Modeling and Performance Evaluation of Wireless Body Area ...

Modeling and Performance Evaluation of Wireless Body Area

Networks for Healthcare Applications

A Dissertation submitted to the

Division of Research and Advanced Studies

of the University of Cincinnati

In partial fulfillment of the requirements

for the degree of

DOCTOR OF PHILOSOPHY

in the Department of Electrical Engineering and Computing Systems

of the College of Engineering and Applied Science

University of Cincinnati

July, 2015

by

Amitabh Mishra

M.Tech. Instrumentation

Devi Ahilya University, Indore, India

July, 2001

Thesis Advisor and Committee Chair: Dharma P. Agrawal

Dissertation Advisor Author

Dharma P. Agrawal Amitabh Mishra

ii

Abstract

Wireless Body Area Network (WBAN) is a low-power Personal Area Network involving sensor nodes

(SNs) that sense and relay physiological data to a central station. WBANs are new and still evolving. We

try to address three open research areas involving WBANs.

The limited energy budget in WBANs necessitates energy conservation to prolong the network lifetime.

The first challenge we try to address is related to improvement of the lifetime of a WBAN, given the

small sizes of body sensor nodes (SNs) and the limited battery power that they run on.

We proposed a dual-prediction framework for improvement of network lifetime. The framework allows

for minimizing data transmission involving four important body parameters by reconstructing their

information by time series prediction at reception. A sample elimination algorithm further optimizes the

framework performance. We enhanced the framework by reducing the sampling frequency and

implementing the algorithm on top, increasing the network lifetime further. The missing samples were

reconstructed by interpolation at the receiver. We probed the effects of adaptive sampling and evaluated

the increase in battery lifetime in WBANs.

We then tried to test the behavior of a WBAN in the presence of other WBANs around it and check the

issues faced by WBANs. Wireless systems can face severe interference problems if they use the same

communication channels at a time. There are issues related to data routing because the critical nature of

WBAN data requires assured communication of body data. For optimum network utilization, efficient

scheduling of transmissions in multiple co-existing WBANs is important in order to avoid intra and inter-

WBAN interference and for a graceful coexistence. We propose that inter-WBAN interference can be

avoided by a QoS based MAC scheduling approach and that intra-WBANs interference can be

circumvented by fuzzy scheduling of intra-WBAN transmissions.

We also propose to use interference to the benefit of WBANs through a framework in which neighboring

WBANs communicate for cooperative packet routing. This lets the WBANs use some spare transmission

iii

slots from their neighbor WBANs when required. This can happen when a WBAN with more or sudden,

emergency data is strapped for transmission slots while its neighbor has some to spare. A routing tree is

created using a weighted two-pass algorithm involving an assisting WBAN that can accommodate routing

requests from its neighbor WBAN.

We further evaluated the possibility of relaying data from a mobile WBAN through small scale networks

for voice communication. The scheme uses dynamic virtual cells that grow and shrink in order to provide

uninterrupted service, while reducing handovers.

Although wireless systems are reliable in conveying sensor data but their use for control applications is

still nascent. We tried to probe if WSNs in general or WBANs in particular could be used for wireless

control. We have evaluated the performance of ON-OFF control involving a wireless sensor network for

musical entertainment applications. We further extended our work and tested the feasibility of control in

WSNs and in more critical real life applications in WBANs.

iv

c 2015 - Amitabh Mishra

All rights reserved.

v

Acknowledgments

I find myself lacking in expression in extending my profound sense of respect and deepest gratitude to Dr.

Dharma Agrawal for his precise guidance, gracious encouragement and being proactive in counseling and

helping me whenever required. His valuable advice and teaching helped me immensely by steering me

towards my goals leading up to this proposal. Moreover, his optimistic attitude, vision and appreciation

was such as to give impetus to my own thoughts and understandings, making me believe that, all that was

accomplished was of my own efforts for which I will ever remain indebted to him.

My sincere and special thanks are also due to Dr. Raj Bhatnagar, Dr. Prabir Bhattacharya, Dr. Chia Han,

and Dr. Marepalli Rao for accepting to be the members of my committee and for taking out the time

required for the procedure out of their immensely busy schedules. I am also indebted to them and to all

my instructors for their valued encouragement, utmost co-operation and continued interest in the shaping

of this dissertation. I appreciate this from my heart.

I reserve words of special gratefulness towards Dr. Jung Hyun Jun, Dr. Hailong Li, Dr. Kuheli Louha, Dr.

Nishan Waregama for their support and advice and helping me with my research.

I appreciate the affirmative attitude and the active motivation that I received from the members of my

research group. Their constructive criticism and valuable suggestions have helped me improve on the

quality of my work. Everyone is a winner in such a kind of civic involvement.

Dr. Prudhvi Janga, Dr. Vineet Joshi, Dr. Namita Mishra, Dr. Preeti Ramchandran, Divya Sardana and

Karthik V.M. deserve a special mention in this note for their help with tips and important pointers that

helped in getting me to this stage of research.

Finally I am thankful to my family for standing by me, supporting me, bearing with the difficult times,

my busy schedule and allowing me to work on my research while sacrificing on my company for the time

that rightly belonged to them. Had it not been for their contribution, I would not have been able to meet

the requirements and complete the job to my satisfaction.

Amitabh Mishra

vi

Table of Contents

Title Page ………………………………………………………………………………... i Abstract ………………………………………………………………………………... ii Acknowledgments ………………………………………………………………………………... v Table of Contents ………………………………………………………………………………... vi List of Figures ………………………………………………………………………………... ix List of Tables ………………………………………………………………………………... xi List of Abbreviations ………………………………………………………………………………... xii

1 Role of Wireless Body Area Networks in the IoT paradigm and challenges involved . . . . 1

1.1 The Internet of Things and WBANs within their broad scheme . . . . . . . . . . . . . . . . . . . . . 1

1.2 The IoT framework and pressing issues in healthcare . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Wireless Body Area Networks for healthcare monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.4 Motivation for research and related prior work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.4.1 Healthcare networks and potential applications of WBANs . . . . . . . . . . . . . . . . . . . . . . . . 7

1.4.2 Issues in WBAN measurements and sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.5 Energy conservation in WBANs: Prior work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.6 Behavior, functioning and challenges of coexistence for WBANs . . . . . . . . . . . . . . . . . . . 12

1.6.1 Addressing interference issues through scheduling and queuing mechanisms: Prior

work

. . . . 13

1.7 Wireless Sensor based systems for control applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.7.1 Control through Wireless Sensor Networks: Prior Work . . . . . . . . . . . . . . . . . . . . . . . . . 15

2 Network Lifetime enhancement in WBANs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.1 Constraints faced by sensor nodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.2 Error minimization and energy conservation in WBANs . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.2.1 Proposed framework and optimization using Linear Elimination Algorithm (LEA) . . . . 20

2.2.2 Performance check on error minimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.3 Evaluation of the impact of sample rate reduction on wave information . . . . . . . . . . . . . . . 28

2.3.1 Sample rate reduction and its relation with network energy saving . . . . . . . . . . . . . . . . . . . . 28

2.3.2 Impact of artifact filtering on amount of data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2.3.3 Prediction by numerical interpolation techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2.3.4 Results of signal processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.4 Energy aspects and evaluation of Network Lifetime . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . 37

2.5 Energy efficient design for WBAN monitoring using next generation Cellular

Channels

. . . . 41

2.5.1 WBAN data through existing communications systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.5.2 Motivation for WBAN energy enhancement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

2.5.3 Prior Work in WBAN energy enhancement . .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.5.4 Encoding the data and errors in encoding . . . . . . . . . . . . . . ... . . . . . . . . . . . . . . . . . . . . . . 45

2.5.5 Proposed Architecture and Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

2.5.6 Results: Spectral translation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2.5.7 Results: Data Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

2.6 Data Acquisition System (DAS) for WBANs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

vii

3 Behavior, functioning and challenges of coexistence in Wireless Body Area

Networks

. . . . 60

3.1 The importance of QoS in functioning of WBANs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.1.1 The 802.15.6 superframe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.2 Addressing the interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.3 Priority-based scheduling schemes for suppressing interference in WBANs . . . . . . . . . . 61

3.3.1 Intra-WBAN scheduling: Problem formulation and analytical modeling . . . . . . . . . . . . . . 63

3.3.2 Inter-WBAN interference avoidance: Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.4 Framework for a Cooperative, Neighbor-aware WBANs . . . . . . . . . . . . . . . . . . . . . . . . 70

3.4.1 Algorithm for Cooperative, Neighbor-aware WBANs. . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

3.4.2 Weight estimator for candidate node selection for cooperative inter-WBAN routing . . . . 71

3.4.3 Simulation schematic for Cooperative Inter-WBAN Routing. . . . . . . . . . . . . . . . . . . . . . . 72

3.4.4 Proposed framework for Cooperative Inter-WBAN Routing. . . .. . . . . . . . . . . . . . . . . . . . 73

3.4.5 Results and performance evaluation for Cooperative, Neighbor Aware Inter-WBAN

Routing

. . . . 75

3.5 Dynamic Virtual Cells with Multiple Multicast Trees for routing WBAN data . . . . . . . . . 78

3.5.1 A Dynamic Virtual Cell. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

3.5.2 Simulation of DVC based Multicast tree model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

3.5.2.1 Simulation Results from the DVC based Multicast tree model . . . . . . . . . . . . . . . . . . . . . 80

4 Evaluation on feasibility of control in Wireless Networks with a focus on WBANs . . . . 83

4.1 Reliability issues in Wireless Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4.2 WSNs and Computer Music . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4.2.1 Four dance performances using wireless control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4.2.2 Water Birds: Compositional Collaboration with Clarinets and Wireless Sensors . . . . . . 86

4.3 Wireless Control in an Industrial setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

4.4 Industrial applications with potential for Wireless Control. . . . . . . . . . . . . . . . . . . . . . . . 94

4.4.1 HART Protocol and WirelessHART . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

4.4.1.1 Simplicity of control through WirelessHART . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

4.4.1.2 Reliability of WirelessHART control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

4.4.1.3 Security of control in WirelessHART . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

4.4.2 The WirelessHART standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

4.4.2.1 Predecessor: the HART Standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . 99

4.4.2.2 WirelessHART Standard details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

4.5 WirelessHART Industrial Application model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

4.6 WirelessHART based framework for Control in WBAN: possible applications . . . . . . . 104

4.6.1 Proposed Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

4.6.2 Performance evaluation and feasibility check: Channel performance with sample

cut

. . . . 107

4.6.3 Evaluation of BER Confidence Level for WirelessHART Channel for the model . . . . . . 109

4.6.4 BER Confidence–level for WirelessHART model tested . . . . . . . . . . . . . . . . . . . . . . . . . 110

4.6.5 Performance evaluation: delay and throughput . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

4.7 Performance check on intrinsically safe routing models . . . . . . . . . . . . . . . . . . . . . . . . . . 115

4.7.1 Performance evaluation on traffic parameters of the model . . . . . . . . . . . . . . . . . . . . . . . 117

viii

5 Future Work . . . . 120

5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

5.2 Data Acquisition System and data compression for WBANs . . . . . . . . . . . . . . . . . . . . . 120

5.3 Evaluation of high concentration presence of WBANs . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

5.3.1 Dynamic virtual coverage for WBANs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

5.3.2 Cognitive routing of critical data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

5.4 Security in WBAN and IoT systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

Bibliography 127

Appendix 138

ix

List of Figures

1.1 Communication from Body transducers via the CSS to the Base Station (BS) . . . . . . . . . 2

1.2 IEEE vision of IoT framework and its applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Communication from Body transducers via the (CSS) to the Base Station (BS) . . . . . . . . 6

2.1 Result of comparison of four prediction algorithms based on sample history . . . . . . . . . . . 19

2.2 Performance of prediction for four cardiac signals using PID and NLR-Neural

Network

. . . . 24

2.3 Graphs indicating errors in prediction for the CVP and ECG Lead-II for PID and

NLR

. . . . 25

2.4 Plots for linear rebuild of the PulmAP signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

2.5 Plots for linear rebuild of the ECG Lead-II signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

2.6 Error plot for linear rebuild of the PAP signal . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . 32

2.7 Error plot for linear rebuild of the ECG Lead-II signal . . . . . . . . . . . . . .. . . . . . . . . . . . . . . 33

2.8 MSE for different sample cuts for the signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.9 Maximum percentage error for sample cuts for the signals . . . . . . . . . . . . . . . . . . . . . . . . 35

2.10 MSE for different techniques for the signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

2.11 Maximum percentage error for different techniques for the signals . . . . . . . . . . . . . . . . . . 36

2.12 Battery model for four commercial batteries and three SN types . . . . . . . . . . . . . . . . . . . . 40

2.13 5-G Enhancement for a typical WBAN BSN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

2.14 Frequency spectrum of the baseband signal for the ECG Lead II signal . . . . . . . . . . . . . . . 51

2.15 Spectral translation of the ECG Lead II signal to the human voice range . . . . . . . . . . . . . 52

2.16 Frequency spectrum of the baseband signal for the CVP signal . . . . . . . . . . . . . . . . . . . . . 53

2.17 Spectral translation of the CVP signal to the human voice range . . . . . . . . . . . . . . . . . . . . 54

2.18 The data being received by the computer and the smartphone . . . . . . . . . . . . . . . . . . . . . . 57

2.19 The experimental setup involving the microcontroller board and the sensor . . . . . . . . . . . 57

2.20 Accelerometer data received over email . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

2.21 Accelerometer data from three axes . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . 59

2.22 Frequency analysis of the accelerometer data to get the steps . . . . . . . . . . . . . . . . .. . . . . . 59

3.1 The structure of an IEEE 802.15.6 Superframe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.2 Zadeh’s basic Fuzzy logic system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3.3 Decision considering BER and Eb/N0 . . . . . . . . . . . . . . . .. . . . . . . . . . .. . . . . . . . . . .. . . . . 65

3.4 Decision considering SNR and Eb/N0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . 66

3.5 Decision considering BER and SNR . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . .. . . . . . . 66

3.6 Split of data packets from second WBAN at the first . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . 69

3.7 Cooperative routing from WBAN ‘A’ via ‘B’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.8 Creation of routing tree . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . .. . . . . . . . . . . . . . 74

3.9 Effective transmissions per packet with distance between SNs. . . . . . . . . . . . . . . . . . . . . . 75

3.10 Packet loss in routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . 76

3.11 Packet transmission time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . 76

3.12 Transmission time for hop distances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

3.13 Dynamic Virtual Cells in the SCN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

3.14 Throughput performance of DVCs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . 81

3.15 Delay performance of DVCs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

x

4.1 Flow of data into sound processing software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

4.2 A traditional, wired Instrumentation Control System . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . 88

4.3 A wireless Instrumentation Control System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . 89

4.4 Architecture of an Instrumentation and Control System built with wireless nodes . . . . . . . 92

4.5 A WirelessHART Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

4.6 The front panel for the liquid level control system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

4.7 The block diagram for the liquid level control system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

4.8 A plot of number of errors v/s the confidence levels with increasing time of

measurement

. . . . 112

4.9 Packet transmission delay v/s the payload in bytes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

4.10 Maximum throughput v/s the payload in bytes . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . 116

4.11 Amount of retransmitted packets for the three models . . . . . . . . . . . . . . . . . . . . . .. . . . . . . 119

4.12 A comparison of reduced lifetime hours for the three models . . . . . . . . .. . . . . .. . . . . . . . . 120

xi

List of Tables

2.1 Mean Square Error values for the two algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.2 Log of additional savings from LEA. . . . . . . . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . . . . . . 27

2.3 Signal specifications for the four parameters . . . . . . . .. . . . . . .. . . . . . . . . . . .. . . . . . . . . . 33

2.4 Maximum Error with sample reduction for the four physiological parameters . . . . . . . . . . 34

2.5 Error values for ECG-Lead II from the five numerical interpolation techniques . . . . . . . . 34

2.6 Life in days for the different battery models - capacities and node power

requirements

. . . . 39

2.7 Signal specifications for the four vital sign BSN parameters . . . . . . . . . . . . . . . . . . . . . . . . 47

3.1 The Fuzzy inference table for transmission error parameters . . . . . . . . . . . . . . . . . . . . . . . 65

3.2 Priority in IEEE 802.15.6 and traffic type . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . .. . . . . 67

3.3 Expected performance gain of the DVCs . . . . . . . . . . . . . . . . . . . .. . . . . . .. . . . . . . . . . . . . 80

4.1 Bitrate QoS requirements for common WBAN parameters . . . . .. . . . . . .. . . . . . . . . . . . . . 106

4.2 Performance results of WirelessHART model for the 24-channel WBAN model . . . . . . . 108

4.3 BER Measurement and confidence evaluation with changes in time window . . . . . . . . . . 112

xii

List of Abbreviations

AP . . . . . . . Access Point

ART . . . . . Arterial Pressure

BER . . . . . . Bit Error Rate

BS . . . . . . . . Base Station

CAP . . . . . . Contention Access Period

CFP . . . . . . . Contention Free Period

CDF . . . . . . Cumulative Distribution Function

CDMA . . . . Code Division Multiple Access

CR . . . . . . . Cognitive Radio

CSS . . . . . . . Coordinating Sink Station

CVS . . . . . . . Central Venous Pressure

DFS . . . . . . . Depth First Search

DQCA . . . . . Distributed Queuing Collision Avoidance

DQRAP . . . . Distributed Queuing Random Access MAC protocol

DVC . . . . . . Dynamic Virtual Cell

ECG . . . . . . Electro Cardio Gram

EV-DO . . . . Evolution-Data Optimized

FBS . . . . . . . Femto Base Station

FCC . . . . . . Federal Communications Commission

FCFS . . . . . First Come First Serve

FDMA . . . . Frequency Division Multiple Access

FRD . . . . . . Flow Rate Distribution

FUE . . . . . . Femto User Equipment

GSM . . . . . . Global System for Mobile Communication

HSDPA . . . . High-Speed Downlink Packet Access

HWN . . . . . Heterogeneous Wireless Networks

ICI . . . . . . . . Inter-cell Interference

IGW . . . . . . Internet Gateway

IP . . . . . . . . . Internet Protocol

ISM . . . . . . . Industrial, Scientific and Medical

LEA . . . . . . Linear Elimination Algorithm

LQI . . . . . . Link Quality Indicator

LTE . . . . . . . Long Term Evolution

MBS . . . . . . Macro Base Station

MIDI . . . . . . Musical Interface Digital Interface

MR . . . . . . . Mesh Router

MSE . . . . . Mean Square Error

OFDM . . . . Orthogonal Frequency Division Multiplexing

OFDMA . . . Orthogonal Frequency Division Multiple Access

PAP . . . . . . Priority Access Period

PulmAP . . . Pulmonary Artery Pressure

PU . . . . . . . Primary User

xiii

QoE . . . . . . . Quality of Experience

QoS . . . . . . . Quality of Service

RSSI . . . . . . Received Signal Strength Indicator

SCBS . . . . . Small Cell Base Station

SCC . . . . . Strongly connected component

SCN . . . . . Small Scale Networks

SDR . . . . . . Software Defined Radio

SFR . . . . . . . Soft Frequency Reuse

SINR . . . . . . Signal to Interference and Noise Ratio

SN . . . . . . Sensor Node

SNR . . . . . . Signal to Noise Ratio

SU . . . . . . . Secondary User

TBM . . . . . Transferable Belief Model

TDMA . . . . Time Division Multiple Access

UE . . . . . . . User Equipment

UMTS . . . . Universal Mobile Telecommunication Services

UWB . . . . . Ultra Wide Band

WiFi . . . . . Wireless Fidelity

WIMAX . . Worldwide Interoperability for Microwave Access

WLAN . . . . Wireless Local Area Network

WMN . . . . Wireless Mesh Network

WBAN . . . . Wireless Body Area Network

Chapter 1

Role of Wireless Body Area Networks in the IoT

paradigm and challenges involved

1.1 The Internet of Things and WBANs within their broad scheme

The Internet of Things (IoT) is a ground-breaking model that abstracts a pervasive

presence of a variety of objects with unique identification and communications

capability such as Radio-Frequency IDentification (RFID) tags, sensors, actuators, and

mobile phones around us at home, in workplace, or anywhere we go [1]. Using unique

addressing schemes for identification and their in-built communications capability,

such objects or Internet nodes would have the capability to interact with each other

and to cooperate with their neighbors in order to reach common goals [2]. It is

projected that by the year 2025, Internet nodes may reside in everyday objects like

furniture, paper documents, supermarket articles, food packages, and others used in

our day to day lives. For the business users, the most obvious outcomes will be visible

in fields such as automation, intelligent transportation of people and goods logistics,

industrial manufacturing, and healthcare. In this context, domestics, e-health, assisted

living, enhanced learning are only a few examples of possible application set-ups in

which the IoT paradigm will play an important role in the near future [3]. The WBANs

of the future are anticipated to be an important subset within the IoT paradigm, as

shown in figure 1.1.

