Wireless Body Area Networks Technology, Implementation, and Applications PDF

Wireless.
indd 1 10/19/11 12:53:03 AM

This page intentionally left blank
Technology, Implementation, and Applications
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2012 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works

Version Date: 20111202
International Standard Book Number-13: 978-9-81424-157-1 (eBook - PDF)
This book contains information obtained from authentic and highly regarded sources. Reason-
able efforts have been made to publish reliable data and information, but the author and publisher
cannot assume responsibility for the validity of all materials or the consequences of their use. The
authors and publishers have attempted to trace the copyright holders of all material reproduced in
this publication and apologize to copyright holders if permission to publish in this form has not
been obtained. If any copyright material has not been acknowledged please write and let us know so
we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information
storage or retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, please access www.
copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc.
(CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organiza-
tion that provides licenses and registration for a variety of users. For organizations that have been
granted a photocopy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and
are used only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
October 19, 2011 15:50 PSP Book - 9in x 6in 00-Mehmet-prelims
Contents
Preface xvii
1 Introduction to Wireless Body Area Network 1

Mehmet Rasit Yuce and Jamil Y. Khan
1.1 Introduction 1
1.2 Applications 4
1.3 Wireless Personal Area Network (WPAN)/Wireless
Local Area Network (WLAN) 6
1.4 Wireless Body Area Network 8
1.5 Design Requirements 12
1.6 Scope of the Book 15
2 Wireless Patient Monitoring in a Clinical Setting 19

Esteban J. Pino, Dorothy Curtis, Tom O. Stair,
John V. Guttag, and Lucila Ohno-Machado
2.1 Introduction 20
2.2 Smart System 23
2.2.1 Architecture 23
2.2.1.1 Hardware 23
2.2.1.2 Software 26
2.2.2 Clinical Implementation 28
2.3 Results 32
2.3.1 Medical Usefulness 32
2.3.2 User Acceptance 34
2.4 Conclusion 36
3 Real-Time Cardiac Arrhythmias Monitoring for

Pervasive Health Care 41
Zhou Haiying, Hou Kun-Mean, de Vaulx Christophe,
and Li Jian
3.1 Introduction 42
vi Contents
3.2 History of PHC Research 44

3.3 Overview of PCC System 47
3.3.1 PCC System Architecture 47
3.3.1.1 Wireless ECG sensor 48
3.3.1.2 Local access server 49
3.3.1.3 Remote access server 50
3.3.1.4 Remote surveillance server 50
3.3.2 PCC Operation Modes 51
3.4 Key Technologies of PCC System 52
3.4.1 Lossless ECG Signal Compression 53
3.4.2 Adaptive Communication Mechanism 54
3.4.2.1 PCC data frame 54
3.4.2.2 PCC communication mechanisms 56
3.4.3 AED Algorithm 58
3.4.3.1 Signal preprocessing and conditioning 59
3.4.3.2 QRS complex detection 62
3.4.3.3 AED performance analysis 65
3.5 Conclusion 66
4 Human Bio-Kinematic Monitoring with Body

Area Networks 75
Roozbeh Jafari, Hassan Ghasemzadeh, Eric Guenterberg,
Vitali Loseu, and Sarah Ostadabas
4.1 Physical Movement Monitoring 76
4.2 Applications 77
4.2.1 Medical Applications 77
4.2.1.1 Gait analysis 78
4.2.1.2 Parkinson’s disease assessment
systems 79
4.2.2 Sports Training Application 80
4.2.2.1 Golf swing training 81
4.2.2.2 Baseball swing training 82
4.3 Hardware and Software Architecture 83
4.4 Signal Processing for Body Area Networks 85
4.5 An Automatic Parameter Extraction Method Based
on HMM 87
4.5.1 HMEM Training and Use 88
4.5.2 Overview 89
Contents vii
4.5.2.1 Preprocessing and feature extraction 90

4.5.2.2 HMM training 90
4.5.2.3 Parametrization and feature selection 90
4.5.3 HMM Training and the Viterbi Algorithm 91
4.5.4 Feature Selection and Model Parametrization
Using Genetic Algorithms 92
4.5.5 HMEM Application Procedure 93
4.5.6 Experimental Analysis 93
4.5.6.1 Examination of per-subject error 94
4.6 System Optimizations 95
4.6.1 Burst Communication 96
4.6.1.1 Task graph 96
4.6.1.2 Problem formulation 98
4.6.1.3 Experimental results 100
5 Signal Processing In-Node Frameworks for Wireless

Body Area Networks: From Low-Level to High-Level
Approaches 107
Francesco Aiello, Giancarlo Fortino, Stefano Galzarano,
Raffaele Gravina, and Antonio Guerrieri
5.1 Introduction 108
5.2 A WBAN Reference Architecture 110
5.3 Software Frameworks for Programming WBANs 111
5.4 Agent-Oriented Platforms for Wireless Sensor
Networks 116
5.5 An Agent-Oriented Design of Signal Processing
In-Node Environments 119
5.6 An Analysis of Agent-Oriented Implementations of
In-Node Signal Processors 123
5.6.1 MAPS-Based and AFME-Based Implementation
of Sensor Agents 125
5.6.2 Agent Implementation Comparison 128
5.7 Conclusions and Future Work 132
6 Hardware Development and Systems for Wireless

Body Area Networks 137
Mehmet Rasit Yuce
viii Contents
6.2 Wireless Body Sensors 138

6.2.1 Sensor Nodes and Hardware Designs 139
6.2.2 Wireless Systems and Platforms 146
6.2.2.1 Wireless transceivers and
microcontrollers 149
6.2.2.2 Existing sensor boards 152
6.2.3 Design of Implanted Sensors Nodes for WBAN 157
6.3 WBAN Systems 162
6.4 A WBAN-Based Multi-Patient Monitoring System 170
6.4.1 Software Programs and Monitoring 175
6.5 Conclusion 179
Appendix 180
7 Wireless Body Area Network Implementations for
Ambulatory Health Monitoring 185
Reza Naima and John Canny
7.1 Design Process 185
7.2 Existing WBAN Implementations 187
7.2.1 Hardware Paradigms 188
7.2.2 Firmware 189
7.2.3 The Data 190
7.3 Signal Acquisition 191
7.3.1 Frequency Bandwidth of Interest 192
7.3.2 Measuring Surface Biopotentials 193
7.3.2.1 The electrode 193
7.3.2.2 Filtering 195
7.3.2.3 Amplifier 195
7.3.2.4 Analog Digital Converter and
Microcontroller 197
7.3.3 Electrocardiograph 197
7.3.3.1 Berkeley Tricorder 201
7.3.4 Electromyogram 201
7.3.4.1 The Berkeley Tricorder 203
7.3.5 Pulse Oximetry (SpO2 ) 203
7.3.6 Respiration 209
7.3.7 Accelerometry 210
Contents ix
7.4 Wireless Interface 211

7.4.1 The Berkeley Tricorder 211
7.4.2 Power Consumption 212
7.4.3 Data Range and Transmit Power 213
7.4.4 Data Rate 216
7.4.5 Human Safety 216
7.4.6 Security 217
7.5 Batteries 218
7.6 Final Thoughts and the Berkeley Tricorder 222
8 Ambulatory Recording of Biopotential Signals:
Constraints and Challenges for Analog Design 229
Refet Firat Yazicioglu, Sunyoung Kim, Tom Torfs,
Julien Penders, Buxi Singh Dilpreet, Inaki Romero,
and Chris Van Hoof
8.1 Introduction: The Need for Portable Medical
Electronics Systems 230
8.2 Basics of Biopotential Signal Acquisition 233
8.3 Constrains and Challanges 235
8.4 Design of Instrumentation Amplifiers for
Biopotential Recordings 239
8.4.1 Uncompensated Instrumentation Amplifiers 239
8.4.2 Compensated Instrumentation Amplifiers 242
8.4.3 Summary and Comparison of Instrumentation
Amplifier Topologies 247
8.5 Signal Integrity Problems in Ambulatory
Measurements 250
8.5.1 Methods Focusing on Motion Artifact
Reduction in Biopotential Recordings 250
8.5.2 Readout Circuits for Adaptive Filtering 251
8.6 Conclusion 255
9 Network and Medium Access Control Protocol
Design for Wireless Body Area Networks 259
Jamil Y. Khan
9.2 Network Topologies and Configurations 262
9.3 Basics of Medium Access Control Protocols 264
9.3.1 WBAN Traffic Characteristics 267
x Contents
9.4 Scheduled Protocols 268

9.4.1 TDMA Protocol 269
9.4.2 Polling Protocol 270
9.5 Random Access Protocols 272
9.6 Hybrid MAC Protocol 275
9.7 Energy Management in a WBAN 276
9.8 Patient Monitoring Network Design 283
9.8.1 Transmission Capacity Requirements 284
9.8.2 PHY and MAC Layer Parameter Selection 285
9.8.3 Network Configuration 286
9.9 Performance Analysis of a WBAN 288
9.10 Conclusions 291
10 Power Management in Body Area Networks for

Health Care Applications 295
Vijay Sivaraman, Ashay Dhamdhere, and Alison Burdett
10.2 Related Work 299
10.3 The Case for Transmit Power Control in Body
Area Networks 301
10.3.1 Normal Walk 303
10.3.2 Slow Walk 304
10.3.3 Resting 305
10.4 Optimal Off-Line Transmit Power Control 305
10.5 Practical On-Line Transmit Power Control 307
10.5.1 A Simple and Flexible Class of Schemes 308
10.5.2 Example Adaptations of the General
Scheme 310
10.5.3 Tuning the Parameters 312
10.6 Prototyping and Experimentation 314
10.6.1 MicaZ Mote Platform 314
10.6.2 Toumaz Sensium TM Platform 316
10.7 Conclusions and Future Work 318
11 Channel Modeling of Narrowband Body-Centric

Wireless Communication Systems 323
Simon L. Cotton and William G. Scanlon
11.1 Introduction to Body-Centric Communications 324
Contents xi
11.2 Channel Modeling for Wireless Body Area

Networks 326
11.2.1 Statistical Distribution of the Fading
Signal in WBANs 329
11.2.1.1 Rayleigh and Rice distributions 330
11.2.1.2 Nakagami distribution 331
11.2.1.3 Weibull distribution 333
11.2.1.4 Lognormal distribution 333
11.2.2 Higher Order Statistics 334
11.2.2.1 Level crossing rate and average
fade duration 334
11.3 Parameter Estimation and Model Selection 335
11.3.1 Maximum Likelihood Estimation 335
11.3.2 Akaike Information Criterion 337
11.3.3 Worked Example 338
11.3.3.1 Model selection 338
11.3.3.2 Level crossing rate 342
11.3.3.3 Simulation of the received
signal envelope 343
12 Antenna Design and Propagation for WBAN

Applications 349
Tharaka Dissanayake
12.1.1 Antenna Gain 350
12.1.2 Return Loss 351
12.1.3 Efficiency 351
12.1.4 Reciprocity 352
12.2 Miniaturized Antennas 353
12.2.1 Planar Inverted-F Antennas 353
12.2.2 Planar Monopoles and Dipoles 355
12.2.3 Planar Slot Antennas 357
12.3 Implanted Antennas 357
12.3.1 Dielectric Loaded Matching of Implanted
Antennas 359
12.3.1.1 Biocompatibility of dielectric
loaded antenna 361
xii Contents
12.4 Volume Conduction Antennas 363

12.5 Summary 364
Appendix A 365
A.1 Function calculating the reflection coefficient 365
A.2 X111 and X514 366
A.3 Function calculating K vectors 366
Appendix B Calculating Frequency-Dependent
Tissue Properties 367
A.4 Cole–Cole function 367
A.5 Calculating the properties 368
A.6 Function for optimization 370
13 Coexistence Issues with Wireless Body

Area Networks 375
Axel Sikora
13.2 Analysis of Interferers 376
13.2.1 Classification 376
13.2.2 Regulation Issues 377
13.2.3 Intrinsic Interference 378
13.2.4 Extrinsic Interference of RF-Stations
within the Same Frequency Band 379
13.2.5 Extrinsic Interference of Other Systems
within the Same Frequency Band 380
13.3 Effect on Transmission 381
13.3.1 Fundamentals 381
13.3.2 Simulation of a Dense Sensor Network
(Intrinsic Interference) 382
13.3.3 Measurement of Real Packet Losses due to
Extrinsic Interference 384
13.3.4 Effects of Coexistence Problems 385
13.4 Countermeasures — An Overview 387
13.4.1 Safety Aspects 387
13.4.2 Classification 387
13.5 Countermeasures on Physical Layer 388
13.5.1 Channel Classification and Selection 388
13.5.2 Frequency Hopping 391
Contents xiii
13.5.3 Frequency Spreading and Code Division

Multiple Access 394
13.5.4 The Promise of Ultra-Wide-Band 395
13.6 Countermeasures on Data Link Layer 396
13.6.1 Basic Medium Access Control 396
13.6.2 Centralized Approach 397
13.6.3 Duty Cycle Management 398
13.6.4 Channel Sensing Methods 398
13.6.5 Persistency and Collision Avoidance 399
13.6.6 Medium Reservation Methods 404
14 Implanted Wireless Communication Making a

Real Difference 411
Henry Higgins
14.2 Why In-body Communication? 412
14.3 Applications 412
14.4 MICS and ISM Bands 412
14.5 RFID Techniques 413
14.6 Propagation Through the Body, Changes in Body
Shape and Posture 414
14.7 Antennas 415
14.7.1 Use of Smith Chart in Coupling Network
Design 417
14.7.2 Design of Antenna Coupling Networks 419
14.7.2.1 Design example 1 420
14.7.2.2 Design example 2 (SAW filter) 420
14.7.2.3 Use of simulation for antennas
and design of coupling networks 426
14.7.3 Physical Body Simulator 427
14.7.4 Body Simulator Measurements and Sample
Results 427
14.7.5 The Role of Automatic Antenna Tuning 428
14.8 Implant Power Constraints and Battery
Considerations 429
14.9 Error Correction 429
14.10 RF Circuit Hardware Options 430
xiv Contents
14.11 Base Station 431

14.11.1 Link Budget 432
14.12 Environment 434
14.13 Manufacture 434
15 Wireless Power and Data Telemetry for Wearable

and Implantable Electronics 439
Zhi Yang, Yu Han, Linh Hoang, YiKai Lo, Kuanfu Chen,
Jian Lao, Mingcui Zhou, and Wentai Liu
15.2 Power Telemetry 441
15.2.1 Mega-Hz and Sub-Mega-Hz Power 443
15.2.2 Inductor Q Boosting 447
15.2.3 Rectifiers and Regulators 449
15.3 Data Telemetry 450
15.4 Design Example 455
Appendix 457
A.1 Equivalent AC Resistance 457
A.2 Coil Model 459
16 Ultra Wideband for Wireless Body Area Networks 467

Mehmet Rasit Yuce and Ho Chee Keong
16.1.1 Background of UWB 468
16.2 Advantages and Limitations of UWB for WBAN 470
16.2.1 Favorable Factors for Use of UWB in
WBAN Applications 471
16.2.2 Limitations of UWB 472
16.3 UWB Hardware Development 473
16.3.1 UWB Antennas for WBAN Applications 473
16.3.2 UWB Transmitters for WBAN Applications 474
16.3.2.1 Effects of pulse width on UWB
spectrum 477
16.3.3 UWB Receiver 480
16.4 PHY Layer for UWB WBAN 481
16.5 UWB WBAN Channel 482
16.6 MAC scheme for UWB WBAN 483
Contents xv
16.7 UWB WBAN Applications 488

16.7.1 Eight-Channel ECG (On-Body) 488
16.7.1.1 UWB pulse generators 490
16.7.1.2 UWB receiver front-end 492
16.7.1.3 Data recovery 492
16.7.2 Implantable UWB WBAN 494
16.7.2.1 Multichannel neural recording
systems 495
16.7.2.2 Electronic pills (wireless
endoscope) 496
16.8 Design and Implementation of an UWB-WBAN
System 500
16.8.1 UWB Receiver Circuitry 502
16.8.2 Experimental Setup and Measurement
Result 503
16.8.3 Summary 506
16.9 Conclusion 507
Index 511
Preface
Sensor-based measurements and monitoring techniques have been

widely used in electronic patient care systems for a long time. The
concept of wireless-sensor-based patient monitoring using wireless
body area network (WBAN) will bring revolutionary changes in
health care systems. WBAN allows flexibility in providing location-
independent and seamless patient monitoring without affecting the
lifestyle of patients. A WBAN system can be deployed at care centers
for elderly people or at homes to look after elderly citizens without
affecting their mobility or activities, thus improving the quality
of life. The WBAN technology has advanced significantly in the
last decade and gradually moved from the research laboratories to
clinical trials and test environments. In future, WBAN applications
will not remain restricted to the field of medicine. They will
register their presence in other areas such as sports and training,
military applications, and human safety. This book covers a range
of topics on the WBAN technology, introduced in a manner that
will be suitable for a broad range of readers. The book will be
a key resource for medical ICT (information and communication
technology) professionals, biomedical engineers, and graduate
and senior undergraduate students in computer, electronic, and
biomedical engineering.
This book discusses the current state of the art by focusing on
the latest research and new design and development methodologies
related to the WBAN technology. The book systematically introduces
basic concepts, hardware, software and system design techniques,
and various WBAN algorithms for wearable and implantable
sensor applications. It also discusses some potential applications
of WBAN in the field of e-health. Low-cost electronics and off-the-
shelf consumer electronics-based wireless embedded systems are
xviii Preface
introduced, which can be used to design basic low-cost wireless

body area networks.
The book offers a comprehensive review of WBAN design,
development, and deployment techniques. All chapters are written
by leading experts in their fields. The contributions by the
authors focus on applications of wireless body area networks,
implementation, channel modeling, signal transmission around and
in the body, antenna design, in-body communication, and hardware
implementation of body sensors and nodes. Chapter 1 starts with an
introduction to the WBAN to help readers understand the topic and
the associated basic techniques. The next three chapters concentrate
on the implementation of WBAN in clinical environments. Chapter
2 discusses clinical applications of WBAN, followed by discussions
on the cardiac arrhythmias monitoring in Chapter 3. Human bio-
kinematic monitoring using body area networks is described in
Chapter 4. Ambulatory health monitoring applications are presented
in Chapter 5. Hardware design, developments, and architectures are
given in Chapter 6. Chapter 7 presents an overview of hardware
implementation for the ambulatory health monitoring systems.
The design of sensor front-end circuits to detect biological
signals is described in detail in Chapter 8. Network architecture and
medium access control techniques used to design sensor networks
are discussed in Chapter 9. Chapter 10 discusses the power man-
agement techniques for wireless body area networks. Chapter 11
concentrates on the radio transmission channel modeling for body-
centric wireless communications. Chapter 12 focuses on antenna
design and signal propagation techniques for WBAN applications.
The effect of interference from other electronic systems such as Wi-
Fi and microwave ovens could influence the performance of WBAN;
these issues are discussed in Chapter 13. Implanted communication
systems have been dealt with in Chapter 14. Chapter 15 describes
the design of implantable devices that could form a sensor network
by connecting with the on-body wireless nodes. The ultra-wideband
(UWB) technology for both wearable and implantable WBAN
applications is studied in the last chapter, Chapter 16.
We hope that this book will be a key resource for researchers
and students who are working in this emerging area of the medical
technology. Considering the rapid progress in the area, we firmly
Preface xix
believe that the wireless body area network technology will play
important roles in future health care and associated areas. Finally we
would like to thank all the authors for their excellent contributions,
which enabled us to develop a key book on wireless body area
network. We also thank the publisher for delivering an important
book in a timely manner for one of the most important technologies
in this century.
Mehmet R. Yuce
Jamil Y. Khan
October 19, 2011 16:39 PSP Book - 9in x 6in 01-Mehmet-c01
Chapter 1
Introduction to Wireless
Body Area Network
Mehmet Rasit Yucea and Jamil Y. Khanb

a Department of Electrical and Computer Systems Engineering,
Monash University, Clayton, VIC 3800, Australia

b School of Electrical Engineering and Computer Science,
The University of Newcastle, Callaghan, NSW 2308, Australia

[email protected]; [email protected]
1.1 Introduction
Recent advances in wireless technologies and ICT (information and

communication technology) systems are enabling the health care
sector to efficiently administer and deliver a range of health
care services. Advanced ICT systems will be able to deliver health
care services to patients not only in hospitals and medical centers,
but also in their homes and workplaces, thus offering cost savings
and improving the quality of life of patients. For example, Internet
is currently available almost everywhere in different forms, using
either cable or wireless networking technologies. With the advance-
ment of mobile and satellite communication technologies along with
broadband communication techniques, e-health services can be de-
livered anywhere at any time. Especially, when wireless devices are
Wireless Body Area Networks: Technology, Implementation, and Applications

Edited by Mehmet R. Yuce and Jamil Y. Khan
Copyright c 2012 Pan Stanford Publishing Pte. Ltd.
ISBN 978-981-4316-71-2 (Hardcover), ISBN 978-981-4241-57-1 (eBook)
www.panstanford.com
2 Introduction to Wireless Body Area Network
integrated with sensors, it is possible to acquire and monitor

human signals at any environment at any time. Hence, Internet
can be used as a major tool to deliver e-health services to both
developing and developed countries. E-health services can take
advantages of wireless body area network (WBAN), which can
act as an enabling technology. WBAN systems could offer great
advancement for a ubiquitous health care, which has the potential to
improve many aspects of everyday living leading to improved quality
of life of many patients. Telemedicine and e-health services could
offer several significant advantages, including speedup of diagnosis,
therapeutic care for emergencies, offering specialist services to
remote and rural locations, and supporting patients’ mobility and
lifestyles [1].
A WBAN monitors physiological signals from some tiny sensors
with wireless transmission capability placed either inside or around
a person’s body, which are used to collect important health data
of a person during a particular activity — medical or sport or
training-related activities. These nodes form a network between the
sensors and a control device. Figure 1.1 shows a generic WBAN
application scenario. Basically, a WBAN system consists of a number
of tiny sensor nodes and a gateway node used to connect them to
remote locations (i.e., hospital, call center) as shown in Fig. 1.1.
In literature, a number of different terminologies or names have
been used for the gateway device; mainly terminologies such
as body control unit (BCU) or central control unit (CCU) or
personnel control unit (PCU) are used. The gateway device can be
a smart phone or any portable device that can aggregate collected
sensor data and forward them to remote stations. The gateway
node can connect the sensor nodes to a range of communication
networks. These communication networks can be either a standard
telecommunication network, mobile/wireless network, a dedicated
medical center/hospital LAN (local area network) or a public WLAN
(wireless local area network) hotspot, commonly known as the
Wi-Fi.
A WBAN allows a user to store collected data in his/her PDA
(personal digital assistant) or iPod or any other portable devices
and then transfer those information to a suitable computer when a
communication link is available. Future applications of WBAN could
Introduction 3
Figure 1.1. A wireless body area network scheme.
introduce numerous possibilities to improve health care and sports

training facilities. In recent years, the WBAN concept has attracted
the attention of medical and ICT researchers. Standard ICT systems
are already in use in medical areas mostly related to patient record
keeping and scheduling tasks. Some ICT systems or equipment are
used to collect patient physiological data for treatments mainly
using on-site wired equipment. The main drawback of the current
system is the location specific nature of the system due to the
use of fixed/wired systems. Introduction of WBANs opens up new
possibilities in patient care and monitoring using e-health and tele-
health systems. Another major area of health care where the WBAN
will find its application is the aged care, where the quality of life
of elderly people can be significantly improved if the health of
older people is invasively monitored and appropriate health care is
provided.
1.2 Applications
With the widespread use of portable devices such as mobile

phones, pagers, PDAs, MP3 players, etc., the concept of collecting
personal data in those devices becomes more realistic. In 1996,
T. G. Zimmerman studied how such electronic devices operate on
and near the human body. He used the term wireless personal area
network (PAN) [2]. He characterized the human body and used it
as a communication channel for intra-body communications. Later
around 2001, the term PAN was modified to body area network
(BAN) to represent all the applications and communications on, in,
and near the body [3]. One of the most attractive applications of
BAN is in medical environment to monitor physiological signals from
patients [4].
In recent years, interest in the application of WBAN has grown
significantly. According to ABIresearch (Allied Business Intelligence
Inc.), “The market for wireless devices that monitor patients’ condition
and report that data to health care providers is on the verge of
explosive growth, according to a new study from ABI Research. Over
the next few years, it will show a remarkable 77% compound annual
growth rate (CAGR) resulting in global revenue of almost $950 million
in 2014” [5]. A WBAN based on low-cost wireless sensor network
technology can significantly enhance patient monitoring systems
in hospital, residential, and work environments. A WBAN system
allows easy internetworking with other devices and networks,
thus offering health care workers easy access to a patient’s
critical as well as non-critical data without or minimal manual
intervention.
Currently many service providers use very basic manual/semi-
automatic remote patient monitoring systems. Most of the current
systems use the standard telephone network where a patient
could send some medical data by manually connecting a medical
sensor to its body that sends information via telephone networks.
Various companies or service providers offer these services. AMAC
(American Medical Alert Corp.) provides a system called health
buddy, which allows patients to send medical data to health care
professionals remotely [6]. Such a system requires manual interven-
tion. According to the leading technology company Frost & Sullivan,
Applications 5
it is expected that the remote patient monitoring technology will be

deployed in the near future to provide the following medical services
[7]:
• Cardiac care
• Diabetes
• Pulmonary diseases
• Pharmaceutical compliance
• Mental health
• Co-morbidities and others
Various medical equipment vendors have started to produce

remote tele-health equipment to support remote patient monitoring
services. One of the examples of such equipment is the TeleStation
produced by Philips [8]. The TeleStation unit acts as a hub that
connects to other physiological data collection devices such as
weighting scale, blood pressure and pulse cuff, ECG/rhythm strip
recorder and the pulse oximeter. The TeleStation collects all the
measurements from different units and sends those data over
the telephone line to a Philips database server. The TeleStation
can also receive instructions from physicians and/or health care
workers. Physicians or health care workers can request more data
from patients to study or analyze their condition. The TeleStation
can be seen as a gateway of a basic WBAN, which comprises
measurement devices. The TeleStation solution is still not a true
WBAN because in this case, patients have to carry out manual
measurements using standard equipment, whereas in a true WBAN
we would expect sensors attached to a human body to automatically
collect physiological data and transmit them over communication
interfaces. Also, the TeleStation is only capable of transmitting data
over a phone line only and doesn’t support mobility because of the
use of bulky measurement equipment.
Recently a European vendor Broek op Langedijk of the Nether-
lands has developed the RS TechMedic, a portable cardiac monitor
[9]. The unit has an internal GSM mobile phone interface as well
as Bluetooth class I and USB connections. Such devices and systems
are yet to conform to WBAN requirements; however, the trend in the
tele-health/e-health sector is moving towards WBAN applications.
Aged care is one of the prime areas of WBAN applications. Aging

population, shortage of medical staff, and high demand of hospital
resources are problems faced in many countries around the world.
A wireless medical sensor network facilitates remote monitoring
system, which allows the medical and health staff to detect early
symptoms of any illness or even they can use those medical data
to alter medications or care pattern. Timely medical intervention
significantly improves a patient’s chances of recovery. WBAN-based
solution will provide invasive, mobile, and cost effective services for
the above applications.
A WBAN-based monitoring system can also be extended to moni-
tor athletes’ performance to assist them in their training activities.
Using sensor nodes, a trainer can get both on and off the track
performance data. For example, for a cricketer or a tennis player,
movement of arms and body postures are very important for their
success. In this case, a trainer can obtain data from a player via a
WBAN and store those data in computers for further analysis.
Another exciting application of WBAN is in space and military
applications. With wireless and wearable sensors, the conditions
and status of astronauts and soldiers can also be monitored. Military
applications of WBAN have different operating environments. In
the case of military application, a WBAN will be formed around a
soldier’s body by interconnecting various sensors, equipment, and
perhaps with different communication devices [10]. In future, de-
ployment of WBAN may not only be limited to medical applications
but may also be extended to many other applications, including
training, industrial safety, military and other logistic applications.
Applications of WBANs to help disabled persons could be another
major application. It is worthwhile to mention here that a WBAN is
a special purpose sensor network; hence, it is possible for a WBAN
to accomplish a range of tasks. Further research and development
work is necessary to develop future WBAN applications.
1.3 Wireless Personal Area Network (WPAN)/Wireless

Local Area Network (WLAN)
Wireless personal and local area networks are low power, short
range wireless networks used to establish small size networks
Wireless Personal Area Network (WPAN)/Wireless Local Area Network (WLAN) 7
Table 1.1. Wireless technologies for short-range data communications
Frequency DataRate Range Standard Transmission

Band Power
Bluetooth 2.4 GHz Up to 1 Mbps 1–100 m IEEE 802. 15.1, 1–100 mW
WPAN
ZigBee 2.4 GHz Worldwide Up to 250 kbps 0–10 m IEEE 802. 15.4 1–10 mW
868/915 Europe/US WPAN
Wi-Fi 2.4 GHz , 5 GHz Up to 400 Mbps 300 m IEEE802.11(b/g/n) 250–1000 mW

6LoWPAN 2.4 GHz 250 kbps ∼ 30m IEEE802.15.4/IETF ∼ 10mW
to exchange data. Using these technologies, it is possible to

develop low-cost, small-size communication networks to support
many applications, including WBAN applications. Table 1.1 lists
some of the current available wireless technologies that could be
used to develop small-scale wireless sensor networks for various
applications. Bluetooth, ZigBee, and 6LoWPAN systems have mainly
been optimized for short range (10 m) communication transmitting
at low power to connect sensors and control nodes. Wi-Fi devices
are generally used for long-range applications and thus consume
more energy due to higher transmission power. In addition, board
dimensions of the wireless platform for Wi-Fi links are larger.
Most of the current WBAN systems tend to use commercially
available WPAN platforms such as ZigBee and Bluetooth as sensor
nodes of a WBAN system. Bluetooth technology can be used to
communicate with new smart phones. It will be important for future
systems to evaluate their designs for multi-patient environment,
especially devices using a wireless platform at 2.4 GHz ISM band due
to the massive presence of other wireless devices and equipment
operating in the same frequency band. Interference from other
devices operating in the ISM band could reduce the reliability of
WBANs.
With the arrival of new smart phones (e.g., iPhones), PDAs , and
other pervasive wireless devices (iPods, iPads), it is now possible
to use many wireless connections to transmit data from anywhere
to the IT unit of a medical center. Smart phones generally have
multiple wireless links, which could be either 2G/3G mobile links
or satellite links or Wi-Fi connections to transfer data. Smart phones
can select the most appropriate link based on availability, quality, or

user preferences.
1.4 Wireless Body Area Network
A body area network is a collection of wireless sensors (i.e., sensor

nodes) placed around or in a human body that are used to exchange
important information from a human body to remote stations.
Although WPAN devices have been used for WBAN applications,
WBAN is a very small-scale network requiring a communication
distance of a few meters between sensors and the control unit. For
most applications, a WBAN node requires a low data transmission
rate as each sensor node will transmit only one physiological signal.
Physiological signals usually occupy small frequency components.
Table 1.2 lists the biological signals that will most likely be the
source of sensors for WBAN applications. Current WPAN standards
are optimized for industrial applications. As an example, ZigBee is
optimized for sensor network applications and Wi-Fi is optimized
for data network with longer transmission ranges. Thus WBAN
systems should be based on a different standard, which should be
developed and optimized specifically for a short-range low power
sensor network around a body [11]. Figure 1.2 shows the standing
of WBAN among other wireless standards.
In addition to wearable vital signal monitoring sensors (ECG,
temperature, heart rate, etc.), the use of implanted sensors is also
Figure 1.2. Wireless standards. See also Color Insert.

Wireless Body Area Network 9
Table 1.2. List of biosensors
Wearable Sensors Implantable

Electrocardiogram (ECG) Pacemaker
Heart Rate Cochlear Implants
Electromyography (EMG) Implantable defibrillators
Electroencephalogram (EEG) Wireless capsule endoscope (Electronic Pill)
Temperature Electronic pill for drug delivery
Pulse oximeter Deep brain stimulator
Blood pressure Retina implants
Oxygen, pH value
Glucose sensor
Movement (accelerometer)
increasing. Some examples are wireless capsule endoscope, cochlear

implant, and implantable defibrillators. Among the biosensors
listed in Table 1.2, except for retina implants and electronic pills,
the remaining sensors require low data transmission rates. The
electronic pill requires the highest data transmission rate, around
10 Mbps, to transmit good quality video from inside the body to a
monitoring device.
In order to monitor the inner organs as well as the status of
medical implants such as pacemakers and defibrillators, a frequency
around 400 MHz has been used as a popular transmission band for
recent systems [12]. To treat a large number of patients wearing
implanted systems in the same environment (e.g., hospital), a
reliable wireless networking is required to monitor and differentiate
each individual implanted device and patient. Thus implanted
wireless nodes in a patient’s body should form a wireless body
area network so that one or more implanted devices inserted in
the bodies of a number of patients in a hospital environment can
be controlled with minimum complexity. It is quite possible that for
some applications both implantable and wearable sensors form a
WBAN system.
A WBAN is mostly likely to incorporate wearable and implantable
nodes operating in two different frequencies. An implantable node
is most likely to operate at 400 MHz using the MICS (Medical
Implantable Communication Service) band, whereas the wear-
able node could operate in ISM/UWB (Instrumentation Scientific
Medical/Ultra Wide Band) or some other specific bands [13].
Both implantable and wearable sensor nodes require the use of

a gateway-body control unit (BCU) to communicate with remote
monitoring stations as depicted in Fig. 1.1.
A WBAN system can interface with or accommodate different
wireless platforms. As an example, a multilink WBAN can have
multiple wireless links. Multiple wireless links could consist of the
standard ZigBee or Bluetooth to cover WPAN areas and a 802.11
based Wi-Fi link to cover a larger transmission range possibly to
connect to a remote station or a database server [14, 15]. Different
configurations of WBAN nodes are illustrated in Fig. 1.3.
Using WBAN nodes after obtaining raw data from a human body,
sensor nodes transmit those data to the control device — BCU
Figure 1.3. Wireless body area network implementation scenarios:

