at each node. After initialization, a node receiving a routing update has to
determine in the state of the updated topology: disconnected, connected (not bi-connected),
or bi-connected. NRP and NAP are activated on each node, depending on the topology
state. If the topology is disconnected, the node increases the transmit power to the maximum possible value. If connected (not bi-connected), special bi-connectivity augmentation
is performed, in where power changes are coordinated with other nodes to avoid all nodes to
react to the topology changes. Finally, if the topology is found to be bi-connected, no action
is taken.
It is important to note that node coordination is achieved by setting a controlled randomized timer for each node, before taking an action. Nodes proceed to take an action only
17

if the topology is still in an undesired state after the timer has expired.

2.4.3

Local Minimum Spanning Tree (LMST)

LMST, described in Li et al. (2003), is a minimum spanning tree (MST) based topology
control algorithm in where each node builds his local minimum spanning tree independently
and only keeps as neighbors those nodes that are one-hop away from it. The topology
constructed under MST preserves the network connectivity and bounds by 6 the degree
of any node. A small node degree is desired because it reduces the MAC-level contention
and interference among nodes. Moreover the resulting topology can be further converted in
one with bi-directional links. Having a topology that only consist of bi-directional links is
important for link level acknowledgments, and medium access control mechanisms such as
RTS/CTS in IEEE 802.11.
Besides the algorithm presented in Li et al. (2003), guidelines to design an effective
topology control are introduced:
• Network connectivity has to be preserved with the use of minimal possible transmission
power on every node.
• Algorithm has to be distributed: each node has to make decision based on the information it has collected from the network.
• Algorithm has to be dependent only from locally information, to have less impact from
mobility.
• Bi-directional links and small node degree are desired on the topology.
Moreover, the network topology constructed under the common maximum transmission
18

range for all nodes, denoted as dmax , is represented as a graph G = (V, E), where V is
the set of the nodes in the network and E is the edges set of G, which represents connection
links between nodes that are within the transmission range dmax . Each node (u) is assigned
with a unique id and it is defined to have a visible neighborhood NVu (G). The visible
neighborhood NVu (G) is the list of neighbor nodes that node u can reach by employing its
maximum transmission power.
LMST algorithm has three phases: information collection, topology construction and determination of transmission power; with an optional optimization phase for the construction
of the topology with bi-directional edges.
The first phase, namely information collection, is the exchange of information by each
node. This information is composed by the node id and position. This aim can be reached
using a periodic message transmitted with the common node maximum transmit power. The
time interval between two of these messages depends on the level of node mobility in the
network.
In the second phase, topology construction, each node applies Prim’s Algorithm independently to obtain its local minimum spanning tree. Two points are fundamental to have
a power efficient minimum spanning tree: the weight of an edge has to be the transmission
power between the two nodes, and the local minimum spanning tree constructed by each
node has to be unique. Since multiple edges with the same weight exist, the minimum spanning tree is not unique. Therefore, a new weight function is defined to guaranteed that in
each step of Prim’s algorithm the choice of minimum weight edges is unique, resulting in an
unique local minimum spanning tree per node.
It is assumed the maximal transmission is known and it is the same to all nodes. With
the message in the first phase every node knows the specific receiver power level of each
19

of its neighbors. In the third phase, the transmission power that a node needs to reach
its neighbors is determined considering two propagation models: the free space propagation
model and the two-ray ground propagation model. In both models, there is a common
relation between Pt (power used to transmit) and Pr (power received):

Pr = Pt · G

(2.1)

where G is a function of the antenna gains of the receiver and transmitter, the antenna
height of the transmitter and receiver, the distance between transmitter and receiver, the
wave length and the system loss. In other words, if we have two nodes A and B, when A
receives the position information from B, it measures the receiving power Pr and obtains G
for node B as:
G=

Pr
Pmax

(2.2)

Therefore, in order for B to be able to receive A’s message, A needs to transmit at least:

Pth · G

(2.3)

where Pth is the minimum power threshold of a node to successfully receive a message.
In the optional optimization phase, a bi-directional network is built with two solutions: to
enforce all the uni-directional links to become bi-directional; or to remove the uni-directional
links and place them with two new topologies. The first one gives more routing redundancy,
while the second one is more efficient in terms of spatial reuse. Furthermore, different
important properties of LMST algorithm are proved, and it is determined how often the
neighbors information should be exchanged for the topology to be updated, by using a
20

probabilistic model of node movement.

2.5

Mobile Robots

The development of new kinds of mobile sensors that can deal with terrain adversities has
always been a stimulating topic of research. Jumping robot technology, also known as
hopping robots, are characterize by their jumping method of movement. Jumping robot
development has improved during the last decade, but working prototypes are still scarce.
Nevertheless, this technology represents an unique opportunity for wireless sensor networks
to be deployed in hard to reach areas or difficult to navigate with traditional wheeled mobile
robot sensors.
Wei et al. (2000), built a prototype no bigger than a 5cm cube size, powered by a battery
source, with no active directional control, resulting in constrained movement. A more in
depth design for a jumping robot, including a working prototype, was later presented and
suggested for usage in celestial body exploration by Burdick and Fiorini (2003).
Increasing jumping height and horizontal displacement, with power source limitations has
been one of the main are of research for this technology. Scarfogliero et al. (2007), developed
a jumping robot inspired from frog locomotion to achieve longer jumps. Similarly, Kovac
et al. (2008) presented a miniature prototype with only 7 grams of mass, able to achieve
jump heights of more than 1 m.
A combustion-driven piston prototype of a hopping robot was developed by Sandia National Laboratories1 . The hopping robot can make jumps up to 3 f t (approx. 0.9 m) of
height, and move horizontally up to 6 feet (approx. 1.8 m) (German (2000)). It is capable of
1 http://www.sandia.gov

21

making 4, 000 jumps on a single fuel tank (20 grams of fuel). Another experimental hopping
robot can make jumps up to 20 f t (approx. 6 m) of height, and can perform about 100
jumps on a single fuel tank.
As hopping robots’ designs evolve, the opportunity to incorporate them into a wireless
sensor network arises. Zhao et al. (2009), discussed the design and development of a manually
loaded, self-stabilizing jumping robot that carries a microcontroller with sensing capabilities.
It uses a 3.7 V , 100 mAh LiPo battery, and is capable to jump up to a height of 0.35 m. An
enhanced design able to perform continuous jumps was later introduced, Zhao et al. (2010).
Research involving hopping sensors has been focused on their relocation. Cen and Mutka
(2008), discussed hopping sensor relocation, considering the impact of air disturbance and
movement inaccuracy. Similarly, Pei et al. (2009); Kim et al. (2010), provided relocation
algorithms for hopping sensors in obstacle abundant terrains. Other relocation algorithms
with balanced migration of hopping sensors, are presented by Kim and Mutka (2009a,b).
Their objective is to avoid the creation of new sensing holes from repeated hopping sensors
requests made to a single supplier.

2.6

State of the Art Sensors

There are different wireless sensor nodes available in the market nowadays. Many of them are
characterize by their small size and low power consumption, Warneke et al. (2001). MICA2,
MICA2DOT, and IRIS are some of the wireless measurement system modules developed by
MEMSIC2 . These modules come with expansion connectors for light, temperature, relative
humidity, barometric pressure, acceleration and seismic activity, acoustic, magnetic field, and
2 http://www.memsic.com/wireless-sensor-networks/

22

others sensor boards. The main differences are on the module size, battery types, data rates,
transmission frequencies, and advertised communication ranges. Some of these specifications
are summarized in Table 2.1.
It is worth to note that experimental communication range for these modules, on open
field scenarios, is much smaller than the advertised. More details on experimental results is
provided in Chapter 3.
Table 2.1: Wireless Measurements Systems Specifications
Battery
Size3 (mm)
Weight3 (grams)
Radio Center Frequency
Data Rate (Kbps)
RF Max. Power (dBm)
Receive Sensitivity (dBm)
Outdoor Range (m)

2.7

MICA2
MICA2DOT
2X AA
3V Coin Cell
58 x 32 x 7
25 x 6
18
3
868/916 MHz, 433 MHz, 315 MHz
38.4
5, 10
-98, -101
150, 300

IRIS
2X AA
58 x 32 x 7
18
2.4 GHz
250
3
-101
300

Wireless Node Localization and Relative Orientation

Global Position Systems (GPS) are perhaps the most popular localization technology available. GPS receivers are able to obtain global coordinates with an accuracy within 3 meters,
and even less than a meter when combined with augmentation systems, Aeronautics and
Space Engineering Board (1995). Nevertheless, the real time operation and high accuracy
of GPS receivers comes with a high power consumption, making this technology infeasible
for many wireless sensor networks mainly composed of nodes with constrained power. For
3 Size and weight of the modules excludes the battery pack.

23

this reason fewer specialized GPS enabled nodes are sometimes deployed to act as beacons.
However, this also comes with complications since in order to establish a coordinate system,
a minimum of three non-collinear beacon nodes are required for two dimensions localization,
and four non-coplanar beacons for three dimensions localization. Therefore, beacon placement must be planned; something that can become impractical for scenarios where sensors
are deployed in remote areas, e.g., volcano monitoring by air-dropped sensors, Song et al.
(2009).
In addition, node localization with beacon nodes requires each node to estimate the
distances to a subset of beacons. Beacon distances from the node are used to determine
nodes’ location with trilateration. Receiver Signal Strength Indicator (RSSI) is perhaps the
less invasive method to estimate the distance between two communicating nodes. Signal
path loss models are used to predict the signal strength over distance, Goldsmith (2005);
Whitehouse (2002). Hence, the RSSI of beacons nodes is collected by the nodes and used to
predict the distances.
Other methods for more accurate distance estimation require additional hardware. Speakers and microphones are used for Time Difference of Arrival (TDoA), in where acoustic signals
are used to measure the signal traveling time and compute the distance Balakrishnan et al.
(2003). On the other hand, Angle of Arrival (AoA) uses an antenna array or microphone
array to determine the direction of propagation of a signal Priyantha et al. (2001). The
drawback of these methods is the added mass and bulkiness they pose to the sensor robot,
making them impractical for most mobile nodes. Moreover, the tight dependency to precise
beacon placement presents a challenge to the design phase of WSN.
Node orientation is usually derived from each nodes’ location coordinates. Location aware
sensors utilize their position coordinates to navigate and reach to target destinations. This
24

approach becomes unfeasible for the vast majority of sensor nodes with limited capabilities.
Furthermore, most relocation operations could rely in partial relative-orientation information
obtained from neighboring nodes.
In Chapter 7, a multi-step procedure to obtain sensors’ relative orientation is presented.
The procedure employs the equipment already available in jumping sensors robot prototypes
without additional hardware that could compromise the design and performance of the sensor
nodes.

25

Chapter 3
WIRELESS COMMUNICATION
RANGE INCREASE

3.1

Preliminaries and Motivations

Wireless communication can be affected by many factors: environment, type and height of
antenna, noise, out of range receivers, etc. These factors can cause signal fading, reducing
effective communication range, and increasing the tendency for packet losses at receiving
nodes. If the communication between two sensor nodes experiences packet losses, retransmission may be required for those packets. Furthermore, if the sensor nodes are to distant
from each other, they can become excommunicated from the network, creating network partitioning. Solutions approaches to deal with network partitioning and connectivity increase
are discussed in Chapter 5.
This chapter studies the changes in sensor communication range, by varying sensors’
heights relative to the surface. In order to increase wireless communication range in sensor
networks, a communication technique that relies on the jumping capabilities of jumping
sensors is proposed. While the sensor is jumping, a change in the altitude enhances, for a
short time, the ability to communicate with other sensors. Base on this, a novel airborne
communication process is defined. Field experiments were conducted and results discussions
are provided.
26

3.1.1

Chapter Organization

First, a discussion of wireless signal strength from a sensor network perspective is given
in Section 3.2. Experimental evaluations are presented and compared to existing signal
path loss propagation models. Section 3.3 introduces airborne communication concepts and
implications. A jumping algorithm and the airborne communication process are defined
for the management of data transmissions in multiple jumps and the synchronization of
transmitter and receiver nodes, respectively. Section 3.4 presents experimental evaluations.
A chapter summary is provided in Section 3.5.

3.2

Received Signal Strength

Received Signal Strength (RSS) is the measured power of the signal when received by a
node. RSS serves as an indicator of quality of signal reception. Therefore, to study communication patterns, a discussion of factors impacting signal strength and their importance to
a jumping sensor communication is given. Among the different factors that can affect signal
strength, this work will focus on the effects of communication distance, sensor elevation, and
transmission power.

3.2.1

Impact of communication distance on RSS

In order to evaluate the impact of communication distance on RSS, experiments were performed in open areas such as a parking lot and a football field with natural grass turf. IRIS
Crossbow sensor nodes were programmed to transmit packets at regular intervals, at their
maximum transmission power (3 dBm). An IRIS base station mounted on a sensor board
was used to measure the RSS of the packets. Measurements were taken at different separa27

tion distances between the transmitter and receiver. Two packet transmission interval rates
(50 msec, 20 msec) were used to see the effect of packet rate over RSS. The impact of packet
rate over RSS is found to be negligible for these experiments. Hence we present results from
just the 20 msec data rate experiment. This is depicted in Figures 3.1 and 3.2. It is worth
to mention that IRIS sensor can receive packets with a signal strength as low as −101 dBm
(IRIS Sensitivity), while it can provide the RSS Indicator (RSSI) as low as −91 dBm (IRIS
Minimum Precision), which means that any received packet with a signal strength between
[−91,−101]dBm will be registered with −91 dBm.

0
−10
−20

RSSI (dBm)

−30
−40

Measurements
Free Space
Simp. Path Loss, exp = 3.5
IRIS Min. Precision
IRIS Sensitivity

−50
−60
−70
−80
−90
−100
−110
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Distance (m)
Figure 3.1: RSSI on Concrete

Figure 3.1 shows the impact of distance over RSSI, at surface level, in a concrete parking
lot (flat surface). It can be seen to follow a simplified path loss model (with exponent 3.5)
and accordingly the RSSI wanes to below measurable values by the time it reaches 10 m
distance. Additionally, it was noticed that a significant amount of packet losses (80 − 90%

28

losses) occurred at distances larger than 10 m between the sender and the receiver.

Pr = Pt K

d0 γ
d

(3.1)

The simplified path loss model is presented in Equation 3.1. Pr is the predicted received
power when transmitting using power Pt from a distance d, and K is a constant (d0 /P0 )
obtained from the received power P0 at a reference distance d0 ; γ is the path loss exponent
(determined by measurements).

0

Measurements
Free Space
Simp. Path Loss, exp = 4
IRIS Min. Precision
IRIS Sensitivity

−10
−20

RSSI (dBm)

−30
−40
−50
−60
−70
−80
−90
−100
−110

−120
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Distance (m)
Figure 3.2: RSSI on Grass
Figure 3.2 shows the impact of distance over RSSI, at surface level, when measured in a
grass turf. The average length of the grass was measured to be 4 cm long. It is worth to note
that it follows again a simplified path loss model with adjusted γ value of 4.0, but the signal
strength reduces to below measurable values at 5 m of distance. This happens because the
thin strands of grass interfere with the signal, dissipating the signal strength, thus, reducing
29

the overall communication range.

3.2.2

Impact of sensor node elevation on RSS

Signal strength is affected by obstructions in the line of sight between the transmitter and
receiver antennas. The Fresnel zone is an elliptical area surrounding the line of sight with
foci collocated at the transmitter and the receiver, Green and Obaidat (2002). Obstructions
within the inner 60% of the Fresnel zone, cause signal diffraction, which reduces signal
strength.
Sensor nodes are often at ground surface level when randomly deployed. Therefore their
transmitted signal strength gets attenuated faster with the increase in distance, resulting
in the loss of all or part of the transmission packets. Experiments show that when the
transmitting sensor can elevate itself, it has an increased opportunity of having successful
communication with a distant receiver. This is attributed to the less interference from the
ground surface irregularities to the Fresnel zone with the increase in altitude of the sensor
node.
Experiments were performed to measure the impact of having the transmitter node at
different heights (with respect to ground level) on communication range, with the receiver
at ground level. For each height, packet transmissions were repeated for different distances
between transmitter and receiver. The maximum communication range for each height was
identified as the farthest distance having zero packet loss. Fig. 3.3 presents the communication ranges obtained for transmitter heights (0 to 1 m).
The obtained results clearly show that increasing the elevation (within practical bounds)
of the sender, increases the communication range with zero packet losses. Furthermore,
experimental results are compared with two signal path loss propagation models: Lee (1986),
30

60

Lee Loss Prediction Model
Green Loss Prediction Model
Experimental (zero loss)

Communication Range (m)

55
50
45
40
35
30
25
20
15
10
5
0

0

10

20

30

40 50 60 70
Heights (cm)

80

90 100

Figure 3.3: Communication Range with Varying Heights
Green and Obaidat (2002), Eqn. 3.2 and 3.3, respectively. Both models take into account
gains from the transmitter and receiver antenna heights (ht , hr ), while Green also considers
loss due to the frequency of operation (f ). Green’s model appears to follow more closely
to our experimental results as the height of the sensor varies from ground level to 1 meter,
although the model was designed for heights above 1 meter.

Ploss = 40log10 (d) − 20log10 (ht ∗ hr )

(3.2)

Ploss = 40log10 (d) − 20log10 (ht ∗ hr ) + 20log10 (f )

(3.3)

It is worth mentioning that most of the path loss models are designed for higher antenna

31

heights, longer communication distances, and lower frequencies of operation, e.g.: Hata
path loss model, Hata (1980), for antenna heights over 30 m; Longley-Rice path loss model,
Longley and Rice (1968), for distances over a Kilometer; Hata/Davidson path loss model,
TIA TR8 Working Group 8.8 Technology Compatibility (1997), for frequencies up to 1500
MHz; Cost231-Hata path loss model, Mogensen et al. (1991), for frequencies up to 2000
MHz. Despite these incompatibilities, Lee and Green path loss models produced the closest
prediction to our experimental results.

3.2.3

Impact of transmission power on RSS

The transmission power has a direct correlation with RSS and the transmission range.
Though transmitting at lower transmission power saves energy, in a hopping sensor network, the energy saving is a small fraction of the entire transmission while hopping process.
Hence, since communication range enhancement is the main focus of this work, it is advisable
to always transmit at maximum transmission power.

3.3

Airborne Communication

This section introduces preliminary concepts for jumping sensor communication while airborne. A jumping algorithm is designed for the successful transmission of packets. An
airborne two-way communication process is presented and evaluated with in-field experiments.

32

3.3.1

Time of Flight

The time of flight of a jumping sensor is the period of time it stays airborne. It is our
goal to use this period of time to reach other sensors that may be distant and unreachable
when staying at ground level. However, as the distance between sensor nodes increases, the
minimum jump height required for successful communication becomes higher, (see Section
3.2.2). We call this minimum height, the threshold height. A jumping sensor needs to jump
higher than the threshold height to provide enough time to communicate with others. The
period of time the jumping sensor stays on and above the threshold height is called the
communication time.
In order to calculate the required jump height of a jumping sensor such that it has enough
communication time, let us study the jumping process from the moment the jumping sensor
reaches the maximum height. At this point its velocity is zero and the sensor will begin to
fall. The falling distance can be estimated using free fall Eqn. 3.4, where g is the gravity
acceleration (9.81m/s2 ) and t is time in seconds while falling.

1
y = gt2
2

(3.4)

The required jump height (Hmax ) is obtained by rewriting Eqn. 3.4 to consider the falling
distance until reaching the threshold height (HT h ), Eqn. 3.5.
1
∆y = Hmax − HT h = gt2
2 ∆y
1
Hmax = HT h + gt2
2 ∆y

(3.5)

Note that t∆y is the period of time it takes to fall from Hmax to HT h ; by symmetry, the
33

period of time for the jumping sensor to reach Hmax from HT h is equal to the time it takes
to fall from Hmax to HT h . Therefore, t∆y is half of the communication time.
Figure 3.4 is a plot of ∆y, the height needed above threshold height. As an example,
if 600 ms are required to communicate while airborne, a jumping sensor will need to jump

Height Above Threshold Height (m)

44.2 cm above the HT h to guarantee the communication height.

5
4.6
4.2
3.8
3.4
3
2.6
2.2
1.8
1.4
1
0.6
0.2
0

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Communication Time (s)

2

Figure 3.4: Height Needed Above Threshold Height VS. Communication Time
A jumping sensor may not have the control to jump to different heights, instead, it may
be only able to jump to a fixed height. In such case, the available time will depend on the
maximum height it can jump and the communication threshold height. Figure 3.5 is the
plot of the available time for a jumping sensor that can jump to 1 meter height, given a
threshold height. For example, if the minimal height for transmission is 30 cm, then the
available transmission time will be the amount of time that the sensor takes to go up from
this threshold height to one meter and come back to the threshold height, which is equal
to 408.6 ms. Similarly, if the threshold height required to transmit is 50 cm, the available
34

transmission time would be 264.6 ms.