Chapter 1: Role of Wireless Body Area Networks in the IoT paradigm and challenges involved

2

Figure 1.1: Envisioning the IoT [4]

Wireless Body Area Networks (WBANs) form an important prong of such wearable

technology. They involve the use of low power, low radio range sensor nodes for

sensing of physiological and bio-kinetic parameters and transmission of sensed data

using wireless link hops over a network. WBANs envisage a human-centric use of

wireless technology for personalized telehealth and telemedicine, and remove the

compulsion to stay confined to the bed or under the care of medical attendants or

doctors in a hospital. Apart from monitoring the physiological and bio-kinetic

parameters of patients and athletes, the concept can also be used in life-saving

applications, especially for the personnel who work in hazardous environments, like


3

first responders, fire-fighters, divers and astronauts. The rise in the cost of healthcare

around the world has proportionately increased the need for integrating WBAN

systems into the upcoming information technology and telecom infrastructure,

including the IoT.

1.2 The IoT framework and pressing issues in healthcare

According to the IEEE vision of the IoT framework, the stakeholders who would be

benefited by the advances in the IoT technology include hospitals and doctors,

appliance providers, application developers and consumers. Figure 1.2 depicts the

possible scenario involving applications domains and the stakeholders.

Figure 1.2: IEEE vision of IoT framework and its applications


4

Two of the major challenges in world health today are: increase in the life expectancy

that in turn causes an increase in the number of geriatrics, and rise in the cost of

healthcare itself. Based on current trends [3], studies indicate that the overall

healthcare expenditure of developed and developing countries is projected to reach

20% of the Gross Domestic Product (GDP) of these countries by 2022. This could

adversely affect the global world-wide economy.

It has been proved by research that an early detection in the initial stages can

prevent most of the ailments and diseases. This fact advocates that a proactive

wellness management with a focus on cost reduction should be ensured by the

healthcare systems of the future. Wearable monitoring systems that have started to

appear in the healthcare market could offer a possible solution to proactive and more

affordable healthcare systems. Such systems can affect early detection of abnormal

conditions and provide substantial betterment in the quality of human life.

WBANs contain an important hidden proficiency to remodel the future of

healthcare monitoring not only by doing away with the need for costly in-hospital

monitoring of patients, but also by helping in the diagnosis of several life threatening

diseases [5]. It is estimated that by 2020, cancer death rates might increase by 50%,

taking the toll up to 15 million [6]. WBAN based cancer cell monitoring can affect early

tumor diagnosis without a biopsy and offers promises for a timely analysis and early

treatment. Yet another dominant cause of death in the world is the cardiovascular


5

disease, which is assessed to cause nearly 30 percent of deaths worldwide [7, 8]. A

major portion of our work is focused on cardiovascular WBAN signals for this reason.

1.3 Wireless Body Area Networks for healthcare monitoring

Wireless Body Area Networks (WBANs) are a special implementation of Wireless

Sensor Networks (WSNs) in the field of healthcare and fitness that focus on sensing

and communication of physiological and biokinetic parameters, providing more

precise values at better rates of sampling than conventional patient data systems [9].

They can support biofeedback and interactivity for modern human-centric diagnostic

and fitness applications. A typical WBAN is a low-power Personal Area Network (PAN),

designed as a short-range, preferably single-hop network supporting low data rate

traffic, and is based on the IEEE 802.15 standards [10]. The main advantage of using a

WBAN is that the subject does not have to stay confined to a room or a bed and is free

to move around. WBANs use wireless transceivers for transmission of data over wired

or wireless links (illustrated in Figure 1.3), from body-mounted or implanted sensor

nodes (SNs) with transducers, to a coordinating and aggregating sink station (CSS)

[11], usually worn externally by the human subject. The CSS can then forward the data

to a central repository known as Base Station (BS) for analysis, archiving and decision

on corrective action by the physician if needed, as well as to various processing

stations. Further, personal medical data networks can be combined to scale up to a

much larger network for telemetry purposes [12].

There are five basic objectives of any man-Instrument System and they are

applicable to WBANs too. The objectives are: Information gathering, Evaluation,


6

Diagnosis, Monitoring and Control. Some important considerations applicable to

WBANs too, are as follows.

Figure 1.3: Communication from Body transducers via the Coordinating

Sink Station (CSS) to the Base Station (BS)

1.4 Motivation for research and related prior work

WBANs enable long-term monitoring and detection of physiological events from a

distance by transmission of sensed data to remote locations. The data could then be

used for diagnosis and could be archived if desired. The use of WBANs may not be


7

limited to human applications and extended to research on unrestrained,

unanaesthetized animals in their natural habitat.

1.4.1 Healthcare networks and potential applications of WBANs

WBANs can have broad spectrum usage from monitoring vital parameters of

astronauts in space to patient monitoring in locations away from the hospital or in an

ambulance. They can be particularly useful in patient monitoring where freedom of

movement is desired, say during exercise ECG. In such applications, trailing wires can

be cumbersome and dangerous. Fitness monitoring is fast catching up among healthy

human beings as well.

Special helmets with surface electrodes can be used for monitoring EEGs of

football players during a game or in cases of mentally challenged children, without

causing trauma to the subject.

In certain ailments like cases of gait in geriatric patients, it is commonly observed

that pains or other symptoms that have been the cause of trouble for several days can

disappear just before a medical appointment or during the checkup. Owing to its

unobtrusive setup, prolonged WBAN monitoring can help record such random

symptoms and events.


8

A capsule endoscopy “Radio Pill” can monitor stomach pressure or pH. It contains

the appropriate sensors and a transmitter. After being swallowed, it senses and sends

the data out to be picked up by an external receiver and recorded.

These are some of the abundant avenues where WBANs can find applications.

1.4.2 Issues in WBAN measurements and sensing

Data from any kind of sensor system including WBANs is prone to variability owing to

a number of reasons like the sensing conditions, ambient environments, errors, and

aging. Stochastic evaluation procedures can be used to address variability of data

furnished by WBANs.

There is a lack of knowledge about interrelationships in the process of

measurement. Large tolerances are accepted in order to counter this limitation. There

could also be interactions between physiological systems in the range of interest.

“Cause-and-Effect” relationships could possibly become extremely unclear and difficult

to define.

The transducer itself can affect the measurement, despite all possible

precautions. The measuring system should be prevented from “loading” the source of

the measured variable. There is a limit to the amount of energy that can be associated

with measurement systems. In particular, it could be critical in case of WBANs. Proper

care must be taken that any possibility of energy concentrations that might affect the

measurement or damage the live and healthy cells is avoided.


9

Some safety considerations are highly important because of the living subjects

involved. The process of sensing or transmission should not cause undue pain, trauma,

or any such discomfort to the subject unless essential to save the subject’s life. As the

SNs that form a WBAN are miniature in size and capacity, and have to be battery

powered, they pose several issues and limitations. A major challenge would be to keep

the WBAN running for as long as possible before a SN replacement is essential.

We chose to address three out of the several open challenges associated with the

working of WBANs. The first challenge is related to improvement of the lifetime of a

WBAN, given the small sizes of the SNs mounted on or implanted inside human bodies

in such a network and the extremely limited battery power that the SNs run on.

1.5 Energy conservation in WBANs: Prior work

There can be several approaches to energy conservation, with the prominent ones

being duty cycling, mobility and data driven approaches [13].

There can be two basic approaches for routing of WSN data: proactive and

reactive [14]. In the proactive scheme, data is transmitted by the CSS at a

predetermined fixed rate and is received by the BS, while in reactive approach the data

is sent if and only if it crosses some predefined threshold values [15]. Both of these

schemes have their advantages and limitations.

The methodology proposed by Chu et al. [16] involves mapping of the data into a

random process as a probability density function. It is founded on stochastic


10

characterization of the sensor data. The result, when combined with the actual sample

values can yield predicted values. Temporal and spatial correlations can be modeled as

Markov processes combined with disjoint-clique. Any aberrations or noise from

prediction can be filtered out. The model can be replicated at the SN as well as the sink

end. Jain et al. [17] propose a Kalman filter based model can also be used for data

prediction. The approach involves stream filtering using a Kalman filter. Tulone and

Madden [18] propose that better prediction can be achieved if a trend component that

keeps changing and evolving with prediction run time can be included in the forecast

model. Such stochastic characterization based prediction techniques are observed to

be computationally demanding for the resources at the SN end. None of these

approaches have been applied to the WBAN data or evaluated for performance and

QoS satisfaction.

Le Borgne et al. [19] propose a dual prediction scheme for conventional sensor

networks. The authors have suggested the use of a framework involving predictions at

the SN end as well as at the CSS end. It works on the basis that the SN keeps comparing

the sensed and predicted values if the values are within error bounds, and transmits a

sample only if the difference exceeds the error bound. In the absence of a sample

received, the CSS assumes the values predicted at its end to work fine and uses them to

create the time series. These predictions end up in saving energy by reducing the

number of transmission packets from the SNs to the CSS.

Another extension for the dual prediction in BSNs by Xia et al. [20] suggests the

use of Proportional Integral Derivative (PID), a proven control algorithm at the sensor


11

and the CSS ends for time series prediction. The authors propose the scheme for use in

a WBSN.

All the approaches listed above except for the approach by Xia et al. [20] have

been used for conventional WSNs. Smaller size of SNs and more constraints on

computing and power resources combined with QoS limitations make WBSNs

remarkably different from their WSN counterparts. Hence, except for the referenced

approach by Xia et al. , other approaches would be questionable choices for WBSNs.

Although the approach has been suggested for WBSNs, we compared the performance

of the algorithms used in this paper [20] and found out that the prediction has some

limitations. We suggested an improvised dual prediction framework that utilizes ANNs

for prediction at the CSS end.

Some approaches to energy saving utilize data aggregation. In-network data

aggregation is a popular research area which focuses on minimizing the energy

consumption by communication in a WSN. It is a global data aggregation method [22]

used in multi-hop networks. There are two approaches in this scheme which consider

reducing the size of aggregated data by using some functions as min, max, average,

sum etc., thereby reducing the number of packets flowing in a network, or keeping the

data aggregate size the same just by combining two data packets, without processing

them further [22]. Adjacent nodes in dense WSN may contain similar or redundant

data. But, simply discarding these packets may not be useful as the accuracy of

monitoring may be affected. Thus, Adaptive Data Aggregation Mechanism (ADAM) is

used which makes the use of a sequence number to identify the data packets which are


12

repeated to reduce the total number of packets routed in a network. Another

mechanism in attribute-aware data aggregation is used to combine the packets having

the same attributes so as to improve the accuracy and efficiency of the data

aggregation.

The Routing protocols to forward aggregated packets are different from the

classic routing protocols. Routing of the sensor data packets to a next hop neighbor

could be done by the wireless SNs using content-based algorithms in order to minimize

the consumption of energy involved. Energy consumed by the network has a large

scope of optimization if data aggregation techniques are resorted to. Hybrid data

aggregation is not commonly used and is still a potential research area.

A possible application of data aggregation approach could be in human body

temperature monitoring. The SN can be programmed to take samples of patient’s

temperature at specified intervals. It can store these values in its buffer, compute an

average of the values and transmit the result at periodic intervals, say, twice an hour.

Such averaged value would still convey the information about patient’s current status

to the attending healthcare professionals.

1.6 Behavior, functioning and challenges of coexistence for WBANs

The second challenge we chose to address was to study the behavior of a WBAN in the

presence of other WBANs around it and the issues that all such WBANs would face. In

future, when WBANs become quite common, it is possible that most living beings –

humans as well as animals, would start existing as some or the other type of WBANs


13

that would be in close vicinity to each other. Severe interference problems could be

present in wireless systems if two such systems within the communication range of

each other try to use the same communication channel at a point of time. There could

also be issues related to data routing due to the nature of the data that WBANs deal

with, and that could make it critical to have assured communication of information to

the other end.

1.6.1 Addressing interference issues through scheduling and queuing

mechanisms: Prior work

While the scheduling problem in WSNs has been addressed in prior research at the

MAC level, hardly any approach has ever focused on the QoS based interference

avoidance schemes. Defining and meeting the QoS requirements become important in

WBANs because they deal with the transmission of vital personal data from human

subjects under observation. Most of the Distributed Queuing MAC protocols (DQ-MAC)

for WBANs have been laid out on the same lines as 802.11b MAC. Such DQ-MAC

protocols can adapt themselves while dealing with variations in the number of WBAN

SNs competing for MAC resources and working with different traffic scenarios,

employing seamless transitions. When dealing with light traffic loads, the protocols can

switch to a random access mode while under heavy traffic, they can work in a

reservation mode.

Distributed Queuing Random Access MAC protocol (DQRAP) suggested by Lin

and Campbell’s approach [23] deals with transmission of voice packets through a


14

shared single-channel wireless communication system. The protocol involves a base

station that broadcasts the received signals while providing access for communication

to multiple wireless nodes sharing the same channel. The broadcast from the BS

contains embedded information regarding the status of slots in the next round of

communication. The concept of a frame is not suggested in DQRAP. Data transmission

and collision resolution and are handled using separate queues.

Distributed Queuing Collision Avoidance (DQCA), a MAC protocol proposed by

Alonzo-Zarate et al. [24] is similar to DQRAP. It comes with an additional provision for

cross-layer design for rescheduling based on virtual priority for transmission. Neither

the DQRAP, nor the DQCA consider the performance parameters of the involved

uplinks while scheduling the transmissions. Hence they cannot provide any QoS

guarantees on transmissions involving prioritized data.

Distributed Queuing Body Area Network (DQBAN) protocol for WBANs suggested

by Otal et al. [25] uses a fuzzy-logic based system for informed generation of demands

for “collision-free” transmission slots, or to decide on backing off from transmission on

poor link conditions. It does not deal with a possibility of intelligent scheduling at the

WBAN coordinator, under the new 802.15.6 standards.

For uplink transmissions, the WBAN coordinator schedules the slots in the CFP

for the WBAN SNs that it manages. If the traffic is heavy and there are more requests

for slots than the coordinator can handle in a single superframe, the coordinator would

have to implement a priority scheme for assigning the slots.


15

1.7 Wireless Sensor based systems for control applications

WSNs carry great potential for commercial, industrial, and consumer applications.

Although wireless systems have been around for quite some time, offer several

benefits over traditional wired systems and have earned the trust about conveying

data containing qualitative and quantitative information but when it comes to relying

on them for control applications, the skepticism still prevails and they are not

considered mature enough. As a third challenge, we tried to probe if WSNs in general

or WBANs in particular could be used for wireless control.

1.7.1 Control through wireless sensor networks: Prior Work

Process monitoring and control applications deal with data sensing, its measurement,

display and record, diagnosis, control equipment operation and emergency alarms and

actions. The ISA100 committee has classified these operations into six different classes

with increased priority [26][27].

A WSN self-test technique is introduced to deal with sensor failures. In the WSN

design in [28], the issues due to dynamic RF environment are dealt with by a self-test

technique. RF environment characteristics can be used to predict the performance and

adapt the operational characteristics of the WSNs for control applications. Model

predictive control [29] advocates tackling the issues arising due to wireless technology

within feedback control loops. WSN based control systems can also benefit from

redundant design for increasing their reliability.


16

In a machine-to-machine application [30], an anomaly detected by a sensor can

alert a monitoring middleware that sends a command to an actuator, logs in the

details, and alarms the operations personnel. Instead of the controller, parameter data

transmissions can directly be sent to the control system [31]. The scheme proposed in

[32] can detect faulty sensors.

Entertainment is one of the several fields that can use wireless SNs for control.

Cook has elaborated on new approaches emerging in sensor based sound and music

control for performances [33]. Human gestures can be translated into sound by SNs

tracking human movement. A dancer’s movements can break light beams from wired

photocell circuit which can then be used for MIDI conversion processing control [34].

A better flexibility in performance can be achieved by using wireless SNs due to

removal of limitations in SN placement and their ability to capture subtle dance

movements that are not hindered by cables. Wireless sensors distributed in an area

can communicate the sensed data to a base station that interfaces with music

generating and signal processing patches [35].

There are other avenues related to wireless control like reliability, energy

consumption and security that have seen some previous research. With new

paradigms like IoT just around the corner, there are several issues related to such

innovative implementations that need addressing through research and provide ample

scope for researchers to be actively involved.

Chapter 2: Network Lifetime enhancement in WBANs

Chapter 2

Network Lifetime enhancement in WBANs

2.1 Constraints faced by sensor nodes

The transducers involved in WBSNs are very small in size, have limited computing

power and limited memory. They run on very small batteries that cannot generate a lot

of power or for a very long time. The limited energy budget in WBSNs necessitates

energy conservation in order to prolong the network lifetime. With the advent of newer

and faster communication techniques that can handle a lot of data, it becomes prudent to

probe for possibilities on using such techniques for relaying of physiological and bio-

kinetic data, as it can be seen that the efforts have already begun.

2.2 Error minimization and energy conservation in WBANs

We have attempted to focus on the data driven approaches, coming up with a framework

combining two approaches that we modified for further improvement of network

lifetime [36]. In our approach, we suggest transmission of all essential data while

minimizing data transmission by time series prediction from available samples at the

CSS. This eliminates some of the data samples from the transmission sequence. The

samples discarded would be those that can be reconstructed at the BS end using

prediction, without an appreciable error in reconstructing the information from

predicted data. Any kind of data processing on the CSS would consume some power, but

it would help in cutting down the data rate over the network and therefore the power


18

required for transmission. Reducing the data rate has the potential to dramatically

reduce the power consumption by the microprocessor and transceiver circuits, because

data transmission consumes much more energy than actual data processing. This would

help the WBSN save a considerable amount of battery power. The resource constraints

also advocate that the prediction algorithms involved should not be computationally too

expensive.

Based on the selected transmitted values received, if a significant fraction of body

parameter data can be predicted correctly at intermediate time, then data need not be

continuously transmitted. Thus, the prediction makes redundant part of data not to be

transmitted.

The prior approaches in this regard except for the one by Xia et al. [21] could be

questionable choices for WBSNs. Although the approach has been tried on WBSNs, we

compared the performance of the algorithms used in this paper and found out that the

prediction has some limitations. The results of comparison for a small sample size of

EEG data from a healthy human being are plotted in the graphs shown in figure 2.1. We

suggest an improvised dual sensing framework that utilizes ANNs for predictions at the

CSS end.


19

Figure 2.1 Result of comparison of four prediction algorithms based on sample history


20

Predicting future values or events is possible within reasonable limits of accuracy using

mathematical calculations on the data from past and current states. Predictive models

can be built using approximations involving mathematical calculations that could be

computationally intensive.

The feedback or back propagation algorithms have proved to work very well with

applications involving time series prediction.

2.2.1 Proposed framework and optimization using Linear Elimination

Algorithm (LEA)

The framework involves post sense processing at the SN end for predicting the time

series. Using the actual sensed values and the predicted values, the SN performs a

comparison between the values and creates two sets of samples, one for transmission

and the other for exclusion from transmission using one of the simpler algorithms to

reduce the complexity in calculations. The samples that are within allowable error limits

are placed in the discarded set.

The algorithm at the SN end is as follows:

1. Read in the set of sensed value samples S1.

2. Using a subset of values from S1, compute P1, the set of predicted values.

3. If (S1 ~ P1) ≤ |(allowable error)|, generate D1, the set of discarded values.

4. If (S1~P1) ≤ slope in initial set of samples, add such samples to D1.

5. Transmit (S1 – D1).


21

For the sake of simplicity, samples dropped in transmission and the retransmission

criteria are omitted in the framework temporarily, so as to have a simple evaluation on

the savings resulting from the model. The same can be taken up as a future work.

The algorithm at the coordinating sink station end is as follows:

1. Collect the received data as samples.

2. Use the samples to train a neural network for prediction based on non-linear

regression.

3. Collect the predicted values in a set PR1.

4. Forward the set for analog interpretation and further processing.

The use of a neural network for prediction allows for a better accuracy in prediction at

the received end, as will be obvious from the results.

2.2.2 Performance check on error minimization

Although simple, the scheme in [21] for WBSNs using Proportional Integral Derivative

(PID) prediction generates more errors. Some errors could mean that vital details go

missing in the predicted time series. A prediction technique with a lesser error would be

preferred in healthcare applications. This is where the ANN technique involving non-

linear regression (NLR) that we suggested scores better over the approach in [21]. The

CSS is designed with more resources than the SN units in terms of computing power,

available memory, storage, and energy source. While the SN units could use a simpler

algorithm like PID or Moving Average for deciding on the samples to be excluded from


22

transmission, the CSS can perform a more accurate prediction using NLR-ANN

technique. We prove that the error in case of NLR-ANN is less than in the case of PID

based prediction.

The physiological parameters data for our prediction algorithms is from real-life

samples [24]. These samples have been taken from healthy subjects as well as from

subjects suffering from one or more type of heart conditions. The datasets for ECG Lead-

II, arterial pressure (ART), central venous pressure (CVP), and pulmonary artery

pressure (PulmAP) comprise of 3600 samples each.

We performed Linear, PID, Moving Average and PAST algorithms for generating

approximations on the data. For testing the accuracy of these prediction algorithms, we

used a more rapidly varying EEG data for putting stringent conditions on the algorithms.

We then trained an ANN based on NLR in prediction for arterial pressure (ART),

central venous pressure (CVP), pulmonary artery pressure (PulmAP), ECG lead-II signal

and compared the prediction performance with that of PID. The PID algorithm has been

chosen because of relatively better, improved robustness and proven performance in the

field of control.