(a) single WBAN node, (b) a WBAN node with multiple bodies, and
(c) multiple WBAN nodes. See also Color Insert.
Wireless Body Area Network 11
— over a short-distance wirelessly using one of the short-range

wireless systems. The control device can be placed on the body
like a mobile phone as shown in Fig. 1.3a, or it can be placed at
an accessible location. Most suitable technologies for this link are
ZigBee, Bluetooth, 6loWPAN, and WMTS. The main task of the BCU
is to transfer data to a PC or to a smart phone. Wireless technologies
used on this segment (BCU to PC or to an Internet device) could
be a mobile communication network, a satellite link, or a Wi-Fi
link. When the Internet is used, the data collected at this PC can be
transferred to remote stations in remote medical centers across the
network.
As described in Fig. 1.3, the collected sensor data can easily
be transferred to remote stations (i.e., medical centers) with the
existing wireless and information technology infrastructures such
as satellite, mobile communication system, Internet, etc. Accessing
the medical data of injured people through Internet is an effective
solution at the moment, which will allow medical professionals at
the hospital to collect and evaluate data while patients are being
transferred to the hospital in an emergency vehicle. If provisioned,
these data can also be accessed outside the emergency areas as they
will be made available online.
The control device will be similar to smart phones we use
in our daily life to receive and monitor the data obtained from
sensors. They will be like minicomputers, which will most likely be
connected via a wireless technology such as Bluetooth, Wi-Fi, 3G/2G
networks or the satellite (VSAT: very small aperture terminal).
These technologies offer flexible communication links that can be
configured to send data from medical sensors to remote medical
centers (second wireless link in Fig. 1.1 and Fig. 1.3), which can
be accessed by medical professions at any time. The BCU can be
attached to the body as a wristwatch or around belt.
The first scenario sharing a single WBAN node can be used
for home care where there is need for only one body monitoring.
Scenarios presented in Fig. 1.3a,b can be used at a disaster area, in
emergency rooms in hospitals, and in ambulance while patients are
taken to the hospital.
Above discussion shows that WBAN is a specialized sensor
network with definite application requirements. For the commercial
deployment of WBANs, it is necessary to develop an industrial

standard by considering its different application scenarios and
requirements.
1.5 Design Requirements
Design of miniaturized, low-power, reliable, and wearable sensor

node devices is the key requirement of a WBAN design. A WBAN
is a special purpose wireless sensor network designed to transmit
data within a very short distance. The quality of service (QoS) of
a WBAN will depend on applications, which are primarily medical
applications. As discussed earlier, WBANs could be used in other
applications such as sports training, rehabilitation purposes, or
military applications where the QoS will be dictated by the nature
of those applications. For medical applications where the source
of information is physiological signals such as heart rate, blood
pressure, ECG, etc., the system generally demands low latency
and high reliability. Also, most of the physiological signal sources
produce shorter data bursts at a regular sampling frequency. Hence,
a WBAN transmission data rate requirement is low to medium for
each connection. However, one should keep in mind that a WBAN
will comprise multiple sensor nodes where data transmission
should be coordinated by using a reliable and efficient medium
access control protocol [4]. Whatever may be the application of a
WBAN, the main design requirement of a WBAN is to develop a
short-range and reliable wireless sensor network.
A WBAN may need to send data over a longer distance or
may need to support mobility. The range of a WBAN can be
extended by using a multi-hop network using a gateway or a router,
which could connect a WBAN to external networks. The range
of a WBAN network should not affect the sensor node designs
because the gateway will be responsible for seamless long distance
services; hence, a modular design approach should be taken when
a long distance WBAN application design is developed. Similarly,
the mobility feature of a WBAN will be taken care by the gateway
or a router, thus reducing the design complexity of sensor nodes.
The gateway should be able to detect movement of a WBAN or a
Design Requirements 13
person and connect itself to available external networks to exchange

information with the WBAN. The mobility speed support will be
determined by the external network. For example, a Wi-Fi can only
support very low speed mobility up to 5 km/h, whereas a 2G/3G
network could support mobility up to 240 km/h. If we are looking
for mobility within a hospital or inside a home, then a Wi-Fi-
type connection could be sufficient, whereas for a sport training
application, it may be necessary to use a 2G/3G network to obtain
data from the sensors located at an athlete’s body. So the WBAN
design requirements can be classified into basic and advanced
requirements. Advance design requirements will be influenced by
the design requirements of the applications that a WBAN is used.
The basic WBAN design attributes are summarized below:
• WBAN sensor nodes should be able to transfer data over a

distance of a few meters using a single hop connection.
• Sensor nodes should be miniaturized so that they can be
easily wearable.
• Suitable frequency bands should be selected to reduce
interference and thus increase the coexistence possibility of
sensor node devices with other network devices available at
medical centers or in homes.
• WBAN nodes should consume low power so that battery
could last for a very long time. Also, WBAN nodes should
use small size batteries so that light-weight nodes can be
used.
• A WBAN should be designed in a fail-safe manner so that
failure of a node can be automatically detected or failure of
a node should not affect the operation of the network.
• A WBAN should be scalable so that health care workers can
increase or decrease the number of nodes on a patient’s
body without any manual intervention of IT personnel.
To achieve the above design attributes, WBAN designers should

consider the following points in their design work:
Wireless sensors: A transmitter circuit can be designed with a

few components, which may consume extremely low power when
designed with an integrated circuit technology. Appropriate sleep
and wake-up cycles should be incorporated with the designed

hardware to reduce the power consumption to a minimum level.
Reliable data communication: Reliable, error free, and robust

information should be received from sensors. A WBAN will use a
wireless channel to transmit data, which is inherently unreliable.
Error checking and correction mechanisms should be incorporated
so that the unreliability of the transmission channels can be
countered. The wireless technologies used in a medical environment
should operate in the frequency bands that are immune to
interference and thus increase the coexistence of sensor node
devices with other network devices available at the same location.
The wireless technology used should have less interference effect
on other medical equipment. Any network outage should be
automatically detected, and the sensor data should be delivered in a
fail-safe manner, which could be a critical requirement for a patient
monitoring system.
Wireless network security and privacy: Key software components

should be defined and developed to accommodate secure and
effective wireless networking. Protocols should be designed in such
a manner that a WBAN data cannot be collected by intruders. A
WBAN will use a wireless transmission channel, which opens up
the possibility of external intrusion while transmitting data. Data
encryption techniques could be used, but designers should keep in
their mind that undue complexities should not be introduced at the
sensor node level to avoid higher battery power and larger physical
size.
Handover mechanism: Handover mechanism should be integrated

in a WBAN using the gateway or a router. Handover features should
not overload the sensor node design.
Miniaturized antenna: Unobtrusive small antenna design should be

used, which will operate at high frequencies. Directional or narrow
beam antenna design could be considered for specific medical
applications.
Scope of the Book 15
Gateway devices: A WBAN node may have a wearable or im-

plantable node. Gateway devices should be developed to interface
with the existing wireless networks used in health care systems.
Gateway devices should implement advanced algorithms, which are
more power hungry so that sensor node design and requirements
remain simple.
Alarm option: An alarm option should be included when an outage

occurs or a sensor node fails.
Comfort: Sensor node electronics could be designed using flexible

and stretchable technology so that sensor nodes can easily be
embedded in textiles (i.e., patient’s clothes). It can be attached to the
human body using a plaster to eliminate movements.
1.6 Scope of the Book
This book addresses the applications, hardware, and software

design of WBANs. It informs readers about some of the current and
possible future applications of WBAN-based monitoring systems. A
number of chapters have been devoted to WBAN applications as
mentioned in the preface of the book. Following the application
chapters, readers are presented with hardware and software design
techniques, which can be utilized to develop WBAN systems for
medical applications. Hardware design techniques and system level
developments are discussed in detail. An important component
of the WBAN is the antenna design, which is also introduced in
a separate chapter. Protocol and network design techniques are
discussed in several chapters, which will provide readers clear
understanding of WBAN network requirements. It is important
to understand protocol and network design issues because the
efficiency and the performance of a WBAN system could be largely
determined by them. The book also introduces various current
industrial standards, which will inform readers about possible
choice of hardware and software system selections. As mentioned
in this chapter, a number of industrial wireless standards exist,
which could be followed to develop a WBAN by procuring and
configuring them in a suitable manner. Signal transmission around
and in-body, in-body communication techniques, and the design of

implantable sensor nodes are also described in individual chapters.
The book also presents a chapter on coexistence issues. This is
an important issue because nowadays many other wireless devices
are used in many locations in hospitals and homes; hence, it is
necessary to understand the impact of other wireless devices and
choice of appropriate frequency bands. In conclusion, the book
provides a comprehensive coverage of WBAN applications, design,
and deployment techniques.
References
1. ETHEL, http://www.e-wwg.com/Publications/eHealth/EHTEL%20
Briefing%20Paper%20Sustainable%20Telemedicine.pdf.
2. Zimmerman, T. G. (1996) Personal Area Networks: Near-Field Intrabody
Communication, IBM Systems Journal, 35(3–4), 609–617.
3. Dam, K. V., Pitchers, S., and Barnard, M. (May, 10–11, 2001) From
PAN to BAN: Why Body Area Networks?, in Proceedings of the Wireless
World Research Forum (WWRF) Second Meeting, Nokia Research Centre,
Helsinki, Finland.
4. Khan, J. Y., and Yuce, M. R. (2010). Wireless Body Area Network (WBAN)
for Medical Applications, New Developments, in Biomedical Engineering
(ed. Domenico Campolo), ISBN: 978-953-7619-57-2, InTech.
5. http://www.abiresearch.com/press/3244-Healthcare.
6. Health Buddy, http://www.amac.com/remote patient monitoring.cfm.
7. http://www.frost.com/prod/servlet/market-insight-
top.pag?docid=18086184.
8. TeleStation, http://www.healthcare.philips.com/in/products/
telehealth/products/telestation.wpd.
9. RSTechMedic, http://www.rstechmedic.com/.
10. Cotton, S. L., Scanlon, W. G. and Madahar, B.K. (2009) Millimeter-Wave
Soldier-to-Soldier Communications for Covert Battlefield Operations,
IEEE Communications Magazine, 47(10), 72–81.
11. http://www.ieee802.org/15/pub/TG6.html.
12. Bradley, P. D. (2007) Implantable Ultralow-Power Radio Chip Facilitates
In-Body Communications, RF Design, 20–24.
References 17
13. Hanna, S. (2009) Regulations and Standards for Wireless Medical

Applications, ISMICT.
14. Istepanian, R. S. H., Jovanov, E., and Zhang, Y.T., (December 2004)
Guest Editorial Introduction to the Special Section on M-Health: Beyond
Seamless Mobility and Global Wireless Health-Care Connectivity, IEEE
Transactions on Information Technology in Biomedicine, 8(4), 405–414.
15. Jovanov, E., Milenkovic, A., Otto, C., and de Groen, P.C., (March 2005)
A Wireless Body Area Network of Intelligent Motion Sensors for
Computer Assisted Physical Rehabilitation, Journal of NeuroEngineering
and Rehabilitation, 2 (6), doi: 10.1186/1743-0003-2-6.
Chapter 2
Wireless Patient Monitoring

in a Clinical Setting
Esteban J. Pinoa , Dorothy Curtisb , Tom O. Stairc , John V. Guttagb ,

and Lucila Ohno-Machadod
a Electrical Engineering Department, Universidad de Concepción, Concepción, Chile
b Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of
Technology, Cambridge, Massachusetts, USA

c Department of Emergency Medicine, Brigham and Womens Hospital, Boston,
Massachusetts, USA
d Division of Biomedical Informatics, Department of Medicine, University of California
San Diego, La Jolla, California, USA

[email protected]
This chapter presents the technology, implementation, and appli-

cation of a wireless patient monitoring system for triage support.
The SMART project was developed jointly at Brigham and Women’s
Hospital in Boston, MA and at the Massachusetts Institute of
Technology, Cambridge, MA. It was deployed and evaluated at
the Brigham and Women’s Hospital over an 18-month period. We
present a discussion of similar projects, all involving wireless patient
monitoring, and then describe the SMART system design and the
results of our experience with its deployment in the waiting area of
the Emergency Department of the Brigham and Women’s Hospital.

www.panstanford.com
20 Wireless Patient Monitoring in a Clinical setting
2.1 Introduction
According to World Health Organization data, globally there are

about 1.4 physicians per 1000 inhabitants [1]. In most developed
countries, the ratio is only slightly better: 2 to 4 physicians per
1000 inhabitants. It is only natural that patients far outnumber care
providers, particularly during peak demands. In large overcrowded
settings, patient oversight becomes increasingly difficult, such as
in waiting rooms during seasonal outbreaks, at improvised first-
care centers or in other nonstandard settings. To manage a large
concentration of patients, a monitoring system is key for ensuring
high quality of care.
In health care settings, a wireless body area network (WBAN)
can be used to collect patient-relevant information instead of having
a caregiver constantly monitoring a group of patients. A wireless,
automated data collection system provides several benefits:
• Ease of data collection with minimum patient discomfort

• Scalability to a large number of patients
• Real-time monitoring and assessment of changing physio-
logical conditions
• Higher patient-to-caregiver ratio possible, while maintain-
ing the same quality of care (better human resource
utilization)
In current scenarios, caregivers are forced to collect patient

data on a case-by-case basis. Bedside data collection is only used
after patients have been admitted to the hospital. Collecting data
is essential to support good medical decisions. In emergency
departments, the first set of data is collected during triage, where
an initial assessment of the patient status is performed. However, a
single snapshot of the patient vitals may lead to misprioritization.
Furthermore, it is very difficult to notice any changes in patient
status after the triage process has been completed.
In terms of scalability, caregiver–centered systems can easily
collapse during peak demand. A patient–centered system has obvi-
ous advantages. Such a system scales naturally as more nodes are
added to monitor more patients. The only constraint is an eventual
Introduction 21
overload at the network layer or at the monitoring station. To this

end, as much computation as possible should be programmed into
the WBAN components.
In order to assist in the monitoring of multiple patients, we devel-
oped SMART (Scalable Medical Alert and Response Technology). [2–
4] SMART implements a patient-based wireless monitoring system
as an aid for multiple patient scenarios. It can help first responders
in emergency situations by providing a fast on–line triage system
requiring little set-up time. In overloaded waiting rooms, it can help
in re-prioritization of patients. In nursing homes, it can facilitate
the supervision of a large number of people, without degrading
their quality of life. SMART provides a steady flow of data from the
patients and a real-time alarm system to prompt re-triage in the case
where a patient’s condition deteriorates.
Physiological signals such as electrocardiogram (ECG) and pulse
oximetry (SpO2 ) are simultaneously acquired from different patients
and processed at a central server where a single operator can
oversee a large number of patients. Location information is also
collected wirelessly to facilitate locating a particular patient who
needs attention.
Similar projects originate from the military. The Artemis project
[5] uses a WBAN to evaluate the physiological state of soldiers in
the field. A fuzzy logic based algorithm [6] gathers information from
the SpO2 sensor to trigger a distress signal or it can be initiated by
a fellow team member. The distress signal is forwarded to medical
personnel, along with location information provided by GPS, to
guide the caregivers to the injured soldier. Konoske et al. present
an evaluation of a Mobile Medical Monitor (M3) [7] for soldier care.
The mobile unit is very complete, incorporating an assortment of
physiological sensors. However, its main drawback and complaint
among users is its inability to supervise multiple patients and save
the data in relational databases, features SMART can support.
There are many proposals from academic institutions [8] for
pervasive, wireless response systems. They all implement solutions
for multiple patient monitoring. In the WIISARD project, a different
solution is proposed for mass casualty incident response [9], with
a clear focus on logistics. Mobile caregivers are equipped with
wireless PDAs to evaluate patients in the field and relay that
information to a central command post. Patients only had basic

online physiological monitoring, unlike SMART, preventing adequate
detection of adverse events. An intelligent triage tag is proposed for
WIISARD [10] as a replacement for the paper tag, mainly for notifica-
tion and prioritization in the field. The wireless infrastructure [11]
is provided using commercially available equipment.
AID-N project, [12] an extension of CodeBlue, [13] presents a
mobile patient monitoring prototype for an electronic triage system.
The sensors are similar to SMART: SpO2 , blood pressure, ECG and
GPS/MoteTrack for outdoors/indoors location. The primary goal is
triage management for mass casualty events. [14, 15] To this end,
a central station collects data from mobile units, caregivers carry
wireless PDAs to communicate with the server, and patients are
equipped with physiological sensors and visual and audible alarms
to alert caregivers to emergent conditions. A survey conducted
after preliminary usability tests shows that the most important
feature for caregivers is triaging support. SMART also provides
triaging support in the main server, alerts and location information.
A difference between AID-N and SMART is that our system is easily
extensible to regular settings, such as a waiting room, helping ease
the learning curve involved in using a new system.
Another application of WBAN systems is concerned with patient
monitoring during transport [16] among units in a hospital. In this
case, ECG and SpO2 sensors are used, but no location information
is necessary. The patient equipment is similar to SMART, based on
an iPaq H5450. Another pre-hospital patient monitoring system
[17] shows that body sensor networks can also prevent over- or
under-triage of trauma patients. The authors propose an iPaq–based
system that records eight different vital signs during transport to
improve triage. They finally conclude that SpO2 and ECG heart rate
(HR) are a better predictors of the need for abdominal surgery
than reports from pre-hospital clinicians. A low SpO2 and high HR
combination is a good predictor for internal bleeding injury. A smart
vest can also be used as support for WBAN. [18] In this case, sensors
are embedded into a vest or T-shirt. Wires are woven into the fabric
and data and power are transmitted to and from a main processing
hardware with wireless capability. Sensors used in the smart vest
acquire ECG, photoplethysmogram (PPG), blood pressure (BP), skin
temperature, galvanic skin response (GSR), and ECG–derived HR.
Smart System 23
A project conducted at Pennsylvania Hospital [19] evaluated an

ECG-based alarm system. Using a commercially available wireless
ECG, the authors were able to assess the validity of ventricular
fibrillation, asystole, tachycardia, or bradycardia alarms. The au-
thors reported high patient and clinician satisfaction, even though
there were more false alarms than useful alarms, in line with
SMART’s findings. Their system did not provide an automated
location system, although the authors acknowledge its importance.
A complete survey of wearable medical monitors is discussed
by Raskovic et al. [20] Issues, current proposed systems, and a
general description of different systems and mobile monitoring
implementations are presented by these authors.
2.2 Smart System
SMART is designed as a mass casualty response system, but is also

easily used in everyday environments. The rationale is that a system
that is used regularly by physicians and nurses will have a much
easier transition into use in a disaster situation. In SMART, the
wireless component allows the subject to move freely, and has little
negative impact on their regular behavior. Furthermore, the patients
perceive a higher sense of care as shown by our post-study survey.
In terms of medical performance, the system helped detect several
cases that would have otherwise gone undetected.
The SMART patient monitoring system was deployed in the
waiting room of the Emergency Department at the Brigham and
Women’s Hospital in Boston, MA. The system was operational from
June 2006 to December 2007, collecting 6815 minutes of data from
145 patients.
2.2.1 Architecture
2.2.1.1 Hardware
On the patient, physiological and location data are collected and
wirelessly sent to a central server. The physiological sensors
measure ECG and SpO2 and are wired to a PDA that acts as a sensor
hub and transmits the data via standard 802.11b Wi-Fi network.
Location is implemented via ultrasound tags and detectors that can
pinpoint a particular patient to a predetermined area. Privacy issues

are considered and patient-sensitive data is encrypted prior to
transmission to caregiver PDAs. Physiological data transmitted from
the patients to the server is not encrypted because it contains no
identifying information.
The ECG sensor’s printed circuit board is the only custom-made
hardware in the project. It implements an instrumentation amplifier
with fixed gain and filters. An 8-bit analog-to-digital converter (ADC)
samples the signal at 200 Hz and sends the data using RS-232 serial
communication. For SpO2 sensing, a commercial sensor from Nonin
Medical Inc. [21] is used. This sensor reports status, SpO2 level, and
heart rate every second, also via serial interface. As shown in Fig. 2.1,
both sensors arrive at a sensor box, connected to a standard PDA
acting as a wireless node. The HP iPaq 5500 is chosen because it
provides Wi-Fi connectivity, extended battery life, and serial ports.
Having a PDA as the basic building block was useful in configuration
and debugging tasks.
Figure 2.1. Patient PDA with ECG leads and SpO2 sensor. See also Color
Insert.
Smart System 25
Figure 2.2. Patient with ECG, SpO2 and location sensor.
Location data is provided by an ultrasonic (US) system from

Sonitor Technologies. [22] US tags attached to patients transmit a
unique code that is read by US receivers mounted on the walls. A
simple signal strength algorithm assigns a tag to its closest receiver.
The advantage of the US system is that the signal is confined by walls.
This way, there is absolute certainty that a patient is in a given room.
Large rooms can be subdivided by properly placing the receivers.
The tags are pen-like and are usually worn on a lanyard around the
neck. Figure 2.2 shows a diagram of the complete gear worn by the
monitored patients.
Fourteen US detectors are installed covering the complete
waiting area and the overflow area. By having detectors outside the
waiting area, we can infer when a patient has left the waiting room
and their general direction.
SMART also implements a caregiver wireless node. Caregivers
can be equipped with a hand-held PDA and a location sensor. This
way they can receive and respond to alarms forwarded to their PDA,
check the status of a patient, and the server can know their location.
The server hardware is a standard PC with dual monitors, wired

to a wireless router. ECG and SpO2 data arrive wirelessly from the
patients to the router. Location information is also received from the
US detectors via wireless adapters. Only three detectors are wired
to the router, due to their close proximity to the server. All wireless
communications use the IEEE 802.11b standard.
The maximum bandwidth provided by the 802.11b network is
11 Mbps, which was enough for our needs. A simple calculation sets
the requirements for 10 wireless nodes (patients and caregivers)
at under 0.5 Mbps. [4] There are some other technical aspects
to consider when using wireless networks. First, the available
bandwidth is reduced as the distance to the wireless router or
access point increases. Proper placement of the wireless router can
minimize the number of nodes with reduced data rate, while serving
the majority of the nodes at higher data rates. Second, wiring the
server to the router reduces by half the number of wireless data
packets on the network. Finally, radio interference can be avoided
by proper coordination of channel usage in the hospital. In our
particular case, our wireless network used a different channel than
the existing network, and we did not experience any problems on
either network.
2.2.1.2 Software
Different programming languages are used in SMART according to
the task and platform, as shown in Fig. 2.3 For the mobile nodes, the
Figure 2.3. Software and data flow in SMART.

Smart System 27
language of choice was Python. The PDA was running a linux-based

OS, Familiar Linux, that provided better control over the hardware,
and an open source platform. There are two Python programs, the
data-collecting and forwarding program running on the patient PDA,
and the data-fetching program running on the caregiver PDA. The
patient PDA program is not intended to be used by patients. It
provides a simple configuration screen to specify serial port usage
and a “live” data screen used in the early stages of the project for
debugging. Data are identified as ECG or SpO2 , timestamped and
sent via TCP to the server. To reduce the wireless traffic, 20 samples
of ECG data are packed together and sent every 0.1 s. Individual
patient data streams are identified according to their originating
PDA identification number. The caregiver PDA program displays a
roster of patients being monitored and their main vital signs. It
also has the ability to query the server for data for a particular
patient and show live or retrospective ECG and SpO2 data. This
communication is encrypted to protect patient privacy, since at this
point patient names are associated with the data. Both programs
present a simple user interface (UI) programmed using Glade and
GTK+ for configuration and monitoring.
At the server, data first arrive at ORnetDB. ORnetDB was devel-
oped in a previous project as an operating room streaming database.
[23] This database is able to log and perform computations on
live feeds of incoming data. Raw data and computed data such as
QRS positions are then saved into a standard relational database,
implemented in PostgreSQL. [24] Finally, the main UI for the
operator is programmed in Java. The operator has an overview of
all patients being monitored and can select one in particular to view
detailed information (Fig. 2.4). There is also a list of issued alarms
for all patients being monitored. The patient roster on the top of
the screen shows: patient number, in/out status, name, SpO2 and
HR from the finger sensor, calculated HR from the ECG data, the
Emergency Severity Index (ESI) [25, 26] assigned upon admission,
the algorithm’s proposed diagnosis, location and current alarm.
Patient management (enrolling in monitoring, and completion of
monitoring) is handled via web page forms. Location information is
shown on a secondary monitor (Fig. 2.5) with colored tags for the
patients and the caregivers, displayed over a map of the waiting area.
Figure 2.4. SMART central main GUI. See also Color Insert.
Figure 2.5. SMART dual monitor display. Left: location data. Right: main
GUI.
2.2.2 Clinical Implementation

The goal of SMART is to enhance response in the presence of a
high number of patients. Introducing a large number of wireless
monitoring nodes requires that the system assist in the detection
Smart System 29
of critical cases. This approach allows scalability with minimal

requirements of extra-trained personnel. To this end, SMART
implements a set of parameter computations and a decision tree
to assist in alarm generation and diagnosis suggestion. To produce
a diagnosis algorithm that will feed the alarming system, several
steps were necessary. [3] We selected an ECG processing algorithm
based on its ability to handle noisy data. 2) The algorithm was tested
against standard ECG databases publicly available from Physionet.
[27] We conducted tests on healthy volunteers to gain experience
handling live and untethered patients. 4) We conducted tests on
patient simulators to get feedback on how the system responded to
different health conditions.
The final algorithm is programmed based on all the tests and
preliminary results. [3] The ECG algorithm is based on the SQRS
algorithm available on the Physionet website. The modifications
introduced include an auto-calibration interval to account for
differences in electrode placement and patient variability, an
adaptive threshold to handle noisy segments, and a “no beat” output
after 3 s of inactivity. The diagnosis suggestion is based on detected
QRS positions and raw ECG statistical properties, sensor integration
with SpO2 , and noise detection. All possible diagnoses and alarms
generated are shown in Table 2.1.
Sinus rhythm, bradycardia, and tachycardia are determined by
simple HR comparison to patient specific thresholds. By default,
the low HR threshold is 60 beats per minute (BPM) and the
Table 2.1. Diagnosis suggestions and alarm generation
Diagnosis Source Validation Alarm
Sinus rhythm ECG or SpO2 - None

Bradycardia ECG or SpO2 - Medical
Tachycardia ECG or SpO2 - Medical
Ventricular tachycardia ECG - Medical
Ventricular fibrillation ECG SpO2 Medical
Irregular rhythm ECG - Medical
Asystole ECG SpO2 Medical
Leads off ECG - Technical
Noise ECG - Technical
Mismatch ECG & SpO2 - Technical
high HR threshold is 100 BPM. The HR is either calculated from

the detected QRS waves in the ECG or reported by the SpO2
sensor. Ventricular tachycardia, fibrillation, and irregular rhythm
can only be computed from the ECG data, since they depend on QRS
width, ECG waveform, and QRS periodicity, respectively. Asystole
status is determined when no QRS complexes can be found in the
ECG.
Technical alarms are generated when an abnormal situation
most likely related to the PDA or sensors (and not the patient)
is detected. Leads off is easily detected from the ECG waveform.
Noise is indicated by the QRS algorithm when a series of QRS-
like peaks are detected in a short period. Finally, the mismatch
technical alarm is issued when the ECG and SpO2 data are
contradictory.
The final decision tree programmed in SMART is shown in Fig. 2.6
The diagnosis with the most votes in the preceding 15 s is used as
the final suggested diagnosis and alarms are issued for abnormal
conditions.
Sensor integration is fundamental for alarm generation. While
SpO2 data proved to be extremely stable, it can’t detect cer-
tain conditions of interest. ECG provides a quicker response
time and better diagnosis capability. However, ECG is highly
affected by noise from electrode movements and muscle activity.
The solution is to validate preliminary ECG findings with SpO2
information. For instance, an ECG finding of asystole is not
possible with a normal O2 saturation and reported HR greater than
zero.
No major problems were found during the clinical implemen-
tation. However, it is clear that if the number of patients were to
increase considerably, the load on the main server might become
unsustainable. One proposed solution is to push most of the load to
the mobile units. This was tested in the early stages of the project but
later discarded in favor of a simpler, centralized system. However,
all signal processing and alarm generation can run on the patient
PDAs. This way, the main server is responsible for only minor tasks.
Alternatively, SMART Central could be implemented using multiple
computers that share the load as the number of patient nodes
increases.
Smart System 31
Figure 2.6. Decision tree programmed in SMART for diagnosis suggestion

and alarm generation.
2.3 Results
During the implementation in the Emergency Department (ED),

some important lessons were learned. Signal quality was worse
than in the healthy volunteer training session data. Wireless,
nonintrusive systems must be prepared to deal with bad quality
data. Despite this, several patients and the system as a whole
benefited from SMART. It was possible to detect unnoticed medical
conditions and to dismiss a medical complaint sooner by using
the data collected from patients in the waiting room. The critical
cases were detected by SMART Central and reported to the
SMART Operator (whose presence was required by the hospital’s
Institutional Review Board) and ED personnel were alerted when
necessary.
2.3.1 Medical Usefulness

ECG, SpO2 , and location data were collected from 145 patients
during 18 months. The amount of data collected per patient depends
how long the patient was in the waiting room, prior to being seen by
a physician. This time ranges from 5 min to slightly over 3 h. Only
patients presenting with symptoms of shortness of breath and chest
pain were approached for enrollment in the study. This selection
was based on which symptoms might deteriorate quickly, resulting
in a negative outcome for the patient, which determined the set of
physiological sensors available in SMART.
The physiological data collected show that in normal conditions
the system works. However, there is an important difference
between real patients and simulators or healthy volunteers. The
noise level encountered on real patients is much higher than
expected, but relatively short-lived. This produces a significant
number of false alarms that quickly become annoying to the
operator. This usually happens on particular patients for different
reasons, the most recurring one being restlessness. However, most
patients present a normal, quiet ECG tracing with minimal number
of alarms. Setting up and removing the equipment from the patients
also causes false alarms. To alleviate this problem, an option is
provided in the main GUI to disable alarms for particular patients.
Results 33
Figure 2.7. Diagnosis estimation and ECG-computed HR of an average

patient. See also Color Insert.
This way, the alarm system is not enabled until the patient has
settled. Figure 2.7 shows a typical tracing and its accompanying
diagnosis suggestion.
During the evaluation period, three patients presented health
problems detected by SMART that required re-prioritization. The
first case was a patient with premature ventricular contractions that
SMART detected as irregular rhythm and tachycardia. The second
case presented a severe bradycardia. In the third case, the ECG-
reported HR was significantly different than the SpO2 - reported HR.
Upon closer examination of the live ECG acquired by SMART, the
patient was admitted sooner to the ED, presenting with junctional
tachycardia.
The last case worth mentioning is a patient who complained that
her pacemaker wasn’t working properly. Figure 2.8 shows the ECG
HR and SpO2 HR and the ECG tracing on the bottom. Every time the
HR dropped below 60 BPM, the pacemaker is engaged and produces
Figure 2.8. ECG – SpO2 HR comparison and ECG pacemaker activity when
patient HR dropped below 60 BPM. See also Color Insert.
a different tracing on the ECG. An added benefit of SMART is that the

physicians are presented with valuable information by the time they
get to see the patient for the first time, helping their diagnosis.
2.3.2 User Acceptance

Acceptance among the patients was very good. Most patients felt
safer as they knew they were being monitored. For them, it was
a large improvement over just waiting to be seen by a physician.
Knowing that their location was known also helped build confidence
in case they needed prompt care.
A survey was conducted among the patients right after they were
admitted into the ED. Seventy patients completed the survey that
evaluated their perception of the system. Figure 2.9 shows that 51%
of the patients would definitively wear the system again. Another
41% would be inclined to do so. Only 7% are against or did not
respond to the question. To the question “Did the monitoring system
Results 35
Figure 2.9. Response to survey question: Would you wear a SMART pouch
again?
Figure 2.10. Response to survey question: Did the monitoring system

make you feel safer?
make you feel safer?,” over 69% of the patients answered positively,
as shown in Fig. 2.10
Even though the patients who participated in this study were
told that the system would not change the attention provided by
the hospital, Table 2.2 shows that most of them felt that they were
getting a higher level of care.
We also asked the patients about their perception of the value of
having their location and their vital signs being monitored. We were
concerned that they might find the location sensor too intrusive.
However, 84.3% considered the location information valuable and
94.3% considered vital sign monitoring valuable (Table 2.3).
Finally, a majority of the patients found the system acceptably
comfortable to wear (Table 2.4).
Table 2.2. Perceived effect on care
Effect of SMART monitoring on care Preferences

Improved it a lot 17%
Improved it a little 24%
No effect 49%
Impaired it a little 0%
Impaired it a lot 0%
Table 2.3. Perceived value of monitoring
Category Location Vitals

Very important 68.6% 78.6%
A little 15.7% 15.7%
Not important 12.9% 5.7%
Not important and actually annoying 0.0% 0.0%
No response 2.8% 0.0%
Table 2.4. Patient evaluation of system

comfort
Was the monitoring system comfortable? Preferences

Didn’t bother at all 82.9%
A little uncomfortable 14.3%
Very uncomfortable 1.4%
No response 1.4%
2.4 Conclusion
The SMART system facilitates the collection, storage, review, and

detection of events that otherwise may go unnoticed regarding the
patients’ medical conditions. To accomplish this, the set of sensors
on each patient in conjunction with the patient’s PDA send their
data wirelessly to a central station where individual care needs are
evaluated.
SMART addresses the need to provide better care to multiple
patients. It can extend care to nonstandard settings such as nursing
homes for the elderly, improvised care facilities in case of mass
References 37
casualty events, locations designed for patients such as hospital

waiting rooms, and homes or offices for people at risk during
their daily activities. As cited, there is concern among health care
facilities regarding patients in unmonitored areas. [19] In standard
care centers, a wireless monitoring system that provides alarms for
critical cases can reduce the workload on the triage staff and provide
a better quality of care for the patients. Initial triage assessment time
can be reduced, and by the time a physician first contacts the patient,
a considerable amount of physiological data is available.
One of the main advantages of wireless networks in medicine is
the ability to provide an enhanced level of care without interfering
with the patient’s quality of life. Even though in SMART the sensors
are wired to the PDA hub on the patient, the wireless transmission
of the data to a central station allows patients be untethered. The
survey conducted among patients after wearing the device shows
with high satisfaction levels, both in terms of care and comfort.
Acknowledgements
This work was supported in part by the National Library of Medicine,

NIH, N01LM33509.
References
1. World Health Organization data and statistics (April 2010), URL:

http://www.who.int/research/en/.
2. J. Waterman, D. Curtis, M. Goraczko, E. Shih, P. Sarin, E. Pino, L. Ohno-
Machado, R. Greenes, J. Guttag, and T. Stair (2005) Demonstration of
SMART (Scalable Medical Alert Response Technology), in AMIA 2005
Annual Symposium, Washington, DC, American Medical Informatics
Association, pp. 1182–1183.
3. E. Pino, L. Ohno-Machado, E. Wiechmann, and D. Curtis (2005) Real–
Time ECG Algorithms for Ambulatory Patient Monitoring, in Proceedings
of the AMIA Annual Symposium, pp. 604–608.
4. D. W. Curtis, E. J. Pino, J. M. Bailey, E. I. Shih, J. Waterman, S. A. Vinterbo, T.
O. Stair, J. V. Guttag, R. A. Greenes, and L. Ohno-Machado (2008) SMART
— An Integrated Wireless System for Monitoring Unattended Patients,
Journal of the American Medical Informatics Association, 15(1), pp. 44–

53. doi: 10.1197/jamia.M2016.
5. S. P. McGrath, E. Grigg, S. Wendelken, G. Blike, M. D. Rosa, A. Fiske, and R.
Gray (2003) Artemis: A Vision for Remote Triage and Emergency Man-
agement Information Integration, URL: <http://citeseerx.ist.psu.edu/
viewdoc/summary? doi>= 10.1.1.128.4959.
6. S. M. Wendelken, S. P. McGrath, and G. T. Blike (2003) A Medical
Assessment Algorithm for Automated Remote Triage, in Proceedings of
the 25th Annual Conference of the Engineering in Medicine and Biology
Society, Cancun, pp. 123–135.
7. P. Konoske, W. Deniston, R. Barker, and D. Moses (1998) Evaluation
of the Mobile Medical Monitor (m3) in a Field Environment, in
Proceedings of the Medical Technology Symposium, Pacific, pp. 83–92,
doi: 10.1109/PACMED.1998.767929.
8. C. Orwat, A. Graefe, and T. Faulwasser (2008) Towards Pervasive Com-
puting in Health Care — a Literature Review, BMC Medical Informatics
and Decision Making, 8(1), pp 26, ISSN 1472-6947, doi: 10.1186/1472-
6947-8-26, URL: http://www.biomedcentral.com/1472-6947/8/26.
9. J. P. Killeen, T. C. Chan, C. Buono, W. G. Griswold, and L. A. Lenert (2006)
A Wireless First Responder Hand-Held Device for Rapid Triage, Patient
Assessment and Documentation During Mass Casualty Incidents, in
Proceedings of the AMIA Annual Symposium, pp. 429–433.
10. L. Lenert, D. Palmer, T. Chan, and R. Rao (2005) An Intelligent 802.11
Triage Tag for Medical Response To Disasters, in Proceedings of the AMIA
Annual Symposium, pp. 440–444.
11. M. Arisoylu, R. Mishra, R. Rao, and L. A. Lenert (2005) 802.11
Wireless Infrastructure To Enhance Medical Response to Disasters, in
Proceedings of the AMIA Annual Symposium, pp. 1–5.
12. T. Massey, T. Gao, M. Welsh, J. H. Sharp, and M. Sarrafzadeh (2006) The
Design of a Decentralized Electronic Triage System, in Proceedings of the
AMIA Annual Symposium, pp. 544–548.
13. D. Malan, T. Fulford-Jones, M. Welsh, and S. Moulton (2004) CodeBlue:
An Ad Hoc Sensor Network Infrastructure for Emergency Medical Care,
in International Workshop on Wearable and Implantable Body Sensor
Networks.
14. T. Gao, M. I. Kim, D. White, and A. M. Alm (2006) Iterative User-Centered
Design of a Next Generation Patient Monitoring System for Emergency
Medical Response, in Proceedings of the AMIA Annual Symposium,
pp. 284–288.
References 39
15. T. Gao and D. White (August 2006) A Next Generation Electronic Triage
To Aid Mass Casualty Emergency Medical Response, in 28th IEEE Annual
International Conference of the Engineering in Medicine and Biology
Society, pp. 6501–6504, doi: 10.1109/IEMBS.2006.260881.
16. Y.-H. Lin, I.-C. Jan, P. C.-I. Ko, Y.-Y. Chen, J.-M. Wong, and G.-J. Jan (2004)
A Wireless PDA-Based Physiological Monitoring System for Patient
Transport, IEEE Transactions on Information Technology in Biomedicine.
8(4), 439–447, ISSN 1089-7771.
17. C. F. Mackenzie, P. Hu, A. Sen, R. Dutton, S. Seebode, D. Floccare, and
T. Scalea (2008) Automatic Pre-Hospital Vital Signs Waveform and
Trend Data Capture Fills Quality Management, Triage and Outcome
Prediction Gaps, in Proceedings of the AMIA Annual Symposium,
pp. 318–322, URL: http://www.pubmedcentral.nih.gov/articlerender.
fcgi? artid=2656089.
18. P. Pandian, K. Mohanavelu, K. Safeer, T. Kotresh, D. Shakunthala,
P. Gopal, and V. Padaki (2008) Smart Vest: Wearable Multi-Parameter
Remote Physiological Monitoring System, Medical Engineering and
Physics, 30(4), pp. 466 – 477, ISSN 1350-4533, doi: DOI:10.1016/
j.medengphy.2007.05.014, URL: http://www.sciencedirect. com/
science/article/B6T9K-4PNF2X4-1/2/733ffbd1f1553e04ed4fc73ef13
31c40.
19. C. V. Pollack (March 2009) Wireless Cardiac Event Alert Monitor-
ing Is Feasible and Effective in the Emergency Department and
Adjacent Waiting Areas, Crit Pathw Cardiol, 8(1), pp. 7–11, doi: 10.1097/
HPC.0b013e3181980f8b, URL: http://dx.doi.org/10.1097/HPC.0b013e
3181980f8b.
20. D. Raskovic, T. Martin, and E. Jovanov (2004) Medical Monitor-
ing Applications for Wearable Computing, The Computer Journal,
47(4), pp. 495–504, doi: 10.1093/comjnl/47.4.495, URL:
http://comjnl.oxfordjournals.org/cgi/content/abstract/47/4/495.
21. Nonin Medical Inc. Nonin Medical Inc., http://www.nonin.com/.
22. Sonitor R home page, Sonitor R Technologies Inc., URL: http://
www.sonitor.com. Accessed April 30, 2010.
23. D. W. Curtis, ORNetDB software, Available on request from dcur-
[email protected].
24. PostgreSQL open source database, URL: http://www.postgresql.org/.
25. P. Tanabe, D. Travers, N. Gilboy, A. Rosenau, G. Sierzega, V. Rupp, Z.
Martinovich, and J. G. Adams (2005) Refining Emergency Severity Index
Triage Criteria, Academic Emergency Medicine, 12(6), pp. 497–501.
26. R. C. Wuerz, L. W. Milne, D. R. Eitel, D. Travers, and N. Gilboy