Communication Time (in sec)

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Threshold height (m)

1

Figure 3.5: Time of Flight for 1 Meter Jump Height

3.3.2

Jumping Algorithm

With respect to the jumping characteristic of a jumping sensor, two cases are considered:
the jumping sensor can either control the height it jumps or always jump to a fixed height;
the last case being a specific case of the previous one. Having control of the jump height
presents the opportunity to optimize the jump energy consumption.
The jumping sensor needs to jump high enough to stay above the communication threshold height for the required communication time, which depends on the transmission data
rate of the sensor. When the required communication time is larger than the maximum
time available for one jump, the data has to be partitioned in multiple jumps in order to be
transmitted. Then, it is on the last jump that the height control capability will take place,

35

lowering the jump height when the remaining data to be transmitted requires a smaller
communication time.
Let D be the amount of data to be transmitted, at a minimum threshold height h. Let Dh
be the corresponding amount of data that can be transmitted on the available communication
time with the highest jump height. The number of jumps is computed with Equation (3.6).

Jumps =

D
Dh

(3.6)

When the jumping sensor can control the jump height, the height of the last jump
can be adjusted depending on the size of the remaining data that has to be transmitted,
Equation (3.7).

Last Jump Data ⇐ D mod Dh

(3.7)

If there is remaining data for the last jump, the new required communication time is
computed, and the new jump height is obtained as presented in Section 3.3.1. It has to be
noted that when the remaining amount of data for the last jump is zero, it means that all
the jumps are performed with the highest jump height possible.

3.3.2.1

Jump Data Partitioning

The flow diagram for the data partitioning process is presented in Figure 3.6. The process
begins with an amount of data that has to be transmitted. The feasibility box checks if
all the data can be transmitted with the maximum jump height possible. If not, the data
is partitioned. Each partition aggregates multiple packets, and the number of partitions is
equivalent to the number of times the hopping sensor needs to jump.
36

Figure 3.6: Jump Data Partitioning Flow Diagram
If the jumping sensor has jump height control, the number of jumps at highest height
is provided, along with the last jump height. Otherwise, if the hopping sensor is limited to
jump to a fixed height, a reduction in computation overhead will be seen by not requiring
the extra calculation of the last jump height based on residual data.

3.3.3

Airborne Communication Process

The change in height from a jumping sensor creates a temporary increase on its wireless
communication range. Next, an airborne communication process is presented to coordinate
the jumping procedure with neighbor sensors, and to enable two-way communications in
37

jumping sensor nodes when airborne.
The airborne communication process is organized in two phases:
• Advertisement phase
• Data communication phase

Elevation

Advertisement
Acknowledgement
Data

X

X
Sender

X

X

Receiver
e
Figure 3.7: Airborne Communication Process

A jumping sensor starts the advertisement phase before it jumps, where it indicates
a potential receiver of this action. The receiver in turn acknowledges the advertisement,
indicating the hopping sensor to proceed with the transmission. To overcome the problem of
out-of-communication-range-at-ground-level receiver, the advertisement is continued during
the initial stage of the jump, until an acknowledgment is received. The advertisement phase
is also used to synchronize the sender with the receiver when communication in multiple

38

jumps takes place. During data communication, if either a packet or an acknowledgment is
lost, the transmitter will retransmit the packet after the expiration of a retransmission wait
period.
The airborne communication process is depicted in Figure 3.7, where only the sender
jumps. From the figure it can be appreciated that due to the long distance between the
nodes, advertisement packets cannot reach the receiver. However, as the sender jumping
sensor gets elevated, its transmission range increases allowing the advertisement packets to
reach the receiver. The receiver acknowledges, and the data communication phase proceeds
as described in Section 3.3.2.

3.4

Experimental Evaluation

In order to evaluate the airborne communication process, experiments were performed to
collect measurements of signal strength, and register packet losses for packets transmitted
while airborne.
IRIS Crossbow sensors were used, with a jumping emulator mechanism. The jumping
action of a jumping robot was emulated using an elastic string mechanism, because of the
unavailability of a jumping robot prototype able to meet desired jump heights. The mechanism was fixed to make the sensor to jump to a height of 1 meter with an error of ±0.05 m
(jump allowed between 0.95 m and 1.05 m). A total time of flight was registered to be on
average 1 second. In most of the experiments, the jumping motion is brought to a stop on
first ground landing, but in some circumstances the sensor had a bouncing effect resulting
in a few additional milliseconds of elevated communication time.

39

3.4.1

Airborne RSS

Experiments were performed with different packet interval rates, 20 msec and 50 msec,
to measure the changes in the received signal strength from packets transmitted airborne.
The impact on RSS due to hopping was evaluated from 25 m onwards, to avoid possible
communication at ground surface level. Multiple sets of experiments were performed and
the results are presented in Figures 3.8, 3.9, and 3.10.

Figure 3.8: RSS Airborne Transmitted Packets (20msec rate)

In Figures 3.8 and 3.9, it can be noticed that during a single jump, the measured RSS
varies between −91 dBm and −88 dBm, with the count of packets diminishing at higher
dBm values. This is further demonstrated in the Figure 3.10, where the corresponding RSS
of each packet transmitted airborne at a packet rate interval of 50 msec, for a distance of 25
m, is presented. The plot shows that with the practical increase in elevation, the received
signal is stronger. Likewise, as the sensor starts descending, a reduction in signal strength

40

Figure 3.9: RSS Airborne Transmitted Packets (50msec rate)

RSSI (dBm)

−88
−89
−90
−91

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Packet
Figure 3.10: RSS Airborne Transmitted Packets at 25 meters (50msec rate)
is registered. The only blemish in the almost uniform plot is the RSS of packet 10, which
for this experiment it is believed to be caused by the change in the sensor node orientation
due to mid-air rotation while jumping.
41

3.4.2

Airborne Throughput

Airborne throughput was measured for packets transmitted while airborne during the data
communication phase. For this experimental evaluation the sensor was set to transmit using
two packet interval rates, 10 msec and 20 msec. The transmitter (while airborne) kept
retransmitting the same packet until it received a valid acknowledgment of receival from
the base station receiver fixed at ground level. On receiving a valid acknowledgment, the
transmitter sent out the next packet in the queue.

Figure 3.11: Throughput Airborne Data Communication (10msec)

Figure 3.11 shows the number of packets received and lost during the data communication
phase for different distances between the jumping sensor and the receiver, for a 10 msec packet
interval rate. As the distance increases, the total amount of transmitted packets (received +
lost) reduces. This is as predicted, because the signal strength of the packets arriving at the
receiver is reduced to a value below the sensing threshold for IRIS, making the advertisement

42

Figure 3.12: Throughput Airborne Data Communication (20ms)
phase to last longer, leaving less time for the data communication phase. The packet loss
can not be fit to a predictable curve due to the rotation movement of the sensor in the air,
which changes antenna orientation. Results from 20 msec experiments followed a similar
trend, but with slightly less packet losses, Figure 3.12.
It is worth to note that the unevenness of the terrain, nearby structures and objects,
and weather conditions have an effect on the signal propagation, and consequently, on the
number of packets received.

43

3.5

Summary

In this chapter a comprehensive study of communication range improvement for a jumping
sensor is presented. A discussion of the primary factors that impact the signal strength of
wireless sensors is provided and studied through experimental evaluations. Field experiments
were performed with wireless sensor nodes to measure the signal strength variance over
distance for different terrains. The impact on communication while a jumping sensor jumps
is presented. Results indicate that transmitting from higher than ground level increases the
wireless communication range of the sensor. For a typical case, the communication range is
increased from 5m-10m to 30m-60m, when the sensor jumps to an average height of 1 meter.
Moreover, a two-way airborne communication process is defined and tested, proving to be
effective as an alternative to reach distant receivers.

44

Chapter 4
HOPPING SENSOR NETWORK

4.1

Preliminaries and Motivations

Sensor nodes are in charge of monitoring their surrounding environment, resulting in generation of data. The gathered data will depend on the target application, e.g., environmental
monitoring, event detection, etc. Furthermore, the data needs to be forwarded to a base
station for data aggregation. Communication range is limited on sensor nodes, therefore,
there is the need to have many deployed sensors in order aid in the routing of data to base
station.
This chapter discusses the impact of network connectivity and area coverage in a jumping
sensor network. A Hopping Sensor Network Model is defined to increase sensing area coverage, with the enhancement of network connectivity. A Hopping Sensor Routing Protocol
is designed from the model, to balance the energy consumption on active hopping (jumping) sensor nodes. Both, the sensor network model and the routing protocol are evaluated
through simulations. Discussion analyses for the obtained results are provided.

4.1.1

Chapter Organization

The relationship between wireless sensor network connectivity and effective area coverage
is discussed in Section 4.2. The impact in network connectivity and area coverage from

45

employing jumping sensors in multiple network topologies is evaluated. The Hopping Sensor
Network Model is defined in Section 4.3. The Hopping Sensor Routing Protocol is presented
in Section 4.4. Performance evaluations of the routing protocol are provided. A chapter
summary and conclusions are provided in Section 4.5.

4.2

Network Connectivity and Area Coverage

A single node is said to have connectivity if itself or within its neighbor nodes exists a direct
or indirect communication path to the base station. Network connectivity is obtained with
Equation (4.1), in where n is the total number of nodes in the network, and c is the number
of nodes having connectivity considering our definition.

NetworkConnectivity % =

c
∗ 100
n

(4.1)

The area that has to be covered by a wireless network can be divided into unit regions.
Each sensor node senses over a set of unit areas. For example if an area is composed of 10
unit areas, and each sensor node is able to sense over 2 unit areas, then theoretically, only
5 sensors are enough to sense the entire area. However, in practice, sensors cannot always
be precisely positioned in needed locations, and not all of them may have connectivity.
Moreover, considering that the resulting area covered is dependent to those sensors that can
transmit data back to the base station, and the redundancy of sensor coverage of unit areas,
the area covered does not increase linearly with the increase in connectivity. Nevertheless,
increase of network connectivity has a significant impact in the total area covered.

46

4.2.1

Connectivity and Area Coverage Evaluation

Different scenarios were simulated to evaluate the impact of network connectivity and sense
area covered by employing hopping sensor nodes. A field was simulated as a two dimensional
grid space of 250 m x 250 m. This grid space is composed of small unit areas of 1 m2 . Sensors
were randomly distributed over the field. The amount of sensors was determined based on
the desired redundancy (R) of sensors covering a unit area, the grid space area (GA), and
the sensing area of a sensor (SA) (eq. 4.2).

Number of nodes =

GA
∗R
SA

(4.2)

A sensing range radius of 10 m was selected, yielding a rounded sensing area (SA) of
314 m2 . Sensor redundancies of 2, 3, 4, and 5 were used, obtaining 398, 597, 796, and 995
sensor nodes respectively. A base-station acting as a sink was positioned in the middle of
one of the grid edges.
Network connectivity and area coverage was measured for twenty (20) network topological
scenarios, for each sensor redundancy value. The case of sensors staying at the surface level
(0 jump height) was considered as the base case for comparison. The succeeding scenarios
are for nodes employing the jumping capability, each for a different maximum jump height.
Only the areas covered by those sensors having connectivity are considered. Figures 4.1 and
4.2 present the network connectivity and area coverage percentages obtained, respectively.
Figure 4.1 shows the average of connectivity percentage results obtained in the simulation
runs. It can be appreciated that for sensors communicating only at surface level, which
means that their communication range is limited to 5 m (lowest packet losss range, see
Section 3.2.2), the network connectivity on average is 0%. By increasing the jump height,

47

Figure 4.1: % Connectivity
the communication range of a sensor increases, resulting in an improved network connectivity.
Another factor affecting the network connectivity is the redundancy coverage. Having high
redundancy values will create high density of sensor per area, resulting in sensors to be closer
to each other, leading to better network connectivity. Additionally, from the results it is seen
that with a jump height of 40cm and higher, more than 90% connectivity was achieved for
redundancies greater than 2.
The percentage of area covered results are shown in figure 4.2. The connectivity on the
network has a critical impact on the area covered, since only those areas sensed by sensors
having connectivity are considered covered. This effect can be seen on sensors communicating
only at ground level, where the network connectivity was nearly 0%, resulting in a 0%
average of area covered. A notable difference in area coverage is obtained when network

48

Figure 4.2: % Area Covered
connectivity is 90% (when the jump height is 40cm) and above. This phenomenon can be
better appreciated in Figures 4.3 (topology with 30cm jump height) and 4.4 (topology with
40cm jump height), where the total area covered is depicted (shaded area) for deployed
topologies having a sensor redundancy value of 3.
The results indicate that a 40cm jump for this type of sensor node is a good threshold
height to have sufficiently high connectivity and increased sensing area coverage. However,
sensors may need to jump higher to obtain the desired amount of transmission time. Another
factor of importance for the area coverage is the sensor topology. Since sensors were randomly
positioned, sensor clusters were likely, i.e., areas with higher densities of sensors, leaving some
other areas with fewer sensors.

49

250

200

150

100

50

0

0

50

100

150

200

Figure 4.3: Topology Area Coverage for 30cm Jump Height

50

250

250

200

150

100

50

0

0

50

100

150

200

Figure 4.4: Topology Area Coverage for 40cm Jump Height

51

250

4.3

Hopping Sensor Network Model

In a mobile sensor network model, the premise for the sensor node mobility is to relocate
the sensor node, thereby increasing the sensing coverage area. In contrast, a Hopping Sensor
Network Model (HSNM) strives to increase the connectivity of the sensor nodes with the
base station in order to increase the sensed area.
The design of a HSNM needs to take into consideration vertical and horizontal displacements, energy consumption, and different application requirements, Figure 4.5.

Figure 4.5: Hopping Sensor Network Model

4.3.1

Vertical Displacement

Vertical displacement is the height that a sensor node can reach hopping. This, along with
the communication threshold height, influences the available time for transmission. With
the knowledge of the available channel bandwidth, the total data that can be transmitted
per hop is computed as in Section 3.3.

4.3.2

Horizontal Displacement

Horizontal displacement is the distance covered by each hop of the node and is difficult to
be controlled. In the HSNM, the goal is to keep this parameter as close to zero as possible in

52

order to maintain the node position relative to its neighbors. In a flat terrain, if the number
of hops is large, then the overall displacement will average out to zero due to random hop
direction. But, this parameter gains importance in an uneven terrain.

4.3.3

Hop Energy Vs. Transmission Energy Consumption

A sensor node consumes energy to hop as well as to transmit data. The relationship between the two can be studied with a simple example. Let there be 10 nodes that are at a
distance of 5 m from each other as shown in Figure 4.6. Results from practical experiments,
presented on Chapter 3, showed the sensor communication range with lowest packet loss at
different jumping heights. Let us consider the two heights, 0cm (no jump) and 60cm, with
communication range of 5m and 20m, respectively.

Figure 4.6: Hopping Sensor Energy Consumption

Let et be the energy consumed to transmit data over one hop and ej60 be the energy
consumed to jump to height 60 cm. As depicted in figure 4.6, to communicate at the ground
level consumes 9et units of energy. Consider the case when the nodes hop and transmit. Node
53

one (1) hops to height 60 cm and transmits the data directly to node 5 while consuming
ej60 + et energy. Similarly, node 5 communicates with node 9 by consuming the same
ej60 + et energy. Node 9 communicates with the base station consuming one et of energy.
In this case, the total energy consumed is 2ej60 + 3et . Similarly, if the node hops to height
25cm while expending ej25 energy and can reach a communication distance of 10m, then the
total energy consumed for the communication will be 4ej25 + 5et . Communicating while
hopping can provide alternative routes for sensors to communicate with the base station,
while avoiding the overload of nodes in high traffic routes.

4.3.4

Application Parameter

The application parameters are the variables that are regulated by the application’s needs.
In a WSN there are two basic parameters of usual interest, namely coverage and connectivity.
Certain non-critical applications such as a temperature sensing in an open environmental
field will tend to require a maximum coverage while requiring an intermittent connectivity.
This is sufficient because the sensor nodes can sense over longer periods of time and then
transmit their data less frequently. In contrast, a real time application (e.g., surveillance
monitoring) requires a maximum coverage while having a continuous connectivity. Application parameters are on of the main factors that influence the impact on performance and
lifetime of the network.

4.4

Hopping Sensor Routing Protocol

A Hopping Sensor Routing Protocol (HSRP) is a centralize approach based on the HSNM.
The base station decides the routing path of the packet within a time frame, and provides the

54

best energy preserving path for the packets. The decision making process involves multiple
stages and is depicted in Algorithm 4.1. Let Pm be the paths available from a node m
to the base station, with each path Pi consisting of a set of nodes with their respective
required jump heights for the given path. A node can participate in more than one path.
The remaining energy for a node k is expressed as ek , the energy consumed to jump to height
h is ejh , and et is the energy for transmitting.
The HSRP finds the minimum energy consuming path while not depleting any node
beyond a threshold τ of energy consumption. This threshold is the maximum desired operational energy cost (fraction of energy required out of their remaining energy) per node in a
path. This value is determined based on application needs and is set prior to execution. The
permissible amount of energy consumption of a node in a path is dependent on the residual
energy in the node. With aging of network life, the remaining energy of the nodes decreases
and will result in some nodes not meeting τ . In order to minimize the impact of this, the
algorithm, as last resource, selects the best among the high energy consumption paths.

55

Algorithm 4.1 Hopping Sensor Routing Protocol
Input: P athsi (n), ek , τ − Energy threshold
Output: Path(): selected path for all n nodes
m=1
while m <= n do
MaxEP athF rac = 1
Pm = Set of paths for node m
for all path Pi ∈ Pm do
for all node k ∈ Pi do
Ei = Ei + ejh + et
EnergyF raction(k) =

ejh + et
ek

if EnergyF raction(k) > τ then
P athV alidity(Pi ) = F alse
if EnergyF raction(k) > MaxEF raci then
MaxEF raci = EnergyF raction(k)
end if
end if
end for
end for
SortedPm = sort(P aths Pm on Energy E)
while (!P athF ound) do
Pi = SortedPm {next path}
if P athV alidity(Pi ) = T rue then
P athF ound = T rue
P athm = Pi
else if MaxEF raci < MaxEP athF rac then
P athm = Pi
MaxEP athF rac = MaxEF raci
end if
end while
P ath(m) = P athm
end while
56

4.4.1

HSRP Evaluation

The HSRP was simulated to evaluates its performance on network energy consumption.
The values obtained from the practical in-field experiments were used to determine the
communication ranges of the nodes. The base case for the communication was when the
sensor node jumped to 30 cm height. At this height, the IRIS Crossbow sensor is able to
communicate over a distance of 12 m as compared to ground level, where it could only
communicate up to 5 m without packet losses. At 5 m communication range most of the
nodes are completely disconnected, hence prompting us to use a sensor jump height of 30 cm
as the default base case. This was compared against the height when the sensor node can
jump either 30 cm or 60 cm. The decision of whether the sensor should jump to 30 cm or
60 cm was taken using HSRP. When the node jumps to 60 cm height, it is simulated to
consume double the amount of energy compared to when it jumps to 30 cm. Irrespective
of whether the node jumps to 30 cm or 60 cm, the transmission cost, in terms of energy,
remains the same as the node always transmits at the highest power (3 dBm). The debate
between the transmission cost and the hopping cost is out of the scope of this evaluation,
since without the hop, the resultant network is mostly disconnected and the sensed data
cannot be transmitted back to the base station.
Every node started with 1000 units of energy. The energy threshold value for the node
was a varying factor and was set to 0.005, 0.01, 0.02, 0.03, 0.04, 0.05. The simulations were
run over 100 time periods, and during each time period 20% of the nodes were randomly
selected to communicate.
Geographical routing protocol was used as base case to compare the energy consumption
performance of HSRP. It uses the shortest distance as the criterion metric to select the node’s

57

communication path to reach the base station. The energy consumption of the network was
evaluated over the complete time length of simulation, Figure 4.7.