A program written for MATLAB r2012 [37] has been used for training the neural

network to predict the time series using non-linear regression involving three different

back propagation algorithms. Of these, the Levenberg Marquardt [38] back propagation

was chosen because it proved to be a faster algorithm as compared to Bayesian

regulation and scaled conjugate techniques across multiple training runs over the data


23

to find the average performance. We have generated regression line plots for these runs

in order to find out the fit of the training algorithm.

In addition to the saving on the sensed data from transducers, the number of

samples can be reduced by various data compression and fusion schemes. Additional

techniques involving sample elimination can be implemented in cascade. We suggest the

use of a linear elimination algorithm on the data that lets the SN decide whether or not

to transmit a particular data sample. Based on the result of a simpler prediction

algorithm (may be PID) at the SN end, the SN decides if it can discard the sample from

transmission. The receiving CSS side would reconstruct this data using NLR-ANN

prediction. As observed, the predicted values would still be in the acceptable error range

without any significant loss in information. Thus, the saving would be in the form of

samples that were not required to be transmitted.

The linear elimination algorithm is implemented in the SN and the CSS. The SN,

after sensing compares the slope across successive samples, and if the slope is the same,

decides not to transmit successive samples. The CSS can predict the missing samples

with a single calculation, having advance knowledge about the interpolation from the

algorithm. The algorithm has been successfully implemented for the data samples from

the four physiological parameters and the savings were recorded.

We adopt the PID algorithm for control and different special cases of PID algorithm

like Linear, PAST and Moving Average for the prediction. History of decades of industrial

control has proved that PID is the algorithm that provides the best compromise for a

stable control. It is also generic and simple.


24

Figure 2.2: Performance of prediction for Arterial Pressure, Central

Venous Pressure, Pulmonary Artery Pressure, and ECG Lead-II using PID

and NLR-Neural Network

Figure 2.1 shows the performance of the NLR-ANN predictor as compared to PID based

predictor for the four parameters. The parameters behave differently from each other,

and have different and characteristic wave-shapes and rates of change. Both the


25

algorithms seem to provide a very good prediction, but the real difference between them

that would be clear from the error plots is not very evident in these graphs.

Regression line plots for training the ANN based on NLR algorithm were generated

for the data from ECG Lead –II. The plots show that the network trains well and that the

fit is good for the training, validation as well as test data. The ECG Lead II waveform has

been chosen for the check on regression because of its unique wave shape that poses

stringent conditions for data prediction. The performance of the ANN-NLR training for

our prediction can be studied from the plots from figures in the appendix from figure

2a.2 to figure 2a.11.

Figure 2.3: Graphs indicating errors in prediction for the Central Venous

Pressure and ECG Lead-II for PID and NLR


26

Figure 2.3 shows the observed variations in error for the two techniques for Central

Venous Pressure and Pulmonary Artery Pressure. Plots for the other two parameters are

similar in nature.

NLR-ANN offers better error performance as compared to the PID predictor. We

also provide a quantitative basis for supporting this assertion in Table-2.1.

Table 2.1: Mean Square Error values for the two algorithms

Table 2.1 shows a comparison of mean square errors for PID and NLR for the four

physiological parameters over 3600 samples. For the NLR, the error values were

calculated over 100 training iterations and the average error reported from those

iterations was considered.

Table 2.1 and the error plots for the four parameters under test prove the clear

supremacy of NLR-ANN technique over the PID technique. This is despite the fact that

the nature and shape of the waveforms is very different for the parameters due to

different frequency components and ranges of values.


27

Artifacts refer to any signal extraneous to any component. Random noise, electrical

interference, cross-talk, and all other unwanted variations in the signal. Most of the

artifacts are generated due to movement of the system and should best be filtered out

for two reasons. For one, they do not contain any real information. The second reason is

that wave-shape variations corresponding to artifacts can erroneously be interpreted as

the real signal. Their inclusion adds up unnecessary and spurious samples that get

encoded and added to the composite digital information of the signal.

The linear elimination algorithm proposed by us was run on the samples from the

four physiological parameters with the results shown in Table 2.2.

Table 2.2 Log of additional savings from LEA

Table 2.2 shows the savings recorded in raw sample values in the row labeled as

‘no filtering’ and the percentage saving in samples obtained. Sample data was then

filtered for noise and the algorithm was run on the filtered set. The rows marked

‘filtering’ and the next row containing the percentage saving show the savings observed

with filtered data.


28

2.3 Evaluation of the impact of sample rate reduction on wave information

In the work so far, we considered the signals in their totality for prediction, considering

complete transmission of all the samples. We then tried to evaluate the effect of sample

rate reduction and study the tradeoff between saved energy and conveyed parameter

information.

This is an improvisation of our previous work with intention of saving more on

energy consumed in data transmission by cutting down further on the samples. This

approach could further save more energy, when applied in combination with the LEA.

2.3.1 Sample rate reduction and its relation with network energy saving

The sampling theorem states that variations in the information need to be sampled at

such a rate that the highest frequency spectral component in the information is sampled

at least twice. Nyquist rate of sampling provides us with the minimal rate of sampling

which is twice that of the highest frequency spectral component. While maintaining the

sampling rate more than this rate ensures that the smallest of variations in the signal

wave-shape are not missed, it could mean overkill in the sampling rate for the signal

ranges where the wave-shape variations do not really convey any significant

information.

Transmitting all the samples would indeed convey a better approximation of the

analog waveform. However, it would also require more data to be transmitted. If the

cutting down on the samples does not compromise on the useful content of information,

it would help us save on a lot of extra energy from samples not transmitted.


29

This would straightaway cut down the data rate. However, it has to be determined

if important or critical piece of information would be lost in the process. This could be

confirmed only after discussions with cardiovascular physicians and surgeons. We

showed the results to the specialists for their opinion to find out what level of

approximation is acceptable to them.

2.3.2 Impact of artifact filtering on amount of data

Artifacts could get introduced due to the movement of the subject and various other

ways. Noise could get into the waveform and superimpose itself on it. A variation due to

artifact or noise should not be erroneously interpreted as important information. Such

artifacts need to be filtered out for inclusion of only pertinent and genuine information

and elimination of samples that can be safely considered redundant.

2.3.3 Prediction by numerical interpolation techniques

The rate of sampling can be decided by the body sensor program depending on the

physical characteristics of the waveform under consideration. Additional algorithms can

be applied for further reduction of SN data prior to transmission. Assuming that the

information was received satisfactorily, the missing information in signals can be

predicted by mathematical techniques like numerical interpolation. If the resulting

waveforms are still adequately close to the original and the diagnosis does not change,

this effort would save a considerable amount of energy from sample reduction.

Moreover, the techniques are computationally not too intensive and can be handled by

the CSS.


30

2.3.4 Results of signal processing

We reduced the sample information for the signals describing four body parameters by

cutting them down to half, one-third, one fourth, and even one-fifth as the extreme case.

The signals comprised of real data from the physiological parameters. Twenty random

sets of 10-second samples (3600 samples in each set) of each signal from healthy

individuals and multiple patients have been chosen for running through the algorithms

for evaluation. The parameters chosen vary a lot in terms of range, variations and shapes

so as to allow a better evaluation of our proposal by diversity in choice.

We also applied our previously proposed LEA to the resulting signal for further

data reduction. We then tried to reconstruct the signals at the receiving coordinator or

base station end by applying five different interpolation algorithms for a comparison.

The algorithms used were: linear, nearest neighbor, linear spline, and two variants of

cubic spline. The results of evaluation of two such sets are shown below in Figures 2.3

and 2.4. The first column shows the signals with progressive reduction in the number of

samples. The second column depicts the reconstruction of the corresponding row at the

receiving end using linear interpolation. For the simulations, the programs were written

in MATLAB r2012 and Java 1.7 [39].


31

Figure 2.4 Plots for linear rebuild of the PulmAP signal

Figure 2.5 Plots for linear rebuild of the ECG Lead-II signal


32

Figures 2.5 and 2.6 show the error plots between reconstructed signals and the original

signal waveforms for the corresponding waveforms in figures 2.3 and 2.4 respectively.

As expected, the error increases with reductions in the sample size. An evaluation on the

amount of maximum tolerable error involves an expert opinion from a practicing

physician. Figures corresponding to the other two parameters are included in the

appendix.

Figure 2.6 Error plot for linear rebuild of the PAP signal


33

Figure 2.7 Error plot for linear rebuild of the ECG Lead-II signal

The following tables show the values of MSE and % Error obtained from the processing

carried out on the signals containing body parameter data.

Table 2.3 Signal specifications for the four parameters

ART CVP PAP ECG-II

Signal Minimum 52.3505 -2.1515 21.4119 -0.5296

Signal Maximum 89.6935 5.6229 43.3432 0.8143

Span 37.343 7.7744 21.93 1.3439


34

Table 2.4 Maximum Error with sample reduction for the four physiological parameters

ART %Error

CVP %Error

PAP %Error

ECG-II %Error

Halved

0.270 0.722

0.134 1.722

0.183 0.835

0.051 3.773

1/3rd 0.317 0.850

0.246 3.168

0.263 1.200

0.083 6.154

1/4th 0.391 1.047

0.211 2.709

0.443 2.018

0.130 9.666

1/5th 0.632 1.694

0.247 3.180

0.481 2.194

0.202 15.001

Table 2.5 Error values for ECG-Lead II from the five numerical interpolation techniques

Linear

%Err

Near

%Err

Spline

%Err

Pchip

%Err

Cubic

%Err

1/2 0.051

3.773

0.195

14.495

0.050

3.691

0.049

3.668

0.050

3.698

1/3rd

0.083

6.154

0.221

16.445

0.086

6.392

0.073

5.447

0.073

5.447

1/4th

0.130

9.666

0.336

25.017

0.122

9.078

0.125

9.316

0.126

9.346

1/5th

0.202

15.001

0.443

32.971

0.125

9.309

0.137

10.202

0.158

11.727

The plots in Figure 2.7 and 2.8 show the nature of mean square error and percentage

error for the various rates of sample reduction for the signals under consideration, while

plots 2.9 and 2.10 show the plots for approximation using the various interpolative

prediction techniques.


35

Figure 2.8 MSE for different sample cuts for the signals

Figure 2.9 Maximum percentage error for sample cuts for the signals


36

Figure 2.10 MSE for different techniques for the signals

Figure 2.11 Maximum percentage error for different techniques for the signals


37

From the above information, the amount of energy saved can be evaluated. A

logical assertion is that the saving would be directly proportional to the sample

reduction ratio. Network lifetime would accordingly be increased. The calculations could

be extended for finding out enhancement of battery life. We extended this work using

additional techniques like data aggregation and adaptive sampling, implemented on top

of the proposed framework for medical applications.

2.4 Energy aspects and evaluation of Network Lifetime

One of the major challenges faced by WBAN designers is the management of energy

consumption, for resourceful operation of the network. In several WBAN applications, it

is practically impossible to change or recharge the batteries when implanted sensors are

being considered. Such scenarios make it significant for a WBAN to implement

mechanisms to efficiently manage energy with the purpose of maximizing the working

time of the system. In turn, this would increase the lifetime for the supported monitoring

applications. The sensor nodes consume energy in their transducer and A/D converter

unit, communications unit for transmission and reception, and in the computing unit for

the processing of data. As the communications unit demands the most power of the

three and exceeds the power requirement of the other units by several orders in

magnitude, the design schemes involve a sleep-awake cycle for conservation of energy.

We have also tried to evaluate the lifetime of the WBAN system in the proposed

framework when some common sensors are used in WBAN nodes. We have focused on

the duration that the sensors would stay powered on with the help of commonly

available batteries.


38

The signals considered by us in this evaluation are sampled at the rate of 360 samples

per second. We have considered two possibilities of encoding the samples using 8-bits

and the more common 12-bits. The overall energy required by a WBAN sensor node

depends on several factors like inter-sensor distances, node sleep-awake cycle, the time

durations that the node stays in a particular mode, and a constant.

On the basis of Heinzelman’s sensor node transceiver model [40], the transmission

energy required to transmit a k-bit message to a distance of d can be computed as:

ETx(k,d) = ETx-elec(k)+ETx-Ampl(k, d) = EElec*k + ε k d 2,

where,

ETx-elec is the energy consumption in transmission electronics,

ETx-Ampl is the energy consumption in the transmission amplifier,

ε is a factor involved in amplification

and, d is the inter-sensor communication distance.

Their model assumes

ETx-elec = ERx-elec = EElec , and ε =100 pJ/bit/m2

In the receiver, to receive a k bit message, the energy consumed is

ERx(k) = ERx-elec(k) = EElec*k

For most sensor nodes, the energy consumed for powering up the transceiver

electronics is the same for transmission and reception circuits and is of the order of tens

of nJ/bit.


39

For continuous operation, the energy required for transmitting all the samples in a

minute comes out to be 8.65 mJ for an 8-bits/sample encoding while it is 12.97 mJ for

samples encode in 12-bits.

We tried to evaluate the lifetimes of networks involving three of the low-power

sensor nodes available commercially using a Qualnet simulation using 24 nodes. The Eco

[41] is a low-power ultra-compact sensor that needs 16 mA of current while

transmitting, 22 mA while receiving and just 2 μA during sleep. The duty cycle involves

10 seconds each of transmission and reception followed by 40 seconds of sleep to

complete the minute-long cycle.

The TI CC3100 [42] fares comparatively in its 1DSSS mode while performs much

better when operated in the 54OFDM mode. Based on the transmission power

requirements of these two sensor modes in the mentioned modes, the following table

emerged in our Qualnet model for three of the commonly used batteries – CR2032,

CR123A and iXTRA, all the three capable of supplying 3.0 volts, 0.5A. Table 2.6 sums up

the performance of the battery model results for continuous operation without any

power management applied.

Table 2.6: Life in days for the different battery models as per their

capacities and node power requirements

Battery → CR2032 CR123A iXTRA

Sensor Node ↓ 225 mAH 1550 mAH, 1700 mAH

ECO (16 mA) 1.76 12.11 13.28

TI – DSSS (21 mA) 1.34 9.23 10.12

TI – OFDM (9.39 mA) 2.99 20.63 22.63


40

From the table, it can be inferred that the advances in battery technology and low-

power sensor design actively improve the lifetime of the network. These are raw results,

before implementing any power management algorithms in the transmission scheme.

Implementation of power management on the top of the evaluated models would

certainly add to the life of the battery as well as that of the network. Our scheme

pertaining to sample reduction for the WBAN parameters reduced on the power

requirements greatly and gave promising results about increasing network lifetime.

Figure 2.12 summarizes the findings.

Figure 2.12: Battery model for four commercial batteries and three SN types


41

2.5 Energy efficient design for WBAN monitoring using next generation

Cellular Channels

We introduce and propose a new architecture tested solely for a prompt sensing and

anytime-connected wireless body area network for the transmission of important

physiological signals over cellular networks. For testing this architecture, we have used

the data from the same four cardiac parameters that are vital for the diagnosis and

monitoring of patients with Cardio-Vascular Diseases (CVDs), which are a major cause of

death today [7]. The physiological data corresponding to the four parameters is sensed

from the subject’s body and is transmitted over a Wireless Body Area Sensor Network to

a WBAN coordinator acting as a sink (CSS). The CSS compresses the data received from

the body sensor nodes, processes it and sends it through GSM communication for

transmission over the existing cellular network to a remote base station or a repository.

The GSM receiving unit at the other end receives the data and directs it to a dedicated

remote server for demodulation. The missing samples in the signals containing original

physiological data are then reconstructed. We have discussed the methods used for the

compression of data in this scheme.

We have also tried to evaluate the means for efficient transmission of data while

keeping the allowable range of error in the physiological content such as to assure that

the original signal characteristics are maintained. We envision and propose to have a

dedicated channel in upcoming generations of mobile communication technology for the

transmission of vital physiological data. If implemented, this would have the potential of

making 24x7 health-monitoring a reality in in forthcoming years. More bandwidth for

everybody is that anticipated for services like data on demand makes our proposal a


42

tangible possibility. In addition, we have tried to evaluate the network lifetime by

focusing on the power consumption in WBAN sensor nodes. Our evaluation is based on

the battery models involving common and newly developed high capacity batteries for

powering up the sensor nodes. Our methodology would facilitate online, round-the-clock

health monitoring for the subscribers of a cellular communication system by working in

partnership with the underlying body sensor networks consuming very little power.

2.5.1 WBAN data through existing communications systems

A battle for WBAN standards have been among Wi-Fi, Zigbee and low power Bluetooth

until the IEEE 802.15.6 standards for WBANs emerged on the scene in late 2012 [43].

The new standards come with QoS provisioning for WBANs, apart from other important

specifications. While the standards define the essentials and protocols for various layers,

compatibility of transmission of the WBAN data utilizing the various available

telecommunication networks is still a gray area. Our position in this regard is that the

aforementioned interworking standards between IEEE 802.15.6 and the evolving 5G

standards need to be developed.

To advocate this further, we have experimented with encoding and communicating

vital sign data corresponding to two parameters via a GSM network and obtained

encouraging results. Our position emphasizes an enormous emerging opportunity for

mobile health sector that holds great promises of reducing the cost of healthcare

monitoring. Such a move could be a great initiator for novel business models in this

nascent sector that would generate a vibrant consumer base. Mature and powerful

consumer mobile technology of today is capable of handling the challenges of innovative


43

healthcare applications, and its emerging standards like 5-G should be further developed

with such applications in mind. None of the previous phases of mobile

telecommunications architectures have considered the need to incorporate the handling

of healthcare applications so far, and hence the authors look towards the 5-G for not

missing on this opportunity while the standards are being framed.

2.5.2 Motivation for WBAN energy enhancement

One of the major challenges related to WBANs involves the energy-fidelity tradeoff.

WBANs have to transform and transmit the sensed parameters into valuable information

of acceptable and appropriate fidelity level, in an energy efficient manner. This feature

calls for selective processing of varied physiological data samples.

IEEE 802.15.6 standards lay down the specifications for WBANs, but they treat

WBANs as standalone networks of a distinct type. Interworking between the IEEE

802.15.6 and the other existing wireless systems like GSM, WCDMA, WiMAX, ZigBee and

HomeRF is an important issue that still needs to be worked on. In our work, we explore

the possibilities of WBAN functions if the mentioned standards and protocols are

developed in the future telecom standards like 5G and provisions of such interworking

are developed.

If an emergency medical condition occurs for a human subject being monitored by

a WBAN, our proposed architecture can send medical data updates to the concerned

medical personnel by means of short messages or encoded as a voice call. While such

updates might not be able to send BSN data corresponding to continuous monitoring,


44

they would still help the medical personnel in an early diagnosis, preparation or decision

on the course of action about the subject.

2.5.3 Prior Work in WBAN energy enhancement

Due to the constraint of a small sensor node size and even smaller batteries, increasing

the lifetime of sensor nodes and consequently that of the networks is always an issue [8,

44]. Body Sensor Nodes (BSNs) are similar to their wireless sensor networks

counterparts [45, 46], but smaller in size, and lower in battery capability, and hence

suffers from severe constraints. The sensor nodes need to keep collecting data samples

and relay them to the CSS. A large number of data samples corresponding to the

physiological parameters are collected by the BSNs. However, the number of samples

collected does not take into account the frequency and nature of variations in the

physiological parameter. In this part of our work, we have tried to reduce this data

content in order to address the energy-fidelity tradeoff [47] by signal processing

methods involving the exclusion of some sample data from transmission and restoration

of missing samples using prediction.

Wagner et al. [48] use ZigBee links and a cable connection for BSN data collection

and processing in an embedded system from which the data is sent over Bluetooth links

to the smartphone for presentation. The approach by Ogunduyile et al. [49] utilizes a

GPRS/Internet connection for uploading the BSN data to a Medical Health Server for

analysis. Baviskar and Shinde’s approach [50] uses BSNs for data logging, processing and

analysis that send the results to the CSS over Bluetooth links. The system proposed by

Bourouis et al. [51] uses GPRS/UMTS link for beaming up BSN data to a server.


45

Although all the approaches involve a WBAN and suggest the use of a smartphone as the

CSS, none focuses on data compression for energy saving and improving on BSN as well

as WBAN life-time, which we address in this paper.

Our work presented here is different from these schemes in that none of the

mentioned approaches explore the possibility of transmitting the sensor data by means

of encoding as a text message or sending it as a voice coded signal by means of digital

modulation. We compress and encode vital parameter data and then package it for

transmitting it as a sequence of short text messages. Our follow up scheme encodes

compressed BSN data as a digital modulated voice signal for transmission over regular

voice channels. Another alternative that we tested was the use of a smartphone as a CSS

that allowed us the use of a combination of Bluetooth [52] and Wi-Fi [53] networks for

quick and easy sending of data anywhere using the Internet.

We have enhanced the sensor network design discussed in [54] by adding the

capability of communicating over commercial wireless voice/data networks to WBANs.

The benefit of our architecture is that it can be directly useful in mitigating inter and

intra-WBAN interference issues, by adding to the solutions proposed by Jamthe et al.