(2000) Reliability and Validity of a New Five-Level Triage Instrument,
Academic Emergency Medicine, 7(3), pp. 236–242, doi: 10.1111/j.1553-
2712.2000.tb01066.x.
27. A. L. Goldberger, L. A. N. Amaral, L. Glass, J. M. Hausdorff, P. C.
Ivanov, R. G. Mark, J. E. Mietus, G. B. Moody, C. -K. Peng, and H. E.
Stanley (2000) PhysioBank, PhysioToolkit, and PhysioNet: Components
of a New Research Resource for Complex Physiologic Signals, Cir-
culation, 101(23), pp. 215–220, doi: http://circ.ahajournals.org/cgi/
reprint/101/23/e215.pdf.
Chapter 3
Real-Time Cardiac Arrhythmias

Monitoring for Pervasive Health Care
Zhou Haiying,a Hou Kun-Mean,b de Vaulx Christophe,c

and Li Jiand
a School of Computer Science and Technology, Harbin Institute of Technology,
N.92, Xi DaZhi Jie, Harbin, China

b Laboratoire LIMOS UMR 6158 CNRS, ISIMA,
University of Blaise Pascal Clermont-Ferrand II, France

c Polytech’Clermont-Ferrand, CUST,
University of Blaise Pascal Clermont-Ferrand II, France

d School of software, Harbin Institute of Technology,
N.92, Xi DaZhi Jie, Harbin, China

[email protected]; [email protected];
[email protected]; lijian [email protected]
Pervasive health care (PHC) is a new health care model that

enables patient mobility, continuous health monitoring, and timely
detection of anomalies. Comparing with the ideal PHC services, the
current available (commercial) remote patient monitoring services
such as portable monitoring equipment and prototype systems
need to be improved in real-time capability, effectiveness, and
reliability of context-sensitive anomaly detection. This chapter
addresses the PHC system architecture and anomaly detection
techniques for chronic diseases; a pervasive cardiac care (PCC)
prototype that supports real-time indoor and outdoor continuous

www.panstanford.com
42 Real-Time Cardiac Arrhythmias Monitoring for Pervasive Health Care
cardiac arrhythmias monitoring service health care models is

presented. Moreover, this chapter explores key PHC technologies
such as architecture model and system, including context-aware
recognition, abnormal detection, and adaptive environment, etc.
3.1 Introduction
Health care, as an essential part of modern society, which con-

tributes to social stability and government reputation, claims a
remarkable percentage of national budgets. Due to global population
aging in the 21st century, health care system worldwide is facing a
serious shortage of financial and human resources [1, 2].
World Health Organization (WHO) reports that the world’s aging
population over the age of 65 is reaching 761 million in 2025. [3, 4]
The white paper of “The Development of China’s Undertakings for
the Aged” declares that China is entering the aging society [5]. By the
end of 2005, the aging population over 60 in China was more than
1.44 million, and will reach 400 million in 2037, ranking first in the
world [6, 7]. Furthermore, China is suffering from a serious shortage
and an extremely unbalanced distribution of health resources: China
takes up 22% of world population, yet only 2% of world total health
resources, among which the 80% of the health resources in China
is concentrated in urban areas, and the 80% of the urban resources
is concentrated in handful large hospitals. Therefore, in China, the
severe contradiction between supply and demand of health care
services is prominent [8, 9].
Many studies have shown that more than 80% elderly people
have chronic diseases as well as lifestyle-related diseases, and
more than 30% of them suffer from two or more types of chronic
diseases [10, 11]. Further studies reveal that long-term continuous
health care observations, rather than clinical care or treatment, are
essential for patients suffering from several chronic diseases. Also
according to epidemiological data and related research findings, we
discover that successful prevention and reduction of acute attack
of diseases depends on the improvement of health care quality,
proactive management of chronic diseases, detection of abnormal
symptoms, and healthy out-of-hospital lifestyles [12].
Introduction 43
To cope with the rapid rise of medical cost and the growing
shortage of health care capabilities, a personalized out-of-hospital
healthcare model, which supports independent lifestyle, is highly
demanded [3, 13]. The model is characterized by its: (1) excellent
chronic disease management in 7 × 24 h, by assisting medical
experts with technologies of pervasive biomedical sensing, com-
puting and communication; (2) excellent abnormality detection
ability, which effectively helps avoid emergencies (sudden death);
(3) timely and accurate health care services and emergency handling
capability, available anytime anywhere [14, 15].
PHC is designed to provide high-quality health care services for
patients or elderly people with more than one chronic disease. By
integrating and balancing health care resources and thus reducing
medical cost, PHC is a strong complement for the existing health
care system in China and developed countries. With the help of
PHC model, health care providers can largely be relieved from
physical and mental pressures, the quality of collaborate-treatment
can be improved and the cost be lowered, and eventually a better
and healthier out-of-hospital lifestyle can be established for the
patients [8].
Patient health care service is a billion-dollar industry [16].
Yet, research on the key technologies of PHC system model and
service provision is still rare [17]. Hence, the major research
purposes of this chapter are: (1) to provide the designing rules of
PHC services and to explore PHC service model; (2) to study the
key technologies of PHC for chronic diseases, including context-
aware computing, abnormal detection and environment adaptive
technologies, etc. Health care is crucial not only to people’s lives and
physical health, but also to national security and harmony. PHC is
the technical foundation of future health care system, particularly
in China. The application of PHC service will help to solve the
conflict between resources and needs, and eventually facilitate
social stability. By adopting pervasive service and IT technology and
integrating pervasive computing concept into health care services,
this chapter proposes an innovative service model for public and
private PHC services provision with no limitation in time and
place.
3.2 History of PHC Research
PHC technology, as a cutting edge of its kind, demonstrates

significant differences in research content, methods, and techniques
from its counterparts such as biomedical engineering technology,
medical informatics, and pervasive computing technology. Due to
the complexity, diversity, and variability of PHC applications, how
to build a ubiquitous ICT-based embedded biomedical technology
health care environment, is a complicated interdisciplinary research
issue [18]. This chapter provides a review of the state-of-the-art of
research and development in the PHC field.
Today the progress in areas of wireless communication tech-
nology and medical telemetry equipment has set forth the
rapid advancement of patients’ clinical care and out-of-hospital
care, and a variety of patient monitoring systems have thus been
developed and applied. Thanks to the technology development in
pervasive computing, medical engineering, medical telemetry, and
other associated disciplines, it is now possible to extract, record,
analyze, and transmit data concerning patients health symptoms
(in the form of physical signal characterization) in a timely way
via medical equipment worn by patients to the computer or PDA
device of medical service providers. In this way, the time required
for medical diagnosis and treatment is largely reduced [15, 19].
Lately, research institutions and organizations involved in health
care studies have made a remarkable achievement in developing
patient monitoring prototype system for accurate monitoring of
patients’ vital signs and timely detection of patients’ abnormal signs
[20]. Recognizing the importance of the aged health care market,
a number of world-renowned medical equipment companies,
including General Electric, Hewlett-Packard, Honeywell, and Intel,
jointly set up Research Center for Aging Services Technology (CAST)
in 2002 in Washington, so as to promote the application of wireless
and remote patient monitoring service model and to encourage the
joint efforts in technology development and cooperation of elderly
people service.
The development of a health care prototype system has
gone through three typical phases [17]. According to the patient
monitoring scene, the prototype can be divided into three different
History of PHC Research 45
types: hospital clinical care, home care as well as outdoor mobile

monitoring [21]. The first generation of monitoring systems, such as
Micropaq [22], collects and transmits multi-parameter monitoring
information via short-range wireless medium (such as Bluetooth)
to the network infrastructures within the building framework in
the hospital. The second generation of surveillance systems, such
as Medtronic [23], supports patient monitoring at home and is able
to make time-appointed collection and transmission of patients’
information. Based on television and cable networks, Motiva
system provides a secure, personalized health care communication
platform, which supports the interaction between patients at home
and remote service providers, so as to achieve monitoring and
management of chronic diseases and to improve patients’ quality of
life [24]. Other second-generation systems, such as CardioNet [25]
and Biotronik [26], provide short-term care for patients at home (7–
14 days). The new generation of surveillance systems is designed
for mobile patients with continuous collection and transmission
of patients’ vital signs via infrastructure-based wireless networks
(WLAN, cellular PCS, satellite, etc.). Relative research includes smart
wearable health care research [27], medical telemetry devices
[28], PDA Mobile Gateway [29], smart clothing care [30] and so
on. References [31–34] present recent researches of the medical
decision-making research for PHC service. There are also some
typical PHC systems for elderly people such as Gator Tech Smart
Room [35], perception Room [36], and the elite care [37].
By now, a large number of health care prototype systems
have been designed, but there is still not much study in the
architecture model of PHC system. The existing literature also
mainly addresses the requirements and challenges in PHC system
design, but not the actual system modeling. Kafeza et al. [38]
analyzed the demand model of warning messages delivery. From
the point of view of ICT infrastructure, Haux et al. [39] defined
four types of health care architectures: people-centered, family-
centered, remote health care service-centered, service provider-
centered. The main functions of health care service are classified
as: emergency state detection and alarm, disease management,
health status feedback and suggestions. It also proposes the main
components of a 24 h medical care environment: indoor home
monitoring system, remote monitoring and location-based outdoor

activities, emergency rescue system, and service support system.
Koch et al. [40] investigated the relevant researches in the health
care field in recent five years and classify them into five categories:
decision support information systems, consumer health informatics,
and remote home healthcare, emergency information technology,
and informatics methods. According to the service controllability
and application location, Doukas et al. [41] divided health care into
two categories: health controlled environment (such as health care
units and hospitals) and immediate health care services unavailable
environment (such as at home or outdoor). Sneha et al. [17]
proposed a component-based system framework of pervasive health
care service and explore the design methods of major system
components (sensing, reasoning, and transferring function). Based
on the wireless sensor network technology, Daramolar et al. [42]
presented a grid-based framework for PHC system. By analyzing
the existing pervasive health care applications, Salvador et al. [43]
attempted to define a general PHC system framework, which is
compatible with its software and classify the PHC application into
three types of typical environment: home, grid space, and mobile
space.
Although some remarkable achievements in the field of patient
care have been made, there are still some major limitations
and constraints: 1) the majority of proposed solutions is not
dedicated to PHC services, and there is no in-depth study of
the application characteristics of PHC, such as patient mobility,
continuous monitoring, and timely detection of anomalies; 2) most
of the solutions do not make full use of mobile computing platforms
for biomedical data analysis, which not only produces large amount
of raw data that leads to high network traffics and more bandwidth
requirements, but also brings huge mental and physical pressure
on medical professionals when performing signal analysis and
abnormalities monitoring.
Between the ideal PHC services and the current remote patient
monitoring services with commercially available portable moni-
toring equipment and prototype systems, a huge gap still exists,
especially in their real-time capability, effectiveness and reliability
of context-sensitive anomaly detection. Partly because the existing
solutions are lack of in-depth understanding, exploration of the
Overview of PCC System 47
complicated processing mode, the relevant system parameters as

well as reasoning and decision-making mechanisms of the PHC
system.
To solve the above problems, this chapter addresses the PHC
system architecture and anomaly detection techniques for chronic
diseases. The objective of this chapter is 1) to provide timely
anomaly detection, prevent sudden attack of illness, and reduce
avoidable medical costs; 2) to provide a PHC service, which is
based upon the mobile computing platform, with the capabilities
of biomedical data analysis and medical intervention, and is able
to reduce network traffics and bandwidth requirements. This
chapter presents a WSN-based pervasive cardiac care prototype that
supports clinical, home, and outdoor models to provide continuous
cardiac arrhythmias monitoring service [44, 45]. Based on this
prototype system, this chapter explores the system architecture
technology [46, 47], context-aware recognition and abnormal
detection technology [48, 49], adaptive environment technology
[50], and so on.
3.3 Overview of PCC System
Heart disease ranks as the top threat to human health. Elderly

people and middle-aged person in long-term sub-health state are
among the high risk population of cardiac sudden death (CSD).
Clinical experiences prove that the most efficient way to prevent
CSD is to detect the cardiac abnormalities in time by making long-
term regular cardiac monitoring. Traditional health care systems
are not suitable for pervasive cardiac monitoring (PCC) due to
the time and space constraints in health care applications, as
well as the inability of real-time continuous ECG monitoring. In
recent years, a new generation of cardiac monitoring system named
pervasive cardiac monitoring is booming. By introducing ubiquitous
computing into traditional healthcare, the PCC system provides
a real-time continuous cardiac monitoring service for different
subjects with no constraints of time or space.
3.3.1 PCC System Architecture

The PCC system integrates the advanced technologies of wireless
communication, embedded system, automatic ECG diagnosis (AED),
Figure 3.1. Architecture of PCC system. See also Color Insert.
and telemedicine, which consists of four main functional compo-

nents: wireless ECG sensor (WES), local access server, remote access
server, and remote surveillance terminal. Figure 3.1 shows the
architecture and operation fashions of the PCC system.
3.3.1.1 Wireless ECG sensor

To minimize cost and to correspond to the latest AHA recommenda-
tions [51], an energy-efficient compact ECG collection device, named
WES, is implemented. WES conforms to the basic characteristics
of wireless sensor device: tiny resource, tiny power-consumption,
and associated tiny cost. Furthermore, the basic embedded software
such as a distributed real-time fault tolerant microkernel [52],
dedicated hardware and firmware [53], and a TCP/IP protocol stack
[54] are implemented and ported into WES.
The WES prototype (Fig. 3.2) is a real-time wireless embedded
portable sensor (size = 70 mm × 100 mm) based on TI MSP430
microcontroller. The key features of WES are
• Gain: 1000
• CMMR (min): 120 dB
• Bandwidth: 0.05 Hz to 125 Hz
Figure 3.2. Wireless ECG sensor. See also Color Insert.
• Programmable sample frequency more than 500 Hz

• Analogue to digital converter (ADC): 12 bit
• Leakage current: 10 μA
WES enables the capture of 4-leads ECG signals sampled at

500 Hz (sample frequency) in real-time. Note that the sample
frequency is reprogrammable (100 Hz to 2000 Hz). The sample data
is sent to the local server over a wireless medium such as Wi-Fi or
Bluetooth. In offline mode, ECG signals can be stored in the flash
memory card. The signal store duration depends on the flash card
capacity, the sample frequency, and the number of ECG leads. For
example, a 128 megabyte flash memory card can store 24 h data of
continuous 4-leads ECG signals sampled at 500 Hz. In this way, the
WES works as Holter or RTEST device.
3.3.1.2 Local access server

Local access server may be implemented on a standard PC, a PDA,
a mobile phone, or a dedicated network access device. It provides
two kinds of network connection services: connection with WES via
a local wireless medium (Wi-Fi or Bluetooth) and connection with
remote system via infrastructure network, such as cable modem,
ADSL, 2G/3G, etc.
In view of the difference of network mediums, local servers, and
transmission speeds caused by network traffics, the PCC system
provides an adaptable communication mechanism to guarantee a
reliable data transmission service between local server and remote
server. Furthermore, in order to provide a real-time data transmis-
sion, it is important to minimize the amount of transmission data
to reduce network traffics over low-bandwidth connection links. A

lossless ECG signals compression algorithm is thus implemented.
If the local access server is a high-performance machine, the ECG
diagnostic module can be migrated from WES into the local server
and the cost of WES will thus be reduced. In addition, patients’ video
information is necessary as an aided method for online diagnosis.
Hence, a ”webcam” can be installed in the local server to provide
patients’ images.
3.3.1.3 Remote access server

The remote access server provides the capability of network
connection between local access server and remote surveillance
server. In view of network types, it can support two types of access
servers: PPP server and WAP server. The PPP server establishes
connections between patients and cardiologists via PSTN, while the
WAP server establishes the connections via wide wireless network.
In addition, if the patients and the cardiologists are located in a
same area (e.g., in hospital or at clinic) and share local network
infrastructures (e.g., high-speed LAN), or they utilize the global
network connection services provided by commercial ISPs (Internet
service providers), this component is no more necessary and can be
removed from the PCC system architecture.
Several function modules can be loaded in the remote access
server, including database service and Web service. The medical
history records of patients are important for the diagnosis of heart
diseases. In the remote access server, the medical records with
multimedia formats are stored in the patient database system, which
include the ECG signals sequences, cardiologists’ diagnostic reports,
patients’ video, voice and individual profile information, etc. The
Web service provides the capacity of Internet access to the patients’
database system.
3.3.1.4 Remote surveillance server

The remote surveillance server has an interactive visualization
graphical user interface (GUI, see Fig. 3.3) which enables cardi-
ologists to diagnose cardiac arrhythmias events in real-time and
Figure 3.3. Remote surveillance server. See also Color Insert.
to respond to alarm messages by monitoring the ECG signals

sequence and patients’ images. This server supports multi-patients
surveillances and one patient on-line diagnosis (by cardiologists)
at one time. The 4-lead ECG signals and related diagnostic results
(by the AED algorithm) are displayed on the screen and stored into
data files with the format of WaveForm DataBase (WFDB) [55]. The
diagnostic reports can be produced automatically and be printed
with the ECG signals sequence and the related statistic results.
When a patient’s number of a PCC application is few, e.g., an
application of small clinic, the cardiologist of the remote surveillance
server can directly connect with the patient in the local access server
via the Internet. In this way, the database service and Web service
can be implemented directly in the remote surveillance server.
3.3.2 PCC Operation Modes

This PCC system enables four operation modes which work together
to make the system always adaptable to different application
scenarios and requirements. Cardiologists can reset the operation
mode by taking into account patients’ physical status and network
medium access bandwidth. The key features of the four operation
modes are illuminated as follows:
• Real-time continuous ECG signal. For the sake of remote

real-time displaying and diagnosing, the data including
continuous ECG signals acquisition and its diagnosis report
will be sent in real-time to the remote system. This
operation mode has the highest alarm level that enables on-
line diagnosis in real-time. Note that this mode does not fit
to monitor a large number of patients due to the limitations
of network bandwidth, system resources and medical
resources, but it has remarkable performance in monitoring
the CSD high risk patient. In practice, each cardiologist can
survey approximately four patients simultaneously. In this
mode, to assure reliable cardiac arrhythmias diagnosis, a
patient’s image is required.
• ECG signal sequence. In order to satisfy remote real-time
multi-patient detection and monitoring, WES is configured
to send automatically a sequence of ECG signals (pre- and
post-abnormality) to the remote system when a cardiac
arrhythmia event defined by cardiologists is detected. This
operation mode is suitable for long-term multi-patient
(lower risk of sudden death than the previous class of
patients) cardiac arrhythmias events surveillance.
• Textual emergency message. In this mode, only a short
textual emergency message will be sent to cardiologists
when a cardiac arrhythmia event is detected. According to
the gravity of the symptom, the cardiologists can decide to
intervene immediately or later. This mode may be operated
on any access medium (wire or wireless).
• Diagnosis report email. It is the lowest level operation
mode. The local server will send periodically a report (like
HOLTER report) attached to an email to the remote server.
The cardiologists predefined the reporting period. This
mode is suitable to monitor a large number of patients.
It is to be noted that cardiologists can remotely reconfigure the

operation mode to adapt to the changes of a patient’s status and
environments.
3.4 Key Technologies of PCC System
In this chapter, we present the key technologies developed for the

PCC system, including the lossless signal compression technique,
Key Technologies of PCC System 53
adaptive real-time communication mechanism, and the adaptive

ECG diagnosis (AED) algorithm.
3.4.1 Lossless ECG Signal Compression

Minimizing the amount of the data transmission is the most effective
technique to reduce network traffics, and it is thus necessary
to compress the ECG signals before data transferring. The 500
Hz sampling frequency in WES offers high quality ECG signals
for the ECG diagnosis, but also increases network traffics for the
data transmission. For example, a 4-lead ECG signals with 500 Hz
sampling frequency and 12 bit sample data (ADC resolution), the
size of its 5 s frame is 20 KB (4 leads * 500 Hz * 5 s * 2 byte). In order
to support real-time data transmission through classical modem,
the network bandwidth must be superior to 32 Kbps (20,000 * 8
bit/5 s) when PCC works in the level-I operation mode. This speed is
obviously unfitted for the low bandwidth networks. The ECG signals
with the high sampling frequency (500 Hz in WES) are absolutely
necessary to guarantee the accuracy of the AED algorithm. Note that
ECG signals with the low sampling rate (often 128 Hz) are acceptable
for the purpose of the ECG observation in the remote visualization
system. In general the ECG signals issued from HOLTER devices are
sampling at 128 Hz. Hence, the sampling frequency of ECG signals
can be sub-sampled to reduce network traffics.
On the other hand, the morphological features of ECG signals
show that the signals have huge data redundancies. A good data
compression algorithm can thus reduce network traffics. In fact,
there are many high ratio data compression techniques such as
MPEG3/4, etc., but they are lossless and CPU time consuming. Thus,
to minimize energy consumption and ease VLSI integration of WES,
we adopt a very simple no-loss compression algorithm: only the
difference value between two consecutive samples is sent to the
remote server. Since the potential amplitudes of the ECG signals in
the cardiac cycle duration are almost equal to the zero isoelectric
value, the difference value between the samples data of the cardiac
cycle duration is thus almost equal to zero. According to the data
sizes of the difference values, the sample compression format is
defined in Fig. 3.4.
Figure 3.4. Data format of signal compression algorithm.
Where the ‘‘ + /−” field indicates the positive/negative property

of the difference value: 1 — positive and 0 — negative; the “type”
field indicates the type of data length, and the “data” field stores the
difference data. The statistical results show that this compression
algorithm has 50%–60% compression ratios. Hence, in case of a
4-lead ECG signals at 500 Hz sampling frequency and 12 bit ADC
resolution, when adopting the sub-sampling frequency technique
(to 125 Hz) and the signal compression algorithm, PCC can reduce
87.5% ∼ 90% network traffics.
3.4.2 Adaptive Communication Mechanism

In order to support a real-time data transmission, the UDP protocol
is adopted in PCC to transmit ECG signals. Because the UDP
protocol does not offer a guaranteed datagram delivery service,
a reliable data transmission mechanism must be implemented
in the application layer. In fact, the network bandwidth of the
PCC communication system is normally fluctuated and affected by
network traffics and some interference factors, an application layer
communication mechanism dedicated to PCC is thus implemented,
which enables reliable real-time data transmission in various
healthcare environments.
3.4.2.1 PCC data frame

The transfer data unit between remote and local peers in PCC
is named a PCC frame, which is used to establish/terminate
connection, to deliver data (ECG signals or images), and to configure
operation modes or other system parameters. Each frame consists
Figure 3.5. Frame format of PCC.
of two parts: a frame header followed by data. Figure 3.4 shows the
PCC frame format.
The type field is used to identify the type of PCC frame. Three
frame types are implemented in PCC: a value of 0 × 01 indicates
a system control frame; a value of 0 × 02 indicates an ECG signal
frame; and a value of 0x03 indicates an image frame. Each type of
PCC frames has a unique system priority identified by the type value
(1 to 3, from high to low).
Figure 3.5a shows the format of an ECG signal frame. Each ECG
frame has a unique identifier specified by the sequence number.
Every time when the local server sends an ECG frame, the sequence
number increments automatically by one for next frame. Basing
upon this value, PCC calculates the surveillance time of heart
monitoring. The value of signal number field indicates the channel
number of ECG signals and the default channel number is four in
WES. The value of QRS number fields represents the number of QRS
complexes detected by the AED algorithm in the ECG frame. The
default sampling frequency of WES is 500 Hz, which is identified
in the sampling frequency field. This value is alterable by the
sub-sampling frequency operation because the signal compression
algorithm changes the size of original ECG frame. Hence this
frequency value is useful to define the length of the uncompressed
original frame in the remote server.
The following area of the ECG frame stores the diagnostic results
of the AED algorithm. It is a QRS structure queue, where the QRS
member number is indicated in the QRS number field. Each QRS
member named QRSResult consists of three elements: QRS position,
QRS length, and QRS state. The QRS position indicates the onset
position of a QRS complex in the uncompressed ECG frame. The
heart rate of each beat can be calculated basing on the positions of
two consecutive QRS. The QRS length represents the time interval of
a QRS. The QRS state shows the type of heart rhythm and its related
heart statue classified by the AED algorithm. The ECG signals field
stores the compressed ECG signals. In the PCC system, each ECG
frame contains 5 s of ECG signals.
Figure 3.5b shows the format of the image frame. The sequence
number field is unused for an image frame. The image field contains
an image in the jpeg format. Figure 3.5c shows the format of the
system control information frame. The value of control code field
indicates the type of control code. The related control information
is stored in the following control information field.
3.4.2.2 PCC communication mechanisms

Three kinds of UDP connections are established in PCC. They
are responsible for the system control (udp CMD), the ECG signal
transmission (udp SIG), and the image transmission (udp IMG).
3.4.2.2.1 PCC system control PCC control frames are responsible

for the system remote configuration, the patient online/offline
notifications, and the ECG frames retransmission management. The
control frame is transferred via the UDP CMD connection. PCC
defines nine types of control frames:
• REQ Connect /ACK Connect; REQ Terminate /ACK

Terminate
• REQ Configure /ACK Configure; REQ Restra
• ACK 5Frames; ACK Image.
The control frames with the code of REQ Connect and

ACK Connect are responsible for the connection establishment
between patients and cardiologists. The control frames with the
code of REQ Terminate and ACK Terminate are responsible for the
connection termination. Both patients and cardiologists have a right
to open and close the UDP CMD connection. The control frames with
the code of REQ Configure and ACK Configure are responsible for
the system configuration. Currently, PCC provides the configurations

of the operation mode and the sampling frequency. The other control
frames with the code of REQ Restra, ACK 5Frames, and ACK Image
will be introduced in the following subsections.
3.4.2.2.2 Signal retransmission mechanism PCC guarantees a reli-

able ECG signal delivery service by providing a signal retransmission
mechanism. Two ECG frame queues are defined, respectively, for the
local access server (Hold Queue) and the remote surveillance server
peer (Wait Queue). The standard length of the two queues is set to
five. The retransmission mechanism is described in Fig. 3.6.
When an ECG frame with the sequence number k is lost during
network transmission, the remote surveillance server sends a re-
transmission requirement frame with a control code of REQ Restra
Figure 3.6. Signal Retransmission Mechanism: Remote Surveillance Peer

(U) and Local Access Peer (D).
and the local access server responses to this requirement by

retransmitting the ECG frame k. When five consecutive ECG frames
are received by the remote surveillance server, a control frame
containing the code of ACK 5Frames will be sent to the local server
to inform the release of the remainder ECG frames in Hold Queue.
3.4.2.2.3 Data competition mechanism Considering the network

quality and communication cost, PCC employs a data competition
mechanism to guarantee those important data (i.e., ECG signals)
being transferred in a higher priority, so that in the remote
surveillance peer, cardiologists can detect and diagnose the ECG
signals in real time.
As mentioned above, PCC has three types of data frames. Each
kind of data frame has a unique transmission priority. The system
control frame has the highest priority in this system. Furthermore,
since the ECG signals are more important than the images data for
the diagnosis of cardiologists, the ECG frame has higher priority
than the image frame. In order to guarantee a real-time ECG signal
transmission service, a data competition mechanism is implemented
in the PCC system, described in Fig. 3.7.
When the queue length of Wait Queue is equal to or greater
than 5, it means that there is at least 25 s of ECG signals stored in
the remote surveillance server. Hence, it is acceptable to allow the
local server to transmit images. Whereas, when the queue length
of Wait Queue is smaller than 5, it means that the network speed
begins to fluctuate and the network quality is lowered; hence,
the system will stop image transmission so as to reduce network
traffics.
3.4.3 AED Algorithm

In recent years, the study of automatic ECG diagnosis (AED)
techniques addressed mainly clinical applications, but rarely PHC
applications, due to two reasons: 1) to be applied on various scenes,
i.e., different objectives and environments, the system signals are
ambulatory, easily being disturbed by interferences; 2) to meet the
needs of portable, the system node has strict resource constraints,
traditional AED techniques are resource-consumption in views
Figure 3.7. System competition mechanism: remote surveillance peer (U)

and local access peer (D).
of PCC services. A novel AED algorithm dedicated to the PCC

applications has thus been developed. By offering the PCC service,
this system can reduce the risk of cardiac sudden death by detecting
cardiac arrhythmia events in time.
3.4.3.1 Signal preprocessing and conditioning

Due to the nonstationary and easy-to-be-disturbed features, the
ambulatory ECG signals must be de-noised before decision-
making. Most of artifacts, such as baseline drifts, electrical noises,
and muscle tremor interferences, can be eliminated or restrained by
adopting suitable filters.
3.4.3.1.1 ECG time series Three ECG signal series, i.e., R(t), AD(t),
and RC(t), are adopted in the AED algorithm. The R(t) series is raw
ECG signals acquired from electrodes. It is generally contaminated
by different kinds of noises. The AD(t) series is the adaptive

differential signals with the processing of the differential filter and
the adaptive filter. The inferences of the baseline drift and the
motion artifacts can be eliminated in the AD(t) series; hence, this
series is used to detect and localize the QRS complexes. The RC(t)
series is the de-noised ECG signals with the operations of the band-
pass filter and the linear amplifier. Since electrical noises and muscle
tremors have been removed from the RC(t) series, the RC(t) series is
used to extract the characteristics of the QRS complexes.
3.4.3.1.2 Adaptive filter Classical filters for the ECG series, e.g.,
notch filter, low-pass filter, and high-pass filter, can effectively
remove or reduce most of interferences. But for motion artifacts,
because of their irregular occurrences and irregular morphological
attributes, these filters cannot eliminate these disturbances. These
artifacts can cause much trouble in QRS detection when encounter-
ing QRS-like artifacts. This algorithm adopts an adaptive filter (AT)
to reduce motion artifacts. The resultant signal series, named A(t),
is generated by performing AT operation in the raw series R(t). The
AT expression is
⎧
⎨ Aecg(0) = R(0)
0 < α < 1, t = 1 · · · N (3.1)
⎩
Aecg(t) = α ∗ Aecg(t − 1) + (1 − α) ∗ R(t)
where α is the balance coefficient of AT, which is a key factor

to influence the performance of the AT filter. If the value of α is
augmented, the previous estimated value Aecg(t – 1) then gives
more proportion in the estimation of current Aecg(t). Thereby, the
achieved A(t) is stable but cannot reveal the changes of ECG signals.
In contrast, if the value of α is diminished, the original value R(t)
then gives more proportion and the achieved A(t) is thus more
dynamic and adaptive to the changes of ECG signals. Hence, the
value of α should be cautiously selected to make A(t) both stable
and adaptive to the changes in R(t). The value of α is set to 0.95
by default in views of the ECG sample frequency in PCC (500 Hz).
Figure 3.8 shows the QRS complex waveform after the adaptive filter
operation with different impact factors.
Figure 3.8. QRS complex waveform after different impact factor in the
adaptive filters. See also Color Insert.
Figure 3.9 shows different ECG series: (a) is the raw signals
R(t), which are seriously polluted by noises. (b) represents the
reconstructed series RC(t) when filtering the R(t) series by
traditional filters, i.e., notch filter, low-pass filter, and high-pass filter.
The RC(t) series still contain the interferences generally caused
by baseline wandering and motion artifacts. (c) is the adaptive
filter signal A(t) when filtering the R(t) series by AT, which has
Figure 3.9. ECG series after different filters.

better signal quality than RC(t). (d) is the reconstructed signal

RC ∗ (t) based on the adaptive filter signal A(t). Obviously, in contrast
to the previous reconstructed signal RC(t), the signal RC ∗ (t) has
better signal quality in which the motion artifacts are effectively
eliminated.
3.4.3.2 QRS complex detection

A new QRS detector has been developed to eliminate noises and
artifacts by exploiting a self-adaptive threshold (SAT) method and
designing state transition recognition (STR) procedure. The SAT
method is used to estimate the peaks of ECG subsegments and
the means of contextual thresholds, which allow estimating the
optimum thresholds in a segment space. The STR procedure traces
the waveform changes of signal series and identifies QRS complexes
based on the optimum thresholds and the rules of state transition.
3.4.3.2.1 Diagnostic segment window (DSW) A short-term redun-

dant data (default 5 s) is important in QRS detection. First, this
short-term segment enables the complex contextual correlative
analysis and reduces the interferences of baseline drift. In view of
the low-frequency baseline drift, a short-term segment has fewer
disturbances caused by baseline wandering than long-term signals.
The redundant data enable the QRS detector to identify current
QRS complex by comparing with fore-and-aft QRS complexes.
Furthermore, in view of the unpredictability and variability of
network quality, redundancy is necessary for data retransmission
and network communication.
3.4.3.2.2 Self-adaptive threshold The QRS complexes of AECG have

rapid changes and high potential amplitudes so that the differential
series D (t) can exactly represent the changes. The QRS signals
have higher absolute amplitudes in a cardiac cycle of the D (t)
series. The solution to QRS detection is to search the optimum
pair-peak for each QRS complex, i.e., the positive and negative
peaks in a cardiac cycle. In Diagnostic Segment Window (DSW),
there are generally multiple pair-peaks because several heartbeats
will occur within 5 s. These pair-peaks make up of a pair-peak
series in a DSW. Based on the pair-threshold extracting from the
Figure 3.10. Mean of pair-peak series in a diagnostic window.
pair-peaks series in a DSW, the STR procedure is then able to locate

QRS complexes. The absolute amplitude of each peak is generally
greater than the associated absolute threshold in D (t). Furthermore,
since the position offset between the D (t) series and the A(t) series
is constant, we can thus obtain the positions of QRS complexes in
A(t) by locating the complexes in D (t).
The SAT method aims to determine the optimum pair-threshold,
which is estimated from two aspects: the mean of the pair-peak
series in a DSW and the pair-threshold of the previous DSW.
The pair-threshold results from the means of the negative and
positive pair-peaks series in a DSW. In order to accurately estimate
these pair-peaks, the diagnostic segment window is divided into
five subsegments with the length of 1 s (see Fig. 3.10). Because
the normal heart rate of a healthy adult is 60–100 bmp, each
subsegment thus contains one heartbeat. Since the differential
signals of QRS complex have the maximum absolute amplitudes in
a cardiac cycle, a pair-peak will indicate a QRS complex and then
can be used to estimate the thresholds. Furthermore, the shorter the
subsegment, the less interference of baseline drift the subsegment
has. A subsegment with the length of 1 s can thus be regarded as a
stationary series.
3.4.3.2.3 QRS location: State transition recognition In view of the

QRS morphology properties in the D (t) series, the complexes
are categorized into two groups: positive and negative. Therefore,
Figure 3.11. Positive states of a QRS complex in D (t). See also Color
Insert.
different states are defined to outline the phases of QRS complex

in D (t). S2∼ S9 represent the positive states of a QRS complex
(see Fig. 3.11); corresponding S20∼ S29 represent negative states.
An adaptive and self-corrected procedure, named STR, is developed
to automatically track the changes of signal series, to correct error
detection, and to record detected complexes. The states transitions
are based on three basic reference lines: the baseline, the positive
threshold, and the negative threshold.
3.4.3.2.4 Feature extraction: geometric analysis method QRS com-

plex has triangular-alike or triangular-component morphological
characteristics (see Fig. 3.12). This chapter thus adopts the geomet-
ric analysis method (GAM) to extract the features of QRS complexes.
GAM has simple operations and low resource consumption, being
able to predict and estimate the key points of QRS complexes
under noisy situations, such as R wave peak, end point of Q wave
(Qt ), and onset point of S wave (Si ). Therein, R wave peak can be
obtained from Tpeak 1 or Tpeak 2 , and it has mono-peak or poly-peaks.
The measurement and detection phases of Qt and Si points are
illuminated as follows.
• Defining two-level thresholds for left and right sides of R

wave (LH = (1/4) × Vpeak 1 , LL = (3/4) × Vpeak 1 , RH = (1/4)
× Vpeak 2 , RL = (3/4) × Vpeak 2 ).
Figure 3.12. Illumination of geometric analysis method. See also Color

Insert.
• Calculating the intersection points between the threshold

values and complex signals. The slopes of two approaching
lines represent two characteristics of QRS complex: SP
(positive slope) and SN (negative slope).
• Obtaining the duration length of QRS (LQRS) which is the
distance of two intersection points between the baseline
and two approaching lines.
3.4.3.3 AED performance analysis