Remaining Network Energy %

100
Geographical Routing
HSRP

95
90
85
80
75
70
65
60
55
0

20

40
60
Time Period

80

100

Figure 4.7: Network Energy Consumption
The energy consumed by the network employing HSRP is 20% lower compared to using
geographical routing. This is because HSRP calculates the improvement of path quality by
the nodes jumping to a higher elevation and selects paths with better quality. The quality
is defined as the amount of energy savings that the path provides to the network by not
depleting any node on the path with a significant portion of its remaining energy.
The second feature of critical importance is the energy balancing within the network. The
number of dead nodes was studied over the course of the 100 time periods. We observed that
the protocol balances the energy used in the network by the different sensor nodes. A node
with a lower amount of energy is used less frequently to forward a packet. This significantly
58

affects the nodes closer to the base station. For example: under certain circumstances, by
employing the protocol, nodes that are two hops away from the base station can hop up
to 60cm height and transmit packets directly to the base station, thereby preventing the
complete drainage of energy from nodes closer to the base station. The number of dead
nodes in the case of HSRP is significantly lower, 1/8 of when using geographical routing.
Also, without the use of the protocol, a large number of nodes closer to the base station
consumed all their energy by forwarding packets for other nodes that were farther away,
resulting in the reduction of connectivity, which affects the number of packets delivered.

59

4.5

Summary

In this chapter the jumping sensor network connectivity and area coverage are evaluated.
A Hopping Sensor Network Model (HSNM) has been introduced in order to increase the
connectivity of the sensor nodes with the base station, and thus, to increase the network
sensed area. Furthermore, a Hopping Sensor Routing Protocol (HSRP) is designed from the
HSNM with the main purpose of manage the energy consumption in the network. In fact,
the protocol chooses the best path in terms of energy consumption, avoiding the fast energy
depletion of active jumping sensor nodes. Simulation results showed that HSRP has lower
energy consumption rate than geographical routing protocol, giving a longer network utile
life.

60

Chapter 5
WIRELESS SENSOR NETWORK
CONNECTIVITY INCREASE
WITH JUMPING SENSORS

5.1

Preliminaries and Motivations

Network connectivity in a Wireless Sensor Network (WSN) is crucial for network operation.
When wireless sensor nodes are randomly deployed in a field, there may be areas that lack
sensor nodes. These areas tend to create communication holes, and in the worst case, break
the topology into disconnected clusters of nodes. Figure 5.1 depicts a WSN topology of
200m x 200m area with a single base station located at the middle of the bottom edge;
nodes with connectivity are marked with diamond shapes, and nodes without connectivity
are marked with asterisks. All the nodes are transmitting at maximum power, however due
to communication holes, only 31% network connectivity is achieved. From the figure, the
formation of clusters of nodes can be identified as those disconnected groups (from base
station) of interconnected nodes.
Jumping sensors can be used to increase connectivity in wireless sensor networks. However, a topology of jumping sensor nodes configured to enhance network connectivity by

61

200
180
160
140
120
100
80
60
40
20
0
0

20

40

60

80

100

120

140

160

180

200

Figure 5.1: WSN Topology at ground level (31% connectivity)
200
180
160
140
120
100
80
60
40
20
0
0

20

40

60

80

100

120

140

160

180

200

Figure 5.2: WSN Topology with gateway jumping sensors (100% connectivity)

62

means of jumping can significantly reduce the network life span, if not well managed. This
chapter presents an approach to deal with communication holes, to enhance the network connectivity using selected jumping sensors nodes located at the boundaries of clusters of nodes.
With the help of selected jumping sensors serving as gateways on disconnected clusters, 100%
of network connectivity is achieved in the same topology (Figure 5.1). This improvement on
network connectivity is depicted in Figure 5.2, where wireless links established by jumping
sensors are represented as dotted lines.
In this chapter, an overview to the jumping sensor network topology is given. A survey on
boundary node identification algorithms is provided. Two decentralized algorithms that use
the jumping capabilities of sensor nodes are designed for the discovery of neighbor clusters
of sensors, Boundary Node Aggressive Neighbor Discovery (aggressive) and Boundary Node
Smart Discovery (smart). Furthermore, energy balanced packet forwarding schemes with the
use of jumping sensor gateways are evaluated for cluster-to-cluster (C-to-C) communication.

5.1.1

Chapter Organization

This chapter is organized as follows. Jumping sensor topology is presented in Section 5.2.
Boundary nodes identification algorithms are reviewed in Section 5.3. Node discovery algorithms are defined in Section 5.4. Evaluation of the algorithms and analysis of network
performance employing C-to-C communications are presented in Section 5.5. A chapter
summary and conclusions are provided in Section 5.6.

63

5.2

Topology

The deployment phase and characteristics of the geographic area impacts a wireless sensor
network topology greatly. A common method of sensor deployment is to locate sensors
precisely in the field to guarantee sensor connectivity with a base station, while maximizing
the covered area. This method becomes infeasible when the network scales to cover difficultto-reach or hazardous areas, e.g., volcano monitoring discussed by Song et al. (2009). As
an alternative, sensor motes are deployed from the air; a technique sometimes referred as
air-dropped-sensors, where sensors are carried and dropped by a helicopter, or an airplane
to the target area. This technique gives the opportunity to reach those difficult-to-reach
or hazardous areas. However, there is chance for sensors to land into segregated groups of
sensors that may not have a connectivity path to the base station.
In this work, a cluster of sensor nodes is defined as a group of interconnected nodes
relying on their wireless communication range at ground surface level, and packet routing
capabilities. A cluster can have direct connectivity, i.e., one or more nodes from the cluster
can reach directly the base station, and provide connectivity to others within the cluster.
Similarly, the connectivity propagation concept applies between clusters. A cluster with
direct connectivity will provide connectivity to its neighbor clusters, and those will continue
in turn to form a cluster-to-cluster (C-to-C) path to base station.
With the existence of clusters in the topology, sensors nodes that lay on the edge (frontier) of a cluster of nodes are classified as boundary nodes. These nodes will help in
the discovery of neighbor clusters and form bridge nodes, i.e., those boundary nodes that
can successfully communicate with nodes from other clusters, by employing jumping sensor
airborne communication. There may be more than one bridge node that shares the same

64

intended cluster receiver. Moreover, a cluster may be able to reach base station trough
different clusters. In order to maximize network functional lifetime, a cluster can distribute
the packet forwarding task among bridge nodes, by selecting a subset of bridge nodes as
gateways to perform C-to-C packet forwarding.

5.3

Boundary Nodes

Cluster’s boundary nodes provide the highest successful discovery probability because of their
position relative to others. Therefore, selecting only boundary nodes to perform discovery,
increases the probability of discovery, and reduces the network energy cost. The identification
of boundary nodes has been addressed by Sahoo et al. (2007), and Fayed and Mouftah (2009).
Two kinds of boundary detection algorithms were introduced by Sahoo et al. (2007):
sequential (SBNS) and distributed (DBNS). Furthermore, using these boundary nodes, a
target detection protocol is used to know the entry and the exit of single target through
the monitoring region. In selecting the boundary nodes, the two algorithms have similar
performance. The main differences between them is the lower communication cost of DBNS
compared to SBNS due to the less amount of information exchange and the faster selection
procedure in DBNS regardless the increase of network size.
Fayed and Mouftah (2009) proposed a local convex view (lcv) as a means to identify nodes
close to the boundary. It is a localized algorithm and can establish neighborhood coordinates
if location information is not prior available. To check if a node is on the boundary, a node is
verified to lie on the convex hull built with its one-hop neighborhood. The most important
result is that lcv seems unaffected by position estimation error because of its geometric
properties. However, it has difficulties identifying boundary nodes under special scenarios

65

where there are areas without sensors (communication holes) within the topology.
Boundary nodes have shown to be beneficial in wireless network topologies. They have
been used to help other nodes in the localization process on wireless sensor networks, Du
et al. (2007). Another approach presented by Aissani et al. (2008), employs boundary nodes
to avoid routing packets to dead-end routes. Han and Poor (2009) introduced a method
based on coalition games, in where boundary nodes are used to cooperate with nodes in the
middle of the network to enhance network performance and connectivity.

5.4

Discovery

When a jumping node jumps, it has the chance to reach farther nodes, by employing airborne
communication (see Section 3.3). Following this approach, jumping sensors can be used to
discover and establish communication with other sensors that are far and without base
station connectivity.

5.4.1

Boundary Node Aggressive Neighbor Discovery

Boundary Node Aggressive Neighbor Discovery is a decentralized algorithm that utilizes all
the jumping sensor boundary nodes (BNs) from clusters, to jump and discover surrounding
neighbor nodes belonging to other clusters.
In order to propagate the connectivity on the wireless network, this process begins on the
cluster with direct connectivity, and continues on other clusters as they become discovered,
in a sequential manner.
Every BN within a cluster will perform the cluster discovery, which consists of hopping
(jumping), transmitting a discovery message, and listening for any reply while airborne. In

66

order for a BN to perform cluster discovery, it has to follow the following phases: Pre-hop,
Hopping, Post-hopping.

5.4.1.1

Pre-hop

Before a BN jumps, it has to coordinate with its one-hop1 away BN neighbors, to avoid
these BNs to jump at the same time and interfere with each other.
Every BN waits for a random amount of time before trying to jump and discover. Once
this time is expired, the BN sends a Hop Advertise packet to its one-hop BN neighbors,
including its NID. Those BNs receiving this packet, reply with a Hop Ack packet, including
their node identification (NID) and the NID of the BN initiating the discovery. These BNs
constrain themselves to perform hopping discovery while the current discovery is taking
place. Upon receive of these Hop Ack packets, the BN discovery initiator verifies if all its
BN one-hop neighbors replied and proceeds to hop.
One or more BN neighbors may fail to generate the Hop Ack packet, due to a packet
loss or due to competition for hopping at the same time. To address the case of missing
acknowledgement due to packet loss, the BN initiator waits a random amount of time and
resends a Hop Advertise packet to those missed BN neighbors, who in turn will reply with
a Hop Ack packet.
A similar procedure applies if the BNs were competing for discovery. Each competing BN
is set to wait before attempting to re-advertise, to provide a chance for the other competitor
to generate the Hop Advertise packet first. Upon receipt of the advertisement packet, any
competitor yields and replies with a Hop Ack packet. The winning BN proceeds to hop, and
those BN neighbors that exclusively received the advertisement packet from BN competitors
1 One-hop communication refers to direct communication between two nodes.

67

and not from the winning BN, will reset their wait timers to start competing among them.

5.4.1.2

Hopping

While the BN is airborne, it transmits a Discovery Msg packet, including its CID and connectivity information, i.e., direct connectivity, or the number of intermediary clusters to reach
the base station. Any BN belonging to another cluster that is able to receive this packet
will attempt to reply with a Discovery Reply packet, including its CID and connectivity
information.

5.4.1.3

Post-hopping

BNs on each cluster with discovery information become bridge nodes, and advertise it to
the BNs of the cluster with a Bridge Advertise packet, including its NID, node remaining
energy, neighbor CID, and connectivity information of the neighbor cluster. In addition,
clusters that are discovered will start to perform discovery, to propagate connectivity.
After the current discovery process finishes on a BN, the BN neighbors resume their wait
timers for discovery. Those BNs that performed the discovery will not compete for discovery
until a new discovery process is triggered, i.e., in the event a cluster gets divided due to
communication holes resulting from missing nodes. It is worth noting that after a jumping
node jumps, it may be possible (depending on the geographic area) to land in a position
within the communication range of the two or more clusters at surface level; in such case,
clusters are merged into one.

68

5.4.2

Boundary Node Smart Neighbor Discovery

In order to reduce the energy consumption of the discovery process, the discovery algorithm
is enhanced to find surrounding neighbor clusters with a subset selection of BNs.

Figure 5.3: Discovered Area Gain of BNs Subset Selection for Discovery

The approach relies on the fact that once a BN is selected to perform discovery, its
one-hop BN neighbors will not contribute much to expand the discovered area. Figure 5.3
depicts the gain in discovered area by only selecting BN that are not one-hop neighbors
of the BNs already selected to perform jump discovery. The communication range of BN
jumping sensors at ground level is represented by a solid line circle, while the communication
range when hopping (jumping) is represented with a dotted line circle.2 From Figure 5.3 it is
also depicted the resulting missed area by not selecting the one-hop BN between the selected
69

BNs. However, the contribution to the communication area of the missed area, in most of
the cases, is minimal.
In order to provide a BN subset selection, the process begins from the cluster having
direct connectivity, where all BNs compete to perform discovery, similarly to the Aggressive
Neighbor Discovery, discussed in Section 5.4.1. The main differences are on the BN behavior
during the Pre-hopping and Post-hopping phases.
In the Pre-hopping phase, once a BN sends a Hop Advertise packet, all its one-hop BNs
will not compete for further discovery until another discovery event is required. BNs that
did not receive an advertisement packet will continue competing for discovery by repeating
the process. In this manner, a subset of BNs is selected to perform the discovery.
During the Post-hopping phase, if the discoverer BN does not reach any other cluster,
it sends a Discovery Null packet to its one-hop BN neighbors, and stops competing to do
further discoveries until a new discovery event is triggered. Upon receipt of this packet, onehop BN neighbors will reset their wait timer to compete for discovery. On the other hand, if a
cluster is discovered, the BN advertises the discovery to all the BNs with the Bridge Advertise
packet. Those one-hop BN neighbors of the discoverer BN will stop competing for discovery.
For the discovered cluster, the set of BNs able to successfully receive the Discovery Msg
packet become bridge nodes. These nodes advertise their new status to all the BNs with
the Bridge Advertise packet. BNs that are a one-hop neighbors of BNs that became bridge
nodes will refrain from discovery until a new discovery process is triggered.
2 Figure 5.3 is meant for visualization purpose only and is not drawn to scale. Jumping

communication range has been measured to increase by a factor of 6 when hopping (see
Chapter 3).

70

5.5

Simulation Results and Analysis

Simulations were performed to test the discovery algorithms in terms of energy consumption
and connectivity propagation for a jumping sensor network. The life span of the network was
tested with different configurations of C-to-C packet forwarding. Experimental data results
were used as input for the communication ranges, 10 meters ground level communication,
and 30 meters for airborne communication (see Section 3.4). The airborne communication
range limit of 30 meters was selected as a conservative measure to account other factors,
such as terrain and weather conditions, that can affect wireless communication.

5.5.1

Experimental Setup

Topology: Twenty (20) sensor network topologies were generated representing a geographic
area of 200m x 200m. Nodes were positioned following a random uniform distribution. Each
node was simulated to provide sensing coverage to a circular area with a 7 m radius. The
number of nodes in the network was selected to provide a sensing coverage of 3 nodes per
unit area (1 m2 ), resulting in 781 nodes (all jumping sensors). A base station node was
located in the middle of the bottom edge of the square topology acting as a sink.
Energy: Each node was simulated to have the same amount of energy as the jumping
sensor prototype built by Zhao et al. (2009). It uses a 3.7 V , 100 mAh LiPo battery, for a
total stored energy of 1332 J. Their prototype can make jumps up to a height of 0.353 m,
consuming 882.6 mJ of energy for each jump. A similar design of a jumping sensor able
to jump 1 m height, is estimated to consume 1.775 J of energy. For simulation purposes,
we considered an energy consumption of 2 J for the jump operation. Other node’s energy
consumption operations were determined following IRIS data sheet specifications as follow:

71

energy to transmit a packet of 36 bytes at maximum transmission power (3 dBm) is 0.09 mJ,
and the energy to receive a packet is rounded to be the same as to transmit, since the different
is relative small.

5.5.2

Discovery Algorithm Performance

In order to study the discovery algorithms performance, energy consumption and connectivity propagation were analyzed on the 20 network topologies after initial discovery. The
algorithms were compared with all jumping sensors randomly employing individual discovery
(referred as ALL). Table 5.1 presents the mean connectivity of network after the discovery
phase, and the mean network energy cost in Joules for each algorithm. It can be appreciated that as the discovery algorithms become smarter with the selection of nodes to perform
discovery, the energy consumption reduces without compromising connectivity.
Table 5.1: Discovery Algorithm Performance
Algorithm
ALL
AGGRESSIVE
SMART

5.5.3

Connectivity %
100
100
100

Energy Consumption (J)
1562.0
670.5
330.6

Network Performance

The network performance was analyzed over 5000 time period events for each of the network
topologies. For every time period, a percentage of the nodes with connectivity was randomly
selected to generate packets. Simulations were run separately for five different percentages
of nodes generating packets (communication loads): 20%, 10%, 5%, 2%, and 1%.
Boundary C-to-C packet forwarding was employed with jumping sensor acting as gateway
nodes to reach the base station. A delay transmission protocol was employed for jumping
72

sensors airborne transmission. Packets arriving at gateways (nodes that need to jump and
transmit) were buffered until transmission buffer became full, but not for more than 3 time
periods after arrival to the node, then transmitted airborne.
Two configurations were simulated to evaluate the network performance employing C-toC packet forwarding: single gateway per cluster, and multiple gateways per clusters. They
were compared with packet forwarding without gateways (no-cluster organization). Single
gateway per clusters is an energy preserving approach, in where every discovered cluster
elects a single gateway node to forward packets to neighbor clusters, to reach the base station.
Shortest path and remaining node energy were used as the gateway selection criteria, smallest
NID was used to break any tie. Multiple gateways per cluster is an extension of the single
gateway approach, but less energy conservative. It provides alternative routes to nodes in a
cluster to obtain shorter paths length to reach base station trough different gateways. With
packet forwarding without gateways, every jumping sensor node takes advantage of reaching
the farthest nodes with airborne communication to attempt shortest paths routes to base
station. It is worth noting that if a jumping sensor is not able to shorten the path with
airborne communication, then it employs ground level communication.
Each C-to-C packet forwarding configuration was evaluated with three metrics:
1. Remaining Network Energy , mean of the remaining energies in all the nodes, at
the end of simulation, with respect to the total initial network energy.
2. Remaining Network Connectivity , percentage of nodes left with connectivity to
base station.
3. Mean Packet Delay Time, mean of the time periods that each packet required to
arrive at base station.
73

Figure 5.4: Remaining Network Energy for Different Communication Loads

Figure 5.5: Remaining Network Connectivity for Different Communication Loads

74

Figures 5.4 and 5.5 present the remaining network energy and connectivity results, respectively, for the different communication loads after 5000 time periods. Single gateway (S.
GTW) C-to-C packet forwarding shows to have a remarkable lower energy consumption, due
to the fact that for any time period the number of sensors jumping to forward packets in
the network is bound by the number of clusters without direct connectivity, thus, consuming
less energy. Multiple gateways (M. GTWs) C-to-C packet forwarding consumed more energy, but proved to be more efficient than packet forwarding without gateways (No GTWs).
Connectivity results show to be highly correlated with network energy for a jumping sensor topology; as more nodes deplete their energy, other dependent nodes are left without
connectivity.

Figure 5.6: Mean Packet Delay Time for Different Communication Loads

A trade-off in energy efficiency and packet delay time is evident in Figure 5.6. Packet
delay time for Single GTW showed to be constant and slightly higher than the others, for

75

communication loads 1%, 2%, and 5%, whereas rapidly increasing for higher communication
loads, due to network saturation and bottleneck effect created from gateways. Multiple
GTWs is more resilient to delay incurred with increased communication load, since it provides
alternative routes to nodes within each cluster.

100

Network Energy %

90
80
70
60
50
40
30
20
10
0
0

S. GTW
M. GTWs
No GTWs
1000

2000
3000
Time Period

4000

5000

Figure 5.7: Network Energy Consumption for 20% Nodes Generating Packets

The energy consumption rates of the network is evaluated for the different communication
loads. Single GTW packet forwarding shows the smallest energy consumption ratio, followed
by Multiple GTWs packet forwarding. For 20% communication loads (Figure 5.7) is noted
that after 3000 time periods there a is an increase in the energy consumption rate for Multiple
GTWs. This change in rate is because more gateways are being used concurrently due to high
packet traffic on the network. The energy consumption rate is reduced later (around 3500
time period), which is a result of the successful reconfiguration of the network to allocate new
gateways. With packet forwarding without gateways (No GTWs), most of the network nodes
have depleted their energy by this time. However, the network energy is not shown to reach
76

100

Network Energy %

90
80
70
60
50
40
30
20

S. GTW
M. GTWs
No GTWs

10
0
0

1000

2000
3000
Time Period

4000

5000

Figure 5.8: Network Energy Consumption for 10% Nodes Generating Packets

100

Network Energy %

90
80
70
60
50
40
30
20

S. GTW
M. GTWs
No GTWs

10
0
0

1000

2000
3000
Time Period

4000

5000

Figure 5.9: Network Energy Consumption for 5% Nodes Generating Packets
zero since for simulation purposes, a node was consider dead after reaching the threshold
energy value of 5J. A similar pattern is seen for 10% communication loads (Figure 5.8) with
slightly lower energy consumption rate. The same increase in energy consumption rate is

77

seen again for Multiple GTWs and No GTWs, but occurring later in time (around 4250 time
period) due to the less packet traffic generation. The energy consumption rates for lower
communication loads (Figures 5.9, 5.10, and 5.11) were comparatively lower to the others,
as expected.