[55]. None of the previous works proposes such a solution for dealing with WBAN

interference.

2.5.4 Encoding the data and errors in encoding

In order to encode the data from the cardiac parameters as short text messages, the

first challenge faced pertains to the limitation that any single message of the mentioned

type cannot contain more data than what would be needed for encoding 160 textual


46

characters. Not much of uncompressed sensor data can be packaged in a single text

message. Multiple text messages in a sequence could be a possible solution, but a lot of

such messages would still need to be packed in a limited amount of uncompressed data.

If the sensor data is compressed, more amount of information could be packaged in the

same number of short messages. Energy-fidelity tradeoff would need to be kept in mind

in the process of cutting down and compressing the data.

For this work, we chose sets of 3600 samples covering 10 seconds of sensor data

for ART, CVP, PAP and ECG signals. A standard short text message can hold 160 encoded

textual characters. If we tried to encode our sample sets as text messages, 3600/160 =

24 text messages would be required. It is important to note that this requirement would

be for raw/unreduced 10 second data. However, if the sample frequency was reduced,

data corresponding to more time could be packaged in the text messages. A reduction in

sample size by a factor of five could encode 10 seconds in 720 samples, which can be

contained in just 5 text messages if each sample was encoded in 8-bits, corresponding to

the ASCII code for the text messages. However, we are not bound to follow such

encoding, and resort to compression in encoding in order to squeeze in even more data

in every message pack.

One such compressed encoding scheme could be delta modulation. One bit

difference delta encoding could approximately hold 160x8 = 1280 data samples in one

short text message pack. This would mean that 10 seconds of data could be encoded in

three text message packs. Lossless compression techniques applied on top of such an

encoding could yield a further tight packaging of data. Data aggregation schemes could

further cut down on the sensors’ transmission load.


47

However, there would be a compression-fidelity tradeoff involved. Maximum allowable

error in an approximation that would result from such a compressed encoding is

debatable and the tradeoff can be best decided by physicians and specialists. The

maximum error in encoding is decided by the step size and the maximum can be half the

step size. In case of PAP, the lower and higher ranges of the signal are 20mV and 45mV

respectively, consequently making the signal span to be 25mV. Encoding this signal into

eight bits yields a step size of 25mV/256 = 97.6 μV, thereby limiting the maximum

allowable encoding error to 48.8μV. The error in case of ECG lead-II signal, similarly, is

3.42μV, because the signal-range from -0.75 mV to 1.0 mV gives a step size of 6.83 μV.

The results for all four signals are summarized in Table 2.7.

Table 2.7: Signal specifications for the four vital sign BSN parameters

Characteristics ↓ ART (mV)

CVP (mV)

PAP (mV)

ECG-II (mV)

Signal Minimum 50.0 -15.0 21.4119 -0.5296

Signal Maximum 90.0 7.0 43.3432 0.8143

Signal Span 40.0 22.0 21.9313 1.3439

2.5.5 Proposed Architecture and Framework

The enhancement proposed by us is intended to send the physiological data via a GSM

network using two different means. However, before we try packaging BSN data in the

form of a text message, we need to reduce the amount of sample data by cutting down

the quantum of transmissions. The number of samples in the data can be reduced if the

missing sample values can be predicted within the limits of acceptable error by signal


48

processing techniques. When it comes to deciding if the level of approximation is fine or

not, the end users need to be consulted. In this case, the end users are the cardiac

physicians and surgeons. In no case can the approximation be allowed to affect or alter

their diagnosis. We followed this practice and consulted the cardiac physicians at each

stage while running these approximations. Dual prediction techniques proposed by us

[56] can then be utilized at the receiving end for missing sample reconstruction apart

from the technique used by us in this paper. We reduced the amount of data by skipping

those samples from the original set that can be predicted at the receiving end. We also

applied delta encoding to further pack more amount of data in every message packet

containing encoded BSN data.

Figure 2.13: 5-G Enhancement for a typical WBAN BSN

Our proposed architecture, shown in Figure 2.13, involves enhancement of body sensor

nodes discussed in [57], with a coprocessor which is an additional microcontroller. This

microcontroller facilitates ease of data logging, processing and temporary storage of


49

data samples from the sensor nodes. The microcontroller has a wireless extension/add-

on that makes the sensor node capable of communicating with the CSS or any other

similar device over GSM or other commercial voice/data network, in addition to IEEE

802.15.6 radio. We call this addition in the extended architecture as the 5G extension for

the sensor node, with a proposal and assumption that the 5G standards would envisage

dedicated channels for communication and processing of sensor data from WBANs. Such

an enhancement would benefit a WBAN by enabling the sensor nodes to communicate

with CSS, which could be a smartphone itself, thus obviating the need for a dedicated CSS

device.

However, even if such provisions are not envisaged in the 5G or future extensions,

the proposed architecture can be designed so as to exploit the current analog or digital

cellular design. We use a smartphone as the CSS in our architecture as smartphones have

become very common these days. They come with a variety of applications available

today. Such applications could also be custom made to cater to the user’s healthcare

needs. This also means that, if needed, the WBAN would then have the capability to send

the user’s physiological data directly to any other smartphone with the physician, or

with emergency or nursing services using the GSM/WCDMA or similar voice/data

network.

Hence the architecture works as follows. The sensor nodes sense and process the

physiological data. The data is then encoded, packed in the desired format, say a text

message or as a voice-coded data message, and passed on to shield. The shield then

transmits the data to the smartphone functioning as the CSS, or any other smartphone as

required. The WBAN CSS can make decisions regarding a need-based use of voice/data


50

network instead of internal WBAN wireless links running on IEEE 802.15.6 depending

on the current scenario with respect to interference, QoS, throughput requirements and

urgency of communication.

2.5.6 Results: Spectral translation

The sample sets corresponding to the four physiological parameters were analyzed for

their frequency spectral components. This information was required in order to assist

with the design of a framework for their possible transmission over the voice channels

in the next generation technologies. After obtaining the baseband in the signals

corresponding to the four physiological parameters, the sample sets for the four signals

were translated in frequency domain. The translation was affected such that the signals

would occupy exactly the same spectral range as the human voice (200 cycles/sec to

3200 cycles/sec) by employing a product-bias algorithm.

The algorithm works by mapping individual spectral components derived from the

signal samples to the new range as specified above. The idea is to shift the low frequency

data information to the human speech window so as to utilize the voice communication

provisions in the smartphone for its transmission. The translation would also provide a

bigger spread while retaining the power information of the baseband, and the resulting

signal could then be transmitted using OFDM over the regular voice channels, as

envisage in the framework. Figures 2.14 and 2.15 show the baseband of the frequency

spread for the ECG lead II signal and figures 2.16 and 2.17 show the spectrums for the

CVP signal.


51

Figure 2.14: Frequency spectrum of the baseband signal for the ECG Lead II signal


52

Figure 2.15: Spectral translation of the ECG Lead II signal to the human voice range


53

Figure 2.16: Frequency spectrum of the baseband signal for the CVP signal


54

Figure 2.17: Spectral translation of the CVP signal to the human voice range

2.5.7 Results: Data Communication

Along with the text-message encoding, another important functionality in our

architecture is to use a speech signal encoding of the physiological sensor data using

digital modulation, and then transmitting the data as a voice call. The coprocessor sends

the related encoding request to the signal processor block in the architecture. The signal

processor block acknowledges the request and generates a digitally modulated output of

the compressed sensor data from the digital modulation used. An important aspect of


55

this encoding is the use of human speech frequencies (100 Hz – 3.3 kHz) in digital

modulation so that the generated output can be treated as a voice signal by the further

processing stages. This would allow the coded signal to be transmitted as a voice call to

the receiving smartphone. The received call can be directed to the right application in

the smartphone for decoding and presentation.

The four cardiac parameters were processed using the techniques discussed earlier

in 2.3. The reduced physiological signals were then compressed, encoded as text

messages and transmitted over GSM network. Our set up for this comprised of an

Arduino microcontroller board with a GSM shield extension. Arduino is a commonly

available open-source electronics prototyping platform that can be used for developing

small embedded applications. The Arduino GSM shield is basically a GSM modem. From

the perspective of the mobile operator, the Arduino GSM shield works like a mobile

phone while from the perspective of the Arduino board, the shield works like a modem.

The Arduino GSM shield lets an Arduino prototyping board to make voice calls, send and

receive SMS, and connect to the Internet using a dedicated library. Using the setup we

successfully received the encoded message signal as a short text message.

The text messages sent by the Arduino were successfully received at the other end

by a GSM cellular handset. The encoded and compressed BSN data was then available for

decoding and rebuilding of the original, uncompressed data samples out of the receiving

handset. For rebuilding the missing BSN sample data at the receiving end, we applied

five numerical interpolation techniques discussed in 2.3.

We also were successful in sending the physiological data from the sensors directly

over a Bluetooth link to the smartphone using an Arduino [58] based microcontroller.


56

The smartphone could then send the same data anywhere in the world using its data or

Wi-Fi connectivity over the Internet. This has been verified as working and is presented

in 2.6.

2.6 Data Acquisition System (DAS) for WBANs

We implemented a raw DAS that could be used for WBANs. The setup involved received

inputs from a three-axis accelerometer sensor transmitting human movement data. The

three parameters from the sensors were transmitted by a Bluetooth-enabled Arduino

microcontroller to a smartphone. The smartphone running on the Android operating

system was used as a CSS. An Android application was developed for acquiring the data

from the sensor and transmitting it to the CSS in real time. The data from the three axes

was packed as a single raw text file and successfully transmitted through the Internet as

an email attachment. The Wi-Fi link available with the smartphone CSS was used for the

Internet access but this could be done over a data connection as well. Figures (17), (18)

and (19) show the experimental setup for the data acquisition system.

The accelerometer was fixed to a shoe worn by the subject who walked around,

trying to simulate various conditions including normal walking, limp, uneven gait and

freeze of gait. The data was acquired by the smartphone CSS and used for trend display.

We tried to use feature extraction for isolation of parameters of interest with respect to

the subject’s movement. The work is still in progress.


57

Figure 2.18: The data being received by the computer and the smartphone

Figure 2.19: The experimental setup involving the

microcontroller board and the sensor


58

Figure 2.20: Accelerometer data received over email

The accelerometer data was then plotted for a trend display and analyzed for frequency

response. We are trying to find out if the Fast Fourier Transformation of the signal FROM

THE MOST relevant axis could help us get more information about the number and

nature of the steps. The trend display of data from the x, y and z axes in the

accelerometer is shown in figure (20). Figure (21) shows the result of the frequency

analysis of the y axis data that was associated with the upward and downward movement

of the foot of the subject wearing the sensor. Extensions to the data acquisition system

are in progress and are proposed as future work. Exploring sensor fusion would not be

applicable because it is not one big picture out of multiple sensors that is required.


59

Figure 2.21: Accelerometer data from three axes

Figure 2.22: Frequency analysis of the accelerometer data to get the steps

60

Chapter 3

Behavior, functioning and challenges of coexistence in

Wireless Body Area Networks

3.1 The importance of QoS in functioning of WBANs

Receiving reasonably accurate data from WBAN SNs is critical in order to ensure that the

quality of service (QoS) level expected from a monitoring application is met. As the

WBANs are distributed randomly, presence of multiple WBANs at the same time in a

restricted space might cause interference, packet loss for critical data and possibly

severe degradation of network performance [60]. Typically not more than 10 randomly

distributed WBANs should co-exist within a volume of 6*6*6 m3 in order to avoid

interference [61]. For WBANs to co-exist gracefully with a fair sharing of bandwidth,

appropriate scheduling mechanisms need to be implemented.

The amount of data, packet generation interval and the QoS requirements do vary

according to the nature of monitoring applications [61]. Real-time data based on

occurrence of events is needed in detection of freezing of gait or detecting a fall in

geriatric patients [59].

3.1.1 The 802.15.6 superframe

Protocol standards for the Physical and MAC layers in WBANs released in May 2012

under IEEE 802.15.6 [61] incorporate the QoS issues not addressed by the earlier IEEE

802.15.4 standard [62]. The standards are specially laid out to support medical WBANs.

They classify WBAN data traffic into 8 different categories depending on a priority table

and have provisions for asymmetric flow of data. The IEEE 802.15.6 MAC layer specifies

a superframe structure as shown in Figure 3.1 to support the standards of QoS in

WBANs. The active portion in the superframe has an embedded Priority Access Period

Chapter 3: Behavior, functioning and challenges of coexistence in Wireless Body Area Networks

61

(PAP) for scheduling sensitive emergency data that comprises of Contention Access

Period (CAP) and Contention Free Period (CFP) windows. CAP uses slotted ALOHA

access while CFP allows for TDMA based access with unique transmission slot

assignment for robust transmission of high priority data to end devices to meet the QoS

requirements.

Figure 3.1: The structure of an IEEE 802.15.6 super-frame [3]

3.2 Addressing the interference

While the scheduling problem in WSNs has been previously researched on at the MAC

level, QoS based interference avoidance schemes have not been in focus due to recent

introduction of new standards. The working of WBANs has to now be evaluated in the

light of the 802.15.6 performance standards and QoS specifications.

3.3 Priority-based scheduling schemes for suppressing interference in

WBANs

To provide optimum network utilization, it is important to efficiently schedule multiple

co-existing WBANs which could possibly suffer from high degree of interference both at

the intra and inter-WBAN interference levels. A graceful coexistence can be made

feasible by appropriately scheduling transmissions from different WBANs. We propose

that inter-WBAN interference can be avoided by a QoS based MAC scheduling approach

and that interference within WBANs can be dealt with by deploying a fuzzy inference

engine for intra-WBAN scheduling [63]. We tested a novel priority scheme that the CSS


62

could use for intelligent slot-schedule generation. Our scheme involves fuzzy based

reasoning and decision making when the CSS has to deal with multiple requests for

allocation of slots. We modeled multiple WBANs with CSS cluster heads, treating them as

groups of SNs trying to acquire transmission time slots. An efficient utilization of the CFP

in MAC superframe helped the WBANs avoid inter-WBAN interference.

The two priority-based scheduling schemes proposed here can address the intra-

WBAN as well as the inter-WBAN interference. In our scheme, a fuzzy inference engine

handles the intra network scheduling. Lofti Zadeh [80] introduced the fuzzy logic

system, which can combine human logic that involves linguistic knowledge as well as

numerical data in parallel for non-linearly mapping an input vector to a scalar output.

Zadeh modified the binary logic involving the set {0,1} to a continuous interval [0,1]. He

removed a distinct boundary and made it possible to transit from a ‘NO’ to a ‘YES’ in a

gradual manner. Use of vague or imprecise inputs is allowed in a fuzzy logic system,

while the goal may be to arrive at a distinct output [64].

Figure 3.2: Zadeh’s basic Fuzzy logic system [80]


63

3.3.1 Intra-WBAN scheduling problem and analytical modeling

The fuzzy inference engine proposed in our scheme helps with the decision about on-

slot allocation or deference of transmission depending on the values of three important

link quality parameters of SNR, BER and Eb/N0, the ratio of per bit energy to noise power

spectral density.

The input parameters to the suggested fuzzy inference engine are experimentally

obtained and accepted values of the three parameters. These values are used to obtain a

decision to schedule or defer the transmission on the slot under consideration for

allocation. The design of the inference system uses the Mamdani Fuzzy model [64]. The

effect of the values of SNR, BER and Eb/N0 on our fuzzy decision is obtained as

visualization.

The system involves three input variables corresponding to all n SNs of the WBAN

in the design. The first input variable SNRk(tk), 1 ≤ k ≤ n, is the sensor Signal-to-Noise

Ratio in dB, with symmetric uplink and downlink conditions assumed at the CSS. Bit

Error Rate of the SN traffic BERk(tk), 1 ≤ k ≤ n forms the second input variable to the

model. The third input variable is Eb/N0k(tk), 1 ≤ k ≤ n . It is the normalized SNR per bit,

and availability of various transmission bands in IEEE 802.15.6 standards make it

important.

No defuzzifier is required because the unique output ‘decision’ is a single value. It

maps to one of the three output possibilities - ‘defer’, ‘schedule’ or ‘forward’.

The minimum value SNRk min corresponding to the normalized BERk(tk) is obtained,

and the input variables are normalized as follows:

BERk *(tk) = BERk(tk) – BERkmin (ratio value),

SNRk *(tk) = SNRk(tk) – SNRkmin (in decibels), and

Eb/N0k *(tk) = Eb/N0k(tk) – Eb/N0kmin (ratio value)


64

The normalized crisp inputs so obtained are applied to the fuzzifier. Each of these

inputs can assume three different states that represent the state of the output. Based on

the proposed fuzzy-logic rules and the input values, a fuzzy decision table can be

created.

BER {too high, acceptable, good},

SNR {dangerous, just-okay, better}, and

Eb/N0 {critical, boundary, superior}

The fuzzy logic system uses triangular membership functions adjusted for

boundary values for creating the mapping. Low processing power and costs dictate the

choice of the function. Limited battery resources make this highly desirable for WBANs.

The linguistic output variable ‘Decision’ is determined as:

Decision {defer, schedule, forward}

Table 3.1 below displays the fuzzy inference table for ‘Decision’. When the SNR is

dangerous, Eb/N0 is critical or the BER is too high, a transmission of packet is deferred

until the next superframe. Only when normalized SNR and BER values are acceptable at

a minimum, a decision is made to transmit the packet on the scheduled slot. If the

conditions did not permit so, the transmitted packet would be lost and packet

retransmission would add to the overall energy cost. Similarly, if the values for the three

link parameters are at least acceptable, a request for more slots can be made for the next

superframe.


65

Table 3.1: The Fuzzy inference table for transmission error parameters

Figure 3.3: Decision considering BER and Eb/N0


66

Figure 3.4: Decision considering SNR and Eb/N0

Figure 3.5: Decision considering BER and SNR


67

3.3.2 Inter-WBAN interference avoidance: Simulation

In the absence of a central coordination manager, avoiding inter-WBAN interference can

be tricky in randomly distributed WBANs, and gets progressively complicated with an

increase in the number of neighboring WBANs. Body movements can cause substantial

variations in the channel gain, thus making interference stronger than the signal

strength even if two WBANs may not be too close to each other [66]. In order to avoid

critical packet loss and enhance the overall WBAN throughput, inter-WBAN interference

needs to be suppressed. Based on traffic classification parameters proposed in IEEE

802.15.6 standard [61], we propose a modified priority preemptive scheduling

algorithm based on PNP-MAC protocol [65] to mitigate inter-WBAN interference.

Our algorithm uses a decentralized approach for scheduling in which the

interfering CSS units communicate with each other before starting their intra-WBAN

communication in order to decide on the channel access slots and sequence for WBAN

SN devices.

The MAC superframe structure supported by IEEE 802.15.6 categorizes the traffic

into 8 different priorities as in Table 3.2 [61].

Table 3.2: Priority in IEEE 802.15.6 and traffic type [61]


68

The QoS based MAC scheduling scheme proposed by us takes these traffic priorities into

account. The channel access in the scheme in various WBAN applications is designed to

consider the diverse nature of QoS. The physiological parameters would be measured

periodically with low data rates, while multimedia data such as video or voice would be

continuous with high data rates [67].

The same superframe structure would be common across participating WBANs for

the bounded beacons. The WBAN CSS broadcasts information about WBAN and CSS

identification and timer information to all SNs in the range just before starting a

superframe.

The SN devices can then identify the right CSS node to request for contention free

period (CFP) transmission slots. Generally, CFP transmission contains periodical SN data

while priority access period (PAP) contains emergency data.

The CSS creates a schedule after sorting the traffic and assignment of slots in CFP

according to descending priorities, to allow higher priority traffic pass early. Before the

start of the next beacon period, the CSS advertises this schedule. Other CSS units within

its communication range can listen to this schedule and save it. Possible interference for

advertised transmission slots can thus be detected and lower priority data in such a time

range is preempted from transmission, yielding the slots to higher priority traffic.

Depending on the priority of traffic around it, the preempted transmission can either be

resumed later or deferred to next superframe. This information is broadcasted to

neighboring SNs at the start of next beacon. The SNs thus know about their transmission

slots in the next superframe, and interference is completely avoided.

We designed and ran an OMNeT++ based simulation extensively to study the

effects of inter-WBAN interference. The design consisted of two WBANs comprising of

six SNs each, with inter-SN distance no more than 3 m. The CSS units for both the

WBANs handled the scheduling of transmissions. The SNs are configured with variable


69

packet rate, involving real-life medical parameters of body temperature, ECG and blood

pressure. Most of the SNs received packets successfully in the first try, while a few

required two or more retransmissions, giving encouraging credence to our proposed

scheduling scheme. A plot of the results indicating packets received by the first WBAN

from the second one in presence of interference can be seen in Figure 3.6.

Figure 3.6: Split of data packets from second WBAN at the first

The work can be extended by evaluating system performance, ascertaining the

behavior of packet traffic across WBANs with more number of active WBANs in the area

of interest. Additionally a check on energy consumption of such WBANs can be

performed, along with possible improvisation on the scheduling decision process.