The AED algorithm has been evaluated on two ECG databases:
MIT-BIH arrhythmia database [56] and CSD database (Clinic STAR
Database). The former contains 48 half-hour excepts of two-
channel ambulatory ECG recordings, and the latter is obtained from
30 patients of C.H.R.U. of Gabriel Montpied’s hospital (Clermont-
Ferrand, France) by using the PCC system named STAR [57]. The CSD
signals are recorded by using WFDB format as the MIT-BIH Database
one.
Dotsinsky et al. [58] defined four performance parameters to
assess the algorithm efficiency (Se: sensitivity and Sp: specificity):
TP (true positive), FP (false positive), FN (false negative), and shifted
SH beats, shown as follows:
TP FP TP
Se = Sp = 1 − = (3.2)
TP + FN + SH TP + FP TP + FP
Table 3.1. Performance evaluations of QRS detection algorithms
Se (%) Sp (%)
Afonso et al. [59] 99.59 99.56
Poli et al. [60] 99.60 99.51
Dotsinsky et al. [58] 99.04 99.62
Kaiser et al. [61] 99.68 99.72
Datex-Ohmeda Corp. [62] 99.86 99.88
Millet et al. [63] Alg 1 94.6 98.0
Alg 2 97.3 98.0
Our algorithm 99.43 99.25 98.55 97.94
MIT CSD MIT CSD
Comparing with the performance results of other algorithms listed

in Table 3.1, the overall results of our detection algorithm, 99.37%
sensitivity and 99.68% specificity on MIT-BIH database, 99.67%
sensitivity and 99.74% specificity on CSD database, show the high
sensitivity and specificity. This detection algorithm has minimal
beat detection latency, low computational consumption, and fast
detection ability.
3.5 Conclusion
Currently, the PCC system has been evaluated on 30 patients who

had acute cardiac arrhythmia disturbances at the CHRU of Gabriel
Montpied’s hospital (Clermont-Ferrand, France), shown in Fig. 3.13.
Since patients and cardiologists are located at the same area (at
the hospital) in this application, the PCC system is thus configured
as clinical care state. The local access server is a laptop on which
the real-time cardiac arrhythmias program is installed. The remote
access server and the remote surveillance server are installed in the
same laptop or a standard PC, which connect with the local server
via the local high-speed network. This system supports continuous
remote cardiac arrhythmias monitoring, and the 4-lead ECG signals
are sent directly to the remote surveillance terminal in real time with
real-time continuous ECG signal operation mode.
The system has also been utilized to evaluate athletes’ cardiac
status during physical exercises, shown in Fig. 3.13. The obtained
Conclusion 67
Figure 3.13. Application example of PCC system. See also Color Insert.
results show this system still provides high quality ECG signals and
accurate QRS detection PCC is an efficient system for diagnosing
cardiac arrhythmias. It provides a new clinical approach adapted
to monitor accurately and efficiently large scale high-risk patients.
Furthermore, it may be used by cardiologists to remotely monitor
patients, to evaluate the efficiency of drugs and to discuss difficult
cardiac pathology cases with other colleagues.
In order to compare with other system, the HP telemetry system
has been applied during PCC evaluation. For each patient, the
evaluation duration is 30 min. The evaluation results obtained by the
PCC system and the HP telemetry system are compared, as shown in
Fig. 3.14. Due to the higher sampling frequency, the ECG signals of
our system have better quality than HP ones. Concerning the clinical
off-line cardiac arrhythmia detection, the two systems have similar
results. The evaluation results prove that the PCC system meets
the requirements of real-time cardiac monitoring and diagnosing
application.
In order to improve this system, the following techniques should
be developed: (i) Embedded RTOS (i.e., HEROS [64]) adopts a full
“modularity” design fashion. The system primitives and tasks of
HEROS will be defined as a set of actions. Thus, it may be configured
according to different applications. (ii) AED algorithm has been
ported into the local access unit and the remote center server, but it
can be integrated as an ECG diagnostic chip. We are working on the
Figure 3.14. PCC test and comparision with HP telemetry.
implementation of an intelligent wireless ECG sensor (IWES) [65]

by integrating the algorithm into a VLSI chip. This chip is currently
under evaluation and test on an ALTERA FPGA board.
Acknowledgment
We would like to thank the MENRT, the ANVAR and the Conseil
Regional d’AUVERGNE for their support, and also the colleagues of
the SMIR group of LIMOS UMR 6158 CNRS for their help.
References
1. Goldberg, S. and Wickramasinghe N. (2003) 21st Century Healthcare:

the Wireless Panacea in Proceedings of the 36th Hawaii International
Conference on Systems Sciences.
2. Varshney, U. (2005) Pervasive Healthcare: Applications, Challenges and
Wireless Solutions Communications of the Association of Information
Systems, 16(3).
3. Tablado, A. Illarramendi, A., Bermudez, J., and Goni, A. (2003) Intelligent
Monitoring of Elderly People, in Proceedings of the 4th Annual IEEE
Conference on Information Technology Applications in Biomedicine.
4. World Health Organization (1999) Death by Cause, Sex and Mortality
Stratum in WHO Regions, WHO General Report.
References 69
5. Chinese National Aging office (2006) Forecast Study of the Devel-

opment Trend of China’s Population Aging Issue, General Report
(in Chinese).
6. Zou, N. J. (2007) Old-Age Security, Old-Age Home and Community
Support: New Option for Old-Age Models, Jiangsu Social Sciences, 4 (in
Chinese).
7. Chinese National Bureau of Statistics (2006) China’s Demographic
Yearbook2006, pp. 289 (in Chinese).
8. Zhu, Z. M. (2010) Health Ubiquitous Service: Exploring Pervasive
Computing in Health/Medical Applications, http://www.bjkp.gov.cn/
bjkpzc/kjqy/it/lsywl/ 250584.shtml (in Chinese).
9. Yang, T. Li, Z. G. and Shi, Y. C. (2007) Study Elderly Care Services by
Integration With the Community Health Services in Rural China, General
Report (in Chinese).
10. Yan, J. L. and Wei, S. (2009) The Health Education Problems and
Countermeasures for Urban Elderly in Community, Modern Nursing,
35(6), pp. 122 (in Chinese).
11. Chronic Care Improvement (2004) A Product of the E-Health Commit-
tee, ITAA E-Health White Paper, pp. 5.
12. Sneha, S. (2008) Patient Monitoring via Mobile Ad Hoc Networks:
Maximizing Reliability While Minimizing Power Usage and Delays,
Doctoral Dissertation, Georgia State University.
13. Website (2010) US Administration on Aging Report on Demographic
Changes, http://www.aoa.dhhs.gov/aoa /stats/aging21 /demography.
html.
14. Chiasson, M. Davidson, E. Kaplan, B., et al. (2004) Strangers in a
Strange Land: Can IS Meet the Challenges and Opportunities of Research
in Healthcare? in Proceedings of the 10th Americas Conference in
Information Systems.
15. Gouaux, F. Simon-Chautemps, L. Adami, S., et al. (2003) Smart Devices
for the Early Detection and Interpretation of Cardiological Syndromes,
in Proceedings of the 4th International IEEE EMBS Conference on
Information Technology Applications in Biomedicine.
16. Website (2010) US Patient Monitoring Industry Overlook., http://
www.frost.com/prod/servlet/report-homepage.pag ?repid=A369-01-
00-00-00.
17. Sneha, S. and Varshney, U. (2009) Enabling Ubiquitous Patient Monitor-
ing: Model, Decision Protocols, Opportunities and Challenges, Decision
Support Systems, 46, pp. 606–619.
18. Bardram, J. E. (2008) Pervasive Health Ccare as a Scientific Discipline,

Methods Information Medicine, 47(3), pp. 178–185.
19. Hung, K. and Zhang, Y. (2003) Implementation of a WAP-Based
Telemedicine System for Patient Monitoring, IEEE Transactions on
Information Technology in Biomedicine 7(2), pp. 101–107.
20. Sneha, S. and Varshney, U. (2005) Wireless ECG Monitoring System for
Pervasive Health Care, in Proceedings of the 11th Americas Conference on
Information Systems.
21. Zhou, H. Y. and Hou, K. M. (2010) Pervasive Cardiac Monitoring System
for Remote Continuous Heart Care, in Proceedings of International
Conference on Bioinformatics and Biomedical Engineering.
22. Website (2010) Patient Monitoring by Welch Allyn, http://
www.monitoring. welchallyn. com/products/wireless.
23. Mendoza, G. G. and Tran, B. Q. (2002) In-Home Wireless Monitoring of
Physiological Data for Heart Failure Patients, in Proceedings of the 24th
Joint IEEE EMBS/BMES Annual Conference.
24. Website (2010) TV-Based Monitoring by Philips http://
www.medical.philips.com/main/ news/content/file 630.html.
25. Website (2010) Cardionet, http://www.cardionet.com/.
26. Website (2010) Biotronik, http://www.biotronik.com/.
27. Bonato, P. (2003) Wearable Sensors/Systems and Their Impact on Bio-
medical Engineering, IEEE Engineering Medicine and Biology Magazine
22(3), pp. 18–20.
28. Gieras, I. A. (2003) The Proliferation of Patient-Worn Wireless
Telemetry Technologies Within the US Healthcare Environment, in
Proceedings of the 4th International IEEE EMBS Special Topic Conference
on Information Technology Applications in Biomedicine.
29. Jovanov, E. O’Donnell Lords, A. Raskovic, D., et al. (2003) Stress
Monitoring Using a Distributed Wireless Intelligent Sensor System, IEEE
Engineering in Medicine and Biology Magazine 22(3), pp. 49–55.
30. Website (2010) Smart Shirt, http://www.smartshirt.gatech.edu.
31. Brahnam, S. Chuang, C. F., Sexton, R. S. and Shih, F. Y. (2007) Machine
Assessment of Neonatal Facial Expressions of Acute Pain, Decision
Support Systems 43(4), pp. 1242–1254.
32. Hu, P. J. H., Wei, C. P. Cheng, T. H. and Chen, J. X. (2007) Predicting
Adequacy of Vancomycin Regimens: a Learning-Based Classification
Approach to Improving Clinical DecisionMaking, Decision Support
Systems 43(4), pp. 1226–1241.
References 71
33. Lin, L. J-HHu, P. and Sheng, O. R. L. (2006) A Decision Support System

for Lower Back Pain Diagnosis: Uncertainty Management and Clinical
Evaluations, Decision Support Systems, 42(2), pp. 1152–1169.
34. Varshney, U. (2008) A Framework for Supporting Emergency Messages
in Wireless Patient Monitoring, Decision Support Systems, 45(4),
pp. 981–996.
35. Website (2010) Gator Tech Smart House, http://www.icta.ufl.edu/
gt.html.
36. Website (2010) Aware Home Project, http://www.awarehome.
gatech.edu/projects/ index.html.
37. Stanford, V. (2002) Using Pervasive Computing to Deliver Elder Care,
IEEE Pervasive Computing Magazine 1(1), pp. 10–13.
38. Kafeza, E., Chiu, D. K. W., Cheung, S. C. and Kafeza, M. (2004) Alerts
Immobile Health Care Applications: Requirements and Pilot Study, IEEE
Transactions on Information Technology in Biomedicine 8(2), pp. 173–
181.
39. Haux, R. Howe, J. Marschollek, M., Plischke M. and Wolf, K. H. (2008)
Health-Enabling Technologies for Pervasive Health Care: on Services
and ICT Architecture Paradigms, Informatics for Health and Social Care
33(2), pp. 77–89.
40. Koch, S. and Hägglund, M. (2009) Health Informatics and the Delivery of
Care to Older People, Maturitas 63(3), pp. 195–199.
41. Doukas C. and Maglogiannis, I. (2008) Intelligent Pervasive Healthcare
Systems, Advanced Computational Intelligence Paradigms in Healthcare
(Springer Press), 107, pp. 95–115.
42. Daramolar, J. O. Osamor, V. C., and Oluwagbemi, O. O. (2008) A Grid-based
Framework for Pervasive Healthcare Using Wireless Sensor Networks:
a Case for Developing Nations, Asian Journal of Information Technology,
7(6), pp. 260–267.
43. Salvador, Z. Larrea M. and Lafuente, A. (2007) Infrastructural Software
Requirements of Pervasive Health Care, in Proceedings of IADIS
International Conference on Applied Computing, pp. 557–562.
44. Zhou, H. Y. (2004) Wireless Sensor Networks Dedicated to Remote
Continuous Real-Time Cardiac Arrhythmias Detection and Diagno-
sis, Ph.D Thesis, University of Blaise Pascal Clermont-Ferrand II
(France).
45. Zhou, H. Y. et al. (2006) A New System Dedicated to Real-Time Cardiac
Arrhythmias Tele-Assistance and Monitoring, Special Issue of Journal of
Universal Computer Science, 12(1), pp. 30–44.
46. Zhou, H. Y. et al. (2004) Remote Continuous Cardiac Arrhythmias De-

tection and Monitoring, Transformation of Healthcare With Information
Technologies, Studies in Health Technology and Informatics (IOS Press),
105, pp. 112–120.
47. Zhou, H. Y. et al. (2005) A Real-Time Continuous Cardiac Arrhythmias
Detection System: RECAD, in Proceedings of the 27 th Annual Inter-
national Conference of the IEEE Engineering in Medicine and Biology
(EMBC05), pp. 875–881.
48. Zhou, H. Y. Hou, K. -M. and Zuo, D. C. (2009) Real-Time Automatic ECG
Diagnosis Method Dedicated to Pervasive Cardiac Care, Wireless Sensor
Network, 1(4), pp. 276–283.
49. Zhou H. Y. and Hou, K. -M. (2008) Embedded Real-Time QRS De-
tection Algorithm for Pervasive Cardiac Care System, in Proceedings
of the 9th International Conference on Signal Processing, pp. 2150–
2153.
50. De Sousa, G. Zhou, H. Y., Hou, K. -M. et al. (2007) Adaptive System for
Wireless Sensor Networks Applications, Journal of Harbin Institute of
Technology 39, pp. 153–157.
51. Bailey, J. J. Berson, A. S. Garson A. et al (1990) Recommendations for
Standardization and Specifications in Automated Electrocardiography:
Bandwidth and Digital Signal Processing, Circulation American Heart
Association 81(2), pp. 730–739.
52. De Vaulx, C. and Hou, K. -M. (2002) DREAM: un micro noyau réparti,
temps réel orienté pour la tolérance aux fautes, Revue Informatique et
santé, Télémédecine et e-santé, 13, pp. 63–69.
53. Gineste, L (2002) Plate-forme de suivi à distance d’arythmie cardiaques,
DEA CSTI report, University of Blaise Pascal Clermont-Ferrand II, France.
54. Palau, P. Hou, K. -M. and Ponsonnaille, J. (2002) TelmedTCP: Protocole
TCP/IP temps réel dédié à la télémédecine, Revue Informatique et santé,
Télémédecine et e-santé, 13, pp. 53–60.
55. PhysioNet (2002) The WFDB Software Package Software for View-
ing, Analyzing, and Creating Recordings of Physiologic Signals,
http://www.physionet.org/physiotools/ wfdb.shtml.
56. Moody G. B. and Mark, R. G. (1990) The MIT-BIH Arrhythmia Database
on CD-ROM and Software for Use With It, Computers in Cardiology 17,
pp. 185–188.
57. Zhou, H. Y. Hou, K. -M. Ponsonnaille, J. et al. (2004) Remote Continuous
Cardiac Arrhythmias Detection and Monitoring, in Proceedings of the 2nd
International Conference on E-health in Common Europe.
References 73
58. Dotsinsky I. A., and Stoyanov, T. V. (2004) Ventricular Beat Detection

in Single Channel Electrocardiograms, BioMedical Engineering OnLine
3(3), doi:10.1186/1475-925X-33.
59. Afonso, V. X. Tompkins, W. J. Nguyen T. Q. and Luo S. (1999) ECG
Beat Detection Using Filter Banks, IEEE Transactions on Biomedical
Engineering, 46, pp. 192–202.
60. Poli, R. Cagnoni, S. and Valli G. (1995) Genetic Design of Optimum
Linear and Nonlinear QRS Detectors, IEEE Transactions on Biomedical
Engineering 42, pp. 1137–1141.
61. Kaiser W. and Findeis M. (2000) Novel Signal Processing Methods for
Exercise ECG, Special Issue on Electrocardiography in Ischemic Heart
Disease, in Proceedings of IJBEM.
62. Datex-Ohmeda Corp. (2002) Bedside Arrhythmia Monitoring Quick
Guide, Internal Web Journal for Medical Professionals, http://
www.clinicalwindow.com.
63. Millet, J. Perez, M. Joseph, G. Mocholi A. and Chorro J. (1997) Previous
Identification of QRS Onset and Offset Is Not Essential for Classifying
QRS Complex in a Single Lead, Computers in Cardiology, 24, pp. 299–302.
64. Zhou, H. Y. Hou, K. -M. de Vaulx C. and Zuo, D. C. (2009) A Hybrid
Embedded Real-Time Operating System for Wireless Sensor Networks,
Journal of Networks, 4(6), pp. 428–435.
65. Ding, H. Hou, K. M. Lecoq J. et al. (2009) Toward a Low Cost and
Single Chip Holter: SoC-Holter, in Proceedings of the 3rd International
Conference on New Technology, Mobility, and Security-2009.
Chapter 4
Human Bio-Kinematic Monitoring with

Body Area Networks
Roozbeh Jafari, Hassan Ghasemzadeh, Eric Guenterberg,

Vitali Loseu, and Sarah Ostadabas
Embedded Systems and Signal Processing Laboratory, Department of Electrical
Engineering, University of Texas at Dallas
Richardson, TX, 75080, USA
[email protected]; [email protected]; [email protected];
[email protected]; and [email protected]
Body Area Networks (BANs), known as enabling technology for

many biomedical applications, are composed of body-worn sensor
devices that can provide mobile and continuous monitoring of
the human body. This chapter presents an overview of platform
design strategies for BANs with applications in physical movement
monitoring. First, an introduction to several compelling applications
is given, which shows BAN versatility in both medical and
recreational fields. Applications are important in the sense that they
help in understanding design requirements of the BAN platform.
An architecture of the system, including hardware and software
components, is then described. It is followed by a description of
typical signal processing for movement monitoring applications.
While this type of signal processing flow can be generally used
in movement monitoring applications, in order to take the most

www.panstanford.com
76 Human Bio-Kinematic Monitoring with Body Area Networks
advantage of the system, the signal processing needs to be custom

tailored for each individual application. An example of this process
is shown based on the Hidden Markov Model (HMM) movement
annotation applications. Finally, the chapter is concluded with
discussion of possible BAN system optimizations. In particular, it is
shown that the energy consumption of the system can be reduced by
using buffers to decrease the number of transmissions.
4.1 Physical Movement Monitoring
A BAN system uses several sensor units each equipped with motion
sensors, processing units, and wireless and memory components.
The system aims to collect sensor readings and transform them into
useful information. In order for a wearable system to be successful,
it is required to be comfortable and not introduce any additional
movement constraints. The main deployment objective of any BAN
is to improve its wearability and the ease of use. This idea has a
few consequences. First, it forces sensor units to be powered by
a battery. Using a large energy source would hinder mobility and,
therefore, is not acceptable. Second, wearability concerns rule out
a wired communication scheme because a set of wires all over the
body is not natural and may constrain some human movements.
Finally, reducing the form factor of sensor units is crucial for
wearability. It can be done via reducing the size of the sensor
unit’s components such as the processor, memories, sensors, and
the battery. While technology made a wide leap in reducing the size
of microprocessors, memories, and sensors, the progress in battery
size reduction has not been as fertile [1]. It suggests that currently
the battery size dominates the form factor of the wearable sensor
units. The requirement of the extreme battery efficiency motivates
the need for light-weight and highly efficient signal processing
algorithms and optimization techniques. The signal processing,
however, needs to exhibit sufficient reliability and sensitivity in
extracting the relevant parameters.
BAN systems are desirable because they provide objective,
quantitative measurements while not restricted to a laboratory
environment. Furthermore, depending on the number and locations
Applications 77
of the sensing units, BAN systems can vary greatly in the scope of
possible tasks from a general action recognition to extracting a very
specific detail about a movement. This property makes BAN systems
extremely useful in a large set of applications. In particular, they
find applications in rehabilitation [2], sports medicine [3], geriatric
care [4], gait analysis [5], balance evaluation, [6] and sports training
[7, 8].
4.2 Applications
BANs can collect a wide range of information that monitor different

aspects of human life from movement monitoring and emotion
recognition [9, 10] to sports training and leisure applications.
Applications can be divided into two categories, including medical
and nonmedical applications. With traditional health care, patients
are either observed by doctors who rely on personal experience
to identify symptoms and the severity of a condition or have to
be examined in a laboratory environment. With wearable mobile
sensors, however, remote and continuous monitoring the patients
is available. Sports training is another important application area of
BANs. Quantitative feedback provided by wearable sensors can be
used to provide atheletes with better instruction on improving their
game without an involved participation from a human expert trainer.
4.2.1 Medical Applications

An emerging application for BANs involves their use in medical
application. BAN area revolutionizes the health care system by al-
lowing inexpensive, continuous, and ambulatory health monitoring
with real-time capabilities. Using human bio-kinematic monitoring
that provides sensor data as a biometric of body functionality is
of increasing interest since it is noninvasive and can be measured
without subject contact or knowledge. In clinical application, as
therapists increase their emphasis on evidence-based practice,
they must increase the use of objective, quantitative methods to
demonstrate efficacy of their method. For instant, gait analysis is
an proper method for demonstrating change from treatment or
from disease progression. Gait analysis by means of BANs for the

person with Parkinsons disease (PD) can be used as a tool for
determining the treatment efficacy of pharmacologic, surgical, or
physical therapeutic interventions. In this section, we review some
of the existence methods for gait analysis and Parkinson’s disease
assessment.
4.2.1.1 Gait analysis

In the human action recognition, gait analysis is the systematic
measurement, description, and assessment of those quantities
thought to characterize human locomotion. Through gait analysis,
kinematic and kinetic data are acquired and analyzed to provide
information, which describes fundamental gait characteristics and
which is ultimately interpreted by clinicians to form a medical
assessment.
For the purpose of gait analysis, wearable sensors of the BAN
can be attached at the externor side of the thigh [11]. The hip angle
is defined as the angle between the thigh and gravity direction.
The swing velocity (angular velocity) of the thigh is defined as
v = dθ dt
. Kalman filter is applied to estimate θ and v, which are
key features of the gait cycle. Another proposed method for gait
analysis is a generic algorithm for temporal parameter extraction
called the hidden Markov event model based on HMMs. This method
constrains the state structure to facilitate location of key events of
gait [12].
In the context of signal processing, gait recognition experiments
using spectral features in terms of magnitude, phase, and phase-
weighted magnitude show that both magnitude and phase spectra
are effective gait signatures. In [13], authors proposed a gait
recognition approach using spectral features of horizontal and
vertical movement of ankles in a normal walk. They used an
integration of magnitude and phase spectra for gait recognition
using AdaBoost classifier. At each round, a weak classifier evaluates
each magnitude and phase spectra of a motion signal as dependent
subfeatures, then classification results of each subfeature are
normalized and summed for the final hypothesis output. Authors
in [14] investigate an ear- worn sensor for the development of a
Applications 79
gait analysis framework. Instead of explicitly defining gait features

that indicate injury or impairment, an automatic method of feature
extraction and selection is proposed.
4.2.1.2 Parkinson’s disease assessment systems

Many efforts have been made to come up with objective measures
for motor impairment and disability assessment in Parkinson’s
disease. Hoehn and Yahr Clinical staging [15], Webster Rating scale
[16], Columbia University Rating scale [17], the short Parkinson’s
evaluation scale (SPES) [18] and the Unified Parkinson’s Disease
Rating Scale (UPDRS) [19] are a few of them. The UPDRS has become
the gold standard of rating scales for PD detection and assessment
[20], as it is one of the most evaluated and reliable, bearing close
correlation to other widely accepted standards at the same time
[21].
Systems have been designed for PD assessment as described in
[22] and [23] albeit for specific activities like recording tremor and
measuring sleep disorder [24]. In [25], Homann et al. examined
a method to measure bradykinesia and akinesia using finger-tap
test on a computer. Rajaraman et al. [26] demonstrated a 3D
tremor measurement system for comparing tremor across patients
and in measuring the efficacy of therapeutic interventions. Their
system employed 3D electromagnetic position sensors to measure
the actual and cumulative displacement of the tremoring finger.
The tremor frequency and amplitude are estimated using spectral
analysis on the data after low-pass filtering at 15 Hz. Systems for
clinical measurement of tremor have also been demonstrated in
[27] and [28], but unrelated to the assessment of any disease in
particular.
The methods used for the measurements have varied from
electromyography (EMG) [29, 30] to accelerometers [23, 31],
electromagnetic sensors, [26] and custom transducers [22]. Hassan
et al. [32] examines light-weight sensor devices which enable on-
body and mobile health-care monitoring. Goetze et al. [20] employed
an at-home testing device for testing motor impairment in PD,
which consisted of a dedicated apparatus for conducting the tests.
The device consisted of a testing panel on the base that contains
a two-key keyboard for finger tapping, two buttons for reaction

time/movement time and repetitive hand tapping assessments, an
eight-peg pegboard, control buttons, and a docking station for an
actiwatch device measuring tremor.
4.2.2 Sports Training Application

Sport training represents the body’s adaptation to conditions of
certain exercises. One can achieve considerable progress in a sport
with the aid of appropriate exercises and training methods [33]. The
biggest challenges of sports training is the evaluation of a movement
and an ability to provide feedback regarding movement’s quality.
Traditionally, these two tasks are performed by a human coach who
uses personal expertise and experience to evaluate exercises and
provide feedback. However, personal coaches are expensive and,
to make training more affordable, often coach multiple students
at the same time, thus diverting attention from detailed diagnosis
of problems of individual students. An automated system that can
assess the overall performance of a learner and pinpoint problem
areas in the learners movements would facilitate performance
assessment, increasing the effectiveness of unsupervised practice.
Traditionally, mechanical training systems [34, 35] have been
used to provide movement feedback. However, these types of
systems lack fine grain details of movements. Authors in [36] use
a motion capture system to record and analyze tennis movements.
They show that alternative forehand and backhand movements
outperform discrete forehand or backhand practices due to the
inertia of the trunk rotation movements between subsequent
strokes. The approach in [37] tracks the movements in rugby
football using close-range or long-range photography. Shoulder
orientation and trunk flexion are demonstrated to be significantly
different in experienced players and novices and, therefore, can be
used to measure players’ performance and training quality. A motion
capture system is also used in [38] to design a virtual baseball
training system. In the system, a batter swings a bat toward the
virtual ball, and the trajectory of the swing is used to calculate the
quality of the swing. While these systems tend to perform well,
they require an expensive laboratory setup and, therefore, are not
Applications 81
desirable. A BAN-based solution to sports training is either based

on a solution where sensors are placed in the environment [39] or
embedded within the sports equipment [40].
Overall, there are two areas of interest in sports training. Training
can focus on a precise execution of each movement individually.
Training can also focus on coordination in execution of series of
movements. While these two components are connected, they can
be considered individually. Next two sections demonstrate how BAN
systems can address both of these cases.
4.2.2.1 Golf swing training

The popular sport of golf requires a complicated sequence of
motions to swing the golf club properly with the primary goal of
propelling the golf ball a certain distance in a desired direction.
A repeatable and consistent golf swing can dramatically improve
a golfer’s score. However, this single movement, which has such a
major impact on the player’s overall game, is difficult to master and
execute consistently for players who are new to the sport or have
little experience.
A full swing is a complex motion of the body aimed at accelerating
the club at great speed. The motion starts at an initial position,
referred to as the address position [41], followed by the swing.
A golf swing can be divided into smaller segments. A general
golf swing model considers a full swing composed of four major
segments: takeaway, backswing, downswing, and follow-through
[42]. Takeaway starts as the first movement after the address
position and ends when the club is approximately parallel to the
target line and at waist level. The backswing follows the takeaway
and continues until the golf club is lifted to its highest point behind
the player. Following this is the downswing in which the club is
brought back down to hit the ball. After impact with the ball, the
follow-through motion brings the club to its stopping point in front
of the player. This model enables more precise analysis of the actions
by reporting the quality of each individual part of the swing.
The goal in achieving a perfect swing is to hit the ball squarely
and straight [43]. This would also give the golfer maximum distance.
Consequently, it is important to investigate actions that prevent
development of a perfect swing. According to the literature, there

are two kinds of common mistakes new players make, resulting
in a poor shot [41]: wrist rotation and out-of-plane movements. To
simplify the system, a virtual coach can address each one of the
issues individually. For example, the system can focus on evaluating
golf swing in terms of the angle of wrist rotation. This information
can be obtained by placing sensing units on the upper body and arms
to capture significant motions during the swing [44] and placing two
sensing units on the golf club to capture its movements [7].
4.2.2.2 Baseball swing training

Baseball batting involves hitting a thrown ball with the primary
objective of transferring maximum force to propel the ball as far
as possible in a desired direction. Numerous baseball players and
coaches have suggested methods for successful batting. A good
swing is the result of a sequence of rotational movements including
foot, knees, hips, shoulder, and hands movements. Generally, the
action of the batter starts in the lower body and moves upward.
Properly performed motions executed at the right time maximize the
power of the swing. Major components of a good swing include bat
speed, bat swing plane, and timing. Common mistakes include late
rotation of lower body, back shoulder dip, and drifting of the front
foot. Late movement of the foot and hips impairs the swing timing.
Dropping the back shoulder affects the bat plane so as the bat does
not pass through the strike zone horizontally, decreasing the chance
of a successful hit. Drifting refers to improper weight transfer from
the back foot to the front foot. One consequence is losing power in
the hips, which decreases the bat speed at impact. Therefore, proper
weight transfer necessitates coordination between different body
segments during the swing.
The considered swing model emphasizes three major events: 1)
Rotation of the lower body (feet, knees, hips) toward the pitcher,
2) rotation of the upper body into the swing, and 3) the swing of
the arms and hands toward the pitcher. These key events should be
executed in a specific and overlapping sequence. The coordination
is extremely important as it ensures that the maximum power from
arms, shoulders, and hips is delivered exactly as the bat crosses the
Hardware and Software Architecture 83
plate [45]. In order to recognize movement coordination, sensing

units placed on the ankle, hip, upper body, and the bat have to
identify body movements and their timings. It can be achieved
with detailed body, choreography modeling with motion transcripts
introduced in [8].
4.3 Hardware and Software Architecture
The purpose of an action recognition system is to classify

transitional movements into pre-defined actions. Given a set of
movements, the system must distinguish between every pair of
motions. These sensors capture inertial information from physical
movements. An example platform used to generate data and achieve
results in this chapter is shown Fig. 4.1. The system consists of
several sensor units; each has a tri-axial accelerometer, a bi-axial
gyroscope, a microcontroller, and a radio. In this specific setup, the
accelerometers are LIS3LV02DQ with 1024 LSb/g sensitivity. The
IDG-300 gyroscopes have 2 mV/◦ /s sensitivity. The processing unit
of each node, or mote, can sample sensor readings at certain rate and
transmit the data wirelessly to a base station. A common protocol
Figure 4.1. (a) A sensor node composed of processing unit and custom-
designed sensor board. The motion sensor board has a triaxial accelerome-
ter and a biaxial gyroscope Sensor node and subject (b) A subject wearing
nine sensor nodes. See also Color Insert.
for this transmission is Time Division Multiple Access (TDMA). For

the purpose of action recognition, authors of [46] use the TelosB
motes [47], which are commercially available from XBow R
. The
sensor board they use is predominantly a custom designed with an
integrated Li-Ion battery that powers both sensor board and mote
as shown in Fig. 4.1a. The sensor nodes can be placed on different
locations on the body to capture movements of their subjects.
Figure 4.1b shows a subject with nine nodes placed on different body
segments.
Signal processing for body sensor network usually comprises
multiple levels of data abstraction, from raw sensor data to data
calculated from processing steps. SPINE (Signal Processing in Node
Environment) [48] is a software framework that aims to enable
efficient implementations of signal processing on sensor nodes
in BAN. SPINE is a framework for distributed signal processing
founded on the following principles:
Open source: SPINE is developed as an open source project
to establish a broad community of users and developers that
contributes to extend the framework with new capabilities and
applications.
Interoperability through APIs: SPINE provides local and remote
applications with lightweight Java APIs that can be used by local
and remote applications to manage the sensor nodes or issue
service requests, and are easily portable to devices of various
capabilities, such as PCs or mobile phones, that can be used as BAN
coordinator.
High-level abstractions: SPINE provides libraries of protocols,
utilities, and processing functions; hence, it simplifies the task
of application developers by raising the level of abstraction. The
layer defined by the SPINE libraries allows designers to focus
on application-specific issues and program at a higher level of
abstraction than TinyOS.
Distributed implementations of classification algorithms: SPINE
simplifies the development of applications that require complex
signal processing algorithms and classifiers. For example, SPINE
supports distributed implementation of classification algorithms
to reduce the amount of data to be transmitted and save
energy.
Signal Processing for Body Area Networks 85
Figure 4.2. SPINE 1.0 Network Architecture [48].
SPINE relies on a BAN architecture with star topology, including

one or more sensor nodes and a BAN coordinator node. The
coordinator typically manages the BAN, collects and analyzes the
data received from the sensor nodes, and acts as a gateway to
connect the BAN with wide area networks for remote data access.
Figure 4.2 visualizes a typical architecture with star topology.
4.4 Signal Processing for Body Area Networks
Figure 4.3 shows several processing tasks typically used for signal
processing and action recognition in BANs. Each processing block is
described as follows.
Sensor data collection: Data is collected from all of the sensors
on each of the nodes at a specified frequency. The sampling rate
can be empirically chosen to provide sufficient resolution while
compensating for bandwidth constraints of the system [49], or it
can be determined to satisfy the Nyquist criterion [50]. Usually,
a 20 samples per seconds would provide fine details of human
movements [51].
Preprocessing: Data is filtered with a small window moving
average to remove high frequency noise. The number of points used
to average the signal can be chosen by examining the power spectral
density of the signals. The filter is required to remove unnecessary
motions (e.g., tremors) while maintaining significant data. With
Figure 4.3. Signal processing flow.
these objectives, authors in [52] test several moving average filters

with varying window sizes and choose the filter that best satisfies
the aforementioned requirements.
Segmentation and annotation: Segmentation algorithms divide
continuous data streams into discrete time intervals of the type
expected by the information processing step, while annotation
algorithms locate and label specific events. To enable real-time
movement monitoring, an automated method is required.
Feature extraction: Statistical and morphological features are
extracted from the signal segment. For example, sensor readings
can be transformed into a set of informative attributes, including
mean, start-to-end, standard deviation, peak-to-peak amplitude,
RMS power, median, and maximum value.
Per-node classification: Each node uses the feature vector
generated during feature extraction to determine the most likely
action. Due to its simplicity and scalability, k-Nearest Neighbor (k-
NN) [53] is a widely used classifier [54].
Final classification: The final decision can be made using either
a data fusion or a decision fusion scheme. In the data fusion,
features from all sensor nodes are fed into a central classifier. The
classifier then combines the features to form a higher dimensional
feature space and classifies movements using the obtained features.
In the decision fusion, however, each sensor node makes a local
An Automatic Parameter Extraction Method Based on HMM 87
classification and transmits the result to a central classifier where

a final decision is made according to the received labels.
In Section 4.5, we focus on the human movement annotation.
Many movement models divide an action into several parts. In
sports, many swings can be divided into phases. The golf swing
is separated into takeaway, backswing, downswing, and follow-
through portions [55]. In walking, the human stride is marked
by several events such as initial stance (the foot placed on the
ground), mid-stance, initial swing (the foot has just been lifted from
the ground), and mid-swing, which repeat indefinitely [56]. These
divisions can be used directly. For instance, high standard deviation
of stride time during walking is indicative of Parkinson’s disease or
Huntington’s disease and can be used to assess the risk of falling
[57, 58]. We introduce a model based on HMMs that divides walking
into aforementioned events. Inertial sensors provide movement
data directly; therefore, these events can be found by looking for
patterns in the sensor data.
4.5 An Automatic Parameter Extraction

Method Based on HMM
Human movement models often divide movements into parts. These

parts are often divided based on key events, also called temporal
parameters. When analyzing a movement, it is important to correctly
locate these key events, and so automated techniques are needed.
There exist many methods for dividing specific actions using data
from specific sensors, but for new sensors or sensing positions, new
techniques must be developed. To address this problem, this section
introduces a generic method for temporal parameter extraction
called the Hidden Markov Event Model based on HMMs. This method
can be quickly adapted to new movements and new sensors/sensor
placements. This method is validated on a walking dataset using
inertial sensors placed on various locations on a human body. The
technique is designed to be computationally complex for training,
but computationally simple at runtime to allow deployment on
resource-constrained sensor nodes.
4.5.1 HMEM Training and Use

The Hidden Markov Event Model (HMEM) is the name of the
introduced key event labeling system, which uses an HMM with a
specific state structure and a modified training procedure designed
to find key events. The model also adds a feature selection and
model parametrization system based on Genetic Algorithms (GAs).
The HMEM makes several assumptions about the underlying data:
1) there are a number of different event types, 2) the events always
occur in a specific order and for cyclical movements they repeat,
3) every single event type is represented in every action, and 4)
there are a number of unlabeled samples between two adjacent
events.
A traditional pattern recognition technique used for time-varying
signals is the Hidden Markov Model. A basic HMM describes a
discrete-time Markov process. At a particular moment the process
is in just one state. At fixed time intervals, the process produces
an output and then transitions to another state (or remains in
the current state). The transitions and outputs are probabilistic
and based exclusively on the present state. The process generates
a sequence of outputs, and a corresponding sequence of states.
The states cannot be directly observed, and are thus hidden. An
HMM is completely described by initial state probabilities, state
transition probabilities, and output probabilities. Algorithms exist
to 1) build an HMM to describe a given set of output sequences, 2)
choose which of several models best describe an output sequence,
3) find the most likely current state of an HMM given the output
sequence up until now, and 4) the most likely state sequence
associated with a particular output sequence for a specified HMM
[59, 60].
For a general HMM, it is possible for any state to transition to
any other state. This is called an ergodic model. Another variant
is to enforce a specific ordering for the states: each state can only
transition to itself or state to the “right” of it in the ordering. This is
called a left-right model [60].
Each key event can be represented by a unique state. Ideal events
occur at a specific time but have no duration. However, given the idea
that the key event might be associated with unique features in the
Figure 4.4. HMEM model and structure. See also Color Insert.
observation sequence, the key event state should have a one-sample

duration. The HMEM encodes this concept into an HMM by removing
the self-transition from states associated with key events, forcing a
transition after one sample. The samples between key events are
represented by transition states which support both self-transitions
and forward-transitions, as seen in Fig. 4.4 states are grouped into
cohorts, which start with a key event state and end with the last
transition state before the next key event state. For any observation
sequence in the training data, the positions of the key event states
are known. This means that training each cohort independently is
identical to training the whole system at once.
4.5.2 Overview
There are several stages required to train the HMEM as shown in
Fig. 4.5.
Figure 4.5. HMEM training procedure.