Network Energy %

100
80
60
40
20
0
0

S. GTW
M. GTWs
No GTWs
1000

2000
3000
Time Period

4000

5000

Figure 5.10: Network Energy Consumption for 2% Nodes Generating Packets]

Network Energy %

100
80
60
40
20
0
0

S. GTW
M. GTWs
No GTWs
1000

2000
3000
Time Period

4000

5000

Figure 5.11: Network Energy Consumption for 1% Nodes Generating Packets

78

5.6

Summary

This chapter discussed the topology and operation of a jumping sensor network. It defined two decentralized algorithms for the discovery of isolated nodes, aggressive and smart
discovery, to propagate base station connectivity. Both algorithms rely on the usage of
jumping sensors on the boundary of a cluster, without prior topology information. The
efficiency of the algorithms is studied in terms of network energy consumption and connectivity propagation. Simulation results from both algorithms showed to increase the total
network connectivity to 100% for each of the evaluated topology scenarios. Also, results
showed significant savings in network wide energy consumption for discovery tasks, 57% energy saving for aggressive discovery and 79% energy saving for smart discovery. Moreover,
cluster to cluster (C-to-C) packet forwarding schemes relying on boundary jumping sensor
gateway nodes are defined and simulated, showing to have remarkable low network energy
consumption through network utility life.

79

Chapter 6
WIRELESS SENSOR NETWORKS
AND TOPOLOGY
REDEPLOYMENT

6.1

Preliminaries and Motivations

Mobile wireless sensor networks are often faced with redeployment-decision problems. Whenever a sensor node lacks persistent connectivity with neighboring nodes, a decision must be
made to either relocate, increase signal strength, or redeploy sensor nodes as an attempt to increase network connectivity. Wireless sensor topologies usually start using node’s maximum
transmission power during network deployment and setup, Ramanathan and Rosales-Hain
(2000). However, network connectivity issues persist mostly because of power-limited communication radios in the sensor nodes and the presence of line-of-sight obstacles reducing
effective communication ranges.
Although specialized types of sensor motes, i.e., jumping sensors, can enhance network
connectivity, in order to have a functional network it is important to address connectivity
issues with respect to the application to be served. For instance, network intensive applications, such as mission-critical security surveillance, might require persistent connectivity

80

among sensor nodes. In such applications, redeployment of sensors is an alternative to
sustain the networking demands. On the other hand, some applications can withstand intermittent connectivity (delay tolerant), meaning that communications between node pairs do
not require long lasting wireless communication links. Instead, the communication occurs
for short periods of time during the network lifespan. For delay tolerant applications, the
network cost for redeployment could exceed the communication gains provided by jumping
sensors.
The network connectivity metric is not enough to represent the network performance
potential, i.e., a network evaluation indicative of network suitability for application’s networking requirements. For example, a network performance metric score should reflect the
existence of intermittent (non-persistent) vs. persistent communication links between nodes
and base station, e.g., the more nodes connected to the base station via intermittent links,
the lower the metric score and the less suitable for network intensive applications. Therefore,
an evaluation metric representing the fitness of a WSN topology in terms of the quality of
its connectivity is required.
This research is focused on analyzing network topologies established with the use of jumping robots sensors, and to defining a Quality of Connectivity (QoC) network performance
metric. Moreover, since efficient node redeployment is a key aspect for enduring sensor networks, the QoC is adapted as an optimization metric for node redeployment. Redeploymentdecision algorithms with different approaches to enhance network performance are presented.
A discussion of the obtained results from evaluations and performance comparisons is provided.

81

6.1.1

Chapter Organization

First, Section 6.2 provides relevant definitions and outlines network connectivity derived
problems. Second, Section 6.3 is dedicated to the discussion of jumping sensor airborne
communication and network performance aspects from the application perspective. Third, in
Section 6.4 we describe the network topology as a network fitness formula considering nodes’
quality of connectivity. Fourth, Section 6.5 discusses cluster merging algorithms aimed to
maximize network performance. Fifth, Section 6.6 presents results from evaluations and
performance comparisons for the proposed algorithms. Finally, Section 6.7 summarizes and
concludes this chapter.

6.2

System Model

In this chapter we consider single sink node topologies where sensor nodes collect field readings and forward them in communication packets to a sink node. We define a node’s connectivity as the ability to communicate with a sink node, directly or through multi-paths
formed by neighbor nodes. For the remainder of this section we refer to the sink node as the
base station node.
Node’s connectivity is one of the factors having the biggest impact on the network performance. If a node cannot forward the field readings to the base station, then it has to wait
until a communication path becomes available. Newer communication paths become available as sensor nodes get discovered by others. Depending on the application requirements,
field readings might become outdated by the time a path to the base station is discovered.
Hence, it is imperative to provide connectivity to as many sensor nodes as possible.
Even though connectivity is crucial for network performance, other factors have to be
82

taken into account when evaluating a network topology. For instance, a fragmented topology
will impact network responsiveness and the capacity to meet application’s flow demands.
On the other hand, bad network configuration and bad reallocation decisions are likely to
deteriorate network utility life.

6.2.1

Connectivity and Sensor Clusters

Network connectivity is evaluated as the ratio of nodes having connectivity with the sink node
out of the total number of deployed nodes. Consider a network represented as a simple graph
N with |N| nodes. We define the nodes with network connectivity as those that belong to
the connected component containing the base station. Call this connected component CB ,
and |CB | the number of nodes having connectivity. Hence, network connectivity can be
defined as Eqn. (6.1).
Network Connectivity =

|CB |
|N|

(6.1)

The connectivity of a node depends on the existence of communication paths to the base
station formed by sensor nodes. Sensor nodes that are within their wireless communication
radius establish link pairs to form communication paths. However, these link pairs are not
always persistence, mostly due to their wireless nature, which creates an impact in node’s
connectivity.
There are a number of scenarios that can disrupt wireless communication links, two of
them are: long distances between node pairs, and obstacles in the network topology. It
is known that wireless signal strength fades with the increase of the propagation distance.
When nodes are randomly deployed in the field, e.g., air-dropped Song et al. (2009), there
is hardly any control on their initial distribution. The problem with this is that some nodes

83

end up too far apart from each other, thus “excommunicating” them from the network. On
the other hand, the presence of obstacles can cause disruption in the signal propagation,
and in some scenarios prevent mobile nodes from getting closer, e.g., river streams, lakes,
boulders, etc.
It is common for a network topology to start divided in clusters of nodes. We define
a cluster as a subset of one or more nodes from the network having stable (long lasting)
communication links. In other words, by removing intermittent links from the network, the
resulting graph’s connected components are exactly the clusters. Many clusters are composed
of sensor nodes distant from the base station. Naturally, in a single base station topology any
cluster, other than the one where the base station is, will have intermittent connectivity or no
connectivity all. Therefore, the number of clusters in the network can have an impact in the
overall connectivity, i.e., persistent (desired in rapid response applications) vs. intermittent
(good for delay tolerant applications).

6.3

Network Performance

We wish to obtain the highest performance possible from a sensor network topology. This
section is dedicated to the discussion of jumping airborne communication and network performance aspects from the application perspective.

6.3.1

Jumping Airborne Communication

Jumping sensor robots allow relocation in rugged terrains where conventional wheeled sensor
robots cannot perform. In addition, jumping sensors while airborne can establish communication links between distant sensor nodes. This is possible because the change in altitude

84

clears the communication line-of-sight allowing the transmitted signal to propagate farther,
thus, enabling communication with distant receivers Cintr´n et al. (2009, 2012).
o
The availability of airborne communication links is dependent on the time window (Tc )
available to communicate during the total time of flight, which is dependent on the threshold
height (HT h ) the sensor must reach for communication and the maximum jump height
(Hmax ) that it can reach. This relation is expressed in Eqn. (6.2) where g is the gravitational
acceleration.

Tc = 2

2(Hmax − HT h )
g

(6.2)

Additionally, we have to consider how long it takes for a jumping robot to perform a complete
jump cycle, Eqn. (6.3). This includes any time required for the jumping robot to prepare
before a jump (e.g., loading of jumping mechanism, self righting and positioning), and the
time of flight obtained from the jump.

Tcycle = Tpre−jump + TT oF

(6.3)

For purpose of analysis we consider the MSU Jumper, a miniature steerable jumping
robot prototype presented by Zhao et al. Zhao et al. (2013). The prototype has a jumping
loading time of 10 seconds, with an estimated normalized jump height (height reached from
vertical jumping) of 87 cm and 73.6 cm when carrying 4 grams and 8 grams of payload
weight, respectively. Figure 6.1 plots both, the jump cycle as a function of time in Fig. 6.1a,
and a zoom-in of the time of flight (TT oF = 775 ms) when jumping to a height of 73.6 cm
in Fig. 6.1b. The communication time window (Tc = 720 ms) is displayed for a threshold
height of 10 cm.
85

0.8
0.7

Height (m)

0.6
0.5
0.4
0.3
0.2
0.1
0
0

2

4

6
Time (s)

8

10

12

(a) Jump Cycle

0.8
0.7

Height (m)

0.6
0.5
Tc Start
Tc End
Hmax = 0.736 m

0.4
0.3
0.2
0.1
0

10.1

10.2

10.3

10.4 10.5
Time (s)

10.6

10.7

10.8

(b) Communication Time (Tc ) Above 10 cm Threshold Height

Figure 6.1: Jumping Robot Jump Cycle and Time of Flight for 73.6 cm Jump Height

86

From Fig. 6.1a, it can be observed that the communication time (Tc ) is small relative
to the entire jump cycle time (Tcycle ) for this particular jumping robot prototype when
carrying a load weight of 8 grams. The jump cycle time can be a determining factor for the
application’s suitability for airborne sensor communication with jumping sensors.
Let us consider a single jump for communication, then the maximum transmission rate
during a jump will be equal to the link speed RT x obtained from the employed wireless
protocol. However, this rate is conditioned to the communication time window during the
jump. Therefore, it is practical to compute the number of communication frames that can
be transmitted during a jump. This can be obtained as follow:

tf rame =

Fbits
RT x

(6.4)

Fjump =

Tc
tf rame + tdelay

(6.5)

where,
tf rame time it takes to transmit a frame.
tdelay inter frame delay time for synchronization between transmitter and receiver.
Fbits number of bits in a communication frame.
RT x transmission rate (link speed) in bits/sec.
Fjump maximum number of frames that can be transmitted per jump.
For instance, using the following networking parameters: frame size of 25 Bytes (10 Bytes
Payload, 6 Bytes physical layer and 9 Bytes MAC sublayer 802.15.4-IEEE-Standard (2006)),
87

RT x = 250 Kbps, tdelay = 7.5 ms, and assuming airborne communication can occur with
same jump height values as presented in Fig. 6.1b for a Tc = 720 ms. Replacing these values
in Eqn. (6.4) and (6.5), the maximum number of frames that can be transmitted per jump
is 86.77 [f rames/jump].
Although the communication time is small when compared to the jump cycle time, it is
enough for short communications. On the other hand, communication that occur with high
frequency or exceeding the available communication time per jump can suffer from delay.
The average frame rates for ground surface (RGx ) and airborne communication (RJx )
with jumping sensors can be obtained with Eqn. (6.6) and (6.7), respectively.

RGx =

1
tf rame + tdelay

(6.6)

RJx =

Fjump
Tcycle

(6.7)

Considering the previous networking parameters, an average frame rate for sensor within
communication range at ground surface level is found to be RGx = 120.48 [f rames/sec].
However, the average frame rate for airborne communication is found to be RJx = 8.05
[f rames/sec]. This lower frame rate is caused by the time it takes for a jumping robot to
complete a jump cycle. The jumping cycle causes packet transmissions to occur in bursts,
resulting in increased network communication latency. It should be noted that these rates are
for the specified jumping characteristics of the MSU Jumper, which is designed to maximize
both jumping height and travel distance.

88

6.3.2

Gateway Airborne Communication

Jumping sensors topologies can employ jumping capable sensors as gateway nodes to disconnected clusters, Cintr´n and Mutka (2010). The jumping frequency of a gateway jumping
o
sensor is determined by the packet forwarding rate the node must achieve. This is related to
the packet arrival rate, which is application dependent. To determine the viability of gateway jumping sensors for a particular application, it is necessary to predict the total packet
rate RCi for clusters.

RCi = ri ∗ ni

(6.8)

where,
RCi the total packet rate for cluster i.
ri the average packet output rate for sensing nodes in cluster i.
ni the number of active nodes in cluster i.
If the jumping node can buffer and transmit Pjump packet in a jump, then it can wait
a maximum of tw before having to forward the corresponding data packets. This relation is
expressed below.

tw =

Pjump
seconds
RCi

(6.9)

Moreover, given an application messaging frequency, the number of sensor nodes a jumping sensor gateway can support can be determined with Eqn. (6.8) and (6.9).

89

Case Study I : Consider an environmental sensing application performing soil moisture
monitoring every 10 minutes ( ri = 0.0017 [pck/s]) as presented by Wu and Liu Wu and
Liu (2012). Each jumping sensor having to wait 10 seconds to perform a jump, and a
Pjump = 86 [pck/jump]. Solving Eqn. (6.9) for the number of nodes, we obtain:

ni =

Pjump
ri ∗ tw

ni =

86
≈ 5058 nodes
0.0017 ∗ 10

(6.10)

In other words, to prevent buffer overflow, under the above mentioned networking characteristics, a single jumping sensor gateway could serve up to 5058 sensing nodes.
Case Study II : In a Moving Object Tracking application, the monitored area is typically
divided into sensing regions Lin et al. (2006). Event messages are sent when an object
exits or enters a sensing region. The event message is a short (about 10 Bytes) message
containing a time stamp, object ID (identifier), exiting region ID, and entering region ID. For
continuous monitoring all event messages are forwarded to the base station. We now consider
an example where a sensor network tracks people running in an area using sensors with a
sensing radius of 5 [m]. A runner with a speed of 20 [Km/hr] will have an area transition rate
of 0.56 [T ransitions/sec]. This results in an event message rate of 1.12 [msg/sec]. While
the average ASC frame rate RJx = 8.05 [f rames/sec] would support the ability to track 7
objects simultaneously, the latency introduced by the jump cycle results in outdated position
information. However, this is not a problem for applications studying animal behavioral
patterns, not requiring position information in real-time, since the data is aggregated for
later evaluation Dyo et al. (2010).
90

It has been shown that the number of sensor nodes, the packet rate and delay tolerance
from the application are key factors to determine whether ASC is going to be able to sustain
the communication load of a cluster. Moreover, jumping sensors have limited power, hence
ASC should be used to provide temporary communication while topology redeployment takes
place to establish long lasting communication links. The network redeployment process is
discussed in the next sections.

6.4

Quality of Connectivity

Network connectivity measures the ratio of nodes able to connect to the base station. However, with this piece of information alone we cannot say much about how the network will
perform under different scenarios, or if it is adequate for a particular application. Hence,
we need a measurement to describe the type of connectivity the network can provide in its
current topology state.
The Quality of Connectivity (QoC) evaluates the network connectivity ratio in terms of
how segregated and distributed are the sensor nodes with connectivity in the topology. In
principle, cluster to cluster communication suffers from intermittent links. The higher the
number of these intermittent communication links on a path to base station, the higher is
the expected delay for a message to arrive at the base station. Therefore, QoC looks at the
entire network topology as levels of connectivity.
The connectivity level of a cluster can be determined with the number of intermittent
links a message must traverse to arrive at the base station. The cluster containing the base
station is considered the first level. One level up are those clusters having at least one
intermittent communication link to the first level. Levels above depend on their immediate

91

lowest level to acquire connectivity, as shown in Fig. 6.2.

C6

C7
C3

C5

C8
C4

C2
C1
L3

L2

L1

L1

L2

L3

Figure 6.2: Single base station clustered topology depicting connectivity levels

To be able to obtain a score for a network topology we have to assign a scoring contribution weight value to each level on the topology. Lower levels are desired, since they are
expected to have lower delay than higher up levels. Therefore, they should contribute more
to the overall network score. This is achieved by assigning a weight that decreases as the
level increases. Also, the level contribution to the QoC has to be proportional to the number
of nodes on it.
We use a geometric distribution to assign diminishing weights to each connectivity level
away from the base station, Eqn. (6.11).

λi = r i

92

(6.11)

1
In order to form a probability distribution we chose a common ratio, r = 2 . This satisfies:

k
k→∞

r
=1
1−r

ri =

lim

i=1

(6.12)

This probability distribution will be applied to the connectivity levels. We define the connectivity levels to be subsets of CB such that:
ℓ

Li

(6.13)

|Li |

CB =

(6.14)

i

where,
ℓ

|CB | =
i=1

There are ℓ connectivity levels indexed as L1 , . . . , Lℓ . Each Li contains |Li | nodes. The
geometrically weighted contribution of each level is represented as:
ℓ

λi |Li | = λ1 |L1 | + λ2 |L2 | + · · · + λℓ |Lℓ |

(6.15)

i=1

In order to provide a representative score of how fragmented the network topology is,
an average over the number of connected nodes is computed. However, because the number
of connectivity levels is finite, the sum of the weights is less than one. For this reason, we
normalize each weight and take the average using:
ℓ

|CB |

λi

(6.16)

i=1

This introduces the network QoC score, Eqn. (6.17). The QoC score is near one (zero) if and
only if the nodes having connectivity are distributed in few (many) clusters. A score value
93

of one is equivalent to having all the nodes with connectivity in a single cluster network.
ℓ

QoC =

i=1

λi |Li |

|CB |

6.4.1

(6.17)

ℓ
i=1

λi

QoC Enhancement

One of the network managing tasks is to identify node’s actions that can increase or reduce
the overall performance. For example, the decision of relocating a set of nodes to merge
two clusters. Such operation creates an impact in network expected life as the sensor nodes
consume more energy while relocating, but can potentially increase network performance.
Therefore, we want to evaluate the potential performance gain before any cluster merging
operation.
Network QoC allows to evaluates the network topology on its current state. The contribution to the QoC from each level depends on the number of nodes on the level, and the
contributions from levels above them. Hence, a merging operation between two clusters in
different connectivity levels represents a QoC increase. This is true because the merging
operation increases the lower level size, and brings dependant clusters from above levels one
level closer to the base station.
c
A cluster contribution to the QoC can be represented as a recursive function QoCi ,

presented in Eqn. (6.18),
c
c
QoCi = λi (|Lc |) + QoCi+1
i

(6.18)

where i indicates the current connectivity level the cluster c is at, |Lc | the size of the level
i
considering only those dependent clusters of c. The recursion will stop at the last dependant

94

c
c
connectivity level of cluster c, where QoCi+1 = 0. The QoCi function results in a rational

number that has to be normalized before it can be used to evaluate which cluster merge
operation has the greatest impact on the QoC.
Since we are using a geometric distribution (Eqn. (6.11)) to assign the connectivity levels
weights λi , we can simplify Eqn. (6.18) to represent all level contribution as:

Qc = λ1 (|Lc | + Qc )
i
i
i+1

(6.19)

c
Then, the QoCi contribution for i ≥ 2 is obtained by:

c
QoCi = λi−1(Qc )
i

(6.20)

Equations (6.19) and (6.20) allow for a distributed computation of each cluster contribution, thereby reducing the computational load on a single cluster.
To better understand the process of using QoC as a decision factor for cluster merging,
let us consider the example topology presented in Fig. 6.3. Clusters’ sizes for the example
topology are indicated in Table 6.1.

Table 6.1: Cluster’s Sizes and Distribution by Connectivity Level of Network Topology in
Fig. 6.3
Li
Cj
|Cj |
|Li |

L1
C1
2
2

L2
C2
15

C3
21
36

95

C4
30

L3
C5
8
42

C6
4

L4
C7
20
20

C7
C5
L4

C4
L3

C6
C2

C3

L2
C1

∗
Figure 6.3: Network Topology Example for QoC2 Contribution Evaluation

Using the values from Table 6.1 in Eqn. (6.17) we can obtain the QoC of the topology
on its current state.

QoC =

λ1 (2) + λ2 (36) + λ3 (42) + λ4 (20)
100

4
i=1

λi

= 0.176

We want to evaluate which cluster, C2 or C3 , merging with C1 will result in the highest
QoC increase. For this particular scenario, the number of nodes that will benefit from either
merge operation is equal, which is:

|C2 | + |C4 | + |C5 | = |C3 | + |C5 | + |C6 | + |C7 | = 53

(6.21)

This was selected to demonstrate that the QoC is not only impacted by the number of nodes,
96

but also by how they are distributed in the topology.
c
2
3
Evaluating both cluster’s QoCi contributions, QoC2 and QoC2 , we obtain:.