70

3.4 Framework for a Cooperative, Neighbor-aware WBANs

Developments in wireless and medical sensor technology, combined with clear-cut

standards being laid out, and expected to make the number of WBANs grow in number,

thereby increasing the concentration of such networks. With a number of WBANs in the

vicinity of each other, interference between them could pose to be a real problem

affecting their throughputs and cause degradation in performance [63]. Instead of

treating the interference as an impediment, we propose to use it to the benefit of WBANs

and suggest a framework in which neighboring WBANs communicate with each other

for cooperative packet routing. They do this in order to use some of the spare

transmission slots from their neighboring WBANs if the need arises. Such a situation can

occur when the WBAN with more or sudden, emergency data traffic is strapped for

transmission slots while its neighbor WBAN has some to spare. The suggested

framework implements a weighted two-pass algorithm for creating a routing tree

involving an assisting WBAN that can accommodate routing requests from its neighbor

WBAN. The first pass queries and creates the possible path while the second pass

determines the connectivity and can also be used for detecting any breaks in the path.

We define a cooperative inter-WBAN routing scheme through power based weight

assignment in which a combination of RSSI and LQI values are used to determine the

best leaf node among a set of candidate leaf nodes in the neighboring WBANs.

3.4.1 Algorithm for Cooperative, Neighbor-aware WBAN

The challenge to detect and create a route to the reachable leaf nodes in a neighboring

WBAN can be met by modifying the Kosaraju-Sharir algorithm [68]. The SNs and the

wireless links connecting them can be thought of as nodes and edges of a directed graph.

In order to study the connectivity, and possibly extend it, strongly connected component

analysis is important.


71

The algorithm finds all SCCs in a directed graph in O(n+e) time. Two DFSs are

performed on the graph. The first run of DFS finds all nodes in the graph while

computing f(i) where f(i) denotes the order when the DFS algorithm finds the node u.

Next, the transpose of the original graph G, GT is computed and DFS is performed on GT in

the order of f(u) decreasing. On examining the forest output that results after the second

DFS, we find that each tree denotes a strongly connected component. This is a general

algorithm for finding SCCs a network. In order to have some benefits, we modified it a bit

by including weights to grow the tree further.

3.4.2 Weight estimator for candidate node selection for cooperative inter-WBAN routing

The proposed scheme identifies a suitable leaf node in the neighboring WBANs for

forwarding packets to the base station. This is in line with the idea that leaf nodes are

generally idle when they are not transmitting or receiving packets. They also have

relatively empty buffers as compared to non-leaf nodes that assist in forwarding the

packets of leaf as well as other non-leaf nodes. This scheme aims to boost WBAN

throughput and optimize energy consumption by suitably distributing the packet

forwarding tasks to the otherwise relatively idle leaf nodes.

The scheme uses a combination of received signal strength indicator (RSSI) and

link quality index (LQI) to determine the best leaf node among a set of candidate leaf

nodes in neighbor WBANs to forward a packet to the base station. Raju et al. [69] have

determined that RSSI alone is a not a reliable metric for distance estimation as a

particular RSSI value may correspond to a set of possible distance values. Further, their

research has verified that a reliable approximation of the distance can be obtained by

using a combination of RSSI and LQI values. We use this metric (combination of RSSI and

LQI) to determine the priority queue of leaf nodes with the best candidate node at the

front of the queue in a neighbor WBAN with which reliable forward and backward

communication may be established so as to forward packets to the base station. In the


72

event of the best candidate getting overloaded or suffering link quality degradation,

subsequent node in the queue is considered as next preferred candidate node.

Figure 3.7: Cooperative routing from WBAN ‘A’ via ‘B’

Figure 3.7 shows two WBANs A and B in the communication range of each other. SNs 1

to 7 in ‘A’ use the services of their CSS 8 for relaying their data to the base station. Leaf

nodes 9 and 14 in WBAN ‘B’ are the possible candidate nodes for emergency data

requests from SNs 3 and 5 in WBAN ‘A’.

3.4.3 Simulation setup for Cooperative Inter-WBAN Routing

The simulation involves two WBANs designated ‘A’ and ‘B’ within the vicinity of each

other for testing the framework. The WBANs have eight SNs in each of them with


73

identical sensing, data processing and transmission capabilities. The CSS units of ‘A’ and

‘B’ can communicate with each other and form an adhoc inter-WBAN network. Only the

leaf nodes in a WBAN are entrusted with the job of accepting the traffic from

neighboring WBAN/s and carrying over the packets. The WBANs are kept stationary in

this simulation for a basic evaluation. The best candidate leaf node from neighboring

WBAN ‘B’ queue is selected for routing using free slots for meeting the routing request

from ‘A’. All WBAN nodes are assumed to be in-vivo (on the body surface, none

implanted) and are in line of sight with the others for uniformity in radio properties. The

simulations are performed on the Java based ‘Shox’ simulator [75] with the WBANs

working under 802.15.4 MAC.

3.4.4 Proposed framework for Cooperative Inter-WBAN Routing

The CSS units in various WBANs implement a discovery mechanism through beacon

signals. Information about the BAN-MAC addresses of the leaf nodes, the channels they

would listen on for accepting traffic from neighbor WBANs and their sleep/awake sync

information is interchanged between neighboring CSS units through beacon broadcasts

before the superframe.

After a successful handshake, the neighboring WBANs are able to use the packet

forwarding services through each other’s leaf nodes. The leaf nodes are chosen for acting

as relay candidates due to enhanced availability of buffer, better resource utilization, and

improved throughput while offering cooperative working of multiple coexisting WBANs.

Whenever the packet forwarding services of the neighbor WBAN ‘B’ are required

by the sender WBAN ‘A’, the RSSI (and LQI) of the leaf nodes in ‘B’ are used for

determining the weights assigned to the leaf nodes in WBAN ‘B’ for including them in a

priority queue that is referred to when choosing a candidate node from ‘B’ for relaying

data. While RSSI is an indicator of signal power, LQI focuses on the BER in the link.

Together, they are used to evolve a metric system for making the queue. The choice of

these two basic parameters for generating the queue is also justified in the light of the


74

fact that they can be combined to give a rough estimate of the distance [69] between the

nodes in ‘A’ and those in ‘B’. Depending on these weights, a weighted routing tree would

be created between ‘A’ and ‘B’. An example of the routing tree is depicted in Figure 3.8.

RTS/CTS mechanism could be used for mitigating channel reservation requests from

neighboring WBANs to be served by ‘B’. Surplus/Emergency packets generated by ‘A’ are

transmitted via ‘B’ and the overall performance of this kind of cooperative network

sharing is evaluated.

Figure 3.8: Creation of routing tree between WBANs ‘A’ and ‘B’

The above functioning does not apply only to the two WBANs ‘A’ and ‘B’ in the

simulation, and could be extended to include more WBANs in the tree. To keep it simple,

we tried evaluating the proposed framework with a minimal number of WBANs initially.


75

Once we have an initial network graph, we can use that as our reference. Later, if

we have a graph with a disconnected component, running the second pass of the

algorithm on it gives us all the connected components. We can compare the results with

our reference and locate the faulty link. In this scenario, we assume that there are only

single links or edges between nodes or graph vertices in each WBAN.

3.4.5 Results and performance evaluation for Cooperative, Neighbor Aware

Inter-WBAN Routing

The following graphs show the results obtained at the end of the simulation.

Figure 3.9: Effective transmissions per packet with distance between SNs (meters)

The effective transmissions per packet vs distance plot in Figure 3.9 helps in establishing

the fact that over varying distance the overhead per packet does not vary significantly.

This very clearly demonstrates the robustness of a network topology that is multi-hop in

nature.

The above fact is also strengthened by the lost packets v/s distance plot shown in

Figure 3.10. It is very clearly evident that the packet loss is not very high even over


76

longer distances. This is due to the fact that the network topology is multi hop in nature

and hence the packet loss due to reduction in energy over long distances is avoided.

Figure 3.10: Packet loss in routing based on distance between SNs (meters)

Figure 3.11: Packet transmission time


77

The effective time taken per packet plotted in Figure 3.11 shows the total time taken in

transmission of a packet. It takes into account the number of retransmissions needed.

Therefore it also includes the additional time required for a successful transmission.

This also shows that on an average the additional/ overhead time per transmission is

not high.

The plot in Figure 3.12 shows the transmission time for particular distance/hops is

shown to stick to a near linear progression. This indicates that the network performance

is optimal when it comes to transmission time over different distances. There are no

outliers in such a network scenario.

Figure 3.12: Transmission time for hop distances

Although the best candidate leaf is decided on the basis of the RSSI-LQI based metric, it

is found to be the same from an intuitive guess. The Euclidean distance-wise nearest leaf

is the best candidate to send the packets through. This is also proven from the

simulation results. A check on fault localization could be performed with simulations on

a larger network to check if the system is scalable.


78

3.5 Dynamic Virtual Cells with Multiple Multicast Trees for routing WBAN

data

So far we have tested the performance of WBANs in a stationary setup. In this effort, we

tried to check a what-if scenario on WBANs involving Small Scale Networks (SCNs) if a

piggyback of WBAN data through SCNs would work. SCNs themselves are relatively new

and their performance for relaying voice data is still under evaluation and improvement.

SCNs involving femtocells are increasing becoming common and popular among service

providers and end consumers. The SCNs find it difficult to manage the mobile users

owing to frequent handovers that reduce the overall performance drastically. The

constraints apply to any WBANs that try to piggyback on the SCNs for routing their data.

High number of handovers would also mean reduced throughput for WBAN data with

QoS guarantees not being probably met. We proposed and tried to evaluate a simple

framework for decreasing the number of handovers using a novel concept of dynamic

virtual cells [72]. Such virtual cells have the capability to dynamically grow or shrink in

accordance with performance requirements. They can also move geographically with the

movements of mobile users. Handovers are observed to reduce in number because they

occur only between those virtual cells that are parts of different multicast trees.

3.5.1 A Dynamic Virtual Cell

If a group of cooperating stations utilize the same dynamic virtual cell (DVC) ID [70] and

the same channel, they can be treated together as a single, large DVC. Such stations

would collectively appear like a large macro cell for the user, thus limiting the handovers

to occur only at the virtual cell boundary, increasing the inter-handover distance [70],

[71].

The use of SCNs would offer benefits like reduction in power consumption, better

frequency reuse, improved performance and low installation cost [71]. The SCNs were

assumed to be connected to a Wireless Mesh Network (WMN) backbone. The traffic


79

between the WMN backbone and the Internet is relayed by multiple Internet Gateways

(IGWs).

The scheme constructs individual virtual cells for each user based on their mobility

modifies them with movement. An optimal multicast tree that would maximize

achievable performance for handover had to be created. The multicast tree would either

need to grow bigger to enhance the coverage of the user or the user would have to

switch over to a new smaller multicast tree as long as the performance gain

compensated for the handover cost. The proposed DVC design [74] could be extended to

cover multiple users by multiple access techniques such as FDMA or TDMA. The

movement of users could be detected by the virtual cell members from RSSI values.

3.5.2 Simulation of DVC based Multicast tree model

Manhattan Mobility Model was used to simulate the mobile user’s movement in a

serpentine pattern at a slow downtown cruise speed of 15 mph. An 802.11 based WMN

was created, in the Qualnet Network Simulator [73] with four Mesh Routers (MRs)

located in the four corners of the simulation area boundary. The MRs used in our

example would act as root nodes for four different multicast trees.

Assuming square blocks, The Small Cell Base Stations (SCBSs) are placed at

intersections and provisioned to have an average coverage radius of about half the block

length. Each of the metropolitan blocks would be covered for service by four SCBSs

except for the locations near the boundary of the network simulation area.


80

Figure 3.13: Dynamic Virtual Cells in the SCN

We have resorted to On-demand Multicast Routing Protocol (ODMRP) in our simulation

in order to construct the multicast mesh network. Handover decisions were based on

maximum throughput and minimum end-to-end delay. A handover would mean an

overhead of cost to create a new multicast tree for a user and the associated signaling for

tree creation and session restart.

3.5.2.1. Simulation Results from the DVC based Multicast tree model

Table 3.3 shows expected system performance of dynamic virtual cells with different

mobility management methods.

Table 3.3 Expected performance gain of the DVCs

Mobility Management

Expected Packet Success Rate

Expected End-to-End Delay

Number of Handovers

Single Root 67.65% 66.63 ms 0

Multiple Roots 90.32% 32.88 ms 21


81

The DVC implementation achieved 67% packet delivery rate without handovers. If there

is an optimal switchover to another DVC, above 90% of average packet delivery success

rate is observed. Although the movement affected 21 handovers, 74% of saving on

handovers is achieved as against a traditional SCN involving a minimum of 81

handovers.

Figure 3.14: Throughput performance of DVCs

Figure 3.15: Delay performance of DVCs


82

Figure 3.14 shows that for a DVC with a single multicast tree, the throughput fluctuation

gradually decreases while the end-to-end delay gradually increases. DVCs with multiple

multicast trees are observed to maintain a constant delay fluctuation and throughput. A

switchover to a better DVC is effected when the throughput gradually decreases,

maintaining a constant performance. Figure 3.15 shows the delay performance from

end-to-end for the user packets generated.

Further investigations into mobility and performance metrics with QoS can be

performed on realistic 3-D human models.

83

Chapter 4

Evaluation on feasibility of control in Wireless

Networks with a focus on WBANs

4.1 Reliability issues in Wireless Control

Although wireless systems have been around for quite some time, and have earned the

trust in conveying data containing qualitative and quantitative information. Skepticism

still prevails when it comes to relying on them for control applications, the. As the third

challenge, we tried to probe if WSNs in general or WBANs in particular could be used for

wireless control.

4.2 WSNs and Computer Music

We worked with the University of Cincinnati College-Conservatory of Music in

developing an interactive computer music generation system for dance performances

[76]. As many as sixteen dancers performed in a series of events where they generated

music by their dance movements using a WSN at the helm of music control. The WSN

involved the use of Tmote SNs with optical and acceleration transducers that provided

control signals to the computer music system implementing Java code for audio-visual

selection and signal processing.

4.2.1 Four dance performances using wireless control

One of the performances used a WSN involving twenty three Tmote SNs connected to a

computer music system. Dancers wore these SNs with optical transducers and 3-axis

Chapter 4: Evaluation on feasibility of control in Wireless Networks with a focus on WBANs

84

accelerometers that triggered sound depending on their location, lighting levels, and

proximity of the dancers to each other. The Tmotes ran TinyOS-2.x [78] for

computational processing and forwarding serial data from the SNs into Java objects for

controlling sound selection and processing.

Figure 4.1: Flow of data into sound processing software

Received Signal Strength Indicator (RSSI) based localization was used to estimate a

rough location of the dancers. RSSI is a measure of signal strength (in voltage) of the

received wireless signal. The signal strength variation can be modeled to estimate the

approximated distance between the transmitter and receiver as it is inversely

proportional to the distance between them. The omnidirectional antennae used in

Tmotes makes it easy to apply simple triangulation for an estimation of the receiver

location. Based on the incoming RSSI data, the base station determines whether a dancer

is close to a stationary perimeter sensor, or if the dancers are close to each other, and

generate different sounds accordingly. The sound-triggering programs react differently


85

depending on the light intensity data received and the location of the dancers. The

variations in light intensity are used to control sound amplitude levels.

The lighting problems cause some issues in sound generation, and at times it is

difficult to have a connection between movement and sound. The RSSI scheme for

location tracking turned out to be not too reliable.

In the second performance, acceleration data from a dancer’s movements is used to

control digital signal processing more explicitly and in this case more successfully.

Acceleration data on the three axes allowed the dancer’s arm movements to control

timbre and other parameters of the music. A better link is observed between the location

of the dancer and specific kinds of sounds.

The dancer’s movements are detected by comparing the stream of acceleration

information from the 3-axis accelerometer data from the sensor on the dancer’s arm is

compared to predetermined and stored movement data from a training set, and is used

for detection of the dancer’s movements. A simple fuzzy scheme is used for this

comparison.

In the third performance, an accelerometer with a Tmote sensor is attached to a ring

around a pianist's fingers. The accelerometer data is relayed through the WSN to the

base station and used to control the sound of the piano. The transformed piano sound is

observed to be clearly related to the pianist’s hand movements.

RSSI and sensed physical parameter values about the current and previous locations

in relation to the SNs, plus a timestamp are conveyed to the base station. This makes it

possible to track the path and speed of the subject in the area of interest.


86

The use of laser pointers with wireless SNs yielded a more precise localization of a

dancer. Localization is achieved using a detection of disruption of laser beam by a

wireless sensor and the RSSI values. The base station collects RSSI values from all the

SNs near the user to determine the user that is closest to the interrupted laser beam.

The user’s location and timestamp of the event are then reported.

By keeping a track of values two consecutive locations that the beam is interrupted

and the time spent between two locations, localization information, average travelling

time and speed of the dancers is ascertained. This provides improved localization over

plain RSSI based localization.

The fourth performance experiment uses wireless pressure SNs for more accurate

triggering of events. Using the pressure sensors, the sound events could be started,

sustained or cut off by the dancers, thus improving the control and allowing the dancers

to change the composition at will. A tracking of dancers’ movements is attempted using

3-axis accelerometers with wireless SNs inside a foam ball that could be rolled. In

general, the complexity and reliability of the WSN control improved. However, it still

remains ON-OFF control at best.

4.2.2 Water Birds: Compositional Collaboration with Clarinets and Wireless

Sensors

This work involved controlling a clarinet’s sound [77]. A WSN with IR sensors is tied up

with a clarinet and responds to the clarinetist’s movements. The WSN then sends data

into a base station computer for signal processing control. A Java program running on

the computer would receive the data from the programmed Tmote SNs.


87

While the use of WSNs for control in the field of entertainment have been encouraging,

possibilities of performing industrial and medical control using WSNs can be probed.

4.3 Wireless Control in an Industrial setup

There is a great potential for the use of WSN technology in the areas of industrial,

commercial, and consumer applications. Process data for monitoring and control from

measurements corresponding to physical parameters such as pressure, temperature,

flow, level, humidity, vibration and several other parameters can be obtained from

sensors and transferred over wireless links to a control system for maneuver and

management. For process monitoring and control, there are several advantages of opting

for WSNs over the traditional, wired automation systems.

A combination of architectures, mechanisms, and algorithms is used in the industry

for monitoring and control of the process activities to achieve a specific goal, usually of

industrial production of required quality. The field of process monitoring and control

has traditionally involved wired instrumentation and equipment in industrial

applications. An example of such an application could be one that involves adjusting the

flow rate of the coolant through the cooling jacket surrounding a reactor for maintaining

a specific, desired temperature. This outcome would be of sustaining a constant, preset

temperature over a period of time, where temperature is the input variable as well as

the controlled variable. As shown in Figure 4.2 for such a system, all the devices are

connected through wires.


88

Figure 4.2: A traditional, wired Instrumentation Control System

The temperature is measured by a specific thermal sensor and is used in a control

function to decide the adjustment of a valve to manipulate the coolant flow rate through

the cooling jacket. The desired temperature is the setpoint for the control loop. The valve

opening position (e.g. the setting of the valve allowing cooling material to flow through

it) is called the manipulated variable since it is subject to control actions. The value of

temperature is sensed and transmitted to the controller that implements the functions

and calculations, transmits the output to control the valve and issues alarm/s on faults

and errors. All of the physical parameter data throughout the process is archived for

future reference. This archival helps maintain the history of process trends and is useful

for operational modifications. If the scheme changes from wired to wireless, sensing and

action devices will communicate on wireless channels through an access point (a

gateway or router), which is connected to the control room using wireless or wired links

using Ethernet or Modbus. Our example from figure 4.2 would then transform to the

schematic in figure 4.3.


89

Figure 4.3: A wireless Instrumentation Control System

Most of the process control applications are mission critical and have stringent

control requirements. Failure of a control loop in process plants can potentially cause

unscheduled shutdowns or serious accidents. Rigorous researches have indicated the

compelling paybacks of embracing wireless communication technology combined with

sensors, but a diversity of challenges at the end-user level has caused a guarded attitude

to be adopted by the technology frontrunners. Most of the early applications of wireless

automation have primarily concentrated on process monitoring instead of confidently

going in for closed-loop process control. The transition has taken some time but the

confidence of plant operations personnel on wireless automation has been on the rise,

due to standards like WirelessHART maturing up.

An open-standard wireless networking technology called WirelessHART (Wireless

Highway Addressable Remote Transducer) [79] for wireless automation in industrial

control came into the market in late 2007. Since then its use has been continuously on


90

the rise. Another open-standard wireless networking technology for industrial

automation called the ISA100.11a, was released in September 2009 [80] and continues

to stay in the race.

Wireless automation has been used for monitoring the well-head annular pressure

and pressures inside a heat exchanger on an offshore platform in the ocean [81] over a

self-organizing mesh field wireless network. Continuous remote monitoring of these

pressures reduces the need for everyday visits to the wellhead for a manual noting-

down of the gauge readings. It also helps the operations personnel identify unusual

readings earlier and lets them take corrective action taken by investigating and

correcting the errors before they grow into serious complications. The installation of

wireless automation systems takes just a few hours as compared to several man-days

required for deploying wired instrumentation and control.