4.5.2.1 Preprocessing and feature extraction

The signal is filtered with a moving average filter to remove high
frequency noise. Then it is normalized by subtracting a large-
window mean and dividing by a large-window standard deviation.
Several parameters representing the action data inside the signal,
referred to as features, are extracted at each sample. These features
are further quantized with a 10-level uniform quantizer based on
the range of the features in the training data [10].
4.5.2.2 HMM training

The HMM is effectively a finite state machine with probabilistic
transitions and certain emission probabilities. These probabilities
must be specified as part of the mathematical model defining
an HMM. The exact locations of all key event states and what
observations they emit are known from the annotations in the
training data. However, the number of transition states and their
transition and emission probabilities are unknown and must be
trained. There are several well-known techniques for training
HMMs, including Baum-Welch and Viterbi Path Counting [59, 61].
The training data is segmented using the canonical annotations.
Each cohort is trained independently using a set of segments that
start with a sample that should be labeled with the cohorts event
and end just before the next labeled event. According to this model,
the first state must be the cohort’s event state, and the last sample
must be associated with the last transition state in the cohort. During
training, it is important to make sure that all considered paths meet
this constraint. Viterbi Path Counting produces a single path for
each event that can be edited to meet the constraints if necessary,
while Baum-Welch can also be constrained in this way, VPC is much
faster, which is important given the already high training times.
This feature is one of the primary reasons for choosing VPC over
Baum-Welch. The details of the training process are discussed in
Section 4.5.3.
4.5.2.3 Parametrization and feature selection

HMMs are trained to represent a process, not to minimize
segmentation error. It is possible to explicitly attempt to increase
classification accuracy by choosing model parameters with that goal

in mind. A genetic algorithm with uniform crossover is used to train
the model. The population fitness is evaluated using the training
model, and then at the end, the model that gives the best results for
the cross-validation data is selected. Further discussion follows in
Section 4.5.4.
4.5.3 HMM Training and the Viterbi Algorithm

A Markov process has N states S = {s1 , s2 , · · · , s N }, and can emit
M observations X = {x1 , x2 , · · · , x M }. For a given observation
sequence O T = (o1 , o2 , · · · , oT ) with T observations, there is a
corresponding state sequence Q (T ) = (q1 , q2 , · · · , qT ). The HMM
λ = {π, A, B} is defined by three sets of probabilities: initial state
probabilities π = {πi |πi = P (qo = si )}, state transition probabilities
A = {ai j |ai j = P (q = s j |q prev = si )}, and observation probabilities
B = {b j (k)|b j (k) = P (o = xk |q = s j )}.
The most common training algorithm is the Baum-Welch algo-
rithm [59], however a newer algorithm, the Viterbi Path Counting
(VPC) algorithm is more appropriate for this work [61]. Both
algorithms follow the training procedure shown in Algorithm 4.1.
They start with a fixed number of states and an initial set of model
parameters, then extract probabilistically weighted state sequences
Algorithm 4.1 HMM Training Procedure

Require: λ0 , ˆ = {O (T 1
1)
, O (T
2
2)
, · · · , O (T
Y
Y)
}
1: λ ← λ0
2: for i ← 1 to K do {estimates from ˜}
3: ˜←∅
4: for all O (T ) ∈ ˆ do
5: Q (T ) ← collect weighted state sequences using λ
6: ˜ ← ˜ ∪ Q (T )
7: end for
8: π̄i = number of times in state si at time 1
number of times at time 1
number of transitions from state si to state s j
9: āi j = number of transitions from state si
number of times in state s j observing symbol xk
10: b̄ j (k) = number of times in state s j
11: λ ← {π̄, Ā, B̄}
12: end for
using the model parameters. Next, the transition and emission

probabilities are updated based on the transitions and observations
associated with each state sequence. The initial model is updated
with these new probabilities. This process repeats until some
desired level of convergence is reached. It will implicitly converge
to a local minima.
The key to VPC is extracting the most likely state sequence.

Q (T ) max = arg max P Q (T ) , O (T ) (4.1)
Q (T ) ∈S T
A dynamic programing algorithm, called the Viterbi algorithm [59],

solves this problem. Using the most likely state sequence extracted
using the Viterbi algorithm, the transition and emission probabilities
are found simply by counting the occurrences in all the most
likely state sequences for each observation sequence in the training
set. Since every conhort is trained individually, each observation
sequence is a sequence starting on the key event and ending right
before the next key event.
Maximizing the probability is equivalent to maximizing the log
probability; therefore, we use log probabilities to prevent numerical
underflow and facilitate faster computing. The order of the Viterbi
algorithm for a left-right model with independent features is
O (Viterbi) = O (T x N ) · O (P rest ), where O (P rest ) is the order of
algorithm required to estimate probability. This order is constant
time for the state transition probability, but based on the number
of features for the observation probability estimation. This means
O (Viterbi) = O (T x N x| F |).
4.5.4 Feature Selection and Model Parametrization Using

Genetic Algorithms
The choice of whether or not to include each feature and the choice
of the number of transition states for the cohorts are all tunable
parameters of the HMEM. The feature selection = {ψk |ψk ∈
{0, 1}, i = 1, . . . , |F |} represents a choice of which features out of
an exhaustive list are to be included and which are to be discarded.

The number of selected features is || = ψk . Feature k is selected
k
if ψk = 1 and is discarded if ψk = 0. The other parameter is
the number of transitions states for each cohort, = {ωe |ωe ∈

{1, . . . , 5}, e = 1 . . . E }, where E is the number of key events. The
full HMEM model is represented by λ H M E M = {λ H M M , , }.
In essence, parametrization consists of choosing several good
models based on one or more objective functions applied to the
training set. These models are then compared against each other
using the same objective function(s) applied to the cross-validation
set. The best model on the cross-validation set is chosen. Because
the cross-validation set exclusively contains data from subjects not
in the training set, generalizability of the models to new subjects is
improved. Genetic algorithms are used to generate the list of “good”
models.
4.5.5 HMEM Application Procedure

After the HMEM is trained, it can be used to find key events in
a data stream for the movement it has been trained on. The data
flow for the algorithm is shown in Fig. 4.6. First, the data is filtered
using the procedure described above, then features are extracted
and quantized. Next, the feature selection is applied, and finally the
most likely state sequence is extracted using the Viterbi algorithm.
The annotation converter finds all the event states in this sequence
and outputs an ordered set where each element consists of a time
and event label.
4.5.6 Experimental Analysis

The results discussed in this section are based on an experiment
reported in [62] and the node placement as described in Section 4.3.
Figure 4.6. Application of the HMEM as described in Section 1.5.5.

The results are reported with precision (P), recall (R), Quality
(RMSE). The first and last annotations were ignored because the
algorithm needs context to determine annotations, and we are
interested in the steady-state performance only. We show error for
each mote using just the accelerometer readings, look at per-subject
error for a poor performing sensor node, and a well-performing
sensor node.
4.5.6.1 Examination of per-subject error

One of the goals of the HMEM system is good generalization to
new subjects. Table 4.1 shows per-subject error for the sensor on
the right thigh. Initially subjects two to four were in training, five
to seven in cross validation, and eight and nine in test. However,
subjects five and seven have walking patterns that differ significantly
from the others, but are similar to each other. Therefore, subjects
four and five were exchanged. It is likely that with a larger dataset
the system could generalize better to such subjects. All the results
are shown from the portion of the subjects’ data that was in the test
dataset.
The sensor on the right thigh, as shown in Table 4.1, performs
well. Subjects eight and nine perform a little worse than subjects in
the training and cross-validation sets. The worst per-subject error
comes from subjects five and seven. The reason for this is not
entirely clear. The use of the GA does not significantly reduce the
discrepancy in per-subject error. Since the final selection criteria
Table 4.1. Subject error for R thigh with Accel and TP

Default (132 Features) GA (38 Features)
Subject P R RMSE P R RMSE
Sub 2 Train 100 100 1.44 100 100 1.62
Sub 3 Train 100 100 1.33 100 100 1.11
Sub 4 Cross 100 100 1.36 100 100 1.38
Sub 5 Train 100 100 3.54 100 100 3.43
Sub 6 Cross 100 100 1.03 100 100 1.02
Sub 7 Cross 100 100 3.25 100 100 3.09
Sub 8 Test 100 100 1.74 100 100 1.76
Sub 9 Test 100 100 1.55 100 100 1.66
System Optimizations 95
Table 4.2. Cross-validated subject error (R thigh)

Subject P R RMSE Fsel
Sub 2 100 100 1.47 6.7

Sub 3 100 100 1.38 8.7
Sub 4 100 100 1.68 4.7
Sub 5 99.8 99.8 3.55 4.7
Sub 6 100 100 1.20 6.7
Sub 7 100 100 3.09 28.7
Sub 8 100 100 1.52 10.7
Sub 9 100 100 1.59 16.0
for the solution is minimum total error, not minimum worst-case

subject error, this is not surprising.
Manual partitioning of the data into training, cross-validation,
and testing sets can artificially bias the results. Therefore, we
performed an experiment for each subject where the subject was
placed exclusively in the testing set, and the training and cross-
validation sets were selected randomly from the remaining subjects.
The results shown in Table 4.2 are the average of three tests
after the GA. These results demonstrate that the model has good
generalization to many subjects, but performs poorly on some.
4.6 System Optimizations
In Section 4.1 we discussed some of the constraints of the BAN

systems. It was pointed out that energy constraint dominates the
wearability of BAN systems. In general, energy can be used for local
computation and communication. Previous studies of embedded
sensor nodes have shown that data communication is expensive in
terms of energy consumption, whereas data processing is relatively
inexpensive [63].
The energy cost of the system can be improved in multiple ways.
First, communication protocols used in embedded devices tend to be
very general. Protocols, such as ZigBee [64], do not take advantage of
the BAN specific features such as node proximity, data requirements,
and signal processing flow. For example, due to proximity of the
sensor units on the body, it may not be required for the radio to
operate at full range. Sensors nodes can be optimized to operate at

the minimum required range to save energy. Second, not all of the
sensor nodes may be required to contribute to a given computation.
If that is the case, then the system can be optimized to select only
required nodes at any given time. Finally, the signal processing
technique can explicitly address the energy problem by selecting
less expansive operations. In the following section, we describe how
in the case of applications that do not have hard data deadlines it
is possible to optimize energy expenditure of the communication
system.
4.6.1 Burst Communication

Most BAN applications have specific signal processing requirements
as can be seen in Section 4.2. The typical signal processing flow
involves several processing blocks, which may require collaboration
among the sensing units. Intensive signal processing may require
a lot of inter-node communication. Frequent inter-node commu-
nication leads to increased energy consumption. Therefore, if an
application does not have a strict deadline, transmissions in short
bursts at a high bit rate can help minimize transmission energy per
bit [65].
4.6.1.1 Task graph

The signal processing model described earlier is usually imple-
mented in a distributed manner over a BAN. Information flow
across the network can be represented with a task graph model.
In a system of n sensor nodes, task graph is composed of n
subgraphs connected through inter-node links. Each subgraph
represents static information dependencies within a node as shown
in Fig. 4.7, where uiR , uiS , uiF , and uiC denote sensor reading and
Figure 4.7. Intra-node task subgraph.

preprocessing, segmentation, feature extraction, and classification

blocks, respectively.
Given a set of n sensor nodes, intra-node task subgraph for
node si is defined by G i = (Vi , E i ), where Vi is the set of four
vertices and E i is the set of four edges. Each vertex, denoted by
μ
ui , corresponds to a processing unit, and each edge is denoted by
μη
eii (μ, η ∈ {R, S, F , C }) representing intra-node dependencies.
Inter-node links that represent dependencies across sensor nodes
are used to connect subgraphs and form a task graph. An inter-
node link is determined according to dependencies induced by
application.
Given a set of n sensor nodes s1 , . . . , sn and inter-node dependen-
cies, task graph G = (V , E ) is formed by connecting n intra-node
task subgraphs G 1 , . . . , G n through dependency links E b defined by
application criteria. The set of edges E is given by
E = Ew ∪ Eb (4.2)
where the set of intra-node edges E w is given by

n
Ew = Ei i = 1, ..., n (4.3)
i =1
and the set of vertices V is defined by

n
V = Vi i = 1, ..., n (4.4)
i =1
where each Vi is the set of vertices within subgraph G i .

As an abstract example, Fig. 4.8 shows a task graph for a network
of two sensor nodes where processing units are connected through
five inter-node dependency links.
Figure 4.8. Sample task graph with five internode links.

In the context of the task graph, the idea behind using buffers
is to transmit the maximum amount of data on the inter-node links
in short time intervals. The large number of data blocks produced
by each processing unit is stored locally and is transmitted using
available bandwidth. This would conform to real situations of health
care systems where physicians are interested in receiving reports
on daily activities rather than immediate reports. By assuming no
immediate deadlines in the system, individual buffers on each link
are maintained and the data blocks are transmitted separately for
each inter-node link. The optimization problem can be defined as
follows:
Optimization Problem: Given task graph G , each inter-node link
μη μη
ei j is associated with a number xi j denoting the number of actions
μ
for which data blocks produced by the source unit ui are buffered
μη
prior to every transmission. The objective is to find values xi j
that minimize the number of transmissions subject to memory
constraints on nodes.
4.6.1.2 Problem formulation

μη
Given the task graph G as described earlier, let ai j be a binary that
represents existence of inter-node links and is given by

μη 1, if si and s j are dependent through uiμ and uηj
ai j = (4.5)
0, otherwise
μ
and let di be another binary denoting whether or not a processing
unit is slower than its predecessor.
μ μ−1
μ 1, if ri < ri
di = (4.6)
0, otherwise
The number of actions A occurred at action rate k during time period
T is given by
A =k×T (4.7)
and the number of transmissions can be calculated by

μη
ai j
Z = A μη (4.8)
i, j
xi j
The total size of the intra-node buffers Wi for node si is given by

μ
Wi = bi (4.9)
μ
and the total size of the inter-node buffers for node si is determined
by
μη μ μη ημ η ημ
Bi = (max(ai j bi xi j ) + max(a j i b j x j i )
j,η j,η
μ
ημ μ−1 ημ
+ max(a j i b j x j i )) (4.10)
j,η
and the total size of delay buffers for node si is given by

μ μ−1 μ μ−1
ημ η
Di = T di ((ri − ri )bi + aji bj ) (4.11)
μ j,η
Let M i be the size of the memory on node si . The problem of

minimizing the number of transmissions can be formulated as a
convex optimization problem as follows:
Min Z (4.12)
subject to
Wi + Bi + D i ≤ M i ∀i ∈ {1, ..., n} (4.13)
μη
xi j ∈ Z + ∀i, j, μ, η (4.14)
The non-linear constraints due to the max functions in Eq. (4.10)
can be transformed into linear equations by expanding every
function over all values taken by the function. The integrality
condition (4.14) can be relaxed to solve the problem using common
convex programming tools.
μη
xi j > 0 ∀i, j, μ, η (4.15)
The solution obtained due to integrality relaxation will not carry
the optimality condition, but it is possible to find a lower bound
on the size of memory for which the result is optimal. For each
sensor node si , the integer relaxed convex optimization finds optimal
solutions for memories of size
μ μ μ
Mi − (1 − ε)αi βi γi (4.16)
μ
where
μ μη μ μη μ μη
αi = ai j bi s.t. j, η = arg max(ai j bi xi j )
j,η
μ ημ η ημ η ημ
βi = a j i b j s.t. j, η = arg max(a j i b j x j i ) (4.17)
j,η
μ ημ μ−1 ημ μ−1 ημ
γi = aji bj s.t. j, η = arg max(a j i b j x j i )
j,η
Inter-node buffers assigned to each processing unit can have at

most three components as given in Eq. (4.10). Integer relaxation
μη
would increase size of each optimal buffer associated with xi j by
μ μ μ
factor 1 − ε of the unit data (αi ,βi ,γi ).
4.6.1.3 Experimental results

The results discussed in this section are based on the experiment
reported in [66]. During the experiment three communication
networks are considered. Each network is based on a particular
signal processing requirements, which define specific inter- and
intra-connection networks. For each network, Table 4.3 shows the
number of network transmissions required, including the number
of packets exchanged with and without using the buffer approach
and highlights the improvement of applying buffers.
Based on the results in Table 4.3, a few conclusions can be made.
In addition to communication simplification, the approach can
reduce the communication coordination and enhance the system
lifetime. Since the considered metric is based on the number of
packets, it is independent of the communication protocol. However,
the choice of packet size is important for the system. A larger
Table 4.3. Number of transmissions for different communication net-
works
# Packets
Configuration Action Rate (H z) Without Buffer With Buffer Improvement
1.8 185760 74485 59%
1
0.0167 1440 80 94%
1.8 46440 31226 32%
2
0.0167 1440 153 89%
1.8 69660 37297 46%
3
0.0167 1440 17 98%
References 101
packet size would enhance energy per bit transmitted ratio. Finally,
the discussed solution assumes that buffers are preallocated for
a particular implementation. An approach that would dynamically
update the buffer size to accommodate for the system requirements
at any given time would be potentially more beneficial.
References
1. S. Park, A. Savvides, and M. Srivastava (2001) Battery capacity

measurement and analysis using lithium coin cell battery, in Proceedings
of the 2001 international symposium on Low power electronics and
design, pp. 387, ACM.
2. V. Jones, R. Bults, D. Konstantas, and P. Vierhout (2001) Healthcare PANs:
personal area networks for trauma care and home care, in Proceedings
of the Fourth International Symposium on Wireless Personal Multimedia
Communications (WPMC), pp. 9–12.
3. O. Bott, M. Marschollek, J. Bergmann, K. Wolf, U. Tegtbur, and R. Haux
(2007) Sensor-enhanced health information system architectures for
home and telecare: concept and prototype, in European Conference on
eHealth, 118, pp. 193–203.
4. S. Po, G. Dagang, M. Hapipi, N. Naing, W. Shen, A. Ongkodjojo, and F. Hock
(2006) Overview of MEMSWear II-incorporating MEMS technology into
smart shirt for geriatric care, Journal of Physics: Conference Series, 34,
pp. 1079–1085, Institute of Physics Publishing.
5. E. Jovanov, A. Milenkovic, C. Otto, and P. De Groen (2005) A wireless
body area network of intelligent motion sensors for computer assisted
physical rehabilitation, Journal of NeuroEngineering and Rehabilitation,
2(1), pp. 6.
6. R. Ramachandran, L. Ramanna, H. Ghasemzadeh, G. Pradhan, R. Jafari,
and B. Prabhakaran (2008) Body sensor networks to evaluate standing
balance: interpreting muscular activities based on inertial sensors,
inProceedings of the second International Workshop on Systems and
Networking Support for Health Care and Assisted Living Environments,
pp. 4, ACM.
7. H. Ghasemzadeh and R. Jafari (2009) sport training using body sensor
networks: a statistical approach to measure wrist rotation for golf
swing, in The Fourth International Conference on Body Area Networks
(BodyNets 09), Los Angeles, CA.
8. H. Ghasemzadeh and R. Jafari (2010) Body sensor networks for

baseball swing training: coordination analysis of human movements
using motion transcripts, in Proceedings of the Eight IEEE International
Conference on Pervasive Computing and Communications Workshops,
pp. 792–795.
9. A. Van Halteren, R. Bults, K. Wac, D. Konstantas, I. Widya, N. Dokovsky,
G. Koprinkov, V. Jones, and R. Herzog (2004) Mobile patient monitoring:
The mobihealth system, The Journal on Information Technology in
Healthcare, 2(5), pp. 365–373.
10. K. Kim, S. Bang, and S. Kim (2004) Emotion recognition system using
short-term monitoring of physiological signals, Medical and Biological
Engineering and Computing, 42(3), pp. 419–427.
11. L. Dong, J. Wu, X. Bao, and W. Xiao (2006) Extraction of gait features
using a wireless body sensor network (BSN), in Proceedings of the
Sixth International Conference on ITS Telecommunications, pp. 987–
991.
12. E. Guenterberg, A. Yang, H. Ghasemzadeh, R. Jafari, R. Bajcsy, and
S. Sastry (2009) A method for extracting temporal parameters based on
hidden markov models in body sensor networks with inertial sensors,
IEEE Transactions on Information Technology in Biomedicine, 13(6),
pp. 1019, (2009).
13. A. Lie, R. Shimomoto, S. Sakaguchi, T. Ishimura, S. Enokida, T. Wada, and
T. Ejima (2005) Gait recognition using spectral features of foot motion,
Lecture Notes in Computer Science, 3546, pp. 767–776.
14. L. Atallah, O. Aziz, B. Lo, and G. Yang (2009) Detecting walking gait
impairment with an ear-worn sensor, in Proceedings of the 2009 Sixth
International Workshop on Wearable and Implantable Body Sensor
Networks 00, pp. 175–180, IEEE Computer Society.
15. M. Hoehn and M. Yahr (1967) Parkinsonism: onset, progression and
mortality, Neurology, 17(5), pp. 427–442.
16. D. Webster (1968) Critical analysis of the disability in Parkinson’s
disease, Modern Treatment, 5(2), pp. 257.
17. G. Montgomery, N. Reynolds Jr., and R. Warren (1985) Qualitative as-
sessment of parkinson’s disease: study of reliability and data reduction
with an abbreviated columbia scale., Clinical Neuropharmacology, 8(1),
pp. 83.
18. J. Rabey, H. Bass, U. Bonuccelli, D. Brooks, P. Klotz, A. Korczyn, P. Kraus,
P. Martinez-Martin, P. Morrish, W. Van Sauten, et al. (1997) Evaluation
of the short parkinson’s evaluation scale: a new friendly scale for
References 103
the evaluation of Parkinson’s disease in clinical drug trials, Clinical

Neuropharmacology, 20(4), pp. 322–337.
19. S. Fahn and R. Elton, Members of the UPDRS Development Committee
(1987) Unified Parkinson’s disease rating scale, Recent Developments in
Parkinson’s Disease, 2, pp. 153–163.
20. C. Goetz, G. Stebbins, D. Wolff, W. DeLeeuw, H. Bronte-Stewart, R. Elble,
M. Hallett, J. Nutt, L. Ramig, T. Sanger, et al. (2009) Testing objective
measures of motor impairment in early Parkinson’s disease: Feasibility
study of an at-home testing device Potential conflict of interest: Nothing
to report., Movement Disorders, 24(4), pp. 551–556.
21. C. Ramaker, J. Marinus, A. Stiggelbout, and B. Van Hilten (2002)
Systematic evaluation of rating scales for impairment and disability in
Parkinson’s disease, Movement Disorders, 17(5), pp. 867–876.
22. J. Ackmann, A. Sances, S. Larson, and J. Baker (1977) Quantitative
evaluation of long-term Parkinson tremor, IEEE Transactions on
Biomedical Engineering, pp. 49–56.
23. E. Van Someren, B. Vonk, W. Thijssen, J. Speelman, P. Schuurman,
M. Mirmiran, and D. Swaab (1998) A new actigraph for long-term
registration of the duration and intensity of tremor and movement, IEEE
Transactions on Biomedical Engineering, 45(3), pp. 386–395.
24. B. Van Hilten, J. Hoff, H. Middelkoop, E. Van der Velde, G. Kerkhof,
A. Wauquier, H. Kamphuisen, and R. Roos (1994) Sleep disruption
in Parkinson’s disease. Assessment by continuous activity monitoring,
Archives of Neurology, 51(9), pp. 922–928.
25. C. Homann, K. Suppan, K. Wenzel, G. Giovannoni, G. Ivanic, S. Horner,
E. Ott, and H. Hartung (2000) The bradykinesia akinesia incoordination
test (BRAIN TEST),c an objective and user-friendly means to evaluate
patients with Parkinsonism, Movement Disorders, 15(4), pp. 641–647.
26. V. Rajaraman, D. Jack, S. Adamovich, W. Hening, J. Sage, and H. Poizner,
(2000) A novel quantitative method for 3D measurement of Parkin-
sonian tremor, Clinical Neurophysiology, 111(2), pp. 338–343.
27. P. Bain, (1998) Clinical measurement of tremor, Movement Disorders, 13,
pp. 77–80.
28. J. Matsumoto, D. Dodick, L. Stevens, R. Newman, P. Caskey, and
W. Fjerstad (1999) Three-dimensional measurement of essential
tremor, Movement Disorders, 14(2), pp. 288-294.
29. G. Deuschl, P. Bain, and M. Brin (1998) Consensus statement of
the Movement Disorder Society on tremor, Movement Disorders,
13, pp. 2–23.
30. D. Wright, K. Nakamura, T. Maeda, K. Kutsuzawa, K. Miyawaki, and

K. Nagata (2008) Research and development of a portable device
to quantify muscle tone in patients with Parkinson’s disease, in
Proceedings of the 30th IEEE Annual International Conference of the
Engineering in Medicine and Biology Society, pp. 2825–2827.
31. P. Bain, L. Findley, P. Atchison, M. Behari, M. Vidailhet, M. Gresty,
J. Rothwell, P. Thompson, and C. Marsden (1993) Assessing tremor
severity, British Medical Journal, 56(8), pp. 868–873.
32. H. Ghasemzadeh, E. Guenterberg, and R. Jafari (2009) Energy-efficient
information-driven coverage for physical movement monitoring in body
sensor networks, IEEE Journal on Selected Areas in Communications,
27(1), pp, 2.
33. A. Viru and M. Viru (2001) Biochemical Monitoring of Sport Training,
Human Kinetics.
34. M. Rizzo (2001) Softball/baseball training machine, United States
Patent, US 6,305,366 Bl (45).
35. Kuster (20 November 2003) Golf swing training device, US Patent App.
10/717,841.
36. Y. Yamamoto (2004) An alternative approach to the acquisition
of a complex motor skill: Multiple movement training on tennis
strokes, International Journal of Sport and Health Science, 2, pp. 169–
179.
37. A. Chong (March 2009) A photogrammetric application in virtual sport
training, The Photogrammetric Record, 24, pp. 51–65(15).
38. T. Komura, A. Kuroda, and Y. Shinagawa (2002) NiceMeetVR: facing
professional baseball pitchers in the virtual batting cage, in Proceed-
ings of the 2002 ACM Symposium on Applied Computing, pp. 1065.
ACM.
39. I. Karliga and J. Hwang (2006) Analyzing human body 3-d motion of
golf swing from single camera video sequences, in Proceedings of the
IEEE International Conference on Acoustics, Speech and Signal Processing
(ICASSP 2006).
40. A. Baca and P. Kornfeind (2006) Rapid feedback systems for elite sports
training, IEEE Pervasive Computing, 5(4), pp. 70–76.
41. A. Cochran, J. Stobbs, and Golf Society of Great Britain (1968) Search for
the Perfect Swing, Triumph Books.
42. P. A. Hume, J. Keogh, and D. Reid (2005) The role of biomechanics in
maximising distance and accuracy of golf shots, Sports Medicine, 35,
pp. 429–449(21).
References 105
43. J. Tartagni and P. Schmid (6 January 2004) Putting stroke training aid,
US Patent 6,672,974.
44. M. Bellagamba (18 June 1991) Athletic swing practice apparatus, US
Patent 5,024,443.
45. C. Welch, S. Banks, F. Cook, and P. Draovitch (1995) Hitting a baseball:
a biomechanical description, The Journal of Orthopaedic and Sports
Physical Therapy, 22(5), pp. 193.
46. H. Ghasemzadeh, E. Guenterberg, and R. Jafari (2009) Energy-Efficient
Information-Driven Coverage for Physical Movement Monitoring in
Body Sensor Networks, IEEE Journal on Selected Areas in Communica-
tions, 27, pp. 58–69.
47. J. Polastre, R. Szewczyk, and D. Culler (April 2005) Telos: enabling ultra-
low power wireless research, in Fourth International Symposium on
Information Processing in Sensor Networks (IPSN 2005), pp. 364–369,
doi: 10.1109/IPSN.2005.1440950.
48. R. Gravina, A. Guerrieri, G. Fortino, F. Bellifemine, R. Giannantonio, and
M. Sgroi (2008) Development of body sensor network applications
using SPINE, in Proceedings of the IEEE International Conference on
Systems, Man, and Cybernetics (SMC 2008), Singapore.
49. H. Ghasemzadeh, J. Barnes, E. Guenterberg, and R. Jafari (2008) A
phonological expression for physical movement monitoring in body
sensor networks, in Fifth IEEE International Conference on Mobile Ad Hoc
and Sensor Systems (MASS 2008), pp. 58–68.
50. N. Stergiou, (2003) Innovative Analyses of Human Movement: Analytical
Tools for Human Movement Research, Human Kinetics.
51. H. Ghasemzadeh, E. Guenterberg, K. Gilani, and R. Jafari (2008) Action
coverage formulation for power optimization in body sensor networks,
in Design Automation Conference, 2008 (ASPDAC 2008), Asia and South
Pacific, pp. 446–451.
52. H. Ghasemzadeh, V. Loseu, and R. Jafari (2009) Structural action
recognition in body sensor networks: Distributed classification based
on string matching, IEEE Transactions on Information Technology in
Biomedicine, 14(2), pp. 425 - 435.
53. R. Duda, P. Hart, and D. Stork (1973) Pattern Classification and Scene
Analysis, Wiley, New York.
54. R. Jafari, R. Bajcsy, S. Glaser, B. Gnade, M. Sgroi, and S. Sastry (June
2007) Platform design for health care monitoring applications, in Joint
Workshop on High Confidence Medical Devices, Software, and Systems and
Medical Device Plug-and-Play Interoperability, pp. 88–94.
55. S. Newell and E. Els (2001) The Golf Instruction Manual, Dorling
Kindersley.
56. J. Perry (1992) Gait analysis: normal and pathological function, 1st edn.
SLACK incorporated.
57. J. Hausdorff, M. Cudkowicz, R. Firtion, J. Wei, and A. Goldberger (1998)
Gait variability and basal ganglia disorders: stride-to-stride variations
of gait cycle timing in Parkinson’s disease and Huntington’s disease,
Movement Disorders, 13(3), pp. 428–437.
58. J. Hausdorff, D. Rios, and H. Edelberg (2001) Gait variability and fall risk
in community-living older adults: a 1-year prospective study, Archives of
Physical Medicine and Rehabilitation, 82(8), pp. 1050–1056.
59. L. Rabiner and B. Juang (1986) An introduction to hidden Markov
models, IEEE ASSP Magazine, 3(1 Part 1), pp. 4–16.
60. L. Rabiner (1989) A tutorial on hidden Markov models and selected
applications in speech recognition, in Proceedings of the IEEE, 77(2),
pp. 257–286.
61. N. Liu, B. Lovell, and P. Kootsookos (2003) Evaluation of HMM
training algorithms for letter hand gesture recognition, in Proceedings
of the Third IEEE International Symposium on Signal Processing and
Information Technology (ISSPIT 2003), pp. 648–651.
62. E. Guenterberg, H. Ghasemzadeh, and R. Jafari (2009) A distributed
hidden Markov model for fine-grained annotation in body sensor
networks, in Proceedings of the 2009 Sixth International Workshop on
Wearable and Implantable Body Sensor Networks, 00, pp. 339–344, IEEE
Computer Society.
63. V. Raghunathan, C. Schurgers, S. Park, and M. Srivastava (March 2002)
Energy-aware wireless microsensor networks, Signal Processing Maga-
zine, IEEE, 19(2), pp. 40–50, ISSN 1053-5888. doi: 10.1109/79.985679.
64. Z. Alliance, ZigBee specification, ZigBee Document 053474r06, Version
1, 2005.
65. J. M. Kahn, R. H. Katz, and K. S. J. Pister (1999) Next century
challenges: mobile networking for “smart dust,” in Proceedings of the
Fifth Annual ACM/IEEE International Conference on Mobile Computing
and Networking (MobiCom 1999), pp. 271–278, ISBN 1-58113-142-9.
66. H. Ghasemzadeh, N. Jain, M. Sgroi, and R. Jafari (2009) Communication
minimization for in-network processing in body sensor networks: a
buffer assignment technique, in Proceedings of the ACM/IEEE Design,
Automation and Test in Europe (DATE09).
Chapter 5
Signal Processing In-Node Frameworks

for Wireless Body Area Networks: From
Low-Level to High-Level Approaches
Francesco Aiello, Giancarlo Fortino, Stefano Galzarano,

Raffaele Gravina, and Antonio Guerrieri
Department of Electronics, Informatics, and Systems (DEIS),
University of Calabria, Via P. Bucci, cubo 41C, 87036 Rende (CS), Italy
[email protected]; [email protected]; [email protected];
To develop applications based on wireless body area networks

(WBANs), either low-level or high-level programming approaches
can be adopted. Low-level approaches are based on low-level
APIs (application programming interfaces) offered by sensor
platforms, whereas high-level approaches are based on high-level
programming models and frameworks that facilitate programming
and increase productivity. This chapter proposes a high-level
approach based on the agent-oriented programming model to
flexibly design and efficiently implement signal processing in-
node environments supporting WBAN applications. The approach
is exemplified through a case study concerning a real-time human

www.panstanford.com
108 Signal Processing In-Node Frameworks for Wireless Body Area Networks
activity monitoring system, which is developed through two

different agent-based frameworks: MAPS (Mobile Agent Platform for
Sun SPOT) and AFME (Agent Factory Micro Edition). A comparison
of the effectiveness and efficiency of the developed systems is finally
presented.
5.1 Introduction
Wireless sensor networks (WSNs) are currently emerging as one

of the most disruptive technologies enabling and supporting next
generation ubiquitous and pervasive computing scenarios [24].
WSNs are capable of supporting a broad array of high-impact
applications in several domains such as disaster/crime prevention,
military, environment, logistics, health care, and building/home
automation. WSNs applied to human body monitoring are usually
called wireless body area networks (WBANs) [29]. WBANs are
conveying notable attention as their real-world applications aim at
improving the quality of life of human beings by enabling continuous
and real-time, non-invasive medical assistance at low cost. Health
care applications where WBANs could be greatly useful include early
detection or prevention of diseases, elderly assistance at home, e-
fitness, rehabilitation after surgeries, motion and gestures detection,
cognitive and emotional recognition, medical assistance in disaster
events, etc.
However, programming WBAN applications is a complex task
due to the hard constraints of wearable devices in terms of limited
resources (computing power, memory, and communications) and
due to the lack of proper and effective software abstractions. To
deal with these issues, several software frameworks have been
developed such as CodeBlue [18], Titan [16], and SPINE [12, 9]. They
aim at decreasing development time and improving interoperability
among signal processing intensive applications based on WBANs
while fulfilling efficiency requirements. In particular, they basically
rely on a star-based network architecture, which is organized into
a coordinator node and a set of sensor nodes. Moreover, they are
developed in TinyOS at the sensor node side and in Java at the
coordinator node side.
Introduction 109
Apart from the adoption of software frameworks, we believe

that the exploitation of the agent-oriented programming paradigm
to develop WSN applications could provide more effectiveness as
demonstrated by the application of agent technology in several key
application domains [8, 17]. In this chapter, we, therefore, propose
high-level, agent-oriented design and agent-based implementation
to support the development of WBAN applications. In particular,
the approach is applied in the context of a real-time human
activity monitoring system through two different Java-based agent
platforms running on the Sun SPOT sensor platform: MAPS (Mobile
Agent Platform for Sun SPOTs) [4, 2] and AFME (Agent Factory
Micro Edition) framework [21]. The system architecture relies on
a typical star-based WBAN composed of a coordinator node and
two sensor nodes, which are located on the waist and the thigh
of the monitored human being, respectively. The coordinator is
based on an enhancement of the Java-based SPINE coordinator [12,
27] and allows configuring the sensing process, receiving sensed
data features, and recognizing predefined human activities. Each
sensor node executes either a MAPS-based agent or an AFME-based
agent that performs sensing of the on-board 3-axial accelerometer
sensor, computation of significant features on the acquired data,
and aggregation and transmission of features to the coordinator.
Finally, a comparison between the MAPS-based and the AFME-based
implementations is discussed.
This chapter is organized as follows. Section 5.2 describes
a reference architecture for WBAN applications from network
and functional perspectives. Section 5.3 details and compares
the state-of-the-art on signal processing in-node frameworks and
applications based on TinyOS for the development of WBAN
applications: CodeBlue, Titan, and two versions of SPINE. In
Section 5.4, the most known available agent-oriented platforms
for developing WSN-based applications are described. Section 5.5
presents an agent-oriented approach for the construction of WBAN
applications, whereas Section 5.6 exemplifies it by describing the
implementation and evaluation of an agent-based, real-time human
activity monitoring system through aforementioned different agent
platforms. Finally conclusions are drawn and future research
delineated.
5.2 A WBAN Reference Architecture
The network architecture of the reference WBAN for signal process-

ing is organized into multiple sensor nodes and one coordinator
node according to the scheme reported in Fig. 5.1. The coordinator
manages the network, collects, stores, and analyzes the data
received from the sensor nodes, and also can act as a gateway to
connect the WBAN with other networks (e.g., Internet) for remote
data access. Sensor nodes measure local physical parameters and
send raw or preprocessed data to the coordinator. In this system,
sensor nodes only communicate with the coordinator according to
the star network topology.
The software architecture of the system consists of two main
components installed, respectively, on the coordinator (e.g., a PC or
a smartphone) and on the sensor nodes. Figure 5.2 shows a schema
of the software architecture from a functional point of view.
At the coordinator side, an interface to the WBAN, which is placed
between user applications and the hardware and software host
platform, is made available. User applications manage the WBAN
through a system API. The top level of the software architecture
at the coordinator allows registered applications being notified
of the following events generated by the WBAN: discovery of
new nodes, sensor data communication, node alarms, and system
messages such as low battery warnings. Commands issued by user
applications and network-generated events are, respectively, coded
in lower-level messages and decoded in higher-level information by
Figure 5.1. WBAN reference network architecture.