2
QoC2 = λ2 (15) + λ3 (38))

(6.22)

= 8.5

3
QoC2 = λ2 (21) + λ3 (12) + λ4 (20)

(6.23)

= 8.0

The contributions have to be normalized in order for cluster C1 to compare them. We
c
normalize each QoCi contributions by the sum of connectivity weights from the resulting
a,k

graph depth, should a merge with the cluster takes place. Let us call this the gain Gi

of

dependent cluster Ck merge with anchor cluster Ca , and ℓk the depth of the resulting graph.

a,k
Gi

=

k
QoCi+1
ℓk
i=1

(6.24)

λi

Evaluating the gain for each cluster merge, we obtain:

1,2

G1 =

8.5
4
i=1

λi

97

= 9.07

(6.25)

1,3

Gi

=

8.0
3
i=1

= 9.14

(6.26)

λi

Hence, the merge between cluster C1 and C3 will provide the higher increase to the QoC
in the network. This can be verified by computing the QoC after the merge. By merging
clusters C1 and C3 , the QoC of the network topology is increased to:
λ1 (2 + 21) + λ2 (15 + 8 + 4) + λ3 (30 + 20)

QoC =

100

3
i=1

λi

= 0.28

Instead if we chose to merge clusters C1 and C2 , the increase in the QoC of the network
topology was going to be less:

QoC =

λ1 (2 + 15) + λ2 (21 + 30 + 8) + λ3 (4) + λ4 (20)
100

4
i=1

λi

= 0.2666

c
It is worth noting that even with the individual QoCi contributions favoring the merge of

clusters C1 and C2 , the impact from reducing a whole connectivity level by merging clusters
c
C1 and C3 is taken into account when QoCi values are normalized in Eqn. (6.24).

6.5

Cluster Merging Decision Algorithms

Cluster merge operations imply relocation, or even additional deployment, of sensing nodes
to establish persistent communication links between clusters. Once the persistent commu98

nication links are established, the involved clusters become unified into a single cluster.
Persistent communication links are desired to enhance network connectivity. The network
QoC performance metric takes into account the existence of intermittent links, since they
tend to cause delay in network communications.
This section presents different algorithms for merging clusters in mobile sensor network
topologies. The merging of clusters in the network topology requires either, node relocation
or redeployment. The scope of this section is to establish the order of merging operations to
occur in the network topology. The QoC is suited for this purpose. The algorithms goal is
to maximize the QoC of the network by smartly choosing clusters to perform merging operations. The cluster merging decision algorithms are divided into centralized and distributed
categories, Section 6.5.1 and Section 6.5.2, respectively. Furthermore, the algorithms differ mainly in the criteria used to select a merging candidate, and the cluster’s subset used
for candidate selection (the merging candidates’ scope). All the algorithms terminate when
there are no more cluster merge options.
There are three merging selection criterion, namely:
• Maximum QoC yield (MQY): Cluster merging action that yields maximum QoC
in the network topology.
• Largest cluster size increase (LCSI): Cluster merging action that results in the
largest cluster size increase.
• Random: Random selection of cluster merging action. Used as a base performance
comparison.
Each algorithm is characterized by a unique combination of the merging criteria, scope,
and computing approach (centralized or distributed). An overview of the algorithms’ charac99

Distributed

Centralized

Table 6.2: Algorithms’ Characteristic Overview
Algorithm
BMN
BMN-BS
LIMN
LIMN-BS
RMN
RMN-BS
DBMN
DBMN-BS
S-DBMN
DLIMN
DRMN

MQY

LCSI

Random

Scope
All
L1
All
L1
All
L1
All
L1
L1 , L2
All
All

teristics is provided in Table 6.2. The scope of the algorithms is expressed as the connectivity
level where cluster’s merging evaluations are performed. For example, an L1 scope indicates
merging operations between level 1 and 2 are allowed, whereas a scope of both, L1 and
L2, indicates merging between level 1, 2, and 3 are allowed. Detailed descriptions for the
algorithm are provided in Section 6.5.1 and 6.5.2.

6.5.1

Centralized Approaches

The centralized approaches rely on a common reachable entity, such as the base station, to
collect data from sensor nodes, perform analysis, and make decisions based on an overall
view of the network topology status. Decisions are taken with the objective to create a
global improvement in the network QoC.

6.5.1.1

BMN and BMN-BS

Best Merge Next (BMN) selects the cluster pair to perform the merging operation, resulting
in the maximum QoC yield. BMN-BS in contrast, only evaluates candidate clusters that

100

are neighbors to the base station cluster, and selects the one with maximum QoC yield to
perform the merging operation. BMN can be considered as a brute-force approach to find
the sequence of cluster’s merging operations yielding the highest improve in QoC.

6.5.1.2

LIMN and LIMN-BS

Large Increase Merge Next (LIMN) selects the cluster merging operation that results in the
largest cluster’s size increase. LIMN-BS in contrast, only evaluates candidate clusters that
are neighbors to the base station cluster, and selects the one having the largest cluster size
to perform the merging operation.

6.5.1.3

RMN and RMN-BS

Random Merge Next (RMN) and Random Merge Next from base station (RMN-BS) are the
study base-cases for performance comparison. RMN, randomly selects cluster pairs in the
network topology to perform merge operations. RMN-BS, randomly selects cluster neighbors
to the base station cluster to merge.

6.5.2

Distributed Approaches

In contrast to the centralized approaches, the distributed approaches work by applying the
merging selection criterion locally to each cluster. Any clusters having dependent1 clusters
communication links will perform the merging selection process, by applying the selection
criteria of the algorithm employed. Due to the distribute nature of the approaches, and
as a measure to limit any possible control communication overhead, there is no particular
1 Dependent clusters are those clusters at farther connectivity levels from base station that

depend on other clusters routing capabilities to be connected to the network, see Section 6.4.

101

order of precedence for cluster merges to occur. Hence, cluster merge operations can occur
simultaneously by default.

6.5.2.1

DBMN, DBMN-BS, and S-DBMN

Distributed Best Merge Next (DBMN) and the other variants are similar to BMN in the
sense that they seek to make a direct impact on network QoC by selecting the maximum
QoC yield merging operations on each cluster having dependents. However, since these
approaches are designed to be carried in a distributed manner, and the QoC is an overall
(global) network performance measurement not available to all the clusters, each cluster
computes the individual Gain as presented in Section 6.4.1.
The difference between distributed variants is in the cluster’s subset used for candidate
selection. In DBMN all the cluster having dependent clusters execute merging operations.
In DBMN-BS, only merge operations between the base station cluster and its neighbors are
allowed. Special DBMN (S-DBMN) allows base station clusters neighbors to execute merging
operations with their dependents as well. In other words, merge operations in DBMN-BS
can occur only between connectivity level2 1 and level 2, while in S-DBMN merge operations
are allowed between connectivity levels 1, 2 and 3.

6.5.2.2

DLIMN

Similar to LIMN, Distributed Large Increase Merge Next (DLMN) uses largest cluster’s size
increase to select among the candidate clusters. The main different is that all clusters having
dependent clusters will employ the selection procedure locally. In other words, clusters with
dependents select the local LIMN, and proceed with the merging procedure.
2 Connectivity level 1 has only one cluster with the network base station.

102

6.5.2.3

DRMN

Similar to RMN, in Distributed Random Merge Next (DRMN) each cluster with dependents
executes a random selection to perform merge operations.

6.6

Evaluations

Simulations were performed to evaluate the performance of the cluster merging decision
algorithms. First, a detail description of the generated simulation environments is provided.
Second, obtained results are presented and evaluated.

6.6.1

Simulation Environment

Network topologies were generated and evaluated using a custom simulator in Matlab. Variable inputs were established for the generation of network topologies: number of connectivity
levels ℓ, number of clusters, and the total number of sensors nodes. The number of clusters
was selected to be five (5) times the number of connectivity levels, and the total number of
sensor nodes was selected to be twenty (20) times the number of clusters. All the algorithms
were evaluated under different number of connectivity levels. Topologies with three (3), nine
(9), and fifteen (15) connectivity levels were found to be representative for all the other
evaluated topology configurations.
The following topology setting requirements were established:
• A connectivity level by definition must have at least one cluster.
• A cluster must be composed of at least one sensor node.
• Every cluster must have at least one communication link with a cluster at a lower
103

connectivity level (with the exception of the Base Station cluster, since by definition,
it is the first and lowest existing connectivity level).
Sensor nodes were randomly distributed into clusters; the clusters were randomly assigned to different connectivity levels. Communication links between clusters were classified
as inter-level (connections between clusters at the same connectivity level), and intra-level
(connections between clusters at different connectivity levels). Aside from the required communication links for each cluster, additional links were randomly established with biased
probability. Intra-level links had 40% chance of occurrence, while inter-level links had 20%
chance.
Following the aforementioned settings, ten (10) topologies were generated for each fixed
number of connectivity levels, totaling thirty (30) network topologies.

6.6.2

Evaluation and Results Analysis

The performance of the cluster merging algorithm was analysed with two metrics:
• Impact to network QoC for each cluster merge operation.
• Number of candidate evaluations associated with each cluster merger operation.
Mean values for QoC and the number of evaluations per cluster merge operation were computed for centralized and distributed approaches. All distributed algorithms, and the algorithms employing random selection functions, such as, RMN, RMN-BS, DRMN, were run
ten (10) independent times for each topology to obtain their performance’s mean value.
For legibility purpose, plots of the results are presented in Section 6.6.3. The results are
grouped by the topology’s initial number of connectivity levels (3, 9, 15). Each connectivity
level has result’s plots for QoC improvement versus the number of cluster merge operations,
104

and the number of evaluations per cluster merger operations. Results are further organized
accordingly to the decisions algorithm approach, i.e., centralized and distributed.

6.6.2.1

BMN, BMN-BS, and DBMN-BS

BMN, BMN-BS, and DBMN-BS produced the maximum QoC yield per merge operation.
Their difference is on the number of evaluations required before each merge operation decision. BMN analyzes each possible merge operation in the network topology, while BMN-BS
and DBMN-BS only consider merge operations directly with the base station cluster. Even
though the number of comparison evaluations is the same for BMN-BS and DBMN-BS (see
plots in Section 6.6.3), they differ in the selection process and candidate evaluation function.
BMN and BMN-BS evaluate each candidate merge operation individually by peeking
into the future and recording the future QoC yield (should that candidate be selected), a
feasible procedure in less resource constrained nodes. On the other hand, in DBMN-BS each
merge candidate cluster computes its expected QoC contribution and provides it to the base
station cluster, who computes and evaluates the candidates’ QoC gains. DBMN-BS is less
computing intensive, and proves to have equal performance as its centralized counterpart.

6.6.2.2

LIMN, LIMN-BS, and DLIMN

The QoC yield per merge operation from LIMN and LIMN-BS was the closest to that
obtained by MQY algorithms. Their QoC maximization strategy relies in choosing the
largest cluster size increase (LCSI). The disadvantage with these approaches is that they
focus only in cluster’s size, neglecting the current level of connectivity of the clusters. The
level of connectivity is heavily weighted by the QoC function. This is because the connectivity
level of a cluster is highly correlated with the expected delay for any communication to reach
105

the base station.
Nonetheless, LIMN is shown to produce a higher QoC yield to that of LIMN-BS due
to the bigger scope of candidate clusters, Fig. 6.4a, 6.6a, and 6.8a. The higher QoC yield
from LIMN comes with a price, a greater number of comparison evaluations, making it more
computing intensive at early stages of redeployment, Fig. 6.5a, 6.7a, and 6.9a. LIMNBS, having less number of comparison evaluations, obtained a QoC yield slightly below of
that from LIMN for topologies with few connectivity levels. However the difference is more
noticeable in topologies having greater number of connectivity levels, Fig. 6.8a.
DLIMN, the distributed counterpart of LIMN, produced a slower increase in QoC, Fig.
6.4b, 6.6b, 6.8b. Its QoC yield is just under that obtained with DBMN, with a slight lower
number of evaluations per merge operation. This is more noticeable in topologies having
greater number of connectivity levels, Fig. 6.9b.

6.6.2.3

S-DBMN and DBMN

S-DBMN came in second place for QoC yield, with a low number of evaluation per merge
operation, when compared to other distributed algorithms, Fig. 6.4b, 6.6b, and 6.8b. A
reduced scope of candidates for merge operations showed to be a key factor on its performance.
DBMN displayed a lower number of evaluations per merge operations than S-DBMN,
Fig. 6.5b, 6.7b, and 6.9b. However, its QoC yield improves significantly only after some
cluster merges. This improvement in QoC yield is more evident after 7 cluster merges for
the topologies with 9 level of connectivity, and 15 cluster merges for the topologies with 15
level of connectivity, Fig. 6.6b and 6.8b, respectively.
It is worth noting that for all the distributed algorithms there is no order of precedence
106

for merging operations to occur. For instance, if DBMN network merge operations order of
precedence were to be sorted by QoC gain, then a QoC yield similar to that obtained by its
centralized counterpart (BMN) would be observed. Such approach would have to be designed
to be semi-distributed, and could benefit from a lower number of evaluations per operations
than its centralized counterpart (BMN). Nevertheless, the communication overhead related
to the cluster’s synchronization for this approach has to be considered.

6.6.2.4

RMN, DRMN, and RMN-BS

RMN and DRMN were selected as the base-case evaluations. All the presented algorithms
show a QoC yield to be far superior to that obtained with RMN and DRMN algorithms.
In spite of its random nature, by significantly reducing the scope of candidates for merge
operations to only neighbor clusters of the base station, RMN-BS shows an almost linear
QoC yield. This is attributed to the repeated increase in the number of sensor nodes to the
base station cluster. The base station cluster size has a higher weight factor in the QoC
function due to the number of persistent communication links associated with it.

6.6.3

Performance Plots

107

1
BMN
BMN−BS
RMN
RMN−BS
LIMN
LIMN−BS

0.9
0.8

QoC

0.7
0.6
0.5
0.4
0.3
0.2
0

1

2

3

4 5 6 7 8 9 10 11 12 13 14
Cumulative Cluster Merges

(a) QoC for Centralized Algorithms

1
DBMN
DBMN−BS
S−DBMN
DLIMN
DRMN

0.9
0.8

QoC

0.7
0.6
0.5
0.4
0.3
0.2
0

1

2

3

4 5 6 7 8 9 10 11 12 13 14
Cumulative Cluster Merges

(b) QoC for Distributed Algorithms

Figure 6.4: QoC in Network Topologies with 3 Connectivity Levels, 15 Clusters
108

Number of Comparison Evaluations

30
BMN
BMN−BS
RMN
RMN−BS
LIMN
LIMN−BS

25

20

15

10

5

0
0

1

2

3

4 5 6 7 8 9 10 11 12 13 14
Cumulative Cluster Merges

(a) Number of Evaluations for Cent. Algorithms

Number of Comparison Evaluations

30
DBMN
DBMN−BS
S−DBMN
DLIMN
DRMN

25

20

15

10

5

0
0

1

2

3

4 5 6 7 8 9 10 11 12 13 14
Cumulative Cluster Merges

(b) Number of Evaluations for Dist. Algorithms

Figure 6.5: Number of Evaluations Per Merge Operation in Network Topologies with 3
Connectivity Levels, 15 Clusters
109

1
BMN
BMN−BS
RMN
RMN−BS
LIMN
LIMN−BS

0.9
0.8
0.7

QoC

0.6
0.5
0.4
0.3
0.2
0.1
0
0

4

8

12 16 20 24 28 32
Cumulative Cluster Merges

36

40

44

36

40

44

(a) QoC for Centralized Algorithms

1
DBMN
DBMN−BS
S−DBMN
DLIMN
DRMN

0.9
0.8
0.7

QoC

0.6
0.5
0.4
0.3
0.2
0.1
0
0

4

8

12 16 20 24 28 32
Cumulative Cluster Merges

(b) QoC for Distributed Algorithms

Figure 6.6: QoC in Network Topologies with 9 Connectivity Levels, 45 Clusters
110

Number of Comparison Evaluations

120
BMN
BMN−BS
RMN
RMN−BS
LIMN
LIMN−BS

100

80

60

40

20

0
0

4

8

12 16 20 24 28 32
Cumulative Cluster Merges

36

40

44

(a) Number of Evaluations for Cent. Algorithms

Number of Comparison Evaluations

120
DBMN
DBMN−BS
S−DBMN
DLIMN
DRMN

100

80

60

40

20

0
0

4

8

12 16 20 24 28 32
Cumulative Cluster Merges

36

40

44

(b) Number of Evaluations for Dist. Algorithms

Figure 6.7: Number of Evaluations Per Merge Operation in Network Topologies with 9
Connectivity Levels, 45 Clusters
111

1
0.9
0.8
0.7

BMN
BMN−BS
RMN
RMN−BS
LIMN
LIMN−BS

QoC

0.6
0.5
0.4
0.3
0.2
0.1
0
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72
Cumulative Cluster Merges

(a) QoC for Centralized Algorithms

1
0.9
0.8
0.7

DBMN
DBMN−BS
S−DBMN
DLIMN
DRMN

QoC

0.6
0.5
0.4
0.3
0.2
0.1
0
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72
Cumulative Cluster Merges

(b) QoC for Distributed Algorithms

Figure 6.8: QoC in Network Topologies with 15 Connectivity Levels, 75 Clusters
112

200
BMN
BMN−BS
RMN
RMN−BS
LIMN
LIMN−BS

Number of Comparison Evaluations

180
160
140
120
100
80
60
40
20

0
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72
Cumulative Cluster Merges

(a) Number of Evaluations for Cent. Algorithms

200
DBMN
DBMN−BS
S−DBMN
DLIMN
DRMN

Number of Comparison Evaluations

180
160
140
120
100
80
60
40
20

0
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72
Cumulative Cluster Merges

(b) Number of Evaluations for Dist. Algorithms

Figure 6.9: Number of Evaluations Per Merge Operation in Network Topologies with 15
Connectivity Levels, 75 Clusters
113

6.7

Summary

This chapter presents an unique fitness evaluation for wireless sensor network topologies.
Network connectivity is crucial for a network topology to be functional. Nonetheless, network
responsiveness and its capacity to meet communication flow demands is crucially important
as well. This work evaluated network performance aspects related to network connectivity
from the application’s perspective. A Quality of Connectivity (QoC) performance metric
was defined to evaluate the network connectivity in terms of how geographically distributed
the sensor nodes are in the topology. Furthermore, the metric served as an indicator factor
of potential communication problems and latency in sensor network topologies.
Centralized and distributed algorithm approaches are presented to maximize network performance. Maximizing network performance with fewer operations as possible is important
for Wireless Sensor Networks, where energy and computing resources are limiting factors.
Therefore, the algorithms make smart cluster merging operations attempting to yield maximum QoC for each operation. Network QoC improvement and the complexity of executing
the approaches was evaluated in multiple and diverse network topology scenarios. From all
the algorithm approaches, DBMN-BS was discussed to be less computing intensive than its
centralized counterpart, BMN-BS, having an equally outstanding network QoC yield per
operation.
As an extension of this work, it would be beneficial to design and evaluate a semidistributed approach with maximum QoC yield decision criteria. Such approach will seek
to combine the network QoC improvement performance obtained with DBMN-BS, and the
low number of comparison evaluations per merge operations obtained with DBMN.

114

Chapter 7
LEVERAGING HEIGHT IN
JUMPING SENSORS
TO OBTAIN RELATIVE
ORIENTATION

7.1

Preliminaries and Motivations

Jumping sensor robots have attracted attention for their unique ability to jump to target
destinations, making them ideal candidates to be deployed in rough terrains where wheeled
robots cannot perform, Cen and Mutka (2008); Pei et al. (2009). It is clear that as a newer
generation of jumping sensors becomes available, opportunities open to solve many existing
Wireless Sensor Network (WSN) problems, which include a node’s relative orientation. We
define a node’s relative orientation as the ability of a node to identify the direction of a target
(e.g., neighbor node) from its current location relative to a reference direction. Obtaining a
reference direction, e.g., the magnetic north, has become simple with the usage of inexpensive
magnetic compass sensors. However, the challenge remains to determine the relative direction
of a neighbor node. While the orientation concept seems relatively simple in nature, it is

115

still a challenge for a highly resource constrained (CPU, power, form-factor, etc.) WSN.

Figure 7.1: Orientation with GPS

Sensor orientation is a crucial area of importance for mobile wireless networks. Generally,
node orientation is obtained from each node’s location coordinates obtained by GPS receivers.
Figure 7.1 illustrates a relocation request in a wireless network composed of a limited number
of mobile nodes. When a node relocation request occurs, the relative orientation of the node
to the target destination is computed from the position coordinates. The computation might
take place in a base station if the node lacks sufficient computational resources. Nevertheless,
localization is a requirement for those mobile networks to work. Enabling every device to be
position coordinate aware can become a costly procedure and inconvenient for most power

116

limited mobile wireless sensor networks.