Another good application of wireless automation is the monitoring of temperature on

a rotating drier to make certain that the correct temperature is reached and upheld

during the process of drying [82]. The cost and effort involved of wiring for temperature

sensors on a rotating drier make a wired scheme infeasible. With the savings introduced

by using wireless transmitters, more temperature transmitters can be located on each

drum to increase the amount of process information and offer a factor of redundancy for

better reliability of measurement.

There are no wiring related encumbrances in wireless sensor and control networks.

The reduction in cabling is substantial and it makes maintaining the sensors and the

entire automation system very easy. Such systems suffer from no wiring maintenance


91

problems such as physical wearing of the cable, corrosion of the cable and insulation,

water in the conduit, burned cabling, freezing, or damage from wild animals. Doing away

with the wiring reduces the cost of the cable, the effort and cost of cable-laying as well as

that of maintenance.

Industrial WSNs have the prospective to outperform the traditional process control

networks with wired devices. To start with, they have a higher speed of data

transmission. The Highway Addressable Remote Transducer (HART) control protocol

has a data rate of 1.2 kbps. Foundation Fieldbus (FF) has a data rate of 31.25 kbps, while

WirelessHart which is based on the IEEE 802.15.4 standard, can reach up to a data rate

of 250 kbps.

Secondly, in case of wired control systems, the devices share a single communication

bus, while in case of wireless automation systems; multiple such systems can work

simultaneously if there is no mutual radio interference in the area of implementation

[83]. Thirdly, redundant design with more sensors or data-points can be employed to

beat the performance of traditional wired control system.

Such a design increases the overall system reliability. Also, if the design allows for

two or more different frequencies being used to communicate the values of the same

automation parameter at a time, transmission of parameter data will succeed even if

there is interference on one of the frequencies. Redundancy could also help with

detection of faulty sensors as proposed in [84].


92

Figure 4.4: Architecture of an Instrumentation and Control System built with wireless nodes

Fig. 4.4 shows the architecture of a wireless node used for process monitoring and

control. There are severe limitations on resources such as processing power, memory,

bandwidth and energy capacity in the design of such a device.

The processing module is responsible for supporting the operating system, the

network protocol, and for handling the data processing and control algorithms. The

microprocessor used in the design has to be chosen wisely, and should ideally have large

memory and powerful computation capability. It should consume ultra-low power, and

are should be energy efficient as can be seen in the comparison of available

microcontrollers in [85]. The design discussed in [86] offers a novel, heterogeneous

multiprocessor sensor node capable of staged wakeup that keeps the system energy

efficient. With advances in technology, more powerful processors with large memory

and reduced cost have started to emerge for sensor nodes.

A substantial amount of data from sensor nodes may be redundant, though

significant. Multiple signals generated from two or more temperature monitoring nodes,

or from dissimilar sampling durations within the same sensing unit require handling by


93

the automation system. Samples from similar data type can lend themselves to data

aggregation by use of functions such as maxima, minima, duplicate suppression and

average to reduce the number of transmissions. Data aggregation in wireless automation

systems can help in achieving the required network capacity within the constraints of

limited bandwidth [87]. The act of aggregating multiple packets into a single packet

increases the network throughput with a reduction in the overhead for each sample data

packet, although with an introduction of extra transmission delay. Network capacity

enhancement can be effected by encoding the data before transmission [88][89].

Transmitting processed data over the WSNs instead of sending raw data for estimation

results can improve the data capacity immensely [90].

By far, the performance limiting factor which is the biggest concern for WSNs is

power consumption required for wireless transmission. As on date, most wireless-based

industrial applications claim a battery life of about four to seven years. There is a limited

viability of use of wireless in automation systems if they require relatively large

amounts of power, and need to be powered by batteries. For the part of such a system

that has a wired power source there are no issues related to preference or feasibility.

Resorting to event-driven energy efficient method can help save on battery power, since

fewer data packets need to be sent and the transmission is less frequent. Power usage

can also be cut down by employing small data packets and transmitting only when

parameters change. Sensor nodes can also be programmed to operate with a very small

duty cycle in order to conserve battery power keeping them as inactive as possible for as

long as possible. Yet another approach that could save appreciable amount of energy

involves the use of wireless sensor nodes that can be programmed to work over variable


94

or adjustable transmission ranges. A scheme on these lines using energy-balanced

consumption by varying the transmission ranges of wireless sensors is suggested in

[91]. The total power consumed can also be minimized using an algorithm that searches

for an optimal transmission range through topology management, as proposed in [92].

The use of energy aware protocols in the network can be another approach towards

conserving battery power. The REAR protocol [93] for wireless sensor networks is a

reliable energy aware routing protocol that considers residual energy capacity of each

wireless sensor node for establishing routing paths. It also supports multi-path routing

to increase the reliability of data transmission. Another such protocol [94] considers

both energy and delay metrics to discover an optimal path with minimum end to end

delay and minimum energy consumption for real time traffic in wireless sensor

networks.

A wireless sensing and control network could also use energy harvesting techniques

for improving network lifetime. In addition to conserving on battery power, alternate

powering mechanisms involving solar power, energy harvesting from vibrations,

ambient temperature or surrounding RF power can be used to prolong the WSN lifetime

[86].

4.4 Industrial applications with potential for Wireless Control

The process control industry has witnessed distinct eras of innovation headway. There

have been changes from hydraulic and pneumatic systems to electrical and electronic

operations. With each new era of control equipment, the operator control workstation


95

has turned out to be progressively smarter. Innovative applications provide

progressively better alarming, control, and diagnostics. The devices that make these

applications possible are turning intelligent and niftier. They can now sense and can

convey useful process and diagnostics information. These smarter gadgets effect a

reduction in engineering costs, offer an enhanced understanding of the process, and

effect the refining of the overall functioning and performance of the industry. They

include innovative diagnostics that can inform about the health and maintenance

requirement of the devices as well as report about the process that they are connected

to. Advancements in a by-now-mature wireless tech enable these smart devices to

exploit the competencies of their existing control system organization. Legacy control

systems can utilize the economical alternative communication path offered by wireless

technology. Measurements that were considered cost-prohibitive earlier now can be

measured and relayed through wireless for incorporation in monitoring and control

schemes. Wireless technology can be applied in industrial monitoring, control and asset

management applications without any safety concerns.

Deployment of new measurement and control parameters has become simpler, more

reliable and cost-effective. There are no wiring costs to be considered and existing

systems do not need to be changed completely. Wireless technology can provide an

infrastructure that lets central office users or even mobile users to access the process

parameters and related equipment.


96

4.4.1 HART Protocol and WirelessHART

The need of a global standard for wired solutions for use in the process-control industry

has been served by HART, developed by the HART™ Communication Foundation (HCF)

[95]. Such a standard is welcome, and offers the benefits of insuring that devices from

multiple suppliers work together – thereby lowering risk and cost for both the supplier

and the user.

The ‘HART’ protocol has been in use in smart instruments for process communication

since late eighties. The population of HART devices installed and being used would be

close to 50 million in the world today, totaling up to a 75% population of all the smart

devices deployed. HCF later expanded the capabilities of the HART protocol beyond

simplicity, reliability and security by adding wireless capability to it. The resulting

WirelessHART is backward compatible with its older, wired counterpart, thus

empowering older HART users to speedily and effortlessly reap the benefits of wireless

technology.

4.4.1.1 Simplicity of control through WirelessHART

One can now fix-up instruments on rotating equipment such as kilns and on movable

assets such as trailers and railcars, thanks to wireless technology. The nature of wireless

makes WirelessHART easy to use. The installation and commissioning is easy and less

costly due to reduced wiring and materials costs, which further reduce the cost of labor

and effort. Adding a new instrument to the network is also easy, as WirelessHART senses

the addition and adjusts to changes in plant equipment by itself. This makes the


97

WirelessHART mesh an adaptive network with self-organizing and self-healing

capability.

The network can easily be extended at reduced engineering and project cost, to far-

off areas, as no wiring is needed. It also offers the flexibility of building up the network

one device at a time. It is possible to deploy wired and wireless devices due to

compatibility, and allows seamless integration to existing distributed asset management

or control systems.

Manual data collection is eliminated, making the access to new WirelessHART mesh

measurements fast and stress-free. Remote plant areas are better visible due to wireless

links. Loop troubleshooting and maintenance is streamlined due to added diagnostics

features that make predictive maintenance easy. Setting up temporary configurations

using wireless devices is easy and quick. WirelessHART can be put to use in supervisory

compliance monitoring in the areas of environmental, safety and health.

4.4.1.2 Reliability of WirelessHART control

Industrial environments are unfriendly to wireless technology implementations. Such

environments have dense metallic set-up, recurrent large-equipment movement and

changing surroundings that make it difficult for radio waves to travel. There is

substantial interference to wireless communication from several sources of radio-

frequency and electromagnetic radiations. The use of direct sequence (DSSS) and

frequency hopping spread spectrum (FHSS) techniques by WirelessHART spreads the

communications among various physical communication channels, thus increasing the

reliability of the communication.


98

By virtue of being a mesh network, WirelessHART is redundant and self-healing.

Using diagnostics information available across the network, it keeps an eye on

degradation of communication routes and keeps repairing and adapting the network for

an optimal performance. It discovers alternate routes around obstacles, and keeps

switching channels.

A WirelessHART network can coexist without issues with various kinds of other

wireless networks, in addition to other WirelessHART meshes. It examines the targeted

channel with clear channel assessment (CCA) before actually using it for transmission. If

it finds some channels to be constantly occupied or noisy, it temporarily blacklists such

channels and refrains from using them. It keeps all transmission extremely

synchronized, and this helps in optimizing both, the radio time and the bandwidth.

4.4.1.3 Security of control in WirelessHART

The WirelessHART standard uses standard 128-bit AES encrypted communication at

multiple tiers for robust and highest levels of protection. The data link layer is protected

by a secret network key used for the authentication of each transmission. Different

sessions at the network layer use different keys for encryption and authentication

among communicating peers. A different join key is used for each device to encrypt and

authenticate during device join process. The keys are all, periodically changed by the

network manager as long as the network is active.

Along with channel hopping at the timeslot level, WirelessHART selects the actual

physical channel at the point of transmission along with a control on the transmission


99

power depending on the transmission distance. All these techniques add to its security

and make sniffing of traffic challenging for intruders.

4.4.2 The WirelessHART standard

The WirelessHART networking technology works in the 2.4GHz ISM radio band. It

employs IEEE 802.15.4™ [96] compatible DSSS radios with packet by packet channel

hopping. The application layer is same for wired HART and WirelessHART standards.

The physical layer, data link layer and network layer are distinct for WirelessHART.

4.4.2.1 Predecessor: the HART Standard

The HART Field Communications Protocol standard [95] allowed piggybacking of two-

way communications on a 4-20mA signal while not interfering with the integrity of

sensed parameter values. The current version of the protocol adds wireless mesh

capability with time sync, along with transport and network layers. It supports event

notifications, block transfers, security with encryption/decryption, high speed piped

data communication and advanced diagnostics which convey information about the

HART device, any attached equipment, the process being examined.

4.4.2.2 WirelessHART Standard details

HART 7 includes a major new communication protocol, the WirelessHART protocol [95],

supporting wireless applications. Like wired HART protocol, WirelessHART protocol

targets fixed sensors and actuators. Rotating equipment, such a kiln dryer, and flexible

manufacturing are also target markets. WirelessHART is needed both for protecting


100

user’s investment in their installed base and for new opportunities. We need to preserve

legacy products and applications, to continue existing work practices and trainings; we

also need to use wireless to lower the cost of making a measurement, access advanced

diagnostics information, and have better equipment monitoring.

The WirelessHART standard leverages existing standards such as the HART standard,

the IEEE-802.15.4 standard (www.ieee802.org/15/pub/TG4.html, Gutierrez et al. 2007),

AES-128 encryption (US FIPS Publication 197), and DDL/EDDL (www.eddl.org). It can

do whatever the wired HART part can do and more. The WirelessHART technology is a

secure networking technology operating in the 2.4GHz ISM radio band. It utilizes IEEE

802.15.4 compatible DSSS radios with channel hopping on a packet by packet basis.


101

Figure 4.5: A WirelessHART Network

4.5 WirelessHART Industrial Application model

WirelessHART can be used in building up a variety of control applications. With the data

rate of 250 kbps or more, WirelessHART has the capacity to ferret out control data in

addition to transmitting of physical parameter data without any risks. Redundant

systems could be implemented in order to improve the reliability of control. A simple

example of such process automation could be the control of liquid level in a tank. The

tank has an inlet for liquid at the top of the tank and an outlet at the bottom. In order to

maintain the level of liquid in the tank, it needs to be measured and chosen as the


102

process variable (PV) for operator indication as well as for the control loop. A PID loop

could be designed for automatic control of the liquid level and tuned using standard

methods like the Ziegler Nichols heuristic technique. We tested such a control system

designed in LabVIEW r2014 [97] as shown below. LabVIEW allows a range of features to

its users, from design and simulation of industrial systems, and their evaluation to actual

control applications. LabVIEW application programs are called virtual instruments and

stored as Virtual Instrument (VI) files that appear and operate just like mimics of actual

instruments. A VI has two components: a front panel that acts as the user interface and

displays the secondary indicating instruments and controls, and a block diagram that

lets the programmer choose the functionality of the components for the instrumentation

and control application. Figures 4.7 and 4.8 show the front panel and the block diagram

of the level control system.

The inlet valve into the tank is an automatic valve while the outlet valve is manual

in operation. The operator can set the level control set-point for auto mode of operation.

An Auto-Manual switch allows for toggling between the two modes. Value and trend

indicators provide the operator with a feedback on the process, with information on the

PV, set-point and controller output. Additional controls exist for changing the control

parameters of the PID algorithm.


103

Figure 4.6: The front panel for the liquid level control system [97]


104

Figure 4.7: The block diagram for the liquid level control system [97]

The block diagram shows that the process is modeled for PID control. The variables can

be displayed as well as logged on to a measurement spreadsheet file for archiving and

sharing.


105

While the modeling of such a control system using WirelessHART is straightforward, we

next try to evaluate if WirelessHART could work for an 802.15.6 standards based WBAN,

while making control a possibility in such networks.

4.6 WirelessHART based framework for Control in WBAN: possible

applications

A possible use of wireless control in WBANs could be an insulin injection actuator that

administers the right dosage of insulin to a diabetic patient on the basis of process

variable values from glucose sensors.

A similar control application could be implemented in case of patients suffering with

Perkinson’s disease. In cases of advanced patients suffering from PD, doctors can see an

unsafe condition called “freezing” or “freezing of gait” [98] in which the patient’s may

not be able to move his or her feet and seems to be stuck, or may be unable to move or

stand up from a sitting position. There are no known causes of the ‘freezing’ problem.

The start and end of a ‘freeze’ cannot be predicted and may result in the patient losing

his or her balance and falling down. The situation could be graver if such a patient is

being helped by a companion when the episode occurs. The episodes occur at random

and the frequency may differ from one patient to another. Such PD patients are typically

administered dopaminergic medicines the amount and frequency of which could be

dependent on the patient’s condition. A control application could utilize an actuator

driven injection mechanism for the dopaminergic medicine. While the mechanism can be

programmed to administer doses at fixed intervals, it can be coupled with a


106

measurement and prediction mechanism to decide when the next dose is due and act

accordingly.

Another such application could be in the case of ‘drop foot’ which is not exactly a

disease but has muscular, neurological or anatomical roots [99]. Subjects experiencing

‘drop foot’ find it difficult to move or lift the front part of their foot resulting in its drag

when they walk. This condition could be temporary or permanent in case of different

individuals. The usual management for ‘drop foot’ is to make the subject wear an ankle-

brace that holds the foot in normal position.

The CSS can serve as the gateway node for measurement as well as control. We have

implemented a barebones data acquisition system for a WBAN using a smartphone for

the CSS. The same CSS could also be extended in functionality for control purposes too.

However, for evaluating the feasibility of WirelessHART for our 24-channel WBAN

model and WBAN control framework, we resorted to our simulation written in MATLAB

r2012.

4.6.1 Proposed Framework

The IEEE 802.15.6 standards stipulate the QoS requirements for the throughputs

required to be met for a few important WBAN applications as indicated in Table 4.1.

Table 4.1: Bitrate QoS requirements for common WBAN parameters

WBAN parameter/control Bit Rate 1 Drug delivery 16 kbps 2 Electroencephalogram 86.4 kbps 3 Electrocardiogram 192 kbps 4 Deep brain stimulation 320 kbps 5 Electromyogram 1.563 Mbps


107

The latency requirements for all the five parameters are less than 250 mS. With the

amount of bandwidth available uniformly across the working channels, it can be

observed that the latency requirements would be met for the 24 parameters when three

channels are available. The samples would not need to be eliminated and could be

transmitted in full, for a WirelessHART implementation. However, for satisfying the bit-

rate required for more demanding application like the EMG, we could resort to reducing

the data or by implementing delta modulation for reducing the number of bits.

So far, no protocol that can handle control has been proposed for 802.15.6 WBAN

standards. There are no specific simulators that would let us model the control for a

WBAN system based on the 15.6 stack hence we went with 15.4 radio based system

which uses the same radio spectrum in the 2.4 GHz unlicensed band and for which

robust control protocols like WirelessHART are already available.

We propose a 15.4 Wireless based Instrumentation and Control schematic up to the

CSS, and a stripped down 15.4 WirelessHART implementation within the WBAN, from

the CSS up to the WBAN actuators. We would just need to curtail the signal power within

the WBAN keeping the signal strengths in mind. The modeled system we used involved

24 channels parameters from the subject’s head, heart, internal organs and limbs. We

also included possible control applications like pacemakers, insulin injectors and

movement actuators in the model. While the model had 24 parameters, it worked using

the 15 channels (Channels 11 – 25) available for use in the WirelessHART.


108

4.6.2 Performance evaluation and feasibility check: Channel performance

with sample cut

The HART protocol is built on the 802.15.4 wireless protocol and allows for a maximum

achievable bandwidth of 250 kBps for transmission of sensor and actuator control data.

A HART frame allows for transmission of a maximum of one message/packet of 127

bytes every 10 ms. The rate of packet generation would need a transmission bandwidth

of 12.7 kBps. The standard specifies for 8 measurements per packet, each measurement

data of size 2 to 4 bytes with an additional overhead of 5 bytes, which require 21 to 37

bytes. With Full 4-byte PCM encoding of all samples, we had the results as in table 4.2.

Table 4.2: Performance results of WirelessHART model for

the 24-channel WBAN model

Good/Active Channels

Max BW in kBps realized

# of 2 to 4 byte samples in payload in 10 ms

1 12.7 8

2 25.4 16

3 38.1 24

4 50.8 32

5 63.5 40

6 76.2 48

7 88.9 56

8 101.6 64

9 114.3 72

10 127 80

11 139.7 88

12 152.4 96

13 165.1 104

14 177.8 112

15 190.5 120


109

When only a single channel active was active, it could allow 8 parameters with 100

samples per second to be transmitted. This rate of sampling was perfect for our previous

model for energy saving using sample reduction with 1/3 to 1/4 of samples eliminated

for an approximation. For a total of 24 WBAN parameters in our model, we required at

least 3 channels to be active at a time in order for them to be collectively able to transmit

with the same rate of 100 samples per second.

We started increasing the number of active channels one by one, assuming an equal

distribution of resources among the parameters. To keep the model simple, we assumed

that all of them were being actively transmitted at the same rate. Initially, up to three

active channels, the sample rate was low and was insufficient for the parameters to

satisfy the QoS requirements. However, with the availability of 4th channel onwards,

things start to change, as can be seen in the results. At this stage, with 4 channels in

operation, the 24 WBAN parameters achieved a sample rate of 133.33 samples/sec. This

rate was better than what was required for our approximation with 1/3rd of the total

samples eliminated.

When 11 channels became available and active, we had a transmission rate of 8800

samples/sec, and this is when we reached the break-point for 24 parameters, each

transmitting at a rate of 360 samples/sec containing WBAN parameter and control data.

Hence, we could prove that WirelessHART would certainly work well for WBAN

applications as far as the QoS requirements on the data rate are concerned.

We further tried to extend this work to differential sampling schemes. When we

packed 1-bit Delta Modulation (DM) encoded 12 bit signals in these packets we found

out that 2 bytes could hold 16 samples, 3 bytes could hold 24, 4 bytes 32 samples. For


110

the 1-bit DM case even in the face of random channel failures, we found out that the

WBAN parameter and control data transmission works fine even if up to 4 channels have

been blacklisted at a point of time.

With 1 bit DM, we found out that 16 samples to 32 samples per packet were

transmitted. The resulting error due to approximation was high, but we could get a

capacity multiplication of 16 to 32 times. The scheme allowed for the transmission of

25600 samples for 8 parameters which is overkill. For all the 24 WBAN parameters, the

rate was still a little above 1000 samples/sec. With just one working channel we could

obtain a performance from the model that was sufficient for 24 physiological

parameters. Each of the channels could handle a packing up 1066.67 samples, and for

this we assumed that the data from all WBAN parameters in the model could be

squeezed into the 8 allowed channels. 2-Bit Adaptive Delta Modulation (ADM) reached

almost the half of these result figures, although for a better accuracy. An extension of the

model into a 3-bit ADM system proved out to be just perfect for a 360 samples/sec

system. The scheme would work perfectly to meet the QoS requirements for the 24-

parameter WBAN using IEEE 802.15.6.