Software Frameworks for Programming WBANs 111
Figure 5.2. WBAN reference software architecture.
the host communication manager according to a specific over-the-

air protocol. This component handles the generation and reception
of messages and is interfaced with specific software components
of the host platform to access the physical radio module to
transmit/receive messages to/from the sensor nodes.
At the sensor node side, the software architecture provides:
(i) abstractions of hardware resources such as sensors and the
radio, (ii) a default set of ready-to-use common signal processing
functions, and (iii) a flexible and modular architecture to be
customized and extended to support new physical platforms and
sensors and introduce new signal processing services. In particular,
the Node Communication Manager acts as the counterpart of the
Host Communication Manager. The Sensor Controller manages and
abstracts the sensors on the node platform, providing a standard
interface to the diverse sensor drivers. Specifically, it is responsible
for sampling the sensors and storing the sensed data in properly
defined Buffers. The Node Manager is the central component of
the sensor node side, responsible of recognizing the coordinator
requests and dispatching them to the proper components. Finally,
the Processing Manager offers core processing services, which can
be enriched by user-defined services.
5.3 Software Frameworks for Programming WBANs
WBAN applications can be mainly implemented according to the

following approaches: (i) application-specific code, which aims at
Table 5.1. Comparison among different approaches for WBAN

application development
Approaches/ Application- Domain-specific General-purpose

Features specific code framework middleware
Code Reusability • •
Rapid prototyping • •
Ease of debugging • •
Code efficiency • •
System interoperability • •
Specific support to flexible •
sensing operations
Specific support to in-node •
signal processing
developing a WBAN application by using the low-level API provided

by the sensor node platform; (ii) general-purpose middleware, which
provides a high-level distributed framework for programming WSN
applications for distributed signal processing; (iii) domain-specific
frameworks, which aims at supporting specific development of
WBAN applications according to a reference architecture. Table 5.1
reports the characteristics of the three approaches: code reusability,
rapid prototyping, ease of debugging, code efficiency, system
interoperability, specific support to flexible sensing operations, and
specific support to in-node signal processing. In the following,
specific applications and frameworks belonging to such approaches
are described.
Most of the previous research efforts on WBANs were focused
on proof-of-concept applications with the aim to demonstrate the
feasibility of new context-aware algorithms and techniques, e.g.,
for recognition of physical activity or prompt detection of heart
diseases, considering issues like power consumption and also
radio channel usage but not taking into account code reusability
and modularity. In reference [22], a method for physical activity
monitoring is presented, which is able to detect body postures and
periods of walking in elderly persons using one kinematic sensor
attached to the chest. In reference [13], wearable motion sensors are
used to guide post-stroke rehabilitation by models to predict clinical
scores of motor abilities. In reference [23], activity recognition
is improved by integrating wearable sensors with ambient blob-

based vision sensor data. Another physical activity recognition
system based on wearable devices is presented in reference [19].
In reference [15], it is discussed in detail the interesting issue
of developing a personal activity recognition system, which is
based on data coming from a single body location regardless
of the specific sensor location and able to work with different
individuals. The system can be further personalized to enhance
the activity recognition accuracy. Finally, the most exhaustive effort
on activity recognition based on wearable sensors is presented
in reference [5]. All these research efforts have been focused on
efficient application-specific solutions rather than on the definition
of reusable frameworks facilitating the development of WBAN
applications.
The first important solution provided to define a general
platform able to support various WBAN applications is CodeBlue
[18]. CodeBlue is a framework running on TinyOS specifically
designed for integrating wireless medical sensor nodes and other
devices. CodeBlue allows these devices to discover each other,
report events, and establish communications. CodeBlue is based
on a publish/subscribe-based data routing framework in which
sensors publish relevant data to a specific channel and end-user
devices subscribe to channels of interest. It includes a naming
scheme, a multi-hop communication protocol, authentication and
encryption capabilities, location tracking, and in-network filtering
and aggregation. CodeBlue provides end-user devices with a query
interface for retrieving data from previously discovered sensor
nodes. Although CodeBlue provides a sensor driver abstraction
architecture, which allows an easy integration of new sensors within
the system, selection of sensor types or physical node identifiers as
data sources, tuning of the data rate, and definition of threshold-
based filters to avoid unnecessary data being transmitted, it does
not allow inserting complex signal processing functionalities into
the sensor nodes. It supports just simple threshold-based triggers
on the sensor readings that do not give enough flexibility for the
variety of requirements of the WBAN applications.
A higher-level approach is adopted by Titan [16]. Titan is a
general-purpose middleware implemented in TinyOS that supports
implementation and execution of context recognition algorithms in

dynamic WSN environments. Titan represents data processing by a
data flow from sensors to recognize result. The data are processed
by tasks, which implement elementary computations. Tasks and
their data flow interconnections define a task network, which runs
on the sensor network as a whole. Tasks are mapped onto each
sensor node according to the sensors and the processing resources
they provide. Titan dynamically reprograms the WSN to exchange
context recognition algorithms, handle defective nodes, variations
in available processing power, or broken communication links. The
architecture of Titan is composed of several software components,
which enhance modularity. Although Titan raises the programming
abstraction level by offering a middleware for effectively developing
signal processing applications in WSNs, it is based on too generic
mechanisms for providing efficient solutions in the specific BAN
application domain. In fact, the programming of a feature extraction
operation on the sensor node, which is often carried out in a BAN
application, requires the creation of at least five tasks (sampling,
buffering, loading, feature calculation, and transmission). Moreover,
some overhead can be introduced due to the connections of the
output and input ports among tasks through which data are
exchanged.
The first example of domain-specific framework in the WBAN
area is represented by SPINE (Signal Processing in Node Environ-
ment) [11, 12, 27]. SPINE is a software framework for the design
of collaborative WBAN applications. It provides programming
abstractions, APIs and libraries of protocols, utilities and data
processing functions, which simplify development of distributed
signal processing algorithms for the analysis and the classification of
sensor data. SPINE [27] is distributed in open source under the LGPL
(Lesser General Public License) license to facilitate establishing a
broad community of users and developers that contribute to the
scientific evolution of the framework with new capabilities and
applications. The SPINE framework is composed of two distinctive
parts: a node side runtime system residing on the sensor nodes
and a Java application, the coordinator, residing on a PC and
having functionalities such as nodes configuration and control, data
gathering and data analysis. To date, two releases of SPINE are
available:
• The TinyOS release (version 1.3), which supports differ-

ent kinds of sensor platforms running the TinyOS [28]
operating system (supported platforms are TelosB, MicaZ,
Shimmer).
• The Z-Stack release (version 1.0), which allows the de-
velopment of WBAN applications on the Z-Stack platform
according to the ZigBee standard [30]. In particular, Z-Stack
is the implementation of the ZigBee stack carried out by
Texas Instruments.
The SPINE2 framework [10, 9] is an evolution of SPINE based on the

C language for reaching a very high platform independency for C-like
programmable sensor platforms (e.g., TinyOS, EmberZNet, Z-Stack)
so raising the level of the provided programming abstractions from
platform-specific to platform-independent. SPINE2 is founded on a
software layering approach based on the C language, which is the
language used for programming the majority of embedded systems.
It embodies the following features: (i) execution on commercial
resource-constrained sensor platforms each one having a different
operating system; (ii) minimization of the amount of code that
should be replicated for each specific implementation; (iii) enabling
C developers (eventually C++) to extend the SPINE2 framework
without having to learn low-level details of specific sensor plat-
forms or without having to learn new programming languages;
(iv) enabling, compiling, and simulating the code by using normal
ANSI C tools.
While SPINE is centered on a programming model based on
functions, SPINE2 is based on a task-oriented programming model
in order to best fit the requirements of collaborative distributed
applications in resource-constrained environments. Distributed and
collaborative applications can then be programmed as a dynamically
schedulable and reconfigurable set of tasks. Different tasks can be
assigned to each node of the network, and tasks can be controlled
at execution time via proper message exchange; in this way the
network can overall adapt to changes in context, in overall goals, in
the state of each single node, and it can better balance load and task
types between each element of the network. Dynamic distribution
of tasks also allows preprocessing of sensed data directly on the
node, a significant reduction of data transmission and battery
consumption, and an overall increase in the network lifetime. Due to

task-oriented programming, application developers do not need to
program in tiny environments but only configure tasks on the WBAN
coordinator.
To summarize, we promote the use of the domain-specific frame-
work approach, which stands in the middle between application-
specific code and general-purpose middleware approaches. It
specifically addresses and standardizes the core challenges of WSN
development within a particular application domain. While main-
taining high efficiency, it allows for a more effective development
of customized applications with little or no additional hardware
configuration and with the provision of high-level programming
abstractions tailored for the reference application domain. In
particular, the SPINE and SPINE2 frameworks are domain-specific
frameworks in the context of signal processing in-node intensive
WBAN applications, whereas CodeBlue and Titan can be seen
as general-purpose middleware solutions for WBAN applications.
SPINE allows for both code efficiency similarly to an application-
specific code approach, and code reusability, rapid prototyping,
easy debugging, and system interoperability as CodeBlue and Titan.
Moreover, it specifically supports functionalities for flexible sensing
operations and easily programmable in-node signal processing.
SPINE2 extends the functionalities of SPINE with the development
of WBAN applications on heterogeneous sensor platforms based on
C-like languages.
5.4 Agent-Oriented Platforms for Wireless

Sensor Networks
In the context of WSNs, it is challenging to develop agent platforms

(APs) for supporting (mobile) agent-based programming [8, 3].
Due to the currently available resource-constrained sensor nodes
and related operating systems, building flexible and efficient agent
platforms is a very complex task. Very few APs for WSNs have been
to date proposed and actually implemented. An AP architecture
(see Fig. 5.3) relies on the services offered by the hosting sensor
node OS and support execution of agents by managing agent
Agent-Oriented Platforms for Wireless Sensor Networks 117
Figure 5.3. Reference architecture of a sensor node agent platform.
lifecycle, migration, communication, sensing capabilities, and sensor

resource access. The most representative APs developed to date are
SensorWare [7], Agilla [8], actorNet [14], MAPS [4], and AFME [21].
SensorWare [7] is a general middleware framework based on
agent technology, where the mobile agent concept is exploited.
Mobile control scripts in Tcl model network participants’ function-
alities and behaviors, and routing mechanisms to destination areas.
Agents migrate to destination areas performing data aggregation
reliably. The script can be very complex, and diffusion gets slower
when it reaches destination areas. The replication and migration
of such scripts in several sensor nodes allows the dynamic
deployment of distributed algorithms into the network. SensorWare
defines, creates, dynamically deploys, and supports such scripts.
SensorWare is designed for iPAQ devices with megabytes of RAM.
The verbose program representation and on-node Tcl interpreter
can be acceptable overheads; however, they are not yet on a sensor
node.
Agilla [8] is an agent-based middleware where each node
supports multiple agents and maintains a tuple space and neighbors’
list. The tuple space is local and shared by the agents residing on the
node. Special instructions allow agents to remotely access another

node’s tuple space. The neighbors’ list contains the addresses of all
one-hop nodes. Agents can migrate carrying their code and state, but
do not carry their own tuple spaces. Agilla is currently implemented
on MICA2, MICAZ, and TelosB motes.
While both Agilla and SensorWare rely on mobile agents, they
employ a different communication model: Agilla’s agent interaction
is based on local tuple spaces, whereas SensorWare’s agent interac-
tion is based on direct communication based on network messages.
In reference [26], another mobile agent framework is proposed.
The framework is implemented on Crossbow MICA2DOT motes.
In particular, it provides agent migration and agent interaction
based both on locally shared memory and network messages. In
reference [25], the authors propose an extension of Agilla to support
direct communication based on messages. In particular, to establish
direct communications, agents are mediated by a middle component
(named landmark) that interact with agents through zone-based
registration and discovery protocols.
In reference [14], actorNet, a mobile agent platform for WSNs
based on the actor model, is proposed. In particular, it provides
services such as virtual memory, context switching, and multi-
tasking to support a highly expressive yet efficient agent functional
language. Currently, the sensor node actorNet platform is specifically
designed for TinyOS on MICA2 sensors.
The aforementioned mobile agent systems for WSNs are all
implemented for TinyOS-based sensor platforms and use ad-hoc
languages for agent programming (Agilla uses a micro-programming
language, whereas actorNet employs a function-oriented language).
Although some supported operations (e.g., migration) are very
efficient, programming complex tasks is not so straightforward
and, moreover, developers need to learn another very specific
language. In the following, the available Java-based agent platforms
are introduced.
The Mobile Agent Platform for Sun SPOT framework (MAPS) [4,
20] is an innovative Java-based framework for wireless sensor net-
works based on Sun SPOT technology, which enables agent-oriented
programming of WSN applications. The MAPS architecture is based
on components that interact through events. Each component offers
An Agent-Oriented Design of Signal Processing In-Node Environments 119
a minimal set of services to mobile agents, which are modelled

as multi-plane state machines driven by ECA rules. In particular,
the offered services include message transmission, agent creation,
agent cloning, agent migration, timer handling, and easy access to
the sensor node resources (sensors, actuators, input switches, flash
memory, and battery).
The Agent Factory Micro Edition (AFME) [21, 1] is a lightweight
version of the Agent Factory framework purposely designed for
wireless pervasive systems and implemented in J2ME, recently
ported onto Sun SPOT and purposely used for implementing agent
communication and migration in WSNs. The AFME agents are
based on a BDI (beliefs-desires-intentions)-like architecture and are
programmed through a high-level logic-based language. AFME also
provides a translator, which converts high-level agent specifications
into Java code.
However, AFME was not specifically designed for WSNs and,
particularly, for Java Sun SPOT. MAPS is conversely specifically
designed for WSNs, and it fully exploits the release 5.0 red of the Sun
SPOT library to provide advanced functionality of communication,
migration, sensing/actuation, timing, and flash memory storage.
Moreover, it allows developers to program agent-based applications
in Java according to the rules of the MAPS framework so that no
translator and/or interpreter need to be developed and no new
language has to be learnt.
5.5 An Agent-Oriented Design of Signal Processing

In-Node Environments
The exploitation of the agent-oriented programming paradigm to

develop WBAN applications could provide more effectiveness as
demonstrated by the application of agent technology in several
key application domains [17]. In this section, we describe an
agent-oriented design of SPINE (see Section 5.3), named ASPINE
[6], which extends the functionalities provided by SPINE and
allows for a more rapid development of signal processing intensive
WBAN applications in terms of agent-based systems. The reference
architecture defined in Section 5.2 was designed by using an
Figure 5.4. The ASPINE high-level architecture.
agent-based approach. Figure 5.4 shows the high-level architecture

of ASPINE through a class diagram.
In particular, the following core agents are defined:
Base station-side
• The CoordinatorAgent is responsible for managing the set

of nodes of the sensor network under control. Management
involves configuring and monitoring nodes.
• The ApplicationAgents are agents implementing
application-specific or domain-specific logics.
An Agent-Oriented Design of Signal Processing In-Node Environments 121
• The CommunicatorAgent allows the CoordinatorAgent and

the ApplicationAgents to interact with the sensor nodes
through an efficient over-the-air application-level protocol.
Sensor node-side
• The SensorManagerAgent manages the sensor/actuator
resources of the node through specific SensorAgents able
to interact with specific sensors (temperature, light, ac-
celerometer, etc.), actuators (LEDs, actuation devices, etc.),
and the battery.
• The CommunicationManagerAgent manages the commu-
nication with the communicator agent and with the
other CommunicationManagerAgents, located at different
sensor nodes, by means of specific CommunicationAgents.
Moreover, such agent manages the radio settings (e.g., duty
cycling, transmission power, etc.).
• The ProcessingManagerAgent supports one or more local
processing tasks or parts of global processing tasks (span-
ning multiple sensor nodes) through processing agents.
They are able to perform computation on sensed data
(e.g., feature extraction, threshold functions, classification
algorithms, etc.) and data aggregation.
An example of in-node signal processor is portrayed in Fig. 5.5 by
means of a data-flow model based on tasks. The in-node signal
processor aims at sampling the on-board 3-axial accelerometer,
computing selected features on sample windows, and sending the
results to the coordinator. In particular, the sensed data periodically
produced by the sensing task acting on the three channels (X, Y, Z)
of the accelerometer are split for the computation of the features
Mean on all three axes (XYZ), and Min and Max on axis X. Each triple
Figure 5.5. Data-flow-based model of an in-node signal processing task.

Figure 5.6. ASPINE-based design of the example in-node signal process-

ing task.
of computed features (<Mean(AccXYZ), Min(AccX), Max(AccX)>)

are aggregated by the aggregation task (Aggr) and sent to the
coordinator node by the data transmission task (Sender) as soon as
the aggregated data is available.
The ASPINE design of the data-flow-based model of Fig. 5.5
is shown in Fig. 5.6. The AccelerometerAgent interacts with the
accelerometer sensor and, according to the set sampling time
(samplingTime), acquires one sample per channel (X, Y, Z), and
stores them into the corresponding buffers chX, chY, chZ. Once
S samples are acquired, the AccelerometerAgent passes them to
the MinAgent, MaxAgent, and MeanAgent. These relationships are
created through the method setSOnCh(AID, Channel, S), where
AID is the agent identifier, Channel refers to the channels to be
considered, and S is the number of samples. In this case, all agents
are based on the same S but on different channels. In particular,
An Analysis of Agent-Oriented Implementations of In-Node Signal Processors 123
MinAgent, MaxAgent, and MeanAgent receive the last acquired S

samples, respectively, from the chX buffer, from the chX buffer, and
from the chX, chY, and chZ buffers. Upon reception of such data, the
agents compute their respective functions and pass the results to the
AggregatorAgent. This waits for the aggregation of the data triple
aggrData=<minChX, maxChX, <meanChX, meanChY, meanChZ
and passes it to the SenderAgent, a specific CommunicationAgent,
which, in turn, transmits it to the CommunicatorAgent located at the
base station.
5.6 An Analysis of Agent-Oriented Implementations

of In-Node Signal Processors
A WBAN-based activity human monitoring system aims at recogniz-

ing postures and movements of assisted livings [22]. The developed
prototype is able to recognize some postures (lying down, sitting,
and standing still) and a movement (walking). The architecture of
the system, shown in Fig. 5.7, is organized into a coordinator and
two sensor nodes according to the reference WBAN architecture
described in Section 5.2.
Figure 5.7. Architecture of the WBAN-based real-time activity monitoring

system.
The coordinator is based on the Java-based SPINE coordinator

developed in the context of the SPINE project [27]. In particular, the
SPINE Manager is used by end-user applications (e.g., real-time ac-
tivity monitoring application) for sending commands to the sensor
nodes. Moreover, the SPINE Manager is responsible for capturing
low-level messages and node events through the SPINE Listener,
which integrates several sensor platform-specific SPINE commu-
nication modules (e.g., TinyOS, Z-Stack, etc.) to notify registered
applications with higher-level events and message content. A SPINE
communication module is composed of a send/receive interface
and some components that implement such interface according
to the specific sensor platform and that formalize the high-level
SPINE messages in sensor platform-specific messages. The SPINE
Listener is enhanced with MAPS/Sun SPOT and AFME/Sun SPOT
communication modules to configure and communicate with MAPS-
based and AFME-based sensor nodes, respectively. Such modules
translate high-level SPINE messages formatted according to the
SPINE OTA (over-the-air) protocol [27] into lower-level MAPS
or AFME/Sun SPOT messages through its transmitter component
and vice versa through its receiver component. Such modules
also integrate an application-specific logic for synchronizing the
operations of the two sensors. The SPINE-based real-time activity
monitoring application was thus completely reused as well as the
SPINE Manager; only the SPINE Listener was modified to account
for such enhancement.
The sensor nodes are based on Java Sun SPOT sensors respec-
tively positioned on the waist and the thigh of the monitored person.
In particular, an agent platform (MAPS or AFME) is resident on the
sensor nodes and supports the execution of the WaistSensorAgent
and the ThighSensorAgent. Such sensor agents have the following
similar step-wise cyclic behavior:
1. Sensing the 3-axial accelerometer sensor according to a given

sampling time (ST).
2. Computation of specific features (Mean, Max, and Min functions)
on the acquired raw data according to the window (W) and
shift (S) parameters. In particular, W is the sample size on
which features are computed, whereas S is the number of new
acquired sample data for calculating a new feature. Usually S is

set to 50% of W.
3. Features aggregation and transmission to the coordinator.
4. Goto 1.
The agents differ in the specific computed features even though the
W and S parameters are equally set. In particular, while the waist
sensor agent computes the mean values for data sensed on the XYZ
axes, the min and max values for data sensed on the X axis, the
ThighSensorAgent calculates the min value for data sensed on the
X axis.
In the following subsections, we describe the implementation of
the WaistSensorAgent carried out with MAPS and AFME and, then,
compare the implemented solutions.
5.6.1 MAPS-Based and AFME-Based Implementation

of Sensor Agents
The architecture of the MAPS-based sensor agent (Fig. 5.8) is
composed of two planes (sensing and feature calculation and
transmission) modeled as finite state machines. Each plane is able
to handle specific events, which are instances of system and user-
defined event types. The sensing plane (see Fig. 5.9a) specifies
the sensing process: the MSG.START event initiates the sensing of
the three channels of the accelerometer; after the sensed data are
returned through the ACC CURRENT ALL AXES event, the sampling
timer is scheduled for the next acquisition; if a MSG.RESYNCH,
formalizing a re-synchronization operation among the sensor agents
sent by the coordinator, is received, the sensing process is paused
until a MSG.RESTART event arrives; finally, the MSG.STOP halts
the sensing process. The feature calculation and transmission plane
(see Fig. 5.9b) specifies the feature extraction process and the
transmission of the computed features to the coordinator, which are
triggered by the MSG.COMPFEATURES event.
The architecture of the AFME-based sensor agent (Fig. 5.10) is
composed of:
• Two perceptors. MTSPerceptor checks for the arrival of a

new message and SensingPerceptor checks for the feature
Figure 5.8. MAPS-based sensor agent.
computation and send the computation results to the

coordinator if such results are available.
• Three actuators. ActivateSensorActuator allows activating
the sensing operation, ResetActuator resets the data buffer
after the reception of the resynch message, and RequestAc-
tuator is used to send a request message to the coordinator.
• Rules. TerImplication contains the rules of the agent
behavior (see below).
Figure 5.9. Planes of the MAPS-based sensor agent: (a) sensing plane;
(b) feature calculation and transmission plane.
Figure 5.10. AFME-based sensor agent.
• One Module. SharedDataModule contains buffers storing the

data sensed from the three accelerometer channels, the
feature extraction parameters, the flag variable (activation)
enabling the sensing process.
• One Service. RadiogramMTS represents the transport ser-
vice for data transmission to the coordinator.
The agent behavior is modeled through the following three basic

rules:
1. message(inform, sender(BaseStation, addresses(BSAddress)), be-

gin) > activateSensors(1);
2. sense(?val), !message(inform, sender(BaseStation, addresses
(BSAddress)), resynch) > request(agentID(BaseStation, addresses
(”radiogram://”+BSAddress)), ?val);
3. message(inform, sender(BaseStation, addresses(BSAddress)), re-
synch) > reset;
Rule (1) enables the sensor reading operation on the sensor node. In
particular, this rule states that the message belief, which is generated
upon the reception of an inform message with “begin” content sent
by the coordinator with address BSAddress, triggers the action

ActivateSensors(1), which will then set the activation flag on.
Rule (2) regulates the agent behavior during the sensing phase.
If a new sense belief having the sensed data is generated and a
message belief generated upon reception of the resynch message
sent by the coordinator does not exist, the request action is carried
out so that the new message containing the feature calculation is
sent to the coordinator.
Finally, rule (3) allows re-synchronizing the sensor agent. In
particular, it states that if the belief message is generated upon
reception of the resynch message, the action reset, which re-
initializes all the data structures in SharedDataModule, is executed.
5.6.2 Agent Implementation Comparison

The comparison of the two different agent implementations are
carried out according to the following two aspects: (i) effectiveness
in prototyping the agent-based solutions; (ii) timing and synchro-
nization of the real-time monitoring.
The development of MAPS and AFME agents is based on different
approaches. MAPS uses state machines to model the agent behavior
and directly the Java language to program guards and actions. AFME
uses a more complex model centered on perceptors, actuators,
rules, modules, and services that define the agent behavior. They
are both effective in modeling agent behavior even though MAPS
is more straightforward as it relies on a programming style based
on state machines widely known by programmers of embedded
systems.
Two important issues to deal with are the timing of the
sensing process in terms of admissible sampling rate and the
synchronization between the operations of the two agents, which
is to be maintained within a maximum skew for not affecting
real-time monitoring. If such skew is overtaken, the two agents
are to be re-synchronized. Indeed such two aspects are strictly
correlated. In particular, as the sensor agents compute a different
number of features, when the sampling rate is high, the agent
computing more features (i.e., the WaistSensorAgent) takes more
time to complete its operations for each S sample acquisition
than the ThighSensorAgent. Re-synchronization is driven by the

synchronization logic included in the developed MAPS/Sun SPOT
and AFME/Sun SPOT communication modules, which sends a re-
synchronization message as soon as it detects that the synchro-
nization skew is greater than a given threshold. Detection is based
on the skew time between the receptions of two messages sent by
the agents that contain features referring to the same interval of
S sample acquisition: if skew ≥ P × S × ST , then resynchronize,
where P is a percentage, S = 0.5 W is the shift, W is the data
sample window, ST is the sampling time. Thus, the evaluation aims
at analyzing the synchronization of the sensor agents and their
monitoring continuity. The defined metrics are:
• The packet pair average time (PPAT), which is the av-

erage reception time between two consecutive pairs of
synchronized packets, having the same logical timestamp
and containing the computed features sent by the sensor
agents.
• The synchronization packet percentage (SPP), which is the
percentage of resynchronization packets, which are sent
by the coordinator for re-synchronizing the sensor agents,
calculated with respect to the total number of received
feature packets.
PPAT should be ideally equals to ST*S, i.e., the packet pair arrives
each monitoring period and so there is no de-synchronization in the
average. SPP should be as much as possible close to 0, i.e., a few or
no re-synchronizations are carried out and so the monitoring can
be continuous as a resynch operation usually takes 600 ms. Thus,
system parameters (W, ST, and P) should be carefully set to optimize
the monitoring process. To this purpose, a set of experiments was
set up to tune the system parameters. In particular, the experiments
were carried out by fixing ST (ms) = [25, 50, 100], W (samples) =
[40, 20], and P (%) = [5, 15, 25]. Figures 5.11 and 5.12 show the
obtained results.
As can be noticed, the system cannot support an ST = 25 ms
because PPAT is always greater than the ideal value and SPP is too
high. This leads to a non-continuous monitoring due to very frequent
Figure 5.11. Analysis of PPAT and SPP of the sensor agents with W = 20,
S = 10, and P = (a) 25%, (b) 15%, (c) 5%. See also Color Insert.
S = 20, and P = (a) 25%, (b) 15%, (c) 5%. See also Color Insert.
re-synchronizations (SPP >= 20% for W = 20 and S = 10). The

best results are obtained with ST = 100 ms, P = 25% or 15% and
W = 20; they guarantee monitoring continuity due to an SPP ≈ 0%
and regularity as the experimented PPAT ≈ the ideal PPAT for W =
20. If W = 20 and P = 5%, ST = 100 ms is not a good value either
because an out-of-limits skew frequently occurs. Although the AFME
implementation performs better than the MAPS implementation,
the AFME implementation collapses in the case W = 20, S = 10,
and P = 5%. It is worth noting that even though a lower ST would
allow a finer monitoring, the considered human activities can be well
captured by an ST = 100 ms as demonstrated by the good accuracy
results (95% in the average) obtained from the carried-out real-time
human activity monitoring [12].
5.7 Conclusions and Future Work
In this chapter, we have presented an agent-oriented approach for

high-level programming of WBAN applications. The agent approach
is not only effective during the design of a WBAN application but
also during the implementation phase. In particular, the presented
frameworks MAPS and AFME allow for a more rapid prototyping
of sensor node code than low-level APIs, which can be effectively
used only by sensor node skilled programmers having knowledge
of sensor drivers, data buffers, radio and energy mechanisms. The
higher level software abstractions provided by MAPS and AFME are
suitable for the development of real-time monitoring systems as
demonstrated by the performance evaluation of the proposed case
study concerning a real-time human activity monitoring system.
Ongoing work is aimed at implementing ASPINE through MAPS
to provide a full-fledged agent-based signal processing in-node
environment.
References
1. Agent Factory Micro Edition (AFME), documentation and software at

http://sourceforge.net/projects/agentfactory/files/ (2011).
References 133
2. Aiello, F., Bellifemine, F., Fortino, G., Gravina, R., and Guerrieri, A. (10–
14 May 2010) An agent-based signal processing in-node environment
for real-time human activity monitoring based on wireless body
sensor networks, in Proceedings of the 1st International Workshop on
Infrastructures and Tools for Multiagent Systems (ITMAS-2010), jointly
held with the9th International Joint Conference on Autonomous Agents
and Multiagent Systems (AAMAS-2010), Toronto, Canada.
3. Aiello, F., Fortino, G., and Guerrieri, A. (25–31 August 2008) Using
mobile agents as an effective technology for wireless sensor networks,
in Proceedings of the 2nd IEEE/IARIA International Conference on
Sensor Technologies and Applications (SENSORCOMM 2008), Cap Esterel,
France.
4. Aiello, F., Fortino, G., Gravina, R., and Guerrieri, A. (2011) A Java-
based agent platform for programming wireless sensor networks, The
Computer Journal, 54(3), pp. 439–454.
5. Bao, L., and Intille, S. S. (2004) Activity recognition from user-annotated
acceleration data, in Proceedings of the 2nd International Conference on
Pervasive Computing (PERVASIVE), pp. 1–17.
6. Bellifemine, F., and Fortino, G. (9–10 July 2009) ASPINE: an agent-
oriented design of SPINE, in Proceedings of the Workshop on Objects and
Agents (WOA’09), Parma, Italy.
7. Boulis, A., Han, C. -C., and Srivastava, M. B. (2003) Design and
implementation of a framework for efficient and programmable sensor
networks, in Proceedings of the 1st International Conference on Mobile
systems, Applications and Services (MobiSys), pp. 1 87–200.
8. Fok, C. -L., Roman, G. -C., and Lu, C. (6–10 June 2005) Rapid
development and flexible deployment of adaptive wireless sensor
network applications, in Proceedings of the 24th International Conference
on Distributed Computing Systems (ICDCS’05), Columbus, Ohio, pp. 653–
662.
9. Fortino, G., Guerrieri, A., Giannantonio, R., and Bellifemine, F. (11–
14 October 2009) Platform-independent development of collaborative
WBSN applications: SPINE2, in Proceedings of the IEEE International
Conference on Systems, Man, and Cybernetics (SMC 2009), San Antonio,
Texas, USA.
10. Fortino, G., Galzarano, S., Giannantonio, R., Gravina, R., and Guerrieri,
A. (2010) SPINE-based application development on heterogeneous
wireless body sensor networks, International Journal of Computing,
9(1), pp. 80–89.
11. Gravina, R., Andreoli, A., Salmeri, A., Buondonno, L., Raveendranathank,
N., Loseuk, V., Giannantonio, R., Seto, E., and Fortino, G. (7–9 June 2010)
Enabling multiple BSN applications using the SPINE framework, in
Proceedings of the International Conference on Body Sensor Networks
(BSN 2010), Biopolis, Singapore.
12. Gravina, R., Guerrieri, A., Fortino, G., Bellifemine, F., Giannantonio, R., and
Sgroi, M. (12–15 October 2008) Development of body sensor network
applications using SPINE, in Proceedings of the IEEE International
Conference on Systems, Man, and Cybernetics (SMC 2008), Singapore.
13. Hester, T., Hughes, R., Sherrill, D. M., Knorr, B., Akay, M., Stein, J.,
and Bonato, P. (3–5 April 2006) Using wearable sensors to measure
motor abilities following stroke, in Proceedings of the 3rd International
Workshop on Wearable and Implantable Body Sensor Networks, 2006
(BSN 2006), MIT, Boston (MA), USA, pp. 5–8.
14. Kwon, Y., Sundresh, S., Mechitov, K., and Agha, G. (2006) ActorNet: An
actor platform for wireless sensor networks, in Proceedings of the 5th
International Joint Conference on Autonomous Agents and Multiagent
Systems (AAMAS), pp. 1297–1300.
15. Lester, J., Choudhury, T., Borriello, G. (2006) A practical approach to
recognizing physical activities, in International Conference on Pervasive
Computing (PERVASIVE), pp. 1–16.
16. Lombriser, C., Roggen, D., Stager, M., and Troster, G. (26 February–2
March 2007) Titan: a tiny task network for dynamically reconfigurable
heterogeneous sensor Networks, in Verteilten Systemen (KiVS 2007),
Bern, Switzerland.
17. Luck, M., McBurney, P., and Preist, C. (2004) A manifesto for agent
technology: towards next generation computing, Autonomous Agents
and Multi-Agent Systems, 9(3), pp. 203–252.
18. Malan, D., Fulford-Jones, T., Welsh, M., and Moulton, S. (June 2004)
CodeBlue: An ad hoc sensor network infrastructure for emergency
medical care, in Proceedings of MobiSys 2004 Workshop on Applications
of Mobile Embedded Systems (WAMES 2004).
19. Maurer, U., Smailagic, A., Siewiorek, D. P., and Deisher, M. (2006) Activity
recognition and monitoring using multiple sensors on different body
positions, in Proceedings of the 3rd International Workshop on Wearable
and Implantable Body Sensor Networks (BSN 2006), MIT, Boston (MA),
USA.
20. Mobile Agent Platform for Sun SPOT (MAPS), documentation and
software at http://maps.deis.unical.it/ (2011).
References 135
21. Muldoon, C., O’Hare, G. M. P., O’Grady, M. J., and Tynan, R. (2008)
Agent migration and communication in WSNs, in Proceedings of the
9th International Conference on Parallel and Distributed Computing,
Applications, and Technologies.
22. Najafi, B., Aminian, K., Ionescu, A., Loew, F., Büla, C. J., and Robert, P.
(June 2003) Ambulatory system for human motion analysis using a
kinematic sensor: monitoring of daily physical activity in the elderly,
IEEE Transactions on Biomedical Engineering, 50(6), pp. 711–723.
23. Pansiot, J., Stoyanov, D., McIlwraith, D., Lo, B. P. L., and Yang, G. Z.
(26–28 March 2007) Ambient and wearable sensor fusion for activity
recognition in healthcare monitoring systems, in Proceedings of the
4th International Workshop on Wearable and Implantable Body Sensor
Networks (BSN 2007), RWTH Aachen University, Germany.
24. Sohraby, K., Minoli, D., and Znati, T. (2007) Wireless Sensor Networks:
Technology, Protocols, and Applications, Wiley.
25. Suenaga, S., and Honiden, S. (14 May 2007) Enabling direct com-
munication between mobile agents in wireless sensor networks, in
Proceedings of the 1st International Workshop on Agent Technology for
Sensor Networks (ATSN 2007), jointly held with the6th International
Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS
2007), Honolulu, Hawaii.
26. Szumel, L., LeBrun, J., Owens, J. D. (30–31 May 2005) Towards a
mobile agent framework for sensor networks, in Proceedings of the 2nd
IEEE Workshop on Embedded Networked Sensors (EmNetS-TT), Sydney,
Australia.
27. Signal Processing In-Node Environment (SPINE), documentation and
software at http://spine.tilab.com (2011).
28. TinyOS website, www.tinyos.net (2011).
29. Yang, G. -Z. (2006) Body Sensor Networks, Springer.
30. ZigBee alliance, technical documents and standard specifications at
http://www.zigbee.org (2011).
Chapter 6
Hardware Development and Systems

for Wireless Body Area Networks
Mehmet Rasit Yuce

Department of Electrical and Computer Systems Engineering,
Monash University, Clayton, VIC 3800, Australia
[email protected]
Miniaturized and wearable sensor nodes are required to form

a wireless body area network (WBAN) to monitor physiological
parameters from human bodies. In this part of the book, we will
present the existing hardware solutions to realize a small wireless
sensor for use in WBAN applications. Implementation issues are
discussed, and details of techniques for the design of wireless
sensors in WBAN applications are presented.
6.1 Introduction
A WBAN network aims to provide efficient and optimized wireless

link for physiological signal monitoring from single or multiple
human bodies. As shown in Fig. 6.1, the future WBAN-based sensor
network requires miniaturized and wearable sensor nodes that can
communicate with the receiving device. A WBAN system consists
of individual wireless sensor nodes that can transfer a person’s

www.panstanford.com
138 Hardware Development and Systems for Wireless Body Area Networks
Figure 6.1. A wireless sensor network system detecting and transmitting

signals from a human body.
physiological data such as heart rate, blood pressure, and ECG via
a wireless link, without the need of any wired connection. Each
sensor will have wireless capability and its design will be optimized
in terms of the physical characteristic of the physiological signal.
Having individual wireless nodes are also very beneficial since not
all patients require all the physiological parameters for diagnosis.
Each sensor can easily be plugged in and plugged out from a human
body.
6.2 Wireless Body Sensors
Figure 6.1 shows a WBAN-based telemedicine system for collecting

medical data from a human body using wireless sensors. This
telemedicine system comprises sensor nodes, a body control unit
(BCU) that transmits data to a receiver station (i.e., remote PC)
that can have communication with remote stations at a medical
center [1].
Wireless Body Sensors 139
The BCU can be a smart phone or any portable device that will
act as a gateway device to pass on the collected data from sensors to
remote stations. The control device will be similar to smart phones
we use in our daily life, which can also be used to monitor the data
obtained from sensors. Such portable electronic devices will mostly
contain Bluetooth, Wi-Fi, and GPS technologies built-in, which can
be connected to medical sensors and to remote medical centers so
that data can be accessed by medical professionals at any time. The
BCU can be carried on the body as a wristwatch or can be placed on
the belt.
For collecting data from a human body, usually short-range
wireless technologies operating at MICS (medical implant communi-
cation service), WMTS (wireless medical telemetry service), 2.4 GHz
ISM (industrial, scientific, and medical) bands have been developed
by designers. However, it is necessary to connect the sensor nodes
to remote stations using Wi-Fi and mobile communication links
for longer range data transfer. In order to provide long-range
remote monitoring, several gateway devices should be developed
to interface with the existing ICT (information and communication
technology) infrastructure in emergency vehicles or in medical
centers. These gateway devices will mainly be used to provide
communication between BCUs and remote computers or mobile
devices.
6.2.1 Sensor Nodes and Hardware Designs