Figure 7.2: Relative Orientation with Jumping Sensors

The objective of this work is to leverage jumping and mid-air spinning to determine the
relative orientation of sensors nodes without the initial need of position within a coordinate
system. Even with the limitations of wireless sensors mentioned earlier, the fluctuations of
the received signal strength as the antenna orientation changes provide sufficient information
to predict relative orientation. The jumping ability allows the jumping sensor to relocate
and while in this process, mid-air rotations can be used to scan the nearby nodes’ relative
orientation. The relative orientation data can be used to guide the relocation sensors as
needed. For example, the destination of a mobile node can be defined as a progression
of dynamic directions adjusted according to the relative orientation of surrounding nodes.
117

Figure 7.2 illustrates the movement progression of a jumping robot relying on the oriented
directions (OD) from neighbor nodes. Every oriented direction is relative to the node itself
and a reference direction. The magnetic north serves as a common fixed reference direction.
Furthermore, with relative orientation, nodes can collaborate to identify network boundaries, uncovered areas, and the existence of obstacles in the topology. Sensor orientation is
discussed further in Section 7.2.
This work presents a multi-step procedure to produce a direction oriented jumping sensor
network. First, the jumping sensor collects and processes signal strength readings to produce
the first prediction set of oriented directions per node. At this stage, the nodes have multiple
possible directions for their neighbors. Second, this set is processed and reduced to two
oriented directions, one being the true direction where the neighbor node lies, and the other
its mirrored direction. Third, the process continues with the execution of a relative direction
agreement algorithm, which selects and validates a single oriented direction per neighbor
node. Finally, the direction oriented network executes a mirror free procedure, which detects
and corrects mirrored oriented directions. Only a pair of beacon nodes is needed for the
procedure to fix nodes’ true orientation from a mirrored orientation.

7.1.1

Organization

This chapter is organized as follows. Section 7.2 introduces the jumping sensor approach
for collecting and processing signal strength data into orientation information. A series
of outdoor experiments are performed and the obtained results are evaluated. In Section
7.3 a directional-orientation decision algorithm is presented. Section 7.4 presents a mirror
orientation identification and correction procedure. In Section 7.6 a summary of the work
and conclusion are provided.
118

7.2

Sensor Orientation

In a three dimensional space, orientation of an object can be measured in terms of rotations
around three axes: vertical (yaw), lateral (pitch), and longitudinal (roll). Figure 7.3 depicts
a three dimension axes configuration where roll and pitch follow the right-handed rule for
positive rotation, while yaw follows the left-handed rule positive rotation.

Figure 7.3: Yaw, Pitch, and Roll

The increase in demand for spacial orientation in technology gadgets, such as, smartphones, remote controllers, digital cameras, etc., have made orientation data acquisition
sensors to be developed in small sized chips at an affordable customer price. Newest generations of wireless sensor robots are designed to have a magnetic compass, gyroscope, and
accelerometer, all integrated as part of their control system. This makes orientation data
available for robot repositioning and movement.
Most wireless sensors presently are equipped with dipole antennas. Dipole antennas
radiate electromagnetic waves in a toroid manner, creating two radiation holes (where the
radiation power drops significantly) along the direction of the antenna endings. Changes in
the antenna orientation can make nearby receivers to register higher or lower signal strength,
depending where the radiation pattern of the antenna is concentrated relative to the receivers.

119

If the antenna orientation is changed intentionally, lower signal strength readings from nearby
receivers can potentially indicate where the radiation holes are, and relative direction from
receivers could be identified.
Jumping sensors are capable of two-way communication with a far receiver while they
are airborne, for relocation or simply for node discovery, as presented in Cintr´n and Mutka
o
(2010). While the jumping sensor is airborne it freely rotates on the air, changing the
orientation of its antenna. Mid air rotations are beneficial because it allows the sensor to
transmit at different antenna orientations. Receiver of airborne communication packets can
process the RSSI of received packets to identify drops in signal power.
This section will present results from a series of experiment conducted where antenna
orientation was manipulated to create fluctuations in the RSSI registered in a receiver node.
The processing of this data makes it possible to identify a node’s relative orientations.

7.2.1

Oriented Direction

In wireless sensors, sensor nodes within radio range of each other are called neighbors. We
define the direction of a node’s neighbor with respect to a common reference as the oriented
direction (OD) of the neighbor from the node. The reference can be the earth magnetic
north pole, a fixed land mark or object with which the sensors can get oriented. Once a
common reference is established, the OD of a neighbor is the vertical axis rotation angle
(yaw) that points in the direction toward the neighbor. Hence, it is worth noting that ODs
are relative to the observer node.
When a jumping sensor jumps, it can broadcast its orientation in communication packets during the entire time of flight. Neighbor receivers register the RSSI for each broadcast
packet. The RSSI is associated to the orientation of the transmitter (at transmission time)
120

that is enclosed in the packet. The OD of a receiver node is obtained by finding the orientation with minimum RSSI.
Due to the nature of dipole antennas, two coplanar 180 ◦ apart directions have to be
taken into consideration when identifying the direction of the neighbor. Both directions, one
identified with minimum RSSI and the other being the 180 ◦ -complement, are assigned equal
probability to be the true direction.

7.2.2

Experimentation

A series of outdoor experiments were performed to obtain Oriented Directions of two wireless
sensor nodes. Two IRIS1 motes where used in the experiments. Maximum transmission
power (3 dBm) was employed for wireless communication. The sensors were separated at
a 5 meter distance on a flat surface. The orientation of the transmitter was changed as
follows. Eight (8) Rotations on a vertical axis (yaw) were set at increments of 45 degrees,
for a complete 360 degree rotation. The antenna in the IRIS is fixed to what we define as
the front side (for experimentation purpose), with a single axis rotation connector, allowing
antenna rotations along the longitudinal axis (roll). Four (4) roll angles were selected over
180 ◦ above the horizontal plane: −45 ◦ , −90 ◦ , 45 ◦ , 90 ◦ . It is worth noting that for any roll
angle, the antenna will be always perpendicular to the longitudinal axis.
Each yaw orientation was combined with all four roll angles to form a total of 32 antenna
orientations, i.e., 8 yaw by 4 roll angles. Two fixed heights were tested: 0 cm and 50 cm above
the surface. Only the height of the transmitting sensor was changed, leaving the receiver at
surface level at all times. All the antenna orientations were tested for each height, repeating
the process for the second node set as the transmitter. Fifty (50) sets of ten (10) packets
1 http://www.memsic.com/wireless-sensor-networks/

121

were transmitted for each antenna orientation. The receiver node was set to register the
RSSI for each packet, and the mean RSSI was computed for each orientation.
In order to obtain the ODs for each node we need to find the antenna orientations having
the minimum RSSI. However, we are only interested in the yaw angle of the antenna itself,
not the sensor. Therefore, it is required to obtain the direction component, from the antenna
orientation, in the horizontal plane of reference. Note that this will not be required if the
antenna cannot change orientation independently from the sensor. In our experiment, the
sensor roll was always kept fixed, and the antenna roll adjusted to the desired angle.
To obtain the antenna yaw angle (in our experiment) we needed to add ±90 ◦ (the sign
is determined by the sign of the roll angle) to the sensor yaw orientation. This is because
the horizontal plane of reference is perpendicular to the vertical axis. Table 7.1 presents the
true antenna yaw orientation for each combination of sensor’s yaw orientation and antenna
roll. For example, if the sensor yaw is 90 ◦ , the antenna yaw angle will be 0 ◦ for a negative
roll, and 180 ◦ for a positive roll. Similarly, a sensor yaw 180 ◦ -complement will result in the
same antenna yaw angles but in opposite direction with respect to the roll sign. Therefore,
we end up with four RSSI measurement corresponding to every yaw angle, as a result from
the translation of the sensor yaw readings to the antenna perspective.
Table 7.1: Antenna True Yaw Orientation for IRIS mote
Roll
-90,-45
45, 90

0
270
90

45
315
135

Sensor Yaw
90 135 180 225
0
45
90 135
180 225 270 315

270
180
0

315
225
45

The RSSI mean was computed for the small samples of ten (10) packets for each antenna
yaw and roll configuration. RSSI results are shown by height for node A at different yaw
angles while transmitting at −45◦ and −90◦ antenna roll, Fig. 7.4 and Fig. 7.5, respectively.
122

RSSI results for node B followed a similar pattern. It should be noted that for experimental
purpose, each node reference (0 ◦ yaw) was fixed to point to toward each other. From Fig.
7.4, it can be appreciated that node A’s lower RSSI were registered at 0 ◦ and 315 ◦ yaw for
a transmitting height of 0 cm, and 0 ◦ yaw for 50 cm height. RSSI readings for −90 ◦ roll
(Fig. 7.5) with low values were registered at 0 ◦ , 270 ◦ , and 315 ◦ yaw angles, for 0 cm height.
Although 315 ◦ yaw RSSI readings had the lowest value when at 50 cm height, both 0 ◦ and

RSSI (dBm)

180 ◦ yaw produced the second lowest RSSI.

−83
−83.5
−84
−84.5
−85
−85.5
−86
−86.5
−87
−87.5
−88
−88.5
−89
−89.5
−90
−90.5
−91

H = 0 cm
H = 50 cm

0

45

90
135 180 225
Antenna Yaw (angle)

270

315

Figure 7.4: Node A’s Antenna Yaw RSSI (w/ Roll −45◦ ) Measured at Node B
The RSSI readings showed a high correlation with the antenna orientation. Some degree
of offset to the expected yaw angle direction was experienced. This was due to a number
of factors. First, the shape of the antenna conductor in the motes is not perfectly straight,
which can create changes in the radiation pattern. Second, the antenna location on the
motes is at its side, not at its center, making it almost half visible from the opposite side,
which can interfere with the signal. A rigid antenna conductor strategically positioned on
123

RSSI (dBm)

−83
−83.5
−84
−84.5
−85
−85.5
−86
−86.5
−87
−87.5
−88
−88.5
−89
−89.5
−90
−90.5
−91

H = 0 cm
H = 50 cm

0

45

90
135 180 225
Antenna Yaw (angle)

270

315

Figure 7.5: Node A’s Antenna Yaw RSSI (w/ Roll −90◦ ) Measured at Node B
the robot’s body can mitigate the influence of these. Third and foremost, when the mote
is at surface level much of the signal is absorbed by the surface, in addition to any lineof-sight obstruction between the transmitter and receivers, e.g., unevenness of the surface.
The ability of a jumping sensor to perform surface and airborne communication makes it
adequate for OD detection. Furthermore, while the mote rotates, transmission can occur at
smaller yaw intervals, providing a higher RSSI resolution per rotation.

7.2.3

Multiple Minima

Multiple RSSI minima can be found at different yaw angles. A solution to this problem in
its simpler form can be obtained by computing the average of yaw angles corresponding to
the minimum RSSI values. However, the yaw angles of the minimum RSSI values may not
always lie close to each other due to the dipole antenna effect. In such case, the average

124

value alone will most likely result to an offset direction far from the true OD. To deal with
this we propose an OD estimator (Alg. 7.1).
Algorithm 7.1 Estimate OD from a set with n directions
1: AOSall ← Angle of separation from all pair of directions.
2: if n = 2 then
3:
if AOS1,2 ≤ 120 then
4:
OD1 ← mean([yaw ∠1 , yaw ∠2 ])
5:
else
6:
OD1a ← yaw ∠1
7:
ODtemp ← yaw ∠2
8:
OD1b ← 180complement(ODtemp )
9:
OD1 ← mean([OD1a , OD1b])
10:
end if
11: else
12:
NearestP air ← Find nearest pair in AOSall
13:
OD1 ← mean(NearestP air)
14:
CurrentSize ← 3
15:
P reviousSize ← 0
16:
while P reviousSize = CurrentSize do
17:
NewSet ← All directions within ±60 ◦ from OD1
18:
OD1 ← mean(NewSet)
19:
P reviousSize ← CurrentSize
20:
CurrentSize ← |NewSet|
21:
end while
22: end if
23: OD2 ← 180comp(OD1)

The OD estimator computes the mean direction from the multiple yaw minima within
120 ◦ yaw range. The yaw range gives enough margin (±60 ◦ ) to include offset readings
from the true direction, while at the same time avoids including 180 ◦-complement readings
created by the dipole antenna effect. If there are only two minima (n = 2) with a separation
angle less or equal 120 ◦ , the mean direction is computed (Alg. 7.1, line 4), else the mean
of one direction (randomly chosen) and the 180 ◦-complement of the other is assigned as
OD1 (Alg. 7.1, lines 6-9). For cases with n ≥ 3 the algorithm starts by computing the
mean direction (OD1 ) from the closest (smallest angle of separation) minima pair (Alg. 7.1,
125

lines 12-13). Then, OD1 is recomputed from the mean of all minima within ±60 ◦ from
the previous OD1 (Alg. 7.1, lines 17-20). This process continues until there is no change
in the minima enclosed in the 120 ◦ range of the computed OD1 . Once OD1 is set, the
180 ◦ -complement OD2 is computed (Alg. 7.1, line 22).
The OD estimator algorithm reduces multiple ODs with minimum RSSI to only two ODs,
the true OD and its mirror. A network-wide OD algorithm to reduce the ODs of each node
to one per neighbor is presented in Section 7.3.

7.3

Orientation Algorithm

This section presents the process to reduce the possible ODs per neighbor sensor node to
a single OD. It is assumed that each node will have two ODs that are 180 ◦ apart for each
neighbor, both having equal probability of being the true OD. Therefore, it is important
that the node’s OD choice is kept consistent and valid with its neighbors.
To illustrate the problem, let’s analyze the ODs for 3 nodes as shown in Fig. 7.6. Each
node has two OD per neighbor. One is the true OD (ODtrue ) where the neighbor really lays,
and the other 180 ◦-complement is what we will refer as the mirror OD (ODmirror ). If node
A chooses node B’s ODtrue , then to keep consistency, node B must choose node A’s ODtrue
from its perspective. The same applies for mirror ODs. Generally, if node B’s OD from A is
ODAB , then node A’s OD from B must be the 180 ◦ -complement of ODAB , as expressed in
Equation (7.1).
ODBA = 180comp(ODAB )

(7.1)

Every nodes’ OD choices must be validated, i.e., comply with an established rule or a
set of rules. If a set of three non collinear nodes are neighbors, then a triangle can be
126

Figure 7.6: ODs Choice Problem
established with nodes as vertices, and lines between directly connected nodes as edges, as
shown in Fig. 7.6. Since each edge is associated with an OD, interior angles can be obtained.
In order for the ODs for three nodes to be valid they must meet the triangle’s sum of interior
angles (180 ◦ ) rule of Equation (7.2). Generally, any triangle formed by three connected and
non collinear nodes labeled A, B, and C, have interior angles ∠BAC, ∠ABC, and ∠BCA,
that can be represented as α, β, and γ, respectively.

α + β + γ = 180

(7.2)

From Fig. 7.6, it can be appreciated that mirror ODs comply also with Equation 7.2. This

127

is possible when the OD choice is kept consistent, i.e., either the ODs are all true ODs or all
mirror ODs. If all ODs in the network are mirrors, then we call it a mirror oriented network.
A mirror oriented network can be transformed into the true orientation by replacing all ODs
with their 180 ◦-complement, which is discussed in Section 7.4. The remaining of this section
discusses the OD selection process.

7.3.1

OD Selection

The OD selection process is presented in Algorithm 7.2. The objective of the algorithm is to
assign valid ODs that keep consistency with those previously selected. A node is selected as
an anchor node (A) to initiate the selection process (Alg. 7.2, line 1). The anchor node can
be selected by quorum agreement, e.g., if unique IDs are used to identify nodes, then a node
with the minimum or maximum ID can be selected. Node A’s neighbors with OD are kept
in OrientedA , while those needing a decision are kept in NonOrientedA . Collinear nodes
are kept in Collinear until a valid OD can be assigned to them. It is likely that a node’s
neighbors are within communication range of each other. We call the common neighbors
(Commons), line 7. If there are neighbors with assigned OD, then the algorithm proceeds to
choose ODs for those other neighbors that are common to them, but not oriented (Comnor ),
line 8.
Once an oriented node is selected (B), the ODs of nodes A and B are divided in two
zones (lines 13-14), using ODAB and ODBA as a zone divider. It is worth recalling that
ODBA is the 180 ◦-complement of ODAB . A third node (C ) is selected from those that
are common with B but still not oriented (line 15). The anchor node (A) makes the first
attempt to assign an OD to C (ODAC ), (line 16). This OD selection is a random choice
between both of the ODs for node C, OD1AC , OD2AC . Next, the anchor selects the OD of
128

node C, from the perspective of node B (ODBC ), that lays in the same zone as (ODAC ).
Validation is done using Eqn. 7.2 with the obtained angles of separation from each pair of
Algorithm 7.2 Neighbor OD Selector
1:
2:
3:
4:
5:
6:
7:
8:
9:
10:
11:
12:
13:
14:
15:
16:
17:
18:
19:
20:
21:
22:
23:
24:
25:
26:
27:
28:
29:
30:
31:
32:
33:
34:
35:
36:
37:
38:
39:
40:
41:

A := Anchor node
OrientedA ← A’s neighbors with OD
N onOrientedA ← A’s neighbors without OD
Collinears ← ∅
repeat
for all B ∈ OrientedA do
Commons ← N eighborsOfA N eighborsOfB
Comnor ← Commons N onOrientedA
Zone1A ← A’s ODs laying within ODAB + 179 ◦
Zone2A ← A’s ODs laying within ODAB − 179 ◦
Zone1B ← B’s ODs laying within ODBA + 179 ◦
Zone2B ← B’s ODs laying within ODBA − 179 ◦
Z1 ← Zone1A Zone2B
Z2 ← Zone2A Zone1B
for all C ∈ Comnor do
ODAC ← OD1AC
if ODAC ∈ Z1 then
if OD1BC ∈ Z1 then
ODBC ← OD1BC
else
ODBC ← OD2BC
end if
else
if OD1BC ∈ Z2 then
ODBC ← OD1BC
else
ODBC ← OD2BC
end if
end if
T ries = 2
repeat
if α + β + γ = 180 then
if T ries > 1 then
ODAC ← OD2AC
ODCB ← 180comp(ODCB )
else
ODAC ← ∅
ODCB ← ∅
Include C in Collinears
end if
else

129

⊲ ODAC ∈ Z2

⊲ Validation

⊲ Collinear nodes!