4.6.3 Evaluation of BER Confidence Level for WirelessHART Channel for the

model

While receiving data, it is not really important to have the value of the true BER of the

WirelessHART network. The threshold value of the BER is important, and sufficient

amount of data needs to be received in order to find out about the confidence about

keeping the BER of the system below the threshold. It becomes important to focus on the


111

number of errors E that result on repeated transmission of N bits in order to find out the

overall percentage of tests for which the obtained BER (E/N) stays under the threshold

bar. This gives us the BER confidence level in percentage. For a Poisson distribution of

errors, it can be calculated using eq. (i). This value indicates the percent confidence

about the WirelessHART network’s BER being below the specified threshold, if the

transmission was repeated an infinite number of times with the BER observed.

(i)

As we cannot quantify for an endless timeframe, the confidence level is constantly under

100%. Before beginning a BER evaluation, one must distinguish an objective level of

confidence. Targets may typically be above 80% confidence level, with 93-95% being the

best levels achieved.

4.6.4 BER Confidence–level for WirelessHART model tested

All industry guidelines indicate a most extreme BER allowed, and WirelessHART is no

exception. According to the specifications, there could be approximately close to one bit

of error on every 1000 received packets or messages, where a maximum message or

packet could comprise of 127 bytes.

In our modeled WirelessHART application, we focused on the amount of time required

to measure the BER for attaining the specific confidence-level with precise number of bit

errors in mind. We checked our model for studying the BER confidence-level observed

with zero, one and two errors, for the boundary data rates of 31.5 kBps and 55.5 kBps.

The results of our findings are detailed in table 4.3 and depicted in figure 4.8. The results


112

provide us with the amount of time required to reach a certain confidence level for the

boundary data conditions. Anything below 75% of confidence level had to be discarded

as it would not be acceptable as per the QoS specifications for most communication

standards.

Table 4.3: BER Measurement and confidence evaluation with changes in time window

#Bit Errors → 0 1 2 Measurement time (sec) ↓

BER Confidence Level (CLx100%)

3 73.61 Very low Very low 4 83.07 Very low Very low 5 89.14 Very low Very low 6 93.03 74.47 Very low 7 95.53 81.64 Very low 8 97.13 86.95 Very low 9 98.16 90.81 Very low

10 98.82 93.58 81.96 11 99.15 95.66 86.5 12 99.63 97.96 90.03 15 99.87 99.01 96.18

Figure 4.8: A plot of number of errors v/s the confidence levels with increasing time of measurement


113

As expected, the measurement time window has to be increased for the model with

increasing number of bit errors for an acceptable level of confidence.

From the above results and modeling, we can safely arrive at the conclusion that

wireless control for WBANs could not only work perfectly well using WirelessHART

protocol, but it would also satisfy the QoS parameters pertaining to sampling frequency

and rate of error in the face of mobility or noisy channels.

4.6.5 Performance evaluation: delay and throughput

Next, we tried to evalute the performance of our 24-parameter WBAN model with

respect to the amount of transmission delay in milliseconds and the maximum

throughput obtained for the two channels of interest.

We chose two of the seven available channels in the IEEE 802.15.6 standard. We

deliberately chose the two at the extreme ends with sufficient bandwidths to

accommodate the 15 channels and may be have room for more. At the lower end of the

spectrum, we chose the 420 – 450 MHz band and at the higher frequency side, we chose

the 2400 - 2483.5 MHz channel. The first band allows for achieving greater range of

communication but over a limited number of channels. It could be used for transmission

of data from the coordinator to the base station due to its bigger transmission range. The

second band is much larger, and allows for more number of channels in a shorter range

of communication. This band would be apt for 'sensor to sensor' and 'sensor to CSS'

communication.

The 420 – 450 Mhz channel would have the advantage of a greater range of

communication and it could be used for communication between the WBAN CSS and the


114

base station. Typically such channels have lower rates of erors due to lower frequencies

of operation, hence we evaluated them for zero error performance.

The 2400 - 2483.5 MHz channel is common between 802.15.4 and 802.15.6

protocol families. A WBAN with control scheme can be implemented using this channel

while extending the performance analysis considering parameter data as well as control

data passing through the available channels. The channel offers an option to choose from

one of the four available data rates of 121.4 kbps, 240.4 kbps, 400.4 kbps and 600.4 kbps

depending on the WBAN application involved. We probed the working of all the four,

while allowing for errors in transmission.

The results pertaining to the end to end delay in packet transmission are shown in

figure 4.9 for variations in payload up to 250 bytes for the two channels. Figure 4.10

shows the maximum throughput obtained for the same packet or payload sizes.

Achievable data rates with the transmission over the 420 – 450 MHz channel for the

model are the same as that for the ideal channel. However, the throughput is

considerably lower than the theoretical maximum for the four data rates due to the bit

errors allowed in our model.

From the results, it can also be seen that with a payload of 127 bytes which is

typically used by WirelessHART, a maximum throughput of 332.3747 kbps can still be

obtained for its WBAN implementation while allowing for errors. Samples encoded in

12-bits can be transmitted for most WBAN applications except for some demanding high

rate of transmission like the EMG.


115

Figure 4.9: Packet transmission delay v/s the payload in bytes


116

Figure 4.10: Maximum throughput v/s the payload in bytes

4.7 Performance check on intrinsically safe routing models

While evaluating the performance of wireless networks for control in WBAN, it becomes

important to check on how it fares as far as the intrinsic safety involving the wireless

system with respect to human cells and tissues is concerned. The wireless energy


117

dissipates itself in the form of heat, which can cause discomfort to the subject and has

the potential of damaging human cells and tissues over prolonged exposures. As an

example, the continuous operation of a pulse oximeter beyond is considered unsafe and

could cause burns as it reaches a temperature of 43 degrees Centigrade. The harmful

effects of the radiation heating can be assessed by using Pennes bio-heat equation [100]

that provides the heat transfer relationship between the temperature of blood in the

vessels and the surrounding tissue. There are certain temperature-aware routing

algorithms [101], [102] and [103] suggested for WBANs that consider parameters like

antenna radiation and the resulting dissipation of power in the surrounding tissue. A

recent work [104] tries to combine thermal routing algorithms with an efficient MAC

protocol to generate duty cycles that allow lesser temperature rise than individual

schemes. While the results in [104] are superior to the other attempts, the work does not

focus on the energy requirements and add-on energy consumption due to

retransmissions. We extended the work and tried to evaluate the performance of the

three models in [104] when it comes to network energy when implemented for the

WirelessHART based WBAN. The models use up to 25 sensors in the human body, which

is very close to our model involving 24 sensors.

All the models study the effect of four probability distributions for network

parameters in addition to temperature rise. The first model used is a sensor-centric

Monte-Carlo (SCMC) model which allows for probability distribution based random

generation of packets and assumes fixed rise and fall in temperatures. The second,

Tissue-based fixed-coordinator (TBFC) model uses a stable solver for a stepped packet

generation involving a fixed CSS and yields a better heat performance than the first.


118

However, the packet loss is higher in this case. As an improvement on packet loss, the

third model (TBFC-1HC) is a modification on the Tissue-based fixed-coordinator model

that implements a ‘one-hop caching mechanism’, in which the data packets are cached

for any delays between their receipts by the one hop neighbor nearest to the CSS, and

transmitted to the CSS upon receipt of the clear-to-send signal.

4.7.1 Performance evaluation on traffic parameters of the model

We tried to evaluate the retransmission overhead resulting from packet loss in case of all

the distributions for the three models. TBFC fared the worst of the three models when it

came to retransmitting packets due to packets dropped, while SCMC was the best.

Among the distributions, the Poisson distribution gave us the lowest retransmission

overhead while Log-Normal had the highest figures among the four distributions. The

results for the best combination of a powerful battery and an advanced SN are plotted in

figure 4.11. The work could be extended for other distributions over a more realistic

human model.


119

Figure 4.11: Amount of retransmitted packets for the three models

We also checked the three routing models for performance in conjunction with the battery

models evaluated earlier by us in 2.3. Of the battery models evaluated, we may recall that

the model based on the 1700 mAH iXTRA battery used with ECO sensor nodes had the

best performance with respect to the network lifetime. We chose the mentioned battery

model for evaluating the impact of retransmissions on the battery and network lifetime.

The TBFC model proved to be most uneconomical on the battery power as compared to

the other two models for all the four probabilistic packet distributions. Figure 4.12 shows

our findings, apart from the details in the tables. The three models pave a way for a study

towards efficient and intrinsically safe, thermal-aware WBANs for wearable computing.


120

Figure 4.12: A comparison of reduced lifetime hours for the three models

121

Chapter 5

Future Work

5.1 Overview

The three challenges taken up and the work done in dealing with them have been discussed

in the previous chapters. The following future work could be taken up as an extension of

the work presented and discussed previously in the dissertation.

5.2 Data Acquisition System and data compression for WBANs

The work can be extended to find out the performance of the 24-parameter WBAN model

when using variable and adaptive rate of sampling for WBAN applications. Compression

and feature extraction techniques hold some promise in energy saving, as we have tried

evaluating their potential in 2.6 and 4.6 for WBAN applications.

Further benefits of saving can be achieved by resorting to differential modulation

schemes. Differential modulation schemes allow for encoding the change from the previous

parameter values in the subsequent samples. As the change itself has smaller range limits

as compared to the entire signal envelope, it is just the amount of increment or decrement

that needs to be conveyed. This change could be as simple as a single bit encoding which

conveys if the subsequent sample is greater than or less than the previous value. Such

differential modulation is known as Delta Modulation (DM). As the step size of the

approximated digital signal for the original analog is assumed to be a constant, DM typically

conveys the next sample to be merely the next higher code value or the previous lower

Chapter 5: Future Work

122

code value through a single bit. The advantage is that the standard 12-Bit encoding for

WBAN parameters could be reduced to single bit encoding after the first few sample values.

DM has the capability to compress the sample information by several orders, and could

mean a ten-fold capacity multiplication in case of WBANs. However, DM comes with an

inherent issue related to its maximum slope being limited by the step size and the number

of bits in the code per sample. The slope overload error could frequently result in a very

poor, may be non-workable solutions to approximating of the analog data this way. The

solution could lie in Adaptive Delta Modulation (ADM). Instead of changing the next

approximated value by a single step, ADM keeps incrementing or decrementing the step

size while trying to catch up with the analog signal values. This allows for much better

approximation as compared to DM, while solving the slope overload problem, resulting in

much lower errors in encoding. Continuously Variable Slope Delta Modulation, CVSDM is a

version of ADM that allows for faster changes in the subsequent variable step sizes for an

even better approximation.

The Android application could be enhanced for handling more WBAN parameters,

performing analysis on them and displaying the results. Features like data compression,

data aggregation and auto-forward of data could be added to the WBAN data acquisition.

The DAS application for the Android operating system could also be written for and

implemented in other commercial smartphone and tablet operating systems like iOS and

Windows 8.


123

5.3 Evaluation of high concentration presence of WBANs

We tried to extend this work by evaluating the performance of moving WBANs in a similar

setup. We tried to model the setup for WBANs using Shox, Castalia, OMNet++ and NS3

network simulators. However, none of the available simulators provided us with a realistic

3-D moving human-like frame that allowed the placement of sensors at different points in

the body. The work was taken up by our research group and a workable model that allows

such a configuration has been recently developed. Currently, the model is under refinement

and evaluation, and the results look positive. The deployment of the moving model in a DVC

environment could possibly be taken up at a later stage, once a mature and reliable 3-D

model is implemented in the available simulators, or is developed within our research

group.

Once such a model is available, it would also be interesting to find out the behavior of

packet traffic across WBANs and the rate of success when the WBAN concentration

increases. The study can be enhanced by adding energy consumption analysis to the

performance evaluation. Additional studies into mobility and performance parameters

with QoS can be carried out when realistic 3-D human models are available for WBANs.

Currently, none of the network simulators offer this feature.

5.3.1 Dynamic virtual coverage for WBANs

The proposed concept behind DVCs evaluated out to be an effective way of handling

handovers in SCNs, from the throughput and handover statistics. The performance analysis

of mobile WBANs is a novel area yet to attract sufficient attention and focus. The behavior


124

of such mobile WBANs while trying to utilize SCNs and other available networks to relay

their data can be studied and evaluated for performance as well as feasibility. This research

work can be performed once realistic 3-D human models WBANs are featured in the

available network simulators. Most network simulators that offer features for simulations

related to WBANs allow for SN placement either in a planar 2-D grid or uniform, ordered

placement in a 3-D space. None of the two is realistic enough.

5.3.2 Cognitive routing of critical data

Development of IPv6 and new concepts like ‘Internet of Things’ point to the emergence of

an IP core infrastructure that will form the basis of wireless networks in not too distant

future. This infrastructure is expected to offer multimedia services with varying QoS levels

and flexibility in demands to users, with mobility. Wireless applications are expected to be

bandwidth hungry. Devices running such applications will probably have the cognitive

ability to sense and select idle networks and channels from primary users to address the

scarcity of bandwidth. This would allow for a better usage of the highly underutilized

licensed spectrum. We propose to model the working of WBANs for such a futuristic

scenario in which SNs have cognitive abilities to sense and select one of the available

network services. WBANs would then be able to select and borrow appropriate radio

resources from primary licensed users for transmission of critical physiological and

biokinetic parameter data. Our proposed model would also find the optimal usage duration

of the borrowed resources, at the same time keeping the cost and interference with the

primary user’s access minimized. This would be done by implementing a classification and

ranking of available resources for borrowing, in order to have in optimal informed


125

selection. An important criterion of categorization would be activity in neighboring

channels, as the available channels with high adjacent channel interference would not be

preferred. Moreover, QoS guarantees need to be taken into account in channel selection,

because of the nature of physiological data involved. The classification and ranking would

keep the rate of data transmission, latency constraints and bandwidths of the various

active wireless technologies into account. Selection of a channel from within the unlicensed

or licensed spectrum of a network would then be based on the requesting WBAN’s

requirements. The model would need to keep an eye on dropped network packets in order

to deal with them separately, with a distinct priority. Provisions for handover to other

channels would be needed, if the primary user returns while the WBAN is using its channel

and so needs to borrow another channel. WBAN applications would also have to be

scheduled according to their ranking based on factors like natural hierarchy or temporary

criticality. The IoT framework is still evolving and so are the WBAN simulators.

5.4 Security in WBAN and IoT systems

Security is also of major concern for WSN applications in process industry. Attacks varies

from eavesdropping on transmissions including traffic analysis or disclosure of message

contents, to modification, fabrication, and interruption of the transmissions through node

capturing, routing attacks, or flooding [107]. When designing the security mechanisms,

both low-level (key establishment and trust control, secrecy and authentication, privacy,

robustness to communication denial-of-service, secure routing, resilience to node capture)


126

and high-level (secure group management, intrusion detection, secure data aggregation)

security primitives should be addressed [108].

It is reported in [109] that this issue is well addressed with existing standards-

based technologies. Breaches in security policies are almost technically impossible unless

caused by unavoidable issues such as disgruntled employees, theft of access passwords, or

bad configuration of the network. AES-128 (Advanced Encryption Standard, with 128-bit

keys and 128-bit block size) symmetric-key cryptography algorithm is used in the IEEE

802.11 and IEEE 802.15.4 standards. In WirelessHart, this algorithm is used in conjunction

with other security services such as key management (rotating keys for added security),

data authentication (for detecting modified data), data freshness service (to avoid replay

attacks) to provide a complete solution with an equivalent or higher level of security

performance than that provided by wired systems. The ZigBee protocol also defines

methods for implementing security services such as cryptographic key establishment, key

transport, frame protection, and device management [110]. A survey of WSN security can

be found in [111]. The paper lists several issues related to security in all kinds of sensor

networks. The IEEE 802.15.6 standards being newer have more constraints. As they have

to be small enough to be wearable, the manufacturers cannot pack a lot of processing

power in them. Computationally intensive authentication and encryption mechanisms

cannot be implemented in WBANs. One positive aspect of WBAN sensors is that they have a

very small range of coverage. Smart algorithms that exploit this feature could work well for

security in such systems. Security and defense against cyber-attacks are the areas that can

provide a lot of basic problems which are open to research.

Bibliography

[1] L. Atzori, A. Iera, and G. Morabito, " The Internet of Things: A survey”, The

International Journal of Computer and Telecommunications Networking archive,

vol. 54 issue 15, October, 2010, Pages 2787-2805

[2] D. Giusto, A. Iera, G. Morabito, and L. Atzori, "The Internet of Things", Springer,

2010. ISBN: 978-1-4419-1673-0.

[3] G. V. Crosby, T. Ghosh, R. Murimi, and C. A. Chin, “Wireless Body Area Networks for

Healthcare: A Survey,” International Journal of Ad Hoc, Sensor & Ubiquitous

Computing, vol 3, no. 3, 2012.

[4] https://sensormonitoring.wordpress.com/2014/03/25/do-i-need-iot-in-

healthcare/

[5] S. Movassaghi, M. Abolhasan, J. Lipman, D. Smith, and A. Jamalipour, “Wireless

Body Area Networks: A Survey,” IEEE COMMUNICATIONS SURVEYS &

TUTORIALS.

[6] K. Kwak, S. Ullah, and N. Ullah, “An overview of IEEE 802.15. 6 standard.” In 3rd

IEEE International Symposium on Applied Sciences in Biomedical and

Communication Technologies (ISABEL), 2010, pp. 1-6, 2010.

[7] http://www.who.int/mediacentre/news/releases/2003/pr27/en.

[8] S. Ullah, H. Higgin, M. A. Siddiqui, and K. S. Kwak, “A study of implanted and

wearable body sensor networks,” In 2nd KES International Conference on Agent

and Multi-Agent Systems: Technologies and Applications, pp. 464-473, Springer-

Verlag, Berlin Heidelberg, 2008.

[9] M. A. Hanson, H. C. Powell, A. T. Barth, K. Ringgenberg, B. H. Calhoun, J. H. Aylor, and

J. Lach, "Body area sensor networks: Challenges and opportunities," in Computer,

vol. 42, no. 1, pp. 58-65, 2009.

[10] IEEE P802.15 WPAN, Channel Model for Body Area Network. [Online] Available:

https://mentor.ieee.org/802.15/dcn/08/15-08-0780-09-0006-tg6-channel-

model.pdf

Bibliography

128

[11] B. Zhen, K. Takizawa, T. Aoyagi and R. Kohno, "A body surface coordinator for

implanted biosensor networks," in IEEE International Conference on

Communications, 2009, pp. 1-5, 2009.

[12] A. Saeed, M. Faezipour, M. Nourani, S. Banerjee, G. Lee, G. Gupta and L. Tamil, "A

Scalable Wireless Body Area Network for Bio-Telemetry," in Journal of

Information Processing Systems, vol.5, no.2, pp.77, June 2009.

[13] G. Anastasi, M. Conti, M. Di Francesco, and A. Passarella, "Energy conservation in

wireless sensor networks: A survey," Ad Hoc Networks, vol.7, no. 3, pp. 537-568,

2009.

[14] D. P. Agrawal and Q. A. Zeng, Introduction to Wireless and Mobile Systems,

textbook, 436 pages, 2003.

[15] A. Manjeshwar and D. P. Agrawal, “TEEN: A Routing Protocol for Enhanced

Efficiency in Wireless Sensor Networks,” in Proceedings of the 15th International

Parallel & Distributed Processing Symposium, pp. 2009–2015, San Francisco, Calif,

USA, April 2001.

[16] D. Chu, A. Deshpande, J. M. Hellerstein, and W. Hong, “Approximate data collection

in sensor networks using probabilistic models,” in Proc. 22nd International

Conference on Data Engineering (ICDE06), pp. 48, Atlanta, GA, April 3–8, 2006.

[17] A. Jain, E. Y. Chang, and Y. F. Wang, “Adaptive stream resource management using

Kalman filters,” in Proc. ACM International Conference on Management of Data

(SIGMOD2004), pp. 11–22, Paris, France, June 13–18, 2004.

[18] D. Tulone, and S. Madden, “An energy-efficient querying framework in sensor

networks for detecting node similarities,” in Proc. 9th International ACM

Symposium on Modeling, Analysis and Simulation of Wireless and Mobile Systems

(MSWIM06), pp. 291–300, October 2006.

[19] I. Lazaridis, and S. Mehrotra, "Capturing sensor-generated time series with quality

guarantees," in Proceedings, 19th International Conference on Data Engineering,

pp. 429-440, 2003.

[20] Y. A. Le Borgne, S. Santini, and G. Bontempi, "Adaptive model selection for time

series prediction in wireless sensor networks," Signal Processing, vol. 87, no. 12,

pp. 3010-3020, 2007.

Bibliography

129

[21] F. Xia, Z. Xu, L. Yao, W. Sun, and M. Li, "Prediction-Based Data Transmission for

Energy Conservation in Wireless Body Sensors," in The 5th Annual ICST Wireless

Internet Conference (WICON), pp. 1-9, 1-3 March 2010.