Low-power wireless technologies should be used to form small,
light-weight, and wearable sensor nodes to collect and sense
physiological data from a human body. Especially, when the number
of sensors used on a single body increases, the weight of each sensor
node becomes an important issue. Wireless sensor nodes are being
developed by biomedical companies and researchers to make them
safe, secure, and light weight so that they can be wearable. Key
parameters of a wireless sensor node can be as follows:
• The wireless sensor nodes should be able to transfer data

over a distance of a few meters.
• Sensor nodes should be miniaturized so that they can
be easily wearable or attachable to a patient body.
Table 6.1. Physiological parameter range and signal frequencies
Parameter Range of parameter Signal frequency

Continuous Signals ECG signal 0.5–4 mV 0.01–250 Hz
EMG (Electromyogram) 10–15 mV 10–500 Hz
EEG 3–300 μV 0.5–70 Hz
◦
Discrete Signal∗ Body Temperature 32–40 C 0– 0.1 Hz
Blood Pressure (BP) 10–400 mm Hg 0–50 Hz
GSR (Galvanic Skin Reflex) 30–3 mV 0.03–20 Hz
Respiratory rate 2–50 breaths/min 0.1–10 Hz
∗
Here the term “discrete” indicates large time intervals.
Alternatively, they can also be inserted in clothes using

textile technology to improve the patient’s comfort.
Wireless sensor nodes are developed specifically to detect and

transmit the physiological signals listed in Table 6.1. These signals
can be put in two categories: discrete and continuous. ECG/EEG
signals are periodic signals, which require continuous monitoring.
Body temperature is an example of discrete physiological data since
it is taken usually at discrete times, e.g., every hour.
Most physiological signals are low frequency in nature and
occupy a small information bandwidth. At such low frequencies and
low amplitudes, some problems inherent to circuits need additional
attention. For reliable information transfer, it is necessary that the
interface electronic in the sensor nodes detect physiological signals
in the presence of noise and increase the signal-to-noise ratio (SNR)
of the detected signal for processing by the subsequent blocks of the
sensor nodes.
Sensor nodes are designed to be small and power efficient so
that their battery lasts for a long time. They collect weak, raw
signals from a human body. Signals from a human body are usually
weak and coupled with noise, and thus they need to go through
amplification/filtering process to increase the signal strength and
to remove unwanted signals and noise. Then an analog-to-digital
conversion (ADC) stage is employed to convert the analog body
signals into digital for digital processing. The digitized signal is
processed and stored in the microprocessor. The microprocessor
(i.e., microcontroller) will then pack the data and transmit over the
Figure 6.2. An example of sensor node hardware: (a) single wireless

sensor node, (b) multi-channel wireless sensor node. See also Color Insert.
air via a wireless transceiver. Figure 6.2 shows two commonly used
hardware implementations of sensor nodes.
A sensor node consists of three main blocks: the sensing front
end, microcontroller, and the radio receiver to undertake the tasks
mentioned above. In addition to these blocks, there is also a battery
and its power management circuitry. The power management circuit
is usually a regulator chip used to distribute the power source to the
individual blocks. It is advised to keep all the sensor blocks’ power
supply level the same (e.g., all at 3 V) so that the regulator will not
consume a large power.
Considering the signal attributes given in Table 6.1, a sampling
rate of around 200–1000 Hz will be necessary for the ADC in the
microcontroller (the sampling rate should be a minimum of twice
the highest frequency in the signal that is digitized). Unlike other
wireless sensor network applications, in a medical scenario, each
sensor node has a different frequency and thus should be optimized
according to its frequency band. The trade-off between the reduction
in sampling rate and the total power consumption of the ADC is thus
determined by the choice of the specific physiological parameter

used in the sensor node.
In case of emergency in a medical application, usually the
monitoring and analyzing of critical physiological signals such as
ECG and blood pressure (BP) will be the top priority. Thus, it is
important if the signals in Table 6.1 are grouped into critical and
noncritical data before attaching sensors on the injured people in a
medical area. The wireless communication from sensors to the BCU
should be designed such that priority should be given to the critical
data. A WBAN-based telemedicine system will result in better data
gathering process if data are prioritized according to the vital sign
and patient situation.
The front end of a sensor node (i.e., interface electronics) is
the first unit connected to sensors and electrodes. It is an analog
electronics circuit that consists of amplifiers and filters. In case
of detecting a continuous physiological signal such as EEG, ECG,
or EMG, the front end should use an instrumental amplifier. An
instrumental amplifier helps to amplify and detect noisy body
signals at the presence of common mode noise (DC noise) at
the input. By looking at Table 6.1 for amplitude and frequency
information on continuous physiological signals, parameter values
could be similar or different in some cases. It mainly depends on the
quality of electrodes used to detect these signals. The same analog
front end can be used to detect these continuous physiological
signals. On the other hand, the required amplification and filtering
can easily be adjusted for a specific signal by modifying capacitors
and resistors in the front-end circuit.
Figure 6.3 shows a front-end circuit that can be used to detect
ECG, EEG, and EMG signals. This circuit utilizes an instrumentation
amplifier (INA321) to detect noisy body signals and a low-pass
filter with an op-amp (LTC6081) [2]. The final stage is a 100 Hz
LPF added for antialiasing. Typically the useful spectrums of ECG
and EEG signals have amplitude of less than 500 μV with a
frequency less than 100 Hz. In this example, the input medical
signals are amplified by 60 dB; INA321 produces a gain of 14 dB, and
the second amplification stage with low-pass filter characteristic-
LTC6081 provides a gain of 46 dB with a cutoff frequency at 100 Hz.
The circuit is running from a 3.3 V source with virtual ground set
Figure 6.3. Schematic of analog front end for continuous physiological

parameter detection.
at 1.25 V. Both components possess high common mode rejection

ratio (CMRR) and low current consumption properties. INA321 has
a CMRR of 94 dB, while LTC6081 has a CMRR of 100 dB. High
CMRR values help obtain better quality signal. The active current
consumption is 40 μA, and when operating in shutdown mode,
it consumes less than 1 μA. The INA321 not only has extremely
good CMRR and meets the low power need, but also has the ability
to shut down when not in use. This is directly controlled by the
microcontroller by connecting the shutdown pin to a CMOS I/O pin.
√
The INA321 instrumental amplifier has 100 nV/ Hz, @ 1 kHz, input
√
voltage noise and 3 fA/ Hz @ 1 kHz input current noise.
To eliminate DC drift caused by ICs, an integrator op-amp circuit
is used as a negative feedback to the reference pin of the INA321.
Any DC buildup is inverted and fed back into the system to cancel it
out. This is only necessary for high gain circuits since any DC drift is
multiplied just like any other signal. With a gain of 1000, a shift of 1
mV will cause the output to jump 1 V, which is close to saturation in
this case. Apart from the integrator feedback, this circuit is relatively
straight forward and has many common features to any other signal
processing circuits.
The 2 M and 100 k resistors are used to limit the current to
provide protection for the human body. In addition to this, it is also
advised to use a safe power supply. Using a portable battery during
Figure 6.4. ECG detection with the AD620 instrumentation amplifier.
the design of this circuit is a better idea rather than using a direct
source from a power line.
The AD620 is another well known low cost, high accuracy
instrumentation amplifier (Analog Devices, USA), which can be used
√
to detect body signals. It has 9 nV/ Hz, @ 1 kHz, input voltage noise
√
and 0.1 pA/ Hz @ 1 kHz input current noise. Similar to INA321,
the AD620 can be configured to have a CMRR value of 100 dB. The
details of circuit configuration as a front end to detect sensor signals
can be found in references [3] and [4]. Figure 6.4 shows a schematic
for ECG detection designed with AD620. As explained before, the
additional filter and a gain stage are added to make the detected
signal ready for digitization at the micro-controller.
Another noise exists in detecting body signals, known as the
power-line interference at 50 Hz (or 60 Hz in some countries). A
notch filter (a combination of low-pass filter and high pass filter)
can be designed at the front end to remove this noise. Alternatively,
this can be done digitally at the microcontroller or at the remote
computer after the continuous signal is received wirelessly [4]. An
analogue notch filter works for most applications, but it requires
more power and increases the board size and is, thus, avoided.
Figure 6.5 shows an ECG signal obtained using the circuit given in
Fig. 6.3. Figure 6.5b shows a clear ECG signal after using a notch filter
to remove the 50/60 Hz noise.
Employing a digital–notch filter at the computer will make the
sensor node less complex. It is also possible to use a combination
Figure 6.5. ECG with 60 Hz (or 50 Hz) noise (a) and without noise (b). See
also Color Insert.
of an analog notch filter at the front end and then digitally via
computation programs like MATLAB. The latter approach will
increase the clarity of the signal further.
Some ECG projects use a “right leg drive” circuit, which is simply
the inverse of the 50/60 Hz injected back into the patient to cancel
the power line noise out. Similar to the above approach, this would
require more op-amps and an extra electrode and is preferred not
to be used. This method also requires a small amount of current,
approximately 1–3 μA, to be passed through the patient.
In case of detecting discrete physiological signals such as body
temperature and respiratory rate, a simple active low-pass filter (a
filter together with a gain amplifier) is sufficient to bring the signal
detected by a sensor to a level to be digitalized by a microcontroller.
Figure 6.2b shows a sensor node based on multichannel
detection. A multiplexer is an array of analog switches operated
with a clock that allows all the input signals to a single output
to be digitized at the microcontroller. A multichannel sensor node
requires the use of wires connected to electrodes for a single sensor
node which should be avoided if possible for a WBAN application.
The existing of wires will restrict human movements and thus
comfort.
Having said that, for some clinical applications, a multichannel
sensor node may be required for certain diagnostic tools. For
example, a multichannel sensor node is used for EEG signals
for brain-computer interface applications [5]. Multichannel signal
detection will increase the data rate requirement. As an example, if

we use 200 Hz sampling frequency for 10 channel EEG signal, the
required data rate is calculated as 200 sample/s per channel × 10
channels × 8–10 bits/sample = 16–20 Kbit/s.
One of the requirements of body area network (BAN) is to
have individual wireless sensor nodes. Even if we integrate a
multichannel EEG signal into a BAN system and considering the
additional preamble bits in a data package, the required data
rate for a WBAN-based physiological monitoring system is still
sufficiently low. A lot of available wireless transceiver will provide
enough bandwidth to handle such data rates. When the number of
sensor nodes increases, the required data rate will depend on the
multi-access communication used. In case of TDMA (Time Division
Multiple Access) and CSMA (Carrier Sense Multiple Access), the
bandwidth requirement is low. For frequency division multiplexing
techniques (e.g., FDM, FHSS, OFDMA) and CDMA schemes, a
wideband will be required for data transmission when the number
of users (here sensors) is increased. Probably this is one of the main
reasons why the WPAN (wireless personal area network) systems
such as ZigBee use CSMA as a multi-access communication. It can
have a higher number of nodes with available low bandwidth. For
more information about multi-access communication techniques in
WBAN, refer to the related chapters of the book.
6.2.2 Wireless Systems and Platforms

In addition to the sensing front end, a sensor node and the BCU
require a microcontroller and a wireless transceiver to coordinate
the transfer of data. Numerous frequency bands can be used in
medical monitoring system. So far there is no available standard
to define operation frequencies specifically targeting a WBAN. It is
important to select a proper and safe (interference free) wireless
band. Especially for medical data transmission, it is crucial for
the patient‘s safety to monitor accurate information. Table 6.2
presents the unlicensed frequency bands that can be used for WBAN
applications.
Low ISM bands, especially the 13.56 MHz frequency, are
widely used for radio-frequency identification (RFID) applications,
Table 6.2. Available frequency bands
RF power
Model Country Frequency BW (dBm)
UWB Worldwide 3.1–10.6 GHz >500 MHz −41
MICS US, Australia 402–405 MHz 300 kHz, 10 −16(25 μW)
Europe, Japan channels
Worldwide
High ISM Worldwide 2.4–2.5 GHz, 20 MHz, 40 MHz >0
5.725–5.875 GHz
Mid ISM Europe US 433.05–434.79 MHz kHz range , Up to 15
Canada, 865– 868 MHz 200–500 kHz
Australia 902–928 MHz 10–15 channels MHz
Low ISM Worldwide 6.765–6.795 MHz kHz range, 14 kHz for 0

13.55–13.567 13 MHz
26.95–27.283 40.66–40.70
WMTS US, Canada 608–614 1395–1400 1.5 MHz 5–6 MHz >10
1427–1432
inductive links for implantable systems, and smart cards for security
systems [6, 7]. Most probably designers will continue to use these
low-frequency ISM bands for inductive link based implantable
systems as they present better performance at low frequencies. For
medium ISM (Mid-ISM) range, countries in Europe, New Zeeland,
and Honk Kong use 865–868 MHz. Australia, Korea, Taiwan, Hong
Kong, and Singapore use a frequency range within and the US and
Canada use the whole band between 902–928 MHz. Japan has 950–
956 MHz ISM band frequencies. These frequencies have widely been
used for cordless headphones and microphones.
In addition to unlicensed ISM bands, there are medical bands
such as MICS and WMTS, which are specifically regulated for
medical monitoring by communication commissions around the
world. The MICS is an ultra-low power, unlicensed, mobile radio
service for transmitting data to support diagnostic or therapeutic
functions associated with implanted medical devices [8]. In order
to monitor medical implant devices and status of inner organs, a
frequency around 400 MHz have been a popular transmission band.
In fact this frequency is an optimal frequency to provide a fully
integrated circuit approach for a wireless transceiver design and

causes relatively insignificant penetration loss (10 dB with 10 mm
penetration) [9]. Using a higher frequency causes higher penetration
loss. Meantime high-level integration becomes difficult at lower
frequency. In addition, a small antenna design is difficult at lower
frequency (less than 400 MHz).
In addition to medical implants, metrological aid service has
primary allocation at 402–405 MHz band and the Earth exploration-
satellite service together with metrological-satellite service has
secondary allocation at 402–403 MHz. Thus transmission in the
MICS has some regulated parameters in terms of channel spacing
and transmission power level. At the moment in some countries, the
MICS is being extended from 402–405 to 401–406 MHz, introducing
additional 2 MHz wider band, which can increase the population of
monitoring patients in hospital environment.
WMTS is used in the United States and Canada but not in Europe.
In Japan, WMTS frequencies are 420–429 MHz and 440–449 MHz.
WMTS has long been used for wireless patient monitoring system in
the United States and Japan [10].
The most common wireless systems such as ZigBee (IEEE
802.15.4), WLANs, and Bluetooth (IEEE 802.15.1) operate at 2.4 GHz
ISM band. The technologies operating at 2.4 GHz ISM band may
suffer from interference issues when they are located in the same
environment. In addition, there may be a variety of other sources
such as microwave and cordless phones operating in 2.4 GHz
band. Latency sensitive WBAN nodes like an ECG node can suffer
significance performance loss from this mutual radio-frequency
interference as a correct submission of a sensor will be delayed.
The electromagnetic interference is another source of interfer-
ence that may limit the use of wireless in certain environments,
especially in medical applications. Electromagnetic energy from
a device can interfere with another when they are operating in
the same environment. There have been already incidents where
telemetry devices have affected the available medical systems in the
hospitals [11].
In addition to the above wireless bands, the ultra-wide band
(UWB) can also be used for monitoring of physiological signals.
High data rate transmission, as commonly known, is not the only
unique property of an UWB-based monitoring system. Some other

advantages of a wide band technology are its low transmitter
power, the physical size which can be extremely small because
of the design at GHz range, the simplicity of transmitter design,
and the band is not crowded as compared to other available
bands [12]. The maximum transmission power is regulated as -
41 dBm. Such a low-signal level will have insignificant RFI/EMI
(radio frequency interference/electromagnetic interference) effect
on medical equipment. The current drawback of UWB technology is
that not enough devices available in the market that can be used as
a complete solution.
The MICS band has also low emission power (25 μW, comparable
to UWB) and lower power consumption, and will thus provide one
of the most suitable medical sensor nodes. It is envisioned that the
sensor nodes of a WBAN will use a specified wireless link, which
will most likely be a narrow ISM band considering the current
wireless technology developments. The UWB band can be used
when interference and EMI are an issue for the environment. For
implantable body area networks, the MICS band may still continue
to be the preferable choice.
6.2.2.1 Wireless transceivers and microcontrollers

Most common wireless chips used in low power sensor network
applications are listed in Table 6.3. The chips that exhibit low power
and have small size are Zarlink’s MICS chip, Nordic’s nRF24E1
device, and CC2420 of TI. Zarlink provides one of the lowest
power wireless chips available today. The low power is achieved by
reducing the supply voltage to a value as low as 1.2 V, while most of
the other transceiver chips operate with a nominal voltage around
3 V. The power figures shown in Table 6.3 are for the maximum
transmission power levels and a nominal supply voltage, which is
3 V. Although these chips are advised to operate at 3 V, their supply
voltage in practice can be reduced to as low as 2 V. This way the
power consumption will be lower than what is shown in the table.
The Zarlink chip has been designed for implantable sensor nodes.
The Nordic device nRF24E1 consumes a transmit current of 10.5 mA
for an output signal level of −5 dBm and 18 mA receiver current for
October 19, 2011
17:4
Table 6.3. Some suitable wireless chips for potential use as a wireless medical sensor node
Power consumption
Model Company Frequency Data rate RF power (dBm) Physical dimension Tx Rx
CC1010 TI1
300 to 1000 MHz 76.8 Kbps −20 to +10 12 × 12 mm (chip)
2
26.6 mA 9.1 mA
CC24xx TI1 2.4 GHz 1 Mbps −25 to 0 7.1 × 7.1 mm2 (chip) 19 mA 17 mA
XE1205 Xemics2 (Semtech) 433, 868, 915 MHz 304 Kbps Up to 15 8 × 8 mm2 (chip) 62 mA 14 mA
PSP Book - 9in x 6in

CX72303 Conexant3 2.4 GHz 1 Mbps −10 to +2 — 34 mW 43 mW
nRF24E1 Nordic4 2.4 GHz 2 Mbps −20 to 0 6 × 6 mm 11 mA 18 mA
AMIS52100 AMIS5 402–405 MHz 16 Kbps −3 to 12 7.5 × 7.8 mm2 (chip) 25 mA 7.5 mA
ZL70250 Zarlink6 402–405 MHz, 433 MHz ISM 800 Kbps <0 7 × 7 mm2 (chip) 5 mA, TX/ RX
1
http://www.ti.com/.
2
http://www.semtech.com.
3
http://www.conexant.com.
4
http://www.nordicsemi.com.
5
http://www.amis.com.
6
http://www.zarlink.com/.
06-Mehmet-c06
a data rate of 250 Kbps. The physical dimensions shown in the table
are based on the packages used by the companies.
The Nordic has another transceiver -nRF401 that operates in the
433 MHz ISM frequency band. It uses frequency shift keying (FSK)
modulation and presents a data rate up to 20 kbit/s. The current
consumptions for both transmitter and receiver are 11 mA and 8
mA, respectively.
The CC1010 transceiver chip from TI (Chipcon) has the capability
to transmit anywhere within 300 and 1000 MHz. It can be used to
generate signals for MICS, Mid-ISM bands, and WMTS frequencies.
The AMIS52100 chip from AMI Semiconductor requires a crystal
oscillator at 12.56 MHz to generate MICS band. A crystal with any
frequency can be used with CC1010 to generate medical and ISM
band frequencies. The effective frequency of the device is arranged
by the external VCO inductor in addition to the crystal oscillator.
For example, an inductor value of 12 nH gives a frequency range of
565–730 MHz. The PLL in chip is able to quickly lock onto WMTS
frequency of 610 MHz. The frequency registers of the device allow
for jumps of 250 Hz, making this device suitable for multiple channel
selection and taking advantage of the available bandwidth by using
FDMA or frequency hopping.
The Semtech’s XE1205 chip can be used for the 433, 868, and
915 MHz license-free ISM frequencies. Similar to CC1010, it can
also operate in other frequency bands in the 180–1000 MHz range.
CX72303 from Conexant operates at 2.4 GHz. This technology is
specifically designed for Bluetooth system solutions
Another Chipcon’s chip CC2420 is a 2.4 GHz IEEE 802.15.4/
ZigBee compliant RF transceiver. It is based on digital direct
sequence spread spectrum (DSSS) baseband modem providing a
spreading with data rate of 250 Kbps. This chip has been widely used
in sensor nodes because of its low-power design and its operation
for ZigBee and Bluetooth applications.
AMIS52100 and TI’s cc10xx chip series can provide MICS and
WMTS signals. Both chips’ output can go as low as −20 dBm
meeting the requirement of −16 dBm transmission power for
the MICS band. Zarlink’s ZL70250, CC1010, and Semtech XE1205
uses FSK data modulation, AMIS52100 amplitude shift key/on-off
key (ASK/OOK) modulation, while CC2420 uses offset quadrature
phase- shift keying (O-QPSK) with half-sine chip shaping, which

is equivalent to MSK modulation. The newer version transceiver
from AMI Semiconductor (AMIS53000) can operate using one of the
following modulation schemes: OOK, FSK, or GFSK. The nRF24E1
operates at 2.4 GHz with GFSK modulation. The transceivers can
work based on both Manchester and NRZ bit coding schemes;
however, the data rate is half in the case of Manchester coding.
6.2.2.2 Existing sensor boards

The unit responsible for processing data, controlling the function-
ality of the components of the sensor node, and coding the sensor
data for the wireless transceiver is the microcontroller. The task of a
microcontroller for a sensor node in WBAN applications is relatively
simple due to the low frequency of biological signals. Thus a less
complex and small size microcontroller can usually be utilized. The
commercially available microcontrollers have a 10-bit and 8-bit ADC
built-in, which eliminates the need for an external off-shelf ADC.
Considering that the sensor node board is required to be light
and small in order to be wearable, a wireless transceiver chip
containing a microcontroller would be advantageous. As an example,
CC1010 from TI has a microcontroller builtin which has a memory
of 2048 + 128, 10-bit ADC, and 22.7 kHz sampling frequency. A
small board can easily be designed, similar to that of Mica2DOT
shown in Fig. 6.6. The Nordic’s nRF24LE1 chip also contains a
microcontroller unit and an ADC. The microcontroller is an 8051
compatible microcontroller, and the ADC is 10 bit with 100 Kbps.
Figure 6.6. A Mica2DOT board (taken from www.xbow.com). See also

Color Insert.
The sensor board with these devices will require only a few external
components: the needed sensing front end given in Fig. 6.3, regulator
IC for power supply arrangement and matching network together
with antenna.
Some of the commonly used wireless sensor platforms (as
a complete sensor boards) are shown in Table 6.4. Crossbow’s
Mica2DOT and T-node are the smallest sensor nodes available in
commercial domain. Another widely used sensor node is Tmote
Sky node [13]. It uses 250 Kbps 2.4 GHz IEEE 802.15.4 Chipcon
Wireless Transceiver (CC2420) and a separate 16-bit 8 MHz Texas
InstrumentsMSP430 microcontroller (10 K RAM, 48 K Flash).
MicaZ node also uses CC2420 transceiver and 10-bit ATmega
128L microcontroller, which has 4 KB RAM and 128 KB FLASH
memories. The MicaZ is a 2.4 GHz mote module used for enabling
low-power wireless sensor networks. TinyNode584 is another ultra-
low power wireless sensor node, which uses Xemics XE1205 for
wireless communication and TI MSP430 microcontroller (10 K RAM,
48 K Flash) as a processing unit [14]. The physical dimension of this
board is 30 × 40 mm2 .
The Crossbow’s Mica2 sensor node works at 868 and 916 MHz
ISM frequencies. Like MicaZ it uses a separate microcontroller —
ATmega128L (4 K SRAM, 128 K Flash, 8 MHz). The total power
consumption of this board at 3.3 supply voltage, including the radio
and microcontroller, is 148. 5 mW (= (27 + 10 + 8)*3.3). During
sleep mode, the power consumption is at μW levels.
The Mica2DOT mote is a third-generation mote module wireless
platform designed for smart sensors similar to the Mica2. Its
quarter-sized (25 mm) form factor makes it suitable for sensor
network applications requiring low size. Mica2DOT is one of the
smallest sensor nodes available in commercial domain (Fig. 6.6).
The power consumption of this node is similar to Mica2 (MPR400).
Mica motes run on TinyOS (TOS) distributed software operating
system v1.0 [15]. They can operate on the frequencies such
as 868/916 MHz, 433 MHz, or 315 MHz multichannel radio
transceiver. The Mica2DOT is Mica2 compatible and uses the same
microcontroller.
Another small node for sensors is T-nodes working in the
868 MHz ISM frequency band. The board uses a separate chip for
October 19, 2011
17:4
Table 6.4. Various hardware sensor nodes configurations
Power consumption†
Model Company Frequency Data rate Trans. power (dBm) Physical dimension Tx Rx
Mica2 (MPR400) Crossbow1 868/916 MHz 38.4 Kbps −24 to +5 58 × 32 × 7 27 mA @ 3.3 V 10 mA @ 3.3 V
18 g (board)
MicaZ Crossbow1 2400 MHz to 2483.5 MHz 250 Kbps −24 to 0 58 × 32 × 7 17.4 mA @ 3.3 V 19.7 mA @ 3.3 V
(IEEE 802.15.4) 18 g (board)

Mica2DOT Crossbow1 868/916 433 MHz 38.4 Kbps −20to +10 25 × 6 mm2 25 mA @ 3.3 × 8 mA @ 3.3 V
3 g (board)
Tmote Sky node Moteiv 2.4 GHz (IEEE 802.15.4) 250 Kbps −25 to 0 66 × 32.6 × 7 mm 19.5 mA @ 3 V 21. 8 mA @ 3 V
T-node SOWNet 868, 433, 915, or 315 MHz 52.2 Kbps −20 to −5 Diameter of 23 mm 25 mA @ 3 V 13 mA @ 3 V
TM
Sensium Toumaz 868/915 MHz 50 Kbps −23 to − 7 90 × 45 × 10 mm 3 mA @ 1 V 2.7 mA @ 1 V
Wibree Nokia 2.4 GHz (Bluetooth) 200 Kbps −6 — — —
1
http://www.xbow.com.
† Transmitter (Tx) power values are usually for the maximum transmission power level. When the transmission power level is configured to lower values,
transmitters will consume a power lower than the values in this Table.
06-Mehmet-c06
Figure 6.7. A T-node sensor node [16]. See also Color Insert.
microcontroller with 10-bit ADC, memory of 128 K flash memory

and 4 Kbyte SRAM [16]. It consumes only 20 μA during sleep mode.
It has a similar size to that of Mica2DOT (23 mm), shown in Fig. 6.7.
TM
The Sensium sensor node uses the chip TZ1030, which uses
a wireless transceiver operating at 868/915 MHz and the 8051
processor for digital processing and with 10-bit ADC. It provides
50 Kbps of data rate [17]. The low power consumption of this node is
resulted from the low-supply voltage of 1 V. The node samples input
signals within a dc to 250 Hz bandwidth.
Another popular node is BTnode [18], which consists of a
Chipcon CC1000 radio (it is CC1010 without a microcontroller built-
in) operating at 433–915 MHz, Atmel AVR microcontroller, and a
Bluetooth radio (Zeevo ZV4002). In other words, a BTnode is made
of a Crossbow Mica2 Mote with SRAM (256 K) and an additional
Bluetooth radio. This node’s power consumption is high due to the
operation of two radios (∼200 mW). However, it gives the advantage
of connecting to another device using the Bluetooth technology. This
node can be used to interface with mobile phones and the Internet
using the Bluetooth link. Its microcontroller is from Atmel (ATmega
128L), having 64 + 180 Kbyte RAM, 128 Kbyte FLASH ROM, and
4 Kbyte EEPROM.
The Conexant is another node based on Bluetooth wireless
technology. It uses CX72303 RF transceiver and CX81400 baseband
controller to establish a Bluetooth communication link [19]. The
CX81400 integrates an RISC processor ARM7TDMI having 192 KB
ROM and 192 KB RAM.
Among the available Bluetooth modules Wibree is probably the
most suitable node for a lowpower sensor network application.
The device has been designed by Nokia to have similar power

consumption and battery life as ZigBee. This device similar to
ZigBee applications, because of its small form factor, can be used in
health care, fitness, security, wrist watches and home entertainment
industries. It operates on coin cell batteries. Wibree operates in
2.4 GHz band with a data rate up to 1 Mbps with a 10 m distance
similar to any other Bluetooth technologies.
In addition to these commercially available sensor nodes, there
are also nodes that have been developed specifically for medical
monitoring. In reference [20], a very compact system has been
developed for a wearable ECG sensor node that can communicate
wirelessly with a base station. The node uses Nordics nRF24E1, a
2.4 GHz RF transceiver together with an embedded 8051-compatible
microcontroller (DW8051) with 512 byte ROM, 4 K RAM, and 9-bit
ADC. The board has the size of 26 (L) × 15 (W) × 7 (H) mm3 .
BSN nodes developed by Imperial College [21] are TinyOS-based
sensor nodes The node also uses CC2420 from TI and MSP430
microcontroller families with 2 KB RAM, 60 KB flash, and 12-bit
ADCs. The node has a size of 26 mm. The power consumption
of these nodes is similar to the power consumption of CC2420
transceiver given in Table 6.3. The sensor board in reference [22]
uses the Nordic nRF905 transceiver however operating at 915 MHz
ISM with 0 dBm and 10 Kbps. This sensor node includes 3-axis
accelerometer SCA3000 and has active power of 7.8 mW and 8.2 mW
for transmitter and receiver respectively. Its dimension is as small as
a one dollar coin.
Another sensor node [23] developed for a medical sensor
network uses the medical bands MICS and WMTS for applications
in the hospital environment. The board uses AMIS and CC1010
transceivers for the radio system and PIC16F785 microcontroller
having 10-bit, 52 kHz sampling clock , and 368 + 256 RAM. The
board has a dimension of 30 mm × 80 mm.
The power consumption given Table 6.4 defines the life time of a
sensor node. For example, a continuously active sensor node using
the CC24XX family (this chip family is used for ZigBee and also for
Bluetooth) can last about seven hours when using a battery with a
capacity of 300 mAh, assuming the transmitter and receiver working
with the same amount of time (the nodes are always active either
there is transmission (the transmitter is active) or the receiving

(the receiver is active)). For monitoring patients in emergency and
disaster events, sensors measuring critical vital signals are required
to be active (like ECG signal) as these vital signals are crucial for
the patient’s life. However, there may be some sensors information
needed to be monitored once a while. The life time of such a sensor
will be similar to current cellular phones we use in our daily life.
It is important to note that the power consumption values given in
Table 6.4 is for the maximum transmitter power. When the distance
between the control unit and the sensor nodes is short, which
is likely the case for the targeted medical applications, a lower
transmission power can be configured in the transceiver of the
sensor nodes to increase the battery life.
6.2.3 Design of Implanted Sensors Nodes for WBAN

Electrocardiogram (ECG) and temperature recording have been
used for more than 50 years in medical diagnosis to understand
biological activities [24] The early implantable devices are con-
structed with simple electronic structures to make the devices
small enough so that they can be inserted in the body. A simple
transmitter connected to a sensor is used to send the signal
from inside the body to external devices for tracking certain
organs’ physiological parameters. Recent implantable systems have
focused on more complex microelectronics systems with the use
of the advanced integrated circuit technology. Some milestones of
implantable devices are given below:
• 1957 — First wearable pacemaker (in 1958 fully implant)