⊲ Valid; must keep Consistency

Algorithm 7.2 Neighbor OD Selector (cont’d)
42:
ODCA ← 180comp(ODAC )
43:
ODCB ← 180comp(ODBC )
44:
Append C to OrientedA
45:
Delete C from N onOrientedA
46:
break
47:
end if
48:
T ries ← T ries − 1
49:
until T ries = 0
50:
end for
51:
end for
52:
Collinears ← N onOrientedA ∩ Collinears
53:
N onOrientedA ← N onOrientedA − Collinears
54:
Lef tCount ← |N onOrientedA |
55:
if Lef tCount > 0 then
56:
DON E ← F ALSE
57:
ODprediction(A, OrientedA , N onOrientedA )
58:
N ewLef tCount ← #N onOrientedA
59:
if Lef tCount − N ewLef tCount = 0 then
60:
DON E ← T RU E
61:
end if
62:
N onOrientedA ← N onOrientedA ∪ Collinears
63:
else
64:
DON E ← T RU E
65:
end if
66: until DONE

⊲ Exit Tries loop

⊲ end for all C ∈ Comnor
⊲ end for all B ∈ OrientedA

⊲ Predictive Assignment

⊲ No new OD

OD per node. If the ODs selections are not valid, the 180 ◦ -complements are selected and
re-evaluated (lines 32-35). If a second re-evaluation fails validation, the node is considered
collinear (line 39). Otherwise, when OD selections are found to be valid, consistency is kept
by assigning the 180 ◦ -complements of ODAC and ODBC to ODCA and ODCB , respectively
(lines 42-43).
There are two circumstances where an OD prediction is needed. First, OD prediction
is needed when a node does not have oriented neighbors. Second, it is needed when the
anchor node and remaining neighbors without ODs do not share common oriented neighbors.
Algorithm 7.2 identifies these cases by keeping track of nodes left without OD (lines 54-65).
At this stage in the algorithm, if there are nodes without OD, and not identified to be
130

collinear, an OD prediction is needed before running a new iteration of the OD selection.
Algorithm 7.3 OD Prediction
1:
2:
3:
4:
5:
6:
7:
8:
9:
10:
11:
12:
13:
14:
15:
16:
17:
18:
19:
20:
21:
22:
23:
24:
25:
26:
27:
28:
29:
30:
31:
32:
33:
34:

A := Anchor node
OrientedA := A’s neighbors with OD
N onOrientedA := A’s neighbors without OD
ODf ound ← F ALSE
OrientedCount ← #OrientedA
if OrientedCount > 0 then
ClustersA ← Sort OrientedA into Clusters
i=1
for all cluster ∈ ClustersA do
{bound1 , bound2 } ← GetBoundaries(cluster)
N oZone[i] ← {bound1 − 60 ◦ , bound2 + 60 ◦ }
i=i+1
end for
for all B ∈ N onOrientedA do
for i = 1 → #ClustersA do
if OD1AB ∈ N oZone[i] then
ODAB ← OD2AB
ODf ound ← T RU E
GoTo FinishLabel (line 32)
else if OD2AB ∈ N oZone[i] then
ODAB ← OD1AB
ODf ound ← T RU E
GoTo FinishLabel (line 32)
end if
end for
end for
end if
if ¬(ODf ound) then
B ← Select N onOrientedA having max #Commons
ODAB ← Random choice {OD1AB , OD2AB }
end if
FinishLabel:
ODBA ← 180comp(ODAB )
return

⊲ Use Oriented nodes

⊲ Consistency

OD prediction is handled in the ODprediction function, described in Algorithm 7.3. The
function first checks for existing anchor’s oriented neighbors. If there are none, the node
with more common neighbors to the anchor is selected and the anchor’s OD to the node
(ODAB ) is selected randomly from the two possible ODs (Alg. 7.3, lines 28-31). Finally,

131

the 180 ◦ -complement of ODAB is assigned to ODBA (Alg. 7.3, line 33).
When there are oriented neighbors, a prediction is done by avoiding conflicting zones.
A conflict zone (NoZone) is the region where the OD of an oriented neighbor (not common
neighbor with the non-oriented nodes) lays. The size of the NoZone is established with two
boundaries, ±60 ◦ to the OD. Similarly, for regions with multiple neighbors (cluster), the
pair of ODs having the greatest angle of separation is used, and ±60 ◦ is added respectively
to establish the two boundaries of the NoZone cluster (Alg. 7.3, lines 10-11). After having
established all the NoZones, both potential ODs from each NonOrientedA neighbor node are
checked to see if any lays in a NoZone. A potential OD found to be in a NoZone is discarded
and its 180 ◦-complement is used as the anchor’s OD for that node (ODAB ), (Alg. 7.3, lines
15-25). Likewise, the 180 ◦-complement is assigned as the node’s OD to anchor (ODBA )
(Alg. 7.3, line 33). Since a prediction can have an impact on other ODs, the ODprediction
function culminates after one OD is predicted, thus resulting in a new oriented node.
The new oriented node is used in Alg. 7.2 to start a new iteration. Also, collinear nodes
have the opportunity to be oriented in further iterations, as more OD are established. Finally,
the algorithm concludes either, when there are no more neighbors left without orientation
(Alg. 7.2, lines 63-64), or when the available information to the anchor is not enough to
predict neighbors’ ODs (Alg. 7.2, lines 59-61).

7.4

Mirror Free Topology

In Section 7.3, an algorithm to choose valid ODs while keeping consistency among the network was presented. However, the resulting topology could be mirrored. In a mirrored
topology, ODs from nodes are pointing 180 ◦ away from the true ODs. This topology still

132

holds validity and consistency, but a transformation is needed to make it usable by sensor nodes. In this section, we will discuss an innovative technique to detect and rearrange
mirrored ODs in an oriented network topology.
We have departed from the notion that the information available to the nodes is not
enough to discriminate between a true and a mirror OD. Therefore, the topology needs
extra information to detect mirrored ODs. Nodes need to have known references in the
field, such as beacons, in order to distinguish true ODs from mirror ODs. However, precise
deployment of references can become difficult in many scenarios, and will increase the overall
cost of deployment and maintenance of the network. Therefore, the usage of beacons should
be limited.
Normally, a set of three to four non collinear beacons are used to triangulate the relative
position of an object. This position can be transformed to a global coordinate system when
these beacons posses geographic coordinates. However, our topology already has relative
orientation information, which can be used by nodes. Distance estimation between beacons
and nodes is not needed. The orientation information can be used in conjunction with a
reduced set of oriented beacon nodes to identify mirrored ODs.

7.4.1

OD Paths and Mirrored Topology Identification

Routing paths between nodes in a network topology with ODs can be used to keep track
of the change in direction, relative to each node in the path, as the path is traversed by
a packet. We define these paths as OD-paths. Each node from the set belonging to an
OD-path is required to have the OD of the previous (incoming) and next (forwarding) node
in path.
Let us consider the nodes in an OD-path to be vertices and the ODs between them be
133

the edges of a graph. If we add an additional edge between source node and destination
node in path, we form a closed loop, which is also a polygon. Polygons share an interesting
characteristic to our interest. It is known that the sum of the exterior angles, with one
adjacent side, of a polygon results in 360 ◦ . Therefore, the sum of the changes in directions
in an closed-loop OD-path equals to 360 ◦.
Now if we can guarantee that the additional edge between source and destination will
always have true ODs, then a mirrored OD-path can be identified as follows. Adding all the
changes in direction from a true OD path will result in 360 ◦. However, when the OD-path
is composed of mirrored ODs, but not the OD’s between source and destination, then the
sum of changes in direction is not 360 ◦ .
A pair of beacon nodes having their location information (e.g. global coordinates) can
compute each other’s true OD to create the closed-loop OD-path. Fig. 7.7a illustrates a true
OD-path with two beacons at the end points, A and F . The mirror identification process
starts with beacon A sending a communication packet to beacon F , including its location

(a) True OD-path

(b) Mirrored OD-path

Figure 7.7: OD-paths
134

information and the OD of the first node in the path (ODAB ). Also, a total change in
direction register (T ) is kept, and updated as the packet traverses the path. Each node in
the path is responsible for updating the change in direction before the packet is forwarded.
Once the packet arrives to beacon F , the location information of A is used to obtain its true
oriented direction from F , ODF A . Beacon F updates the change in direction register with
the two missing direction changes. (1) the change in direction f , as if the packet were to
be forwarded directly to beacon A. (2) the initial change in direction a, i.e., change from
ODF A to ODAB . The path is identified as true after verifying the total direction change,
T = 360 ◦ .
Similarly, Fig. 7.7b illustrates the mirrored OD-path. From the figure, we can see that
even if the path is mirrored, the true ODs from beacons are used, i.e., ODF A∗ and ODAF ∗ .
This is due to the fact that the beacons can obtain their absolute position. Therefore, the
changes in direction considered from F ∗ to A, and from A∗ to F are the supplementary
angles a′ , i.e., a + a′ = 180 ◦ , and f ′ , i.e., f + f ′ = 180 ◦. When adding all the changes in
directions, we have T = 0 ◦ .
Since the oriented topology is kept consistent (see Section 7.3), a true OD-path indicates
that the nodes covered by the beacons are true oriented. It should be noted that a node
can be covered even if it’s not in the communication range of the beacon. Otherwise, if
the network is identified as mirrored, a transformation is performed, in which the ODs are
changed with their 180 ◦-complement.

7.4.2

Special Cases

The orientation in some OD-paths can create false identification of mirrored path. Particularly those OD-path having OD-edges crossing over a straight line drawn between the source
135

and destination node. Figure 7.8 depicts a true OD-path and its mirrored OD-path, with
one OD-edge crossing. When there is an odd number of crossings, the sum of changes in
directions equals 0 ◦ for both true and mirrored OD-paths.

(a) True OD-path, 1 crossing

(b) Mirrored OD-path, 1 crossing

Figure 7.8: OD-path with One Edge Crossing

136

Proof. The sum of direction changes (angles a, b, c and d) from a true OD-path with one
crossing, as shown in Fig. 7.8a, is equal to 0 ◦ :

a+b +c+d = 0◦

Let us consider the triangle interior angles sum theorem, which states that the sum of
the interior angles of a triangle equals 180 ◦ :

a′ + b′ + e′ = 180 ◦
d′ + c′ + e′ = 180 ◦

Therefore:

a′ + b′ + e′ = d′ + c′ + e′
a′ + b′ = d′ + c′

Considering the supplementary angles in Fig. 7.8a, we have:

a′ = 180 ◦ − a
b′ = 180 ◦ − b
c′ = 180 ◦ − c
d′ = 180 ◦ − d

137

(7.3)

Then Equation (7.3) can be rewritten as:

180 ◦ − a + 180 ◦ − b = 180 ◦ − c + 180 ◦ − d
a+b = c+d

(7.4)

Equation (7.4), shows that the magnitude of (a + b) is equal to the magnitude of (c + d).
Additionally, from Fig. 7.8a it is possible to notice that angles c and d have opposite rotation
direction than angles a and b. Therefore we can write that:

a + b = c + d = −(a + b)

Thus:
a + b + c + d = a + b − (a + b) = 0 ◦

Similarly, it can be proven that the sum of direction changes (angles a, b, c and d) in a
mirrored OD-path with one OD-edge crossing, depicted in Fig. 7.8b, is also equal to 0 ◦ .
Proof. The sum of direction changes (angles a, b, c and d) from a mirrored OD-path with
one crossing, is equal to 0 ◦ :
a+b +c+d = 0◦
Let us consider the triangle interior angles sum theorem:

d + c′ + e′ = 180 ◦
a + b′ + e′ = 180 ◦
138

Therefore:

d + c′ + e′ = a + b′ + e′
d + c′ = a + b′

(7.5)

Considering the supplementary angles in Fig. 7.8b, we have:

b′ = 180 ◦ − b
c′ = 180 ◦ − c

Then Equation (7.5) can be rewritten as:

d + 180 ◦ − c = a + 180 ◦ − b
d+b = a+c

(7.6)

Equation (7.6), shows that the magnitude of (d + b) is equal to the magnitude of (a + c).
Additionally, from Fig. 7.8b it is possible to notice that angles a and c have opposite rotation
direction than angles b and d. Therefore we can write that:

d + b = −(a + c)

Thus:
a + b + c + d = a + c − (a + c) = 0 ◦

139

The same can be proven for mirrored OD-paths with greater odd number of crossings
(see Section 7.5.1).
The sum of direction changes is not affected when the number of crossing edges is even,
for which it results as expected, 360 ◦ for true OD-paths, and 0 ◦ for mirrored OD-paths.
Proofs for even crossings cases are provided in Section 7.5.2.
All these cases share a characteristic that is unaffected by the crossings parity (even,
odd). Consider the number of changes in the angle sign for each change in direction in
an OD-path. There are more sign changes in mirrored OD-paths than in true OD-paths.
Changes in angles sign can be tracked in both of the candidate paths. The path having
the least sign changes is selected as the true path. For example, let’s consider both, true
and mirrored, OD-paths presented in Fig. 7.8. Also, assume counter clockwise rotation as
positive rotation. If we traverse the path from node A to node F in the true OD-path (Fig.
7.8a), then the sequence of direction changes is {a,b,-c,-d}. Thus, for this path there is only
one sign change, whereas for the mirrored path (Fig. 7.8b the sequence of direction changes
is {-a,b,-c,d}, resulting in three sign changes.

7.5

Additional Proofs for Special Cases

A discussion of special cases for the identification of mirrored topologies was provided in
Section 7.4.2. In this Section, additional proofs supporting the successful identification of
further special cases are presented.

140

7.5.1

Proof for Odd Crossings

Our objective is to prove that the sum of direction changes in an OD-path, having an
odd number of OD-edge crossings (cx) over an imaginary straight line between source and
destination, is equal to 0 ◦ .

7.5.1.1

OD-paths with cx > 1

Let us consider an OD-path with 3 crossings, Figure 7.9.

Figure 7.9: OD-path with 3 Edges Crossing

141

Proof. Sum of direction changes, angles a, b, c, d, e, f , from OD-path in Figure 7.9, is equal
to 0 ◦ :
a+b+c+d+e+f =0
Considering the triangle interior angles sum theorem:

a′ + b′ + g ′ = 180 ◦
h′ + c′ + g ′ = 180 ◦
h′ + i′ + d′ = 180 ◦
f ′ + i′ + e′ = 180 ◦

We can write:

a′ + b′ + g ′ = h′ + c′ + f ′
a′ + b′ = h′ + c′

(7.7)

h′ + i′ + d′ = f ′ + i′ + e′
h′ + d′ = f ′ + e′
h′ = f ′ + e′ − d′

(7.8)

Replacing h′ in Equation (7.7) with Equation (7.8) we obtain:

a′ + b′ = f ′ + e′ − d′ + c
a′ + b′ + d = f ′ + e′ + c′
142

(7.9)

Considering the supplementary angles:

a′ = 180 ◦ − a
b′ = 180 ◦ − b
c′ = 180 ◦ − c
d′ = 180 ◦ − d
e′ = 180 ◦ − e
f ′ = 180 ◦ − f

Then, Equation (7.9) can be rewritten as:

a+b+d = f +e+c

(7.10)

Equation (7.10), shows that the magnitude of (a + b + d) is equal to the magnitude of
(f + e + c). Additionally, from Figure 7.9 it is possible to notice that angles c, e, and f have
opposite rotation direction than angles a, b, and d. Considering the rotation direction we
can write that:
f + e + c = −(a + b + d)
Then, the sum of the angles:

(a + b + d) + (f + e + c) = (a + b + d) − (a + b + d) = 0 ◦

143

(7.11)

7.5.2

Proof for Even Crossings

Our objective is to prove the following. First, the sum of direction changes in a true OD-path
that has an even number of OD-edge crossings (cx) over an imaginary straight line between
source and destination, is equal to 360 ◦. Second, the sum of direction changes is equal to
zero for mirrored OD-paths with even number of OD-edge crossings.

7.5.2.1

OD-path with cx = 2

Figure 7.10: True OD-path with 2 Edges Crossing

Proof. The sum of direction changes (angles a, b, c, d and e) from a true OD-path with 2
OD-edges crossing, as shown in Fig. 7.10, is equal to 360 ◦:

a + b + c + d + e = 360 ◦
144

Let us consider the triangle interior angles sum theorem:

a′ + b′ + f ′ = 180 ◦
f ′ + c′ + g ′ = 180 ◦
g ′ + e′ + d′ = 180 ◦
g ′ = 180 − d′ − e′

Therefore:

a′ + b′ + f ′ = f ′ + c′ + g ′
a′ + b′ = c′ + g ′

(7.12)

Replacing g ′ in Equation (7.12):

a′ + b′ = c′ + 180 − d′ − e′
a′ + b′ + d′ + e′ = 180 + c′

Considering the supplementary angles in Fig. 7.10, we have:

a′ = 180 ◦ − a
b′ = 180 ◦ − b
c′ = 180 ◦ − c
d′ = 180 ◦ − d
e′ = 180 ◦ − e
145

(7.13)

Then Equation (7.13) can be rewritten as:

a + b + d + e = 360 ◦ + c
c = a + b + d + e − 360 ◦

From Fig. 7.10 it is possible to notice that angle c has opposite rotation direction than
angles a, b, d, and e. Then, we can write:

c = −(a + b + d + e − 360 ◦)

Therefore, the sum of the angles can be expressed as:

a + b + d + e + c = a + b + d + e − a − b − d − e + 360 ◦
a + b + d + e + c = 360 ◦

However, the sum of direction changes (angles a, b, c, d and e) in the mirrored OD-path
with two OD-edges crossing, depicted in Fig. 7.11, is equal to 0 ◦ .
Proof. The sum of direction changes (angles a, b, c, d and e) from a mirrored OD-path with
2 OD-edges crossing is equal to 0 ◦ :

a +b+c+d+e = 0◦

146

Figure 7.11: Mirrored OD-path with 2 Edges Crossing
Let us consider the triangle interior angles sum theorem:

e + d′ + g ′ = 180 ◦
c′ + g ′ + f ′ = 180 ◦
f ′ + b′ + a = 180 ◦
f ′ = 180 − b′ − a

Therefore:

e + d ′ + g ′ = c′ + g ′ + f ′
e + d ′ = c′ + f ′

147

(7.14)

Replacing f ′ in Equation (7.14):

e + d′ = c′ + 180 − b′ − a
a + e − c′ = 180 − d′ − b′

(7.15)

Considering the supplementary angles in Fig. 7.11, we have:

b′ = 180 ◦ − b
c′ = 180 ◦ − c
d′ = 180 ◦ − d

Then, Equation (7.15) can be rewritten as:

a+e+c = d+b

From Fig. 7.11 it is possible to notice that angles b and d have opposite rotation direction
than angles a, e, and c. Therefore, we can write that:

d + b = −(a + e + c)

Hence,
a + b + d + e + c = (a + e + c) − (a + e + c)
a+b +d+e+c = 0◦

148

7.5.2.2

OD-paths with cx > 2

Figure 7.12: True OD-path with 4 Edges Crossing

OD-path with more even number of crossings can be proven to have the sum of direction
changes equal to 360 ◦ in a similar manner to OD-path with 2 OD-edges crossing.
For instance, let us consider a true OD-path with four crossings, Fig. 7.12. It can be
proved that:

a + b + d + f + g = 360 ◦ + c + e

149

From the Fig. 7.12, c and e show to have opposite rotation direction than angles a, b, d,
f , g. Considering the rotation direction we can write:

c + e = −(a + b + d + f + g − 360 ◦ )

Consequently, the sum of the angles is equal to 360 ◦:

a+b+d+f +g+e+c
= a + b + d + f + g − (a + b + d + f + g − 360 ◦ )
= 360 ◦

We have proved that the sum of direction changes in OD-path with even number of
crossings yields the same results as with OD-paths with no crossings: 360 ◦ for true ODpaths, and 0 ◦ for mirrored OD-paths.

150

7.6

Summary

A node’s relative orientation presents a new alternative to solve wireless sensor network
problems otherwise solved with localization technology that is sometimes difficult to incorporate due to network limitations. The solution presented in this chapter takes advantage
of jumping nodes that spin while airborne, to obtain neighbor nodes’ relative orientation.
Results from field experiments show the viability of our approach. Our proposed algorithms
can be adapted in a WSN to work in a distributed manner. Moreover, having an oriented
network topology relieves the network from precise deployment of beacon nodes with location
information, thereby, reducing the number of such nodes in the network.
Future areas of work will seek to enhance the proposed orientation algorithms to be more
noise tolerant. Likewise, oriented networks present an opportunity to solve network-dynamics
problems, such as obstacle and boundary identification.

151

Chapter 8
SUMMARY AND FUTURE WORK

8.1

Summary

This dissertation discussed several research topics addressing issues related to wireless sensor
networks (WSN) connectivity and area coverage problems. First, in Chapter 3, a discussion of
the primary factors that impact signal propagation of wireless sensors is provided and studied
through experimental evaluations. Changes in sensor communication range are studied by
varying sensors’ heights relative to the surface. Results indicated that transmitting from
higher than ground level increases the wireless communication range of the sensor.
The aforementioned findings served as motivation to propose a novel communication
technique that relies on the jumping capabilities of sensors. While the jumping sensor robots
are airborne, the change in elevation enhances their ability for a short time to successfully
communicate with other sensors that are out of communication range at the ground level.
Field experiments were conducted and results showed a considerable improvement in wireless
communication ranges. An increase from 5m-10m to 30m-60m was seen when the sensor
jumps to an average height of 1 meter. Additional, a two-way airborne communication
process is defined and tested with sensor nodes, proving to be effective to communicate with
distant receivers.
Second, in Chapter 4 the relation between network connectivity and area coverage is
discussed. A new Hopping Sensor Network Model (HSNM) is introduced. The model high152

lights the factors that have to be taken into account in a jumping sensor network in order to
increase the network connectivity, and thus increasing the network covered area. Furthermore, a Hopping Sensor Routing Protocol (HSRP) is designed from the HSNM to balance
the energy consumption in the network. Simulation results showed a 20% network energy
savings when HSRP was employed, resulting in a longer network utile life.
Third, in order to improve base station connectivity in a jumping sensor network, and
to overcome communication holes, two decentralized algorithms for the discovery of isolated
sensor nodes are defined in Chapter 5. The algorithms employ only those jumping sensor that
are located on the boundary of the clusters to perform discovery, while saving network energy.
The main difference between the discovery algorithms is on the selection of boundary nodes to
perform discovery, one being more selective and energy preserving than the other. Simulation
results demonstrated the efficiency of the algorithms to increase network connectivity, and
to reduce the network energy consumption for discovery. From all the evaluated topology
scenarios, the algorithms were able to achieve 100% network connectivity. With regards
to network wide energy consumption for discovery tasks, results showed significant savings,
57% energy saving for aggressive discovery and 79% energy saving for smart discovery.
Fourth, Chapter 6 presents an unique fitness evaluation methodology for wireless sensor network topologies. Network connectivity is crucial for a network to be functional.
Nonetheless, network responsiveness and its capacity to meet communication flow demands
are important as well. This work evaluated network performance aspects related to network
connectivity from the application’s perspective. A Quality of Connectivity (QoC) performance metric is defined to evaluate the network connectivity in terms of how geographically
distributed the sensor nodes are in the topology. The metric evaluates and assigns a score
to the network topology, which serves as an indicator of how susceptible is the topology to
153

communication problems and latency.
Centralized and distributed algorithm approaches are presented to maximize network
performance. Maximizing network performance with fewer operations becomes important in
WSN where energy and computing resources are limiting factors. The algorithms make smart
cluster merging operations attempting to yield maximum QoC for each operation. Network
QoC improvement and the complexity of executing the approaches is evaluated in multiple
and diverse network topology scenarios. From all the algorithm approaches, Distributed Best
Merge Next from base station (DBMN-BS) was discussed to be less computing intensive than
its centralized counterpart, Best Merge Next from base station (BMN-BS), having an equally
outstanding network QoC yield per operation.
Last, a node’s relative orientation presents a new alternative to solve wireless sensor
network problems otherwise solved with localization technology that is sometimes difficult to
incorporate due to network limitations. The solution presented in Chapter 7 takes advantage
of jumping nodes that spin while airborne, to obtain neighbor nodes’ relative orientation.
Results from field experiments show the viability of our approach. Our proposed algorithms
can be adapted in a WSN to work in a distributed manner. Moreover, having an oriented
network topology relieves the network from precise deployment of beacon nodes with location
information, thereby, reducing the number of such nodes in the network. Likewise, oriented
networks present an opportunity to solve network-dynamics problems, such as obstacle and
boundary identification.