[22] A. Ray and D. De, “Data Aggregation Techniques in Wireless Sensor Network: A

Survey,” International Journal of Engineering Innovation and Research, Vol. 1, No.

2, March 2012, pp. 20-31.

[23] L. Harn-Jier and G. Campbell. "Using DQRAP (distributed queuing random access

protocol) for local wireless communications," In Proceedings of Wireless' 93,

1993, pp. 625-635.

[24] J. Alonso-Zarate, C. Verikoukis, E. Kartsakli, A. Cateura, and L. Alonso. "A near-

optimum cross-layered distributed queuing protocol for wireless LAN," Wireless

Communications, IEEE 15, no. 1, 2008, pp. 48-55.

[25] B. Otal, L. Alonso, and C. Verikoukis, “Novel QoS Scheduling and Energy-saving MAC

protocol for Body Sensor Networks Optimization,” BodyNets '08 Proceedings of

the ICST 3rd international conference on Body area networks, Article 27, March

2008.

[26] ISA: http://www.isa.org/filestore/ISASP100_14_CFP_14Jul06_Final(2).pdf

[27] DUST NETWORKS:

http://www.dustnetworks.com/cms/sites/default/files/SWP_Industrial_Environ

ments.pdf

[28] P. Jiang, H, Ren, L. Zhang, Z. Wang, and A. Xue, “Reliable application of wireless

sensor networks in industrial process control,” Proceedings of the 6th World

Congress on Intelligent Control and Automation, June 2006.

[29] A. Bemporad, S. D. Cairano, E. Henriksson, and K. H. Johansson, “Hybrid model

predictive control based on wireless sensor feedback: an experimental study,”

Proc. IEEE Conference on Decision and Control, pp. 5062-5067, December 2007.

[30] CISCO:

http://www.cisco.com/web/strategy/docs/manufacturing/swpIIWS_Emerson_w

p.pdf

Bibliography

130

[31] J. Song, A. K. Mok, D. Chen, and M. Nixon, “Challenges of wireless control in process

industry,” Workshop on Research Directions for Security and Networking in

Critical Real-Time and Embedded Systems, April 2006.

[32] T. Banerjee, B. Xie, and D. P. Agrawal, “Achieving fault tolerance in data aggregation

in wireless sensor networks,” Global Telecommunications Conference, pp. 926-

930, 2007.

[33] P. Cook, “Principles for Designing Computer Music Controllers.” Proceedings of the

Conference on New Interfaces for Musical Expression, 2001

[34] M. Helmuth, J. Merkowitz, B. McKinney, K. Shiota, C. Stark, “Center for Computer

Music in Cincinnati in 2007: (ccm)2”, Proceedings of the International Computer

Music Association, 2007.

[35] A. Mostafa, H.Y. Jun, D.P. Agrawal, M. Helmuth, “Dancing with the Motes,” Fifth IEEE

International Conference on Mobile Ad-hoc and Sensor Systems, (IEEE MASS

2008), pp. 538-540. September 29-October 2, 2008, Atlanta, GA.

[36] A. Mishra, S. Chakraborty, H. Li, and D. P. Agrawal, "Error Minimization and Energy

Conservation by predicting data in Wireless Body Sensor Networks using Artificial

Neural Network and Analysis of Error", CCNC-2014, Las Vegas, NV, USA, Jan 10-13,

2014.

[37] MATLAB, version (R2012). Natick, Massachusetts: The MathWorks Inc., 2012.

[38] M. T. Hagan and M. B. Menhaj, “Training feedforward networks with the Marquardt

algorithm,” in IEEE Transactions on Neural Networks, vol. 5, no. 6, pp. 989 – 993,

November 1994.

[39] J. Gosling, “The Java Programming Language”,

http://docs.oracle.com/javase/specs/#237601

[40] W. Heinzelman, “Application specific Protocol Architectures for Wireless

Networks”, PhD Thesis, Massachusetts Institute of Technology, 2010

[41] C. Park, J. Liu, and P. H. Chou, "ECO: an ultra-compact low-power wireless sensor

node for real-time motion monitoring", Conference on Information Processing in

Sensor Networks, 2005. IPSN 2005. Fourth International Symposium on, Pages

398-403, April 2005.

[42] http://www.ti.com/product/CC3100/datasheet

http://docs.oracle.com/javase/specs/#237601

Bibliography

131

[43] The IEEE 802.15.6 standards

[44] E. Dishman, “Inventing wellness systems for aging in place,” Computer, vol. 37, no.

5, pp. 34-41, May, 2004.

[45] N. Jain and D. P. Agrawal, “Current Trends in Wireless Sensor Network Design,”

International Journal of Distributed Sensor Networks, vol. 1, pp. 101–122, 2005.

[46] J. Yick, B. Mukherjee, and D. Ghosal, “Wireless Sensor Network Survey,” in Computer

Networks, vol. 52, issue 12, pp. 2292-2330, 22 August 2008.

[47] M. A. Hanson, H. C. Powell, A. T. Barth, K. Ringgenberg, B. H. Calhoun, J. H. Aylor,

and J. Lach, “Body area sensor networks: Challenges and opportunities,” in

Computer, vol. 42, no. 1, pp. 58-65, 2009.

[48] M. Wagner, B. Kuch, C. Cabrera, P. Enoksson, A. Sieber, “Android based Body Area

Network for the evaluation of medical parameters,” Proceedings of the Tenth

Workshop on Intelligent Solutions in Embedded Systems (WISES), Klagenfurt, pp.

33 – 38, 5-6 July 2012.

[49] O. O. Ogunduyile, O. O. Oludayo, and M. Lall, “Healthcare Monitoring System Using

a Collection of Sensor Nodes,” International Journal of Emerging Technology and

Advanced Engineering, ISSN 2250-2459, vol. 3, no. 2, pp. 632-639, February 2013.

[50] N. R. Baviskar and A. Shinde, “Android Smartphone Based Body Area Network for

the Evaluation of Medical Parameters in Real Time,” International Journal of

Electrical, Electronics and Data Communication, ISSN: 2320-2084, vol. 2, no. 4, pp.

66-69, April 2014.

[51] A. Bourouis, M. Feham, and A. Bouchachia, “Ubiquitous Mobile Health Monitoring

System for Elderly (UMHMSE),” International Journal of Computer Science &

Information Technology (IJCSIT), vol. 3, no. 3, pp. 74-82, June 2011.

[52] Bluetooth Special Interest Group, Bluetooth Core Specification plus Enhanced data

Rate, Specification of the Bluetooth system, 1.2, 2003

[53] Wi Fi standards: http://standards.ieee.org/about/get/802/802.11.html

[54] D. P. Agrawal and Q. A. Zeng, Introduction to Wireless and Mobile Systems,

textbook, 436 pages, 2003.

Bibliography

132

[55] A. Jamthe, A. Mishra, and D. P. Agrawal, “Scheduling schemes for Interference

Suppression in Healthcare Sensor Networks,” ICC-2014, Sydney, Australia, 10-14

June 2014.

[56] A. Mishra, S. Chakraborty, H. Li, and D. P. Agrawal, “Error Minimization and Energy

Conservation by predicting data in Wireless Body Sensor Networks using Artificial

Neural Network and Analysis of Error,” CCNC-2014, Las Vegas, NV, USA, Jan 10-13,

2014.

[57] D. P. Agrawal and A. Mishra, “Designing Wireless Sensor Networks: from Theory to

Applications,” WCSN 2011, Seventh IEEE Conference on Wireless Communication

and Sensor Networks, Dec 5-9, 2011, Panna, India.

[58] The Arduino Microcontroller board:

https://www.arduino.cc/en/guide/introduction

[59] A. Jamthe, S. Chakraborty, S. K Ghosh, and D. P. Agrawal, “An implementation of

Wireless Sensor Network in monitoring of Parkinson's Patients using RSSI,” PWSN

2013, 5th International Workshop on Performance Control in Wireless Sensor

Networks, May 23, 2013, Cambridge, Massachusetts, in conjunction with the 9th

IEEE International Conference on Distributed Computing in Sensor Systems

(DCOSS '13).

[60] E. J. Kim, S. Youm, T. Shon, and C. H. Kang, “Asynchronous internetwork interference

avoidance,” The Journal of Supercomputing for wireless body area networks,

August 2013, Volume 65, Issue 2, pp 562-579.

[61] IEEE 802.15 task group 6. [Online]: http://www.ieee802.org/15/pub/TG6.html

and http://standards.ieee.org/findstds/standard/802.15.6-2012.html

[62] ZigBee Tutorial. [Online] Available: http://www.ifn.et.tudresden.de/~marandin/

ZigBee/ZigBeeTutorial.html.

[63] A. Jamthe, A. Mishra, and D. P. Agrawal, "Scheduling schemes for Interference

Suppression in Healthcare Sensor Networks", ICC-2014, Sydney, Australia, 10-14

June 2014.

[64] J. M. Mendel, "Fuzzy logic systems for engineering: a tutorial," Proceedings of the

IEEE 83, no. 3, 1995, pp. 345-377.

Bibliography

133

[65] J. Yoon, G. S. Ahn, S. S. Joo, and M. Lee, "Pnp-mac: Preemptive slot allocation and

non-preemptive transmission for providing qos in body area networks," in Proc. of

Consumer Communications and Networking Conference, Jan.2010, pp.1-5.

[66] A. Zhang, D. B.Smith, D. Miniutti, L. F.Hanlen," Performance of Piconet Co-Existence

Schemes in Wireless Body Area Networks," Wireless Communications and

Networking Conference (WCNC), 2010 IEEE, 18-21 April 2010, Sydney, NSW.

[67] B. Otal, L. Alonso, and C. Verikoukis, “Novel QoS Scheduling and Energy-saving MAC

protocol for Body Sensor Networks Optimization,” BodyNets '08 Proceedings of

the ICST 3rd international conference on Body area networks, Article 27, March

2008.

[68] T. H. Cormen, C. E. Leiserson, and R. L. Rivest, “Introduction to Algorithms”, The

MIT Press, Cambridge, MA, 1989

[69] M. Raju, T. Oliveira, D. P. Agrawal, “A practical distance estimator through

distributed RSSI/LQI processing - An experimental study”, ICC 2012, 6575-6579

[70] P. Charriere, J. Brouet, and V. Kumar, “Optimum channel selection strategies for

mobility management in high traffic tdma-based networks with distributed

coverage,” in Personal Wireless Communications, 1997 IEEE International

Conference on, dec 1997, pp. 167–172.

[71] J. Hoydis, M. Kobayashi, and M. Debbah, “Green small-cell networks,” Vehicular

Technology Magazine, IEEE, vol. 6, no. 1, pp. 37–43, march 2011.

[72] C.L. Tan, K. M. Lye, and S. Pink, “A fast handoff scheme for wireless networks,” in

WOWMOM, 1999, pp. 83–90.

[73] Qualnet Network Simulator 5.0, Scalable Networking Technologies

[74] N. Weragama, J. H. Jun, A. Mishra and D. P. Agrawal, “Simulation of Mobility Aware

Dynamic Virtual Cells Utilizing Multiple Multicast Trees”, IEEE ComSoc TCSIM

Quarterly Newsletter, vol. 15, pp. 2-4, Dec 2012.

[75 ] J. Lessmann, T. Heimfarth, P. Janacik, “ShoX, An Easy to Use Simulation Platform for

Wireless Networks", Tenth International Conference on Computer Modeling and

Simulation, 2008

[76] M. Helmuth, J. H. Jun, T. Oliveira, J. B. Merkowitz, A. Mishra, A. Mostafa, D. P.

Agrawal, “Wireless Sensor Networks and Computer Music, Dance and Installation

Bibliography

134

Implementations,” International Computer Music Conference 2010, New York,

USA, June 1-5, 2010.

[77] M. Helmuth, R. Danard, J. H. Jun, T. Oliveira, A. Mishra, and D. P. Agrawal, “Water

Birds: Compositional Collaboration with Clarinets, Wireless Sensors, and RTcmix,”

SEAMUS 2011, 26th Annual Conference of the Society for Electro-Acoustic Music

in the United States, January 20–22, 2011, Miami, Florida.

[78] TinyOS 2.x, http://www.tinyos.net/

[79] WirelessHART: http://en.hartcomm.org/main_article/wirelesshart.html

[80] ISA 100 Fieldbus Standard: http://www.fieldbus.org/index.php

[81] EMERSON: http://www.EmersonProcess.com/SmartWireless

[82] Honeywell:

http://hpsweb.honeywell.com/Cultures/en-US/Products/Wireless/Solutions/default.htm

[83] J. Song, A. K. Mok, D. Chen, and M. Nixon, “Challenges of wireless control in process

industry,” Workshop on Research Directions for Security and Networking in

Critical Real-Time and Embedded Systems, April 2006.

[84] T. Banerjee, B. Xie, and D. P. Agrawal, “Achieving fault tolerance in data aggregation

in wireless sensor networks,” Global Telecommunications Conference, pp. 926-

930, 2007. http://dx.doi.org/10.1109/GLOCOM.2007.178

[85] M.A.M. Vieira, C.N., Jr. Coelho, D.C., Jr. da Silva, J.M. da Mata, “Survey on wireless

sensor network devices,” Proc. Emerging Technologies and Factory Automation,

pp. 537-544, September 2003. http://dx.doi.org/10.1109/ETFA.2003.1247753

[86] V. Raghunathan, S. Ganeriwal, and M. Srivastava, “Emerging techniques for long

lived wireless sensor networks,” IEEE Communications Magazine, vol. 44, no. 4,

pp. 108-114, 2006. http://dx.doi.org/10.1109/MCOM.2006.1632657

[87] M. Hanssmann, S. Rhee, and S. Liu, “No wiring constraints,” IEEE Industry

Applications Magazine, pp. 60-65, August 2009.

http://dx.doi.org/10.1109/MIAS.2009.932593

Bibliography

135

[88] J. Le, J. C. S. Lui, and D. Chiu, “DCAR: Distributed coding-aware routing in wireless

networks,” IEEE Transactions on Mobile Computing, vol. 9, no. 4, pp. 596-608,

April 2010. http://dx.doi.org/10.1109/TMC.2009.160

[89] S. Katti, H. Rahul, W. Hu, D. Katabi, M. Medard and J. Crowcroft, “XORs in the Air:

Practical wireless network coding,” IEEE/ACM Transactions on Networking, vol.

16 , no. 3, pp. 497-510, June 2008. http://dx.doi.org/10.1109/TNET.2008.923722

[90] B. Lu, T. G. Habetler, and R. G. Harley, “A novel motor energy monitoring scheme

using wireless sensor networks,” IEEE Industry Applications Conference, October

2006. http://dx.doi.org/10.1109/IAS.2006.256844

[91] H. Liu, P. Tang, and Z. Erdun, “An energy-balanced variable range transmission

scheme in wireless sensor networks,” International Conference on Wireless

Communications, Networking and Mobile Computing, pp. 1-4, 2009.

[92] J. Shin, M. Chin, and C. Kim, “Optimal transmission range for topology management

in wireless sensor networks,” Information Networking, Advances in Data

Communications and Wireless Networks, vol. 3961, pp.177-185, 2006.

[93] K. Shin, J. Song, J. Kim, M. Yu, and P. S. Mah, “REAR: Reliable energy aware routing

protocol for wireless sensor networks,” International Conference on Advanced

Communication Technology, pp. 525-530, 2007.

http://dx.doi.org/10.1109/ICACT.2007.358410

[94] A.H. Mohajerzadeh, M.H. Yaghmaee, “An efficient energy aware routing protocol

for real time traffic in wireless sensor networks,” International Conference on

Ultra Modern Telecommunications & Workshops, pp. 1-9, 2009.

http://dx.doi.org/10.1109/ICUMT.2009.5345536

[95] D. Chen, WirelessHART: Real-Time Mesh Network for Industrial Automation, 3 DOI

10.1007/978-1-4419-6047-4_1, © Springer Science+Business Media, LLC 2010

[96] IEEE Std. 802.15.4, Wireless Medium Access Control (MAC) and Physical Layer

(PHY) Specifications for Low-Rate Wireless Personal Area Networks (WPANs),

IEEE, New York, N.Y., 2006.

[97] http://www.ni.com/labview/

[98] http://www.parkinson.org/Parkinson-s-Disease/Living-Well/Safety-at-

Home/Freezing

Bibliography

136

[99] http://www.mayoclinic.org/diseases-conditions/foot-drop/basics/definition/con-

20032918

[100] H.H Pennes, “Analysis of Tissue and Arterial Blood Temperatures in the resting

human forearm,” Journal of Applied Physiology, Vol. 1, pp. 93-121, 1948.

[101] D. Takahashi, Yang Xiao, and Fei Hu. “LTRT: Least Total-Route Temperature

routing for Embedded Biomedical Sensor Networks”, IEEE Global

Telecommunications Conference, 2007 (GLOBECOM ’07), pp. 641–645, Nov 2007.

[102] M. Tabandeh, M. Jahed, F. Ahourai, and S. Moradi. “a thermal-aware shortest hop

routing algorithm for in vivo biomedical sensor networks”. In Information

Technology: New Generations, 2009. ITNG ’09. Sixth International Conference on ,

pages 1612–1613, April 2009.

[103] Q. Tang, N. Tummala, S. K. S. Gupta, and L. Schwiebert. “TARA: thermal-aware

routing algorithm for implanted sensor networks”, vol. 3560, pp. 206–217, June-

July 2005

[104] A. G. Krishnamurthy, J. H. Jun, and D. P. Agrawal, "Gradient based Temperature-

aware routing in Body Area Sensor Networks", 9th International Conference on

Body Area Networks, {BODYNETS} 2014, London, Great Britain, September 29 -

October 1, 2014

[105] P. Smets and R. Kennes, “The transferable belief model,” Artificial Intelligence, vol.

66, no. 2, pp. 191–234, 1994.

[106] L. A. Zadeh, “Fuzzy logic and approximate reasoning,” Synthese, vol. 30, no. 3-4, pp.

407–428, 1975.

[107] V. C. Gungor and G. P. Hancke, “Industrial wireless sensor networks: Challenges,

design principles, and technical approaches,” IEEE Trans. Industrial Electronics,

vol. 56, no. 10, pp. 4285-4265, October 2009.

http://dx.doi.org/10.1109/TIE.2009.2015754

[108] A. Perrig, J. Stankovich, and D. Wagner, “Security in wireless sensor networks,”

Communication of the ACM, vol. 47, no. 6, pp. 53-57, June 2004.

http://dx.doi.org/10.1145/990680.990707

Bibliography

137

[109] CISCO:

http://www.cisco.com/web/strategy/docs/manufacturing/swpIIWS_Emerson_wp.pdf

[110] Idaho National Laboratory: http://csrp.inl.gov/Documents/Securing ZigBee

Wireless Networks in Process Control System Environments.pdf

[111] J. P. Walters, Z. Liang, W. Shi, and V. Chaudhary, “Wireless sensor network security:

A survey,” in Distributed, Grid, and Pervasive Computing, Yang Xiao Eds. Auerbach

Publications, 2006.

138

APPENDIX

Additional Results and Figures from Chapter 2

Figure 2a.1 EEG signal from a healthy person

Figure 2a.2 Usage of dataset in training of ECG lead-II signal

139

Figure 2a.3 Error Autocorrelation plot for ANN-NLR training of ECG lead-II signal

Figure 2a.4 Regression plot for ANN-NLR training of ECG lead-II signal

140

Figure 2a.5 Input Error Correlation plot for ANN-NLR training of ECG lead-II signal

Figure 2a.6 Training State plot for ANN-NLR training of ECG lead-II signal

141

Figure 2a.7 Training Performance plot for ANN-NLR training of ECG lead-II signal

Figure 2a.8 Time series response for ANN-NLR training of ECG lead-II signal

142

Fig

ure

2a.9

Err

or

His

togra

m f

or

AN

N-N

LR

tra

inin

g o

f E

CG

le

ad-I

I sig

na

l

143

Fig

ure

2a

.10 P

erf

orm

ance

co

mp

ariso

ns o

f G

ola

y a

nd

Me

dia

n filt

ers

art

ifa

ct filte

ring

in

EC

G le

ad-I

I sig

na

l

144

Fig

ure

2a.1

1 T

rain

ing r

esp

on

se f

or

AN

N-N

LR

tra

inin

g o

f E

CG

le

ad-I

I sig

na

l a

fter

Me

dia

n F

ilte

ring

145

Fig

ure

2a.1

2 P

redic

tio

n p

lots

for

art

eria

l p

ressu

re fro

m P

ID a

nd N

LR

-AN

N

alg

orith

ms

146

Fig

ure

2a.1

3 E

rro

r p

lots

fo

r art

eria

l p

ressure

fro

m P

ID a

nd

NLR

-AN

N a

lgorith

ms f

or

AR

T

147

Fig

ure

2a.1

4 E

rro

r p

lots

fo

r art

eria

l p

ressure

fro

m P

ID a

nd

NLR

-AN

N a

lgorith

ms f

or

CV

P

148

Fig

ure

2a.1

5 E

rro

r p

lots

fo

r art

eria

l p

ressure

fro

m P

ID a

nd

NLR

-AN

N a

lgorith

ms f

or

CV

P