• 1984 — The Australian cochlear implant approved by
FDA
• 2000 — First clinical trial wireless endoscope (electronic
pill)
• Study of Bionic eye since 1990
In addition to these projects, there is a vast area of different

implantable systems, being designed by designers, such as car-
dioverter defibrillators, micoelectrode arrays for detecting and
stimulating brain functions, and electronic pills for drug delivery

systems. Moreover, there are implantable biosensors developments
to conduct animal studies.
Electronics for implantable sensors are designed taking into
account very specific requirements for a given wireless telemetry
application. Wireless technologies such as ZigBee, WLANs, and
Bluetooth (IEEE 802.15.1) developed in the commercial domain
cannot be used directly in medical implants because of the following
reasons: (1) they have been optimized for general use; (2) the device
size exceeds the required size limitation of the current implant
technology, (3) their frequency bands are very crowded, and (4)
they do not meet the safety-related radiation. The existing advanced
wireless systems operate at 2.4 GHz ISM band and suffer from
strong interference from each other when they are located in the
same environment [25]. Thus an implant system should have a
different transmission band for an interference-free wireless link
as the transmitted information could be related to patients’ critical
conditions.
In implantable telemetry systems, to optimize the power
and data telemetry links independently, the dual-band (i.e., two-
inductive links) has been used [26]. The power and data signals
are carried on two wireless links. For implantable systems like
the cochlear implant and bionic eye, one wireless data link (i.e.,
forward data link) is used to transmit data from the external
unit to inside the skin to stimulate the nerves in order to help
the patient restore hearing and sight. These devices also require
the transmitting of status information regarding the operation of
the device, which is done based on a third link-back telemetry. There
are also systems such as pacemakers, electronic pill, and biosensors
that send physiological data such as ECG, pressure, or video images
from inside the body to the external unit and then to the monitoring
station in order to track a patient’s condition. The latter implantable
systems may or may not require the forward data link depending on
whether data or a control signal is required from the external unit to
control the implantable system.
Figure 6.8 shows the basic building blocks of implant systems.
The power signal is received and then regulated to provide
supply for the wireless data links, the sensors, and the signal
Figure 6.8. An implanted sensor node with the external unit and wireless
data links. See also Color Insert.
acquisition unit, which processes and amplifies the senor data. The
external unit may contain another wireless link that will control
the communication between the implant devices and the remote
monitoring station. The patient will wear the external unit, which
will be on the surface of the body close to the implanted device.
A remote wireless receiver will be used to receive data from the
external unit for monitoring and analyzing the data. As the required
specifications are not restricted for this link, commercial wireless
devices can be used to undertake this duty. This additional wireless
link will provide patient free in a room in the hospital environment.
Implantable systems have been constructed using inductive links
between the external and implanted units with a distance of a
few centimeters. The main reason behind the short distance and
the inductive link is the need of wireless powering. Inductive
links are mainly used to power implants to eliminate the use of
batteries. The wireless module is designed with a very simple
communication scheme such as binary ASK (amplitude-shift key-

ing), FSK (frequency-shift keying), and PSK (phase-shift keying)
modulators and demodulators [26].
The 13.56 MHz ISM band is usually the most common frequency
for such telemetry systems and is also used for RFID applications.
A low transmission frequency usually less than 20 MHz is utilized
mainly because of the simplicity in the design and to avoid the
use of power-hungry blocks such as mixer, oscillators to maintain
the miniature size of the implant. For example in reference [27],
PSK is used at 20 MHz carrier frequency for the wireless data
link for a retina prosthesis. ASK is used for a bionic eye system in
reference [28] with 1–10 MHz frequencies and for a recent cochlear
implant in reference [29] at 2.5 MHz transmission frequency.
In addition, there are some attempts for general implants and
neural prostheses systems employing FSK [30, 31] operating at
480 kHz and 5/10 MHz frequencies, respectively. The most suitable
frequency for such implantable system requiring low frequency
forward data transmission would be the ISM band 13.56 MHz
as given in Table 6.2. Otherwise, designers would have to obtain
permissions from communication authorities (e.g., FCC) to use
a frequency rather than ISM. In addition to the wireless power
link, utilizing a low frequency transmission has been attractive to
eliminate the absorption of more energy [32].
There are some issues with the communication schemes of
implantable systems for forward data telemetry and back data
telemetry. For example, when many similar types of implantable
systems are used in the same environment or for the same patient,
the simple communication systems mentioned above will face the
problem of interference and lack the multiuser (i.e., multi-access)
capability. For a frequency less than 20 MHz, you are only allowed to
use a few kHz bandwidths, which limit the use of the multi-access
(also known as multiuser) communication techniques to allow
several implantable devices work in the same environment. For
example, allowable bandwidth for 13.56 MHz ISM (the industrial,
scientific, and medical) band is regulated with 14 kHz. Thus a
more advanced wireless technology will be required in the future to
accommodate better radio links for medical implants enabling safe
and reliable communication links [33].
The above issue has become more evident in practice when

more advanced medical implants such as wireless implantable
cardioverter defibrillators and electronic pills (i.e., wireless endo-
scopes) started to appear in clinical environments. These modern
devices need a much longer range for the wireless telemetry
in addition to a higher bandwidth. Moving to higher operation
frequencies was the only way to increase the range and dedicate
enough spectrums for a reliable communication. Thus, in order
to alleviate some of the issues mentioned above, international
authorities have allocated a band at 402–405 MHz with 300 kHz
channels (MICS band). With the use of this band, it is aimed to
deliver high level of comfort, better mobility, and better patient care
[34, 8].
The MICS band has been proposed to permit individuals and
medical practitioners to utilize ultra-low power medical implant
devices such as cardiac pacemakers and defibrillators, particularly
without causing interference to other users of the electromagnetic
radio spectrum. However, unlike the low-frequency inductive links,
at high frequencies like 400 MHZ a wireless transceiver requires
the use of radio-frequency (RF) blocks such as voltage-controlled
oscillators (VCOs), mixer and phase-locked loops (PLLs) to down
convert (or up convert for the transmitter case) the frequency signal
in order to process it using the integrated circuit technology. These
blocks are constructed using inductors and capacitors on chip or off
chip, which increase the physical size of the wireless chip [9, 35].
Medical implants have a physical limitation for the electronics of the
wireless telemetry and cannot afford to accommodate such blocks.
Thus high-level of integration, which results in miniaturization and
low power consumption, should be met for wireless designs at MICS
band. So far only one company, Zarlink, designed a transceiver for
implantable applications. Zarlink’s transceiver still requires some
external components for its full functions [35]. Although with an
MICS link a data could be transferred from or to the implant within a
distance of a few meters, when the implant system requires the use
of a wireless power or wirelessly charging the battery, the external
unit still will need to be attached to the body close or adjacent to
the implant system for an efficient power transfer as indicated in
Fig. 6.8. It is still necessary to utilize the power signal at a lower
frequency, which also eliminates the possible interference issue

from the power signal to the data signals [27].
Medical implant telemetry can be categorized into two groups:
high data rate and low data rate systems. As an example,
the wireless capsule endoscope and the retina prosthesis are
the recent implanted systems requiring a large amount of data
delivered to or outside the body. Implantable multichannel neural
recording systems and multichannel monitoring of continuous
signals electroencephalogram (EEG) also necessitate a high data
rate communication. As an example, scientists aim to achieve the
recording of more than 100 channels in order to simultaneously
record brain functions; a data rate more than 20 Mbps is required.
A similar figure is also useful for a wireless endoscope implant to
obtain higher resolution and detailed real-time video images. MICS
has been divided into channels with 300 kHz bandwidth and thus
cannot provide such data rates. Thus for these types of implantable
systems requiring high data rate, UWB will be the technology to
investigate [12].
6.3 WBAN Systems
The early telemedicine systems have used the WMTS for patient
monitoring at hospitals. As an example, the LX-5160, which is a
WMTS telemetry transmitter (608∼614 MHz WMTS Band), has
been developed by Fukuda Denshi USA [36]. It transmits ECG and
respiration parameters to a central telemetry receiver, which can
be displayed on a remote computer. This device has a dimension
of 5.4 (W) × 2.2 (D) × 8.6 (H) cm and a weight of 80 gram.
Transceiver-608 telemetry device developed by Mindray [10] also
uses a WMTS transmitter for patient monitoring. This telemetry
uses a 12-lead ECG and has a dimension of (13.0 × 7.5 × 3.1
cm). Its weight is 181.4 g. The wireless telemetry is based on
channelized WMTS frequency meaning uses different frequency
for transmit and receive signals. For example, 608–614 MHz
(transmit) WMTS channelized 1395–1400 MHz and 1427–1429.5
MHz (receive) WMTS are used. Welch Allyn is another company
developing patient monitoring system for vital signs monitoring
WBAN Systems 163
for hospital and sportive applications [37]. They have developed

micropaq R
Wearable Monitor that displays heart rate, one or two
ECG leads, motion-tolerant SpO2 and pulse bar. The device can also
connect wirelessly to a central monitoring station. This device has a
large dimension similar to a mobile phone.
Comparing to the wireless module discussed before, these
devices are bulky and are not appropriate sensor node when
considering a WBAN application. Thus new telemedicine systems
are currently being developed using new small size sensor nodes
described earlier.
Recently Corventis, Inc. has developed a small sensor node
TM
called PiiX for wireless cardiovascular solutions [38]. It uses a
TM
wireless gateway device named zLink to communicate with the
monitoring center. The device has some limitations. For example,
PiiX is intended for single patient use and cannot be used for patients
with implantable devices. The unique features of this device are
leadless and water-resistant. It eliminates the cumbersome leads
and wires, and can still operate during showering or sleeping. Piix
sensor is quite small and can easily be attached to the body like
a plaster. Figure 6.9 shows a picture of PiiX sensor node from
Corventis and micropaq R
Wearable Monitor from WelchAllyn.
A UK-based company Toumaz [17, 39] is developing a digital
plaster where the sensor node can be attached to the human
Figure 6.9. Figure PiiX sensor node from Corventis [38] and
micropaq
R
Wearable Monitor from WelchAllyn [37]. See also Color Insert.
body with a plaster. The wireless sensor node is based on ECG

and temperature data. The data received from the plaster can be
displayed on a PDA device. It can also be transferred to a monitoring
station via WLAN link using the PDA as a gateway device. The digital
plaster can communicate with PDA up to 3 meter distance. The
plaster is based on the chip TZ1030 which uses the 8051 processor
for digital processing a wireless transceiver.
The Fitbit has developed a wireless sensor node that a tracks
motion of a human body. The Fitbit sensor contains a 3D motion
sensor that tracks the motion in three dimensions and converts to
useful information such as calories burned, steps taken, distance
traveled, sleep quality, etc. It can be worn on the waist, in pocket
or on undergarments. Nike and adidas have produced similar body-
worn devices Nike+iPod and miCoach Pacer, respectively, for sport
activities [40].
Here in we will discuss some of ongoing projects for WBAN
applications. These projects have mainly used the sensor nodes
provided in Table 6.4. Some have designed their own sensor nodes
by using the wireless transceivers in Table 6.3 together with one
of famous microcontroller — ATMega128L, MSP430, and/or 88051.
There have been a number of different developments for WBAN
systems in the literature. The goal here is to compare the existing
systems in terms of their suitability for a large scale implementation
for WBAN applications. We will discuss whether the available
systems demonstrate a multi-patient or a single-patient monitoring.
A comparison of some of current implemented WBAN systems is
given in Table 6.5.
The project in reference [41] presents the AID-N triage system
for disaster applications. It uses the MicaZ or TmoteSky platform
from Crossbow Technology (2.4 GHz Radios (802.15.4) with CC2420
transceiver chip) in the sensor node for data collection. The sensor
node communicates with another ZigBee device that is attached to
a personal server, which is a laptop and PDA. Software programs
have been developed to gather vital signs from a disaster scene. This
system requires WLAN connections with an IEEE 802.11 link for
transmission of patients’ data to a remote server in a medical centre.
If there is no network connection, the data will be stored in a device
or can be monitored on a PDA in an emergency vehicle until patients
are taken to a hospital.
October 19, 2011
17:4
Table 6.5. WBAN-based telemedicine systems
Existing systems Wireless device Sensors Comments
CodeBlue [42] 802.15.4 (2.4 GHz) Pulse oximeter, EKG, motion-activity It uses MicaZ/T-mote sensor nodes.
WiiSARD [43] 802.11 Pulse oximeter It uses PDA device as a sensor in order to
use Wi-Fi link to connect the Internet.
(Gao, 2007) [41] ZigBee (2.4 GHz) SpO2 , ECG, BP It uses mesh networking for multi-patient,
and WLAN for remote destinations.
(Park, 2006) nRF24E1 radio (2.4 GHz) ECG The ECG sensor communicates to a
computer wirelessly or a base station with
the Wi-Fi link. No multi-patient scenario.

(Espina, 2008) [44] IEEE 802.15.4 (2.4 GHz) ECG, PPG, blood pressure Single-user data is presented.
(Chen, 2009) [1] IEEE 802.15.4 (ZigBee) 8–Channel EEG Single patient. Internet data transmission is
provided. No wireless gateways.
(Zhang, 2009) [2] Bluetooth (2.4 GHz) ECG Although it mentions it is done for two
persons ECG, one continuous ECG
monitoring data is shown. The control
device is directly connected to a home
computer via USB (no wireless gateways).
No data transmission through the Internet.
(Anliker, 2004) [45] GSM Blood pressure, SpO2 , 1-lead ECG Multiple sensors have been integrated in
WBAN Systems
one hand-held device.
(Contd.)
165
06-Mehmet-c06
October 19, 2011
17:4
Table 6.5. (Continued)
Existing systems Wireless device Sensors Comments
(Yuce, 2008) [23] MICS, WMTS Temperature, pulse rate, single lead he design targets monitoring of
ECG/EEG physiological ameters for multi-patient. It
uses multi-hoping technique. Long-range
data transmission through the Internet.
Several wireless gateways to increase
mobility.
TZ1038 Sensium [17,39] 868/915 MHz Single lead ECG, temperature Data received and displayed on a PDA
device. Also transferred to a monitoring

station via WLAN link using the PDA as a
gateway device.
(Jovanov, 2005) [1] ZigBee (2.4 GHz) Activity sensor The control device connects with a
computer for data display directly or
wirelessly using a Wi-Fi link. It is a
multi-hopping system with multi-sensors;
however, one patient data has been
implemented. No Internet transmission has
been indicated.
BSN Node [21] IEEE 802.15.4 (2.4 GHz) 3-lead ECG, 2-lead ECG, and SpO2 Multichannel sensor node. Sensor signals
sensors can be gathered, displayed, and analysed on
a PDA. The PDA works as a wireless
gateway for remote destinations.
06-Mehmet-c06
October 19, 2011
17:4
(Wang , 2009) nRF905 (915 MHz) 3-D acceleration The sensor nodes communicate directly to
a base-station board that includes an
aMSP430F149 microprocessor and an
nRF905. Data from single body is tested.
(Yuce, 2009) [12] UWB Multichannel EEG/ECG The design targets monitoring of
multichannel continuous physiological

ameters for implantable and wearable
systems. No wireless gateways.
Human++ [48] nRF24E1 radio (2.4 GHz) Single-lead ECG, EEG An ECG node is designed using flexible
polyimide material. No wireless gateways.
HeatlhGear [46] Bluetooth (2.4 GHz) Oximeter The sensor communicates with a cell phone
using the Bluetooth link. No wireless
gateways. Individual use.
WBAN Systems
167
06-Mehmet-c06
The CodeBlue project uses sensor nodes based on the popular

MicaZ and T-mote designs [42] for a multi-patient monitoring
system. The WiiSARD team [43] developed a prototype system for
pulse oximeter sensor. It operates with the Wi-Fi link to enable a
medical response through the Internet. It uses a PDA device as a
senor which has a large size which is not attractive for a medical
sensor node.
In reference [44], Espina et al. presented an IEEE 802.15.4-based
wearable telemedicine system that monitors continuous cuff-less
blood pressure and ECG signals. The sensor data was measured
on a single user. Date is displayed from the sensor to a PDA or
a wristwatch device. Anliker and his colleagues [45] designed a
wrist-worn device to measure multiple physiological parameters
from one person. Sensors like blood pressure, SpO2 , one lead ECG
all have been integrated in one handheld device. The data is then
transmitted using a GSM data link to a computer similar to Fig. 6.1.
The device acts the same as a regular cellular phone since it
operates with the GSM network. However, measurements more than
one patient at the same time have not been provided. HealthGear
project uses a cell phone as a central processing unit that receives
signals from body sensors [46]. It uses the Bluetooth technology
to connect a Bluetooth-enabled cell phone. The sensor is a Nonin’s
flex oximeter used to measure a single user’s blood oxygen level
(SpO2 ). The Berkeley Tricorder sensor node is another Bluetooth-
based WBAN system developed at Berkley [47] for ambulatory
health monitoring systems. The system utilizes a multichannel
wireless sensor node, including parameters such as ECG, EMG, blood
oxygenation, respiration, and motion all on one board. The data are
received and stored on a PC via a Bluetooth link.
Another system in reference [20] designs very small, wearable
(probably one of the smallest custom-made sensor nodes available
at the present time) ECG sensor that communicates wirelessly with
a base station connected to a computer. The system works similar
to a multi-hopping network by using the Wi-Fi link. Although a
multi-patient data monitoring is not presented, this system can be
used for measurements and monitoring of vital signs from multiple
patients.
WBAN Systems 169
A wireless patient monitoring system based on an ECG patch has

been developed for cardiac monitoring by IMEC Belgium. IMEC’s
Human++ program develops wearable wireless devices [48]. The
current ECG node electronic is assembled on flexible polyimide
substrate that can easily be integrated in textile.
There are also ongoing projects to develop sensor platforms
operating at UWB (3–10 GHz) [12], MICS, and WMTS bands [49].
A WBAN system for multi-patient medical monitoring is presented
in reference [23]. It uses a multi-hoping technique through wireless
gateways to transfer data from sensors on several human bodies
to remote locations. It uses the Internet to allow access at any
location. In reference [12], a low-power WBAN system has been
developed using UWB technology for implantable multichannel
neural recording for brain–computer interface applications and 8-
channel ECG for patient monitoring. The ECG data is monitored at a
remote station.
Table 6.5 shows the current progress in WBAN applications. Real-
time and simultaneous monitoring is required from multi-sensors
on multi-bodies for future WBAN systems. System should take the
advantage of the Internet for data transmission to remote locations,
which will provide an access to a person’s data anywhere in a city.
Future WBAN systems should also develop small-size sensor boards
to meet the size and power constraints. Most of the current systems
use the commercially available ZigBee (IEEE 802.15.4) devices and
boards, which are designed for many other applications and, thus,
are not entirely wearable.
For longer range communications, the wireless nodes should
be able to communicate to remote stations through a wireless,
portable gateway (it could be a watch, PALM, iPhone, mobile phone,
etc.). The wireless transmission technique between the gateway
device and remote monitoring stations could be done using one of
wireless protocols such as ZigBee, Bluetooth, or Wi-Fi for using the
Internet.
The implanted device will most likely use the MICS band or a
frequency around 400 MHz. The wearable nodes in a WBAN system
will use an ISM band for low data rate communication and UWB
when a high data rate transmission is required.
6.4 A WBAN-Based Multi-Patient Monitoring System
A WBAN system that has been designed for health care applications
is presented in this section. The system is based on a multi-hopping
network technique that can be used in medical environments
for remote monitoring of physiological parameters from multiple
patient bodies [52]. The system is depicted in Fig. 6.10. The data is
transferred to remote stations through the local area network or the
Internet already available in medical centers.
An example of sensor node design for medical monitoring is
depicted in Fig. 6.11. The analog front-end of the sensor node for
EEC, EEG, and EMG uses the circuit given in Fig. 6.3. The pulse
rate sensor consists of an infrared emitter (SFH487-880nm) and a
phototransistor (SFH309FA). Light is shone through the tissues, and
the variation in blood volume alters the amount of light falling on the
detector. When the heart beats, a pressure wave moves out along the
arteries at a few meters per second. This pressure wave can be felt at
the wrist, but it also causes an increase in the blood volume, which
can be detected by a plethysmograph. Human heartbeat ranges
between 50 bpm and 200 bpm. These figures indicate that the
frequency range of human heartbeat is between 0.83 Hz and 3.33 Hz:
A high-pass filter and a low-pass filter are used to form a band-pass
Figure 6.10. Multi-hopping WBAN-based multi-patient monitoring.

A WBAN-Based Multi-Patient Monitoring System 171
Figure 6.11. A 4-channel sensor node. See also Color Insert.
filter to remove unwanted environment signals. The optical part of

the pulse sensor consumes quite large power [49]. It is also possible
to detect the pulse rate from the ECG signal as explained later in
Fig. 6.16. The temperature sensor uses the IC LM 35 for temperature
measurement. The IC produces an analog voltage with respect to
temperature. The ADC in the microcontroller samples the voltage
and converts it to a digital/data for RF wireless transmission.
The primary function of BCUs is to collect data from sensor nodes
via the first wireless link and forward these data to a remote PC
(i.e., control station) for further analysis. Two types of BCU devices
are needed in order to provide a complete WBAN transmission
coverage in a medical center. One BCU is designed to be connected to
a computer (Fig. 6.12a) via the USB port, while the other BCU is used
to function as an intermediate device (Fig. 6.12b) that presents a
second wireless link for a longer range wireless sensor network. The
latter case is more suitable for large medical centers and functions
as a portable gateway device. Although both BCUs can be used for
multi-patient monitoring, the first BCU type (BCU-1) can also be
useful for private usage at home or in a room of a hospital for single-
patient monitoring.
The sensor nodes/BCU hardware requires a microcontroller and
a wireless transceiver to coordinate all activities. The BCU-1/sensor
nodes consist of a microcontroller PIC16F87 and a wireless
transceiver (AMIS52100 IC and CC1000). Both AMIS and CC1000
can generate MICS band frequencies. Especially when CC100 is used,
the operation frequency can be configured to 433 MHz ISM, WMTS
bands as well as MICS.
Figure 6.12. Block diagram of BCUs: (a) wired BCU, (b) portable BCU. See
also Color Insert.
In addition to these chips, we use another transceiver, the

CC1010 chip from TI (this chip contains CC1000 and a microcon-
troller built in), on the intermediate BCU (BCU-2) board to develop
a wireless transmission and networking between BCUs and the
remote base station. The CC1010 and CC1000 transceiver chips
have the capability to transmit anywhere within 300 and 1000 MHz
(It was tuned to WMTS band for the second wireless link in our
prototype system). The AMIS IC has a data rate capability of 19 Kbps,
while CC1010 provides 76 Kbps. The BCU-2 device in a WBAN can
also be composed of standards such as ZigBee and the 802.11 Wi-
Fi standards to accommodate and interface with different wireless
platforms and to connect to the Internet for remote monitoring. A
summary of devices used in BCUs and sensor node boards is given
in Table 6.6. The hardware implementations of BCU-1 and BCU-2 are
given in Figs. 6.13 and 6.14.
The modulation technique used in CC10xx chips is FSK with a
frequency step of 64 kHz, which is acquired by loading the crystal.
FSK has better interference rejection and causes less spectrum
splatter normally seen in ASK systems. The picture in Fig. 6.15
Table 6.6. Summary of devices used in sensor nodes and BCUs
Device features AMIS (52100) CC1000 CC1010* PIC16F887

Size (mm) 7.5 × 7.8 9.7 × 6.4 12.9 × 12.9 17.53 × 17.53
Modulation ASK/OOK FSK/OOK FSK —
Sensitivity −117 dBm −109 dBm −107 dBm —
Power: Transmitter TX: 25 mA Tx: 26.7 mA Tx: 26.6 mA < 0.6 mA (active)
Receiver Rx: 7.5 mA Rx: 7.4 mA Rx: 9.1 mA
Data Rate 19 Kbps 76.8 Kbps 76.9 Kbps —
Sniffing** 500 nA 200 nA ∼ 0 mA —
Memory (RAM) — — 2048 + 128 368 + 256
Additional Features — — 10-bit, 22.7 kHz 10-bit 52 kHz
sampling frequency sampling frequency
*Note: CC1010 has a microcontroller built-in.

** Sniff mode enables the receiver to wake up or operates at times to “sniff” received RF signals
and then return to “sleep” or “wait” mode if a signal is not detected.
Figure 6.13. A wired BCU (BCU-1). See also Color Insert.
Figure 6.14. Intermediate body control unit (BCU-2). This device is

shared by more than one patient and is portable. It contains dual wireless
transceiver to support two directional wireless links. See also Color Insert.
Figure 6.15. Spectrums generated for MICS and WMTS frequencies with
Chipcon CC1010.
demonstrates the FSK modulation for MICS and WMTS bands. What
is necessary to realize from this picture is that the bandwidth of
the signal is well within the bandwidth of the WMTS band (608–
615 MHz).
In order to provide communication from sensors to a personal
computer, a wired serial port interface, such as RS-232 standard,
should be used to connect the base station to monitor the received
body signals. The interface system here uses RS-232 serial interface
and USB connections.
Figure 6.16 is an ECG signal obtained from our set up. The ECG
signal is transmitted from a sensor node to the computer. Each
sensor node representing only one patient can only have one ECG.
In order to eliminate the DC noise (50 Hz/60 Hz interference), a
recursive filter has been software implemented to obtain an accurate
ECG signal (see appendix for notch filter implementation). As shown
in Fig. 6.10, by clicking on 50 Hz filter, the recursive notch filter
operates on the received ECG signal.
High computation programs like MATLAB can further be used
to analyze data automatically and to warn medical professionals
when the value of a critical data goes outside the safe margin. A
warning signal could be easily generated by the program at the PC
to warn health professionals to track the sensor node generating
a warning signal. This feature will strengthen the reliability and
safety in implementation, which would be useful for patients’
lives.
Figure 6.16. A wireless ECG monitoring showing pulse rate as well.
A medical data from a single human body can be monitored

within 300 ms for a distance of 1 m when the WBAN system is
implemented in a four–patient scenario. When the active patient
is doing all sort of activities, e.g., sitting standing, turning around
the other patients, the average delay is 1 s up to 5 m and 2 s at
the distance of 10 m. These values show that the WBAN system
exhibits a correct, timely, and reliable communication performance
up to 10 m with a time performance less than 2 s for a multi-patient
scenario [52].
6.4.1 Software Programs and Monitoring

In order to monitor data in body area network applications,
several computer programs should be developed. The necessary
software programs have been identified in Fig. 6.17 for this specific
application. A software called GATEWAY should be developed at
the monitoring PC to control the communication with the BCU to
get readings from sensors and then forward them through another
network/Internet to an application on a remote PC (at a remote
location). While performing this task, the GATEWAY will also verify
the data integrity and schedules retransmission, if required. Another
software program is developed at the remote PC (called BSN),
Figure 6.17. Software programs for the proposed WBAN application. See
also Color Insert.
which gets readings from GATEWAY via the network/Internet. These

readings are stored in the remote PC for analysis. A graphical user
interface (GUI) at the local PC as well as at remote PCs should be
designed to display the human body data. Both the data received
from the BCU and the data sent to the BSN can also been shown by
the GUI in text or graphical formats. In case of medical application,
the physiological signals of patients can be accessed by medical
staff anywhere in the medical center as long as their computers are
connected to the local area network in the building.
The BSN application is designed to collect and store readings
automatically so that no person is required to be stationed at
the application. It gets sensor information and readings from
the GATEAWAY and will ask retransmission if error is detected.
It undertakes the administration of patients’ particulars such as
assigning new sensor ID to patients, segregating sensor readings
from different patients, and storing them into the data base. The GUI
at the monitoring PCs allows medical personnel to enter patients’
information. It can also display live monitoring graphs on the screen
(Fig. 6.18). Every sensor device has a unique sensor ID and must be
registered under a patient name before they are used. In the event
that an unregistered sensor node is used, all its readings received
will be discarded by the BSN application
Since all sensor nodes of a body are communicating with the
same BCU, the data is prefixed with an identifier, which is used
to identify the source of the data. As mentioned earlier, to reduce
collisions further between data sent, a firmware (medium access
control (MAC) protocol) is written to control data transmissions. The
communication between the sensors and the BCU is bi-directional as
to support a multiuser (i.e., multi-patient) communications.
Figure 6.18. Example of a graphical menu registering and viewing

patients.
The GATEAWAY menu shows a GUI at the local PC, which is

designed to configure the source and destination socket and port
numbers for data transfer on the Ethernet (See Fig. 6.19). This GUI
ensures portability if the BSN application or the GATEWAY need
to be relocated, it can be configured to work without additional
changes in the codes. Both the data received from the BCU and
the data sent to the BSN can also been shown here in text format.
The physiological signals of patients can be accessed by medical
Figure 6.19. The GATEWAY menu.

staff anywhere in the medical center as long as their computers are

connected to the local area network in the building.
The software packages used for monitoring and storing data (in
the database) should be user friendly so that medical professionals
utilize them efficiently. The programs should allow a medical
professional to easily set up sensors. When needed, one or more
sensors should be removed without affecting the operating of the
other sensors in the system. Software programs at the monitoring
PC can be designed in such a way that when a sensor stops working,
the data from the remaining sensors should still be received and
monitored. Before a new sensor is deployed for a particular person,
a new sensor ID (i.e., user ID) should be assigned at the monitoring
PC so that the MAC protocol used for the data communication
will enable a communication from that sensor. The MAC protocols
mentioned above can integrate an error-checking mechanism to
allow the monitoring of correct data only and discard when a data
is faulty.
A database server is developed to maintain data integrity, which
is necessary for big medical centers. Monitoring of ECG for an
individual patient is shown in Fig. 6.20. In this GUI, more detailed
patient’s particulars can be seen from the database. Clicking on
Figure 6.20. An example of ECG Monitoring. See also Color Insert.

Conclusion 179
a patient shows the sensors attached to them and their personal

information/picture. Clicking on a sensor displays the sensor’s
information (e.g., interval). It also brings up how many recordings
are available. Single clicking a record shows when the record was
made. Double clicking displays the record on the graph. In the
following figure, it is arranged for an ECG monitoring. It is very
handy to have a data file compatible with a computation software
program like MATLAB for signal analysis as shown in the window.
6.5 Conclusion
We have discussed implementation issues and presented details of

techniques for design of wireless sensors in WBAN applications.
Portable and wireless gateway nodes are used to connect the
sensor nodes to the local area network or the Internet already
available for long-range access points. Wireless standards such as
ZigBee (IEEE 802.15.4) and Bluetooth (IEEE 802.15.1) are popular
low-power technologies for communication between sensors and
the control device. New wireless devices specifically targeting at
WBAN application can be designed based on MICS, WMTS, 433 MHz
ISM as well as UWB bands. Miniaturization of the sensor node
electronics, especially the sizes of the microcontroller, the wireless
chip, the battery, and low-power consumption are current hardware
related issues for small sensor nodes. TI’s CC2420 and CC1010
transceivers and Nordic’s transceivers like nRF24E1 are current
popular choices for wireless module in a sensor node. The successful
implementation of a WBAN system should operate and coexists with
other network devices and should provide wearable, wireless (no
wire connections), easy to remove, and attach sensor nodes, leading
to increased mobility of patients and flexibility.
Any future projects based on WBAN should take under con-
sideration the co-existence of other network devices operating in
similar frequency bands. The majority of WBAN systems cover
only wearable sensor nodes. There is very limited information
on integration of wearable and implantable systems together.
Implanted or wearable sensor nodes should be able to work together
in a WBAN system without any data collisions.
Acknowledgements
I would like to thank Anthony Bott and Ng Peng Choong for their
help in developing software programs and boards.
Appendix
Digital Notch Filter Designs

A digital notch filter can be designed to filter out the 50/60 Hz
noise for continuous physiological parameters. We used the website
in [53] as a reference. Two parameters should be defined: f , the
center frequency, and BW, the bandwidth. Finding the notch filter
coefficients:
R = 1 − 3 × BW
BW = 0.033 = 13.2 Hz
∴ R = 0.901
50
f = = 0.125
400
1 − 2 × 0.901 × cos(2π f ) + 0.9012
K =
2 − 2 × cos(2π f )
∴ K = 0.917731353
a0 = K , a1 = −2K cos(2π f ) = −1.297868126, a2 = K
b1 = 2R cos(2π f ) = 1.27420642, b2 = −R 2 = −0.811801
These are used in the equation
y[n] = a0 x[n] + a1 x[n − 1] + a2 x[n − 2]

+ · · · + b1 y[n − 1] + b2 [n − 2]
Note that BW and f are fractions of the sampling frequency. All

calculations are done in radians. The bandwidth 13.2 Hz, which
could be wide for the application, can easily be decreased to a
nominal value.
The system requires that three previous inputs and the last two
outputs are stored. Whenever a new value is calculated, the array
References 181
is shifted so that x[n − 1] = x[n], etc. It is possible to increase the

number of coefficients but at the expense of CPU resources and the
increased risk of instability. Decreasing the bandwidth of the filter
also has limits and will cause problems if it becomes too small.
References
1. Jovanov, E., Milenkovic, A., Otto, C., and de Groen, P. C. (March 2005)
A Wireless Body Area Network of Intelligent Motion Sensors for
Computer-Assisted Physical Rehabilitation, Journal of NeuroEngineering
and Rehabilitation, 2(6), doi:10.1186/1743-0003-2-6.
2. http://focus.ti.com/lit/ds/symlink/ina321.pdf, 2010.
3. http://www.analog.com/static/imported-files/data sheets/AD620.
pdf.
4. E. Company-Bosch and E. Hartmann , ECG Front-End Design is
Simplified with Micro-Converter
R
“ Analog Dialogue 37-11, November
(2003).
5. P. A. Pour, T. Gulrez, O. AlZoubi, G. Gargiulo, and R. A. Calvo
(15–18 December 2008) Brain-Computer Interface: Next Generation
Thought Controlled Distributed Video Game Development Platform,
IEEE Symposium on Computational Intelligence and Games, pp. 251–257.
6. S. Hanna (2009) Regulations and Standards for Wireless Medical
Applications, ISMICT 2009.
7. K. Finkenzeller (April 2003), RFID Handbook, 2nd edn, John Wiley and
Sons, New York, USA.
8. FCC Rules and Regulations (January 2003) MICS Band Plan, Table of
Frequency Allocations, Part 95.
9. A. Tekin, M. R. Yuce, and W. Liu (2008) Integrated VCOs for Medical
Implant Transceivers, VLSI Design, 2008, Article ID 912536, 10 pages,
doi:10.1155/2008/912536.
10. http://www.na.mindray.com/pdf/panorama WMTS specifications us.
pdf.
11. http://www.fda.gov/MedicalDevices/Safety/MedSunMedicalProduct
SafetyNetwork/ucm127778.htm, 2010.
12. M. R. Yuce, H. C. Keong, M. Chae (October 2009) Wideband Commu-
nication for Implantable and Wearable Systems, IEEE Transactions on
Microwave Theory and Techniques, 57(2), pp. 2597–2604.
13. Moteiv Corporation (2010) http://sentilla.com/files/pdf/eol/tmote-

sky-datasheet.pdf.
14. http://www.tinynode.com. Ref: Tinynode.
15. http://webs.cs.berkeley.edu/tos/.
16. http://www.snm.ethz.ch/Projects/T-Nodes.
17. A. Wong, G. Kathiresan, C. Chan, O. Eljamaly, O. Omeni, D. McDonagh,
A. Burdett, and C. Toumazou (2008) A 1 V Wireless Transceiver for
an Ultra-Low-Power SoC for Biotelemetry Applications, IEEE Journal of
Solid-State Circuits, 43, pp 1511–1521.
18. http://www.btnode.ethz.ch/Documentation/BTnodeRev3Sensor
Guide.
19. http://www.palowireless.com/database/conexant/101685b.pdf.
20. C. Park, P. H. Chou, Y. Bai, R. Matthews, and A. Hibbs (2006) An Ultra-
Wearable, Wireless, Low Power ECG Monitoring System, in Proceedings
of IEEE BioCAS, pp. 241–244.
21. G. Z. Yang (2006) Body Sensor Networks, Springer-Verlag London.
22. B. Wang, et al. (11–13 June 2009) A Body Sensor Networks Development
Platform for Pervasive Healthcare, in Proceedings of the 3rd International
Conference on Bioinformatics and Biomedical Engineering, pp. 1–4.
23. M. R. Yuce, and C. K. Ho (August 2008) Implementation of Body Area
Networks Based on MICS/WMTS Medical Bands for Healthcare Systems,
IEEE Engineering in Medicine and Biology Society Conference (IEEE
EMBC08), pp. 3417–3421.
24. R. S Mackay, B Jacobson (1961) Radio Telemetering From Within the
Human Body, Science, 134, pp. 1196–1202
25. S. Y. Shin, H. S. Park, and W. H. Kwon (August 2007) Mutual Interference
Analysis of IEEE 802.15.4 and IEEE 802.11b, Computer Networks, 51,
pp. 3338–3353.
26. W. Liu, et al. (September–October 2005) Implantable Biomimetic
Microelectronic Systems Design, IEEE Engineering in Medicine and
Biology Magazine, 24, pp. 66.
27. M. Zhou, M. R. Yuce, and W. Liu (September 2008) A Non-Coherent
DPSK Data Receiver With Interference Cancellation for Dual-Band
Transcutaneous Telemetries, IEEE Journal of Solid-State Circuits, 43,
pp. 2203–2012.
28. W. Liu, et al. (October 2001) A Neuro-Stimulus Chip With Telemetry Unit
for Retinal Prosthetic Device, IEEE Journal of Solid-State Circuits, 35,
pp. 1487–1497.
References 183
29. S. K. An, et al. (June 2007) Design for a Simplified Cochlear Implant
System, IEEE Transactions on Biomedical Engineering, part 1, 54,
pp. 973.
30. P. R. Troyk, and G. A. DeMichele (September 2003) Inductively-Coupled
Power and Data Link for Neural Prostheses Using a Class-E Oscillator
and FSK Modulation, in Proceedings of the IEEE International Conference
on Engineering in Medicine and Biology Society, pp. 3376–3379.
31. M. Ghovanloo, and K. Najafi (December 2004) A Wideband Frequency
Shift Keying Wireless Link for Biomedical Implants, IEEE Transactions
on Circuits Systems II, 51, pp. 2374–2383,.
32. M. A. Stuchly, A. Krazewski, S. S. Stuchly, and A. M. Smith (1982)
Dielectric Properties of Animal Tissues in Vivo at Radio and Microwave
Frequency: Comparison Between Species, Physics in Medicine and
Biology, 27, pp. 927–936.
33. D. Halperin, et al. (March 2008) Security and Privacy for Implantable
Medical Devices, IEEE Pervasive Computing, 7(1), pp. 30–39.
34. Australian Communications Authority, Radio Frequency Planning Group
(October 2003) Planning for Medical Implant Communications Systems
and Related Devices, Proposals Paper, http://www.acma.gov.au/.
35. P. D. Bradley (December 2006) An Ultra-Low Power, High-Performance
Medical Implant Communication System (MICS) Transceiver for
Implantable Devices, in Proceedings of the IEEE Biomedical Cir-
cuits and Systems Conference (BioCAS 2006), pp. 158–161, also at
http://www.zarlink.com/zarlink/hs/4889.htm.
36. http://www.fukuda.com/fukuda usa/lx-5160.html, 2010.
37. http://www.welchallyn.com, Micropaq
R
Monitor.
38. http://www.corventis.com/AP/nuvant.asp, 2010.
39. A. Burdett (2006) Low Power Device Communication, IET Seminar on
Biocompatible Materials and Devices, http://www.toumaz.com/.
40. www.micoach.com.
41. T. Gao, et al. (September 2007) The Advanced Health and Disaster
Aid Network: A Light-Weight Wireless Medical System for Triage,
IEEE Transactions on Biomedical Circuits and Systems, 1(3), pp. 203–
2007.
42. V. Shnayder, B. Chen, K. Lorincz, T. R. F. F. Jones, and M Welsh (2005)
Sensor Network for Medical Care, Technical Report TR-08-05, Division of
Engineering and Applied Sciences, Harvard University.
43. https://wiisard.org.
44. J. Espina, T. Falck, J. Muehlsteff, Y. Jin, M. A. Adán, and X. Aubert

(June 2008) Wearable Body Sensor Network Towards Continuous Cuff-
less Blood Pressure Monitoring, in Proceedings of the 5th International
Workshop on Wearable and Implantable Body Sensor Networks, pp. 28–
32.
45. U. Anliker, et al. (2004) AMON: A Wearable Multiparameter Medical
Monitoring and Alert System, IEEE Transactions on Information Tech-
nology in Biomedicine, 8, pp. 415–427.
46. N. Oliver, and F. Flores-Mangas (2006) HelathGear: A Real-Time
Wearable System for Monitoring and Analysing Physiological Signals,
in Proceedings of the IEEE International Workshop on Wearable and
Implantable Body Sensor Networks (BSN 2006), pp. 1–5.
47. R. Naima, and J. Canny (June 2009) The Berkeley Tricorder: Ambulatory
Health Monitoring, in Proceedings of the 2009 6th International
Workshop on Wearable and Implantable Body Sensor Networks, pp. 53–
58.
48. J. Penders, at al. (December 2008) Human++: From Technology
to Emerging Health Monitoring Concepts, in Proceedings of the 5th
International Summer School and Symposium on Medical Devices and
Biosensors (ISSS-MDBS 2008), pp. 94–98.
49. M. R. Yuce, et al. (December 2007) Wireless Body Sensor Network Using
Medical Implant Band, Journal of Medical Systems, 31, pp. 467–474.
50. H. Chen, W. Wu, and J. Lee (March 2009) A WBAN-Based Real-Time
Electroencephalogram Monitoring System: Design and Implementation,
Journal of Medical Systems., 34, pp. 303–311.
51. Y. Zhang, and H. Xiao (November 2009) Bluetooth-Based Sensor
Networks for Remotely Monitoring the Physiological Signals of a Patient,
IEEE Transactions on Information Technology in Biomedicine, 13(6),
pp. 1040–1048.
52. 52. M. R. Yuce (July 2010), “Implementation of Wireless Body Area
Networks for Healthcare Systems,” Sensors & Actuators: A. Physical, 162,
pp. 116-129.
53. http://www.dspguide.com/ch19/3.htm.
Chapter 7
Wireless Body Area Network

Implementations for Ambulatory
Health Monitoring
Reza Naimaa and John Cannyb

a Department of Bioengineering, University of California,
Berkeley, 306 Stanley Hall #1762, Berkeley, CA 94720-1762

b Electrical Engineering and Computer Sciences,
University of California, Berkeley, 387 Soda Hall,

MC1776,Berkeley, CA 94720-1776
The design and implementation of an ambulatory health monitoring

device is a process of matching the medical need to available
technology. The process starts with an understanding of the re-
quirements, which leads to the design of sensors, filters, amplifiers,
etc. There are a number of additional decisions to be made that
range from the battery chemistry to the wireless technology that
must also match the underlying requirements. Finally, some existing
implementations will be reviewed.
7.1 Design Process
The first step in developing a device is to understand the medical

need in detail. It is not sufficient to simply have