154

8.2

Future Work

This research established the basis for sensor communications and exploration techniques
that were beyond the available technology of half a decade ago. Jumping sensors provide an
alternative to address common wireless sensor network problems. Moreover, the development
of newer type of sensor robots will provide opportunities to deploy and study wireless sensor
networks from different perspectives.
Future work is envisioned to further enhance and incorporate the proposed approaches
aimed to improve network connectivity and area coverage with state of the art technology
available today. In particular, the following topics are proposed to serve as future directions.

8.2.1

Heterogeneous Wireless Sensor Network Test-bed

Environmental monitoring with heterogeneous wireless sensor networks is an encouraging, yet
challenging area of research. Heterogeneous networks are characterized for having diverse sets
of specialized nodes, each presenting unique abilities, such as, the type of movement, making
them suitable for adverse scenarios. Moreover, there has been an increase in autonomous
vehicles research recently. It would be interesting to adapt such to a multi-robot testbed, incorporating the use of jumping sensors. Furthermore, jumping sensor prototypes
have advanced to the point where Airborne Sensor Communication (ASC) can start to be
implemented and tested in further challenging scenarios.
The usage of multi-robot environments poses new research questions. For instance, from
the deployment perspective, what factors should be considered to establish the quantity
distribution for different types of sensor nodes? Additionally, how should special node units
be allocated to needed areas?

155

As an initiative, the work introduced in Chapter 6 to increase networks performance
by optimizing sensor nodes deployment can be extended to apply in heterogeneous sensor
networks. Efficient node deployment should be re-defined to consider the diversity in nodes
characteristic presented in such networks.

8.2.2

Neighbor Discovery

This work developed a jumping sensor neighbor discovery protocol, and enhanced it to be
more energy efficient by reducing the selection of nodes to perform discovery (see Chapter 5). The enhanced algorithm performs discovery making use of two-hops boundary nodes
neighbors of the initially selected nodes. It is motivating to focus research to optimize the
selection of nodes to perform discovery, i.e., find the minimum discoverer set.
The discovery node reduction hypothesis relies on the fact that once a boundary node
is selected to perform discovery, its communication airborne range grows (a factor of 6 on
average for 1 meter jump height, see Chapter 3) such that the contribution of two-hop away
boundary neighbors may still be minimal. Finding the minimum discoverer set is a difficult
problem for wireless sensor nodes without GPS (Global Position Systems) capabilities, where
there is not a precise sensor location. However, the availability of sensor’s relative orientation
(discussed in Chapter 7) allows for newer solutions to be explored.

156

BIBLIOGRAPHY

157

BIBLIOGRAPHY
802.15.4-IEEE-Standard (2006), 802.15.4-2006 IEEE Standard for Information technology - Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer (PHY)
Specifications for Low-Rate Wireless Personal Area Networks (WPANs), . (Cited on
page 87.)
Aeronautics and Space Engineering Board, N. R. C. (1995), The
Global Positioning System:
A Shared National Asset:
Recommendations for
Technical Improvements and Enhancements, National Academy Press, URL
http://books.google.com/books?id=FAHk65slfY4C. (Cited on page 23.)
Aguayo, D., Bicket, J., Biswas, S., Judd, G. and Morris, R. (2004), Link-level measurements from an 802.11b mesh network, SIGCOMM Comput. Commun. Rev., vol. 34 (4),
p. 121–132. (Cited on page 12.)
Aissani, M., Mellouk, A., Badache, N. and Djebbar, M. (2008), A New Approach
of Announcement and Avoiding Routing Voids in Wireless Sensor Networks, in Global
Telecommunications Conference. GLOBECOM 2008. IEEE, p. 1–5. (Cited on page 66.)
Balakrishnan, H., Baliga, R., Curtis, D., Goraczko, M., Miu, A., Priyantha,
B., Smith, A., Steele, K., Teller, S. and Wang, K. (2003), Lessons from Developing
and Deploying the Cricket Indoor Location System, Preprint., Massachusetts Institute of
Technology. (Cited on page 24.)
Biaz, S., Yiming, J., Qi, B. and Wu, S. (2005), Dynamic signal strength estimates
for indoor wireless communications, in Proceedings of the International Conference on
Wireless Communications, Networking and Mobile Computing (WiCOM ’05), vol. 1, p.
602–605. (Cited on page 12.)
Burdick, J. and Fiorini, P. (2003), Minimalist Jumping Robots for Celestial Exploration,
The International Journal of Robotics Research, vol. Vol. 22 (No. 7-8), p. 653–674. (Cited
on page 21.)
Camp, T., Boleng, J. and Davies, V. (2002), A Survey of Mobility Models for Ad
Hoc Network Research, Wireless Communications & Mobile Computing (WCMC): Special
Issue on Mobile Ad Hoc Networking: Research, Trends and Applications, vol. 2, p. 483–502.
(Cited on page 14.)
Carbunar, B., Grama, A., Vitek, J. and Carbunar, O. (2006), Redundancy and
Coverage Detection in Sensor Networks, ACM Transactions on Sensor Networks, vol. 2 (1),
p. 35. (Cited on page 11.)
Cen, Z. and Mutka, M. W. (2008), Relocation of Hopping Sensors, in ICRA ’08: Proceedings of The IEEE International Conference on Robotics and Automation, p. 569–574.
(Cited on pages 22 and 115.)
158

Cheng, Y.-M. and Yen, L.-H. (2006), Range-Based Density Control for Wireless Sensor Networks, in Proceedings of the 4th Annual Communication Networks and Services
Research Conference (CNSR ’06), p. 170–180, Washington, DC, USA. (Cited on page 11.)
´
Cintron, F. J. and Mutka, M. W. (2010), Hopping Enhanced Sensors for Efficient
Sensor Network Connectivity and Coverage, in MASS ’10: Proceedings of the 7th IEEE
International Conference on Mobile Ad-hoc and Sensor Systems, San Francisco. (Cited on
pages 89 and 120.)
´
Cintron, F. J., Pongaliur, K., Mutka, M. W. and Xiao, L. (2009), Energy Balancing
Hopping Sensor Network Model to Maximize Coverage, in ICCCN ’09: Proceedings of The
IEEE 18th International Conference on Computer Communications and Networks, p. 1–6.
(Cited on page 85.)
´
Cintron, F. J., Pongaliur, K., Mutka, M. W., Xiao, L., Zhao, J. and Xi, N.
(2012), Leveraging Height in a Jumping Sensor Network to Extend Network Coverage,
Wireless Communications, IEEE Transactions on, vol. 11 (5), p. 1840 –1849. (Cited on
page 85.)
Ding, Y., Wang, C. and Xiao, L. (2007), A connectivity based partition approach for
node scheduling in sensor networks, in DCOSS’07: Proceedings of the 3rd IEEE international conference on Distributed computing in sensor systems, p. 354–367, Springer-Verlag,
Berlin, Heidelberg. (Cited on page 12.)
Du, X., Mandala, D., Zhang, W., You, C. and Xiao, Y. (2007), A Boundary-Node
based Localization Scheme for Heterogeneous Wireless Sensor Networks, in Military Communications Conference. MILCOM 2007. IEEE, p. 1–7. (Cited on page 66.)
Dyo, V., Ellwood, S. A., Macdonald, D. W., Markham, A., Mascolo, C.,
´
Pasztor, B., Scellato, S., Trigoni, N., Wohlers, R. and Yousef, K. (2010),
Evolution and Sustainability of a Wildlife Monitoring Sensor Network, in SenSys ’10:
Proceedings of the 8th ACM Conference on Embedded Networked Sensor Systems, p. 127–
140, ACM, Zurich, Switzerland, URL http://doi.acm.org/10.1145/1869983.1869997.
(Cited on page 90.)
Fayed, M. and Mouftah, H. T. (2009), Localised convex hulls to identify boundary nodes
in sensor networks, Int. J. Sen. Netw., vol. 5 (2), p. 112–125. (Cited on page 65.)
German,
J. (2000), Hop to it:
Sandia hoppers leapfrog conventional
wisdom about robot mobility, Sandia Lab News, vol. 52 (21), URL
http://www.sandia.gov/LabNews/LN10-20-00/hop_story.html. (Cited on page 21.)
Goldsmith, A. (2005), Wireless Communications, Cambridge University Press. (Cited on
page 24.)
Green, D. and Obaidat, A. (2002), An accurate line of sight propagation performance
model for ad-hoc 802.11 wireless LAN (WLAN) devices, in ICC ’02: Proceedings of The
IEEE International Conference on Communications, vol. 5, p. 3424 – 3428 vol.5. (Cited
on pages 30 and 31.)
159

Han, Z. and Poor, H. (2009), Coalition Games with Cooperative Transmission: A Cure
for the Curse of Boundary Nodes in Selfish Packet-Forwarding Wireless Networks, Communications, IEEE Transactions on, vol. 57 (1), p. 203–213. (Cited on page 66.)
Hata, M. (1980), Empirical formula for propagation loss in land mobile radio
services, IEEE Transactions on Vehicular Technology, vol. 29 (3), p. 317–325, URL
http://dx.doi.org/10.1109/T-VT.1980.23859. (Cited on page 32.)
Howard, A., Mataric, M. and Sukhatme, G. S. (2002), Mobile Sensor Network Deployment using Potential Fields: A Distributed, Scalable Solution to the Area Coverage Problem, in DARS ’02: Proceedings of The Sixth International Symposium
on Distributed Autonomous Robotics Systems, p. 299–308, Fukupka, Japan, URL
http://cres.usc.edu/pubdb_html/files_upload/71.pdf. (Cited on page 13.)
Kavak, A., Yang, W., Dandekar, K. R. and Xu, G. (1999), Effects of base station
antenna height and mobile terminal movement on the vector propagation channels, IEEE
49th Vehicular Technology Conference (VTC 1999), vol. 1, p. 777–781 vol.1. (Cited on
page 12.)
Kim, M. and Mutka, M. W. (2009a), Multipath-based relocation schemes considering balanced assignment for hopping sensors, in IROS’09: Proceedings of the IEEE/RSJ international conference on Intelligent robots and systems, p. 5095–5100, IEEE Press, Piscataway,
NJ, USA. (Cited on page 22.)
Kim, M. and Mutka, M. W. (2009b), On Relocation of Hopping Sensors for Balanced
Migration Distribution of Sensors, in Proceedings of the International Conference on Computational Science and Its Applications: Part II, ICCSA ’09, p. 361–371, Springer-Verlag,
Berlin, Heidelberg, URL http://dx.doi.org/10.1007/978-3-642-02457-3_31. (Cited
on page 22.)
Kim, M., Mutka, M. W. and Choo, H. (2010), On Relocation of Hopping Sensors for
Rugged Terrains, in ICCSA ’10: Proceedings of the International Conference on Computational Science and its Applications, vol. 0, p. 203–210, IEEE Computer Society, Los
Alamitos, CA, USA. (Cited on page 22.)
Kovac, M., Fuchs, M., Guignard, A., Zufferey, J.-C. and Floreano, D. (2008),
A miniature 7g jumping robot, in ICRA ’08: Proceedings of the IEEE International Conference on Robotics and Automation, p. 373–378. (Cited on page 21.)
Lee, W. C. Y. (1986), Mobile Communications Design Fundamentals, Howard W. Sams &
Co., Indianapolis, IN, USA. (Cited on page 30.)
Li, N., Hou, J. C. and Sha, L. (2003), Design and Analysis of an MST-based Topology
Control Algorithm, in INFOCOM ’03. Proceedings of The Twenty-Second Annual Joint
Conference of the IEEE Computer and Communications Societies, vol. 3, p. 1702–1712.
(Cited on page 18.)

160

Lin, C.-Y., Peng, W.-C. and Tseng, Y.-C. (2006), Efficient in-network moving object
tracking in wireless sensor networks, Mobile Computing, IEEE Transactions on, vol. 5 (8),
p. 1044–1056. (Cited on page 90.)
Liu, C.-M. and Lee, C.-H. (2004), Power efficient communication protocols for data gathering on mobile sensor networks, Vehicular Technology Conference, 2004. VTC2004-Fall.
2004 IEEE 60th, vol. 7, p. 4635–4639. (Cited on page 11.)
Liu, C.-M., Lee, C.-H. and Wang, L.-C. (2004), Power-efficient communication algorithms for wireless mobile sensor networks, in PE-WASUN ’04: Proceedings of the 1st
ACM international workshop on Performance evaluation of wireless ad hoc, sensor, and
ubiquitous networks, p. 121–122, ACM, NY, USA. (Cited on page 11.)
Longley, A. and Rice, P. (1968), Prediction of Tropospheric Radio Transmission over
Irregular Terrrain, A Computer Method, Technical report, ESSA Tech. Rep. ERL 79-ITS
67, Washington, DC. (Cited on page 32.)
Mogensen, P. E., Jensen, C. and Andersen, J. B. (1991), 1800 MHz Mobile Net
Planning Based based on 900 MHz Measurements, in COST 231 TD(91)-008, University
of Aalborg, Firenze. (Cited on page 32.)
´
Pei, Y., Cintron, F. J., Mutka, M. W., Zhao, J. and Xi, N. (2009), Hopping Sensor
Relocation in Rugged Terrain, in The IEEE/RSJ International Conference on Intelligent
RObots and Systems (IROS 2009), . (Cited on pages 22 and 115.)
Priyantha, N. B., Miu, A. K., Balakrishnan, H. and Teller, S. (2001), The Cricket
Compass for Context-Aware Mobile Applications, in 7th ACM MOBICOM, Rome, Italy.
(Cited on page 24.)
Ramanathan, R. and Rosales-Hain, R. (2000), Topology control of multihop wireless
networks using transmit power adjustment, in INFOCOM ’00: Proceedings of The Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies.,
vol. 2, p. 404–413. (Cited on pages 13, 16 and 80.)
¨
Romer, K. and Mattern, F. (2004), The Design Space of Wireless Sensor Networks,
IEEE Wireless Communications, vol. 11 (6), p. 54–61. (Cited on page 9.)
Sahoo, P., Hsieh, K.-Y. and Sheu, J.-P. (2007), Boundary Node Selection and Target
Detection in Wireless Sensor Network, in Wireless and Optical Communications Networks.
WOCN ’07. IFIP International Conference on, p. 1–5. (Cited on page 65.)
Santi, P. (2005), Topology control in wireless ad hoc and sensor networks, John Wiley &
Sons Ltd. (Cited on page 15.)
Scarfogliero, U., Stefanini, C. and Dario, P. (2007), Design and Development of the
Long-Jumping “Grillo” Mini Robot, in ICRA ’07: Proceedings of the IEEE International
Conference on Robotics and Automation, p. 467–472. (Cited on page 21.)

161

Shah, R. and Rabaey, J. (2002), Energy aware routing for low energy ad hoc sensor
networks, in Wireless Communications and Networking Conference. WCNC2002. IEEE,
vol. 1, p. 350 – 355. (Cited on page 11.)
Song, W.-Z., Huang, R., Xu, M., Ma, A., Shirazi, B. and LaHusen, R. (2009),
Air-dropped sensor network for real-time high-fidelity volcano monitoring, in MobiSys
’09: Proceedings of the 7th international conference on Mobile systems, applications, and
services, p. 305–318, ACM, New York, NY, USA. (Cited on pages 24, 64 and 83.)
TIA TR8 Working Group 8.8 Technology Compatibility (1997), A Report on
Technology Independent Methodology for the Modeling, Simulation and Empirical Verification of Wireless Communications System Performance in Noise and Interference Limited
Systems Operating on Frequencies between 30 and 1500MHz, Technical report, TR-8 Mobile and Personal Private Radio Standards Committee, TELECOMMUNICATIONS INDUSTRY ASSOCIATION (TIA) Mobile and Personal Communications, IEEE Vehicular
Technology Society Propagation Committee. (Cited on page 32.)
Wang, X., Xing, G., Zhang, Y., Lu, C., Pless, R. and Gill, C. (2003), Integrated
Coverage and Connectivity Configuration in Wireless Sensor Networks, in SenSys ’03:
Proceedings of the 1st international conference on Embedded networked sensor systems,
p. 28–39, ACM, New York, NY, USA. (Cited on page 12.)
Warneke, B., Last, M., Liebowitz, B. and Pister, K. (2001), Smart Dust: communicating with a cubic-millimeter computer, Computer, vol. 34 (1), p. 44–51. (Cited on
page 22.)
Wei, T. E., Nelson, G. M., Quinn, R. D., Verma, H. and Garverick, S. L. (2000),
Design of a 5-cm monopod hopping robot, in ICRA ’00: Proceedings of the IEEE International Conference on Robotics and Automation, vol. 3, p. 2828–2833 vol.3. (Cited on
page 21.)
Whitehouse, C. D. (2002), The design of calamari: an ad-hoc localization system for
sensor networks, Master’s thesis, University of California at Berkeley. (Cited on page 24.)
Wu, X. and Liu, M. (2012), In-situ soil moisture sensing: measurement scheduling and
estimation using compressive sensing, in IPSN ’12: Proceedings of the 11th international
conference on Information Processing in Sensor Networks, p. 1–12, ACM, Beijing, China,
URL http://doi.acm.org/10.1145/2185677.2185679. (Cited on page 90.)
Yanmaz, E. (28-30 April 2008), Impact of Topology-dependent and independent Mobility
Models on the Connectivity of Wireless Networks, IEEE Sarnoff Symposium, p. 1–5. (Cited
on page 16.)
Younis, M., Munshi, P., Gupta, G. and Elsharkawy, S. M. (2006), On Efficient
Clustering of Wireless Sensor Networks, in DSSNS ’06: Proceedings of the Second IEEE
Workshop on Dependability and Security in Sensor Networks and Systems, p. 78–91.
(Cited on page 12.)
162

Zhang, H. and Hou, J. C. (2005), Maintaining Sensing Coverage and Connectivity in
Large Sensor Networks, Ad Hoc & Sensor Wireless Networks, vol. 1 (1-2), p. 89–124, URL
http://www.oldcitypublishing.com/AHSWN/AHSWN1.1-2abstracts/Zhangabs.html.
(Cited on page 12.)
Zhao, J., Yang, R., Xi, N., Gao, B., Fan, X., Mutka, M. W. and Xiao, L. (2009),
Development of a Miniature Self-stabilization Jumping Robot, in IROS ’09: Proceedings
of The IEEE/RSJ International Conference on Intelligent RObots and Systems, p. 2217–
2222. (Cited on pages xiii, 4, 5, 22 and 71.)
Zhao, J., Xi, N., Gao, B., Mutka, M. W. and Xiao, L. (2010), Design and Testing
of a Controllable Miniature Jumping Robot, in IROS ’10: Proceedings of The IEEE/RSJ
International Conference on Intelligent Robots and Systems, . (Cited on page 22.)
Zhao, J., Xu, J., Gao, B., Xi, N., Cintron, F. J., Mutka, M. W. and Xiao, L.
(2013), MSU Jumper: A Single-Motor-Actuated Miniature Steerable Jumping Robot,
IEEE transactions on Robotics, vol. 29 (3), p. In Press. (Cited on page 85.)
Zou, Y. and Chakrabarty, K. (2003), Sensor Deployment and Target Localization Based
on Virtual Forces, in INFOCOM ’03: Proceedings of The Twenty-Second Annual Joint
Conference of the IEEE Computer and Communications Societies, vol. 2, p. 1293–1303
vol.2. (Cited on page 13.)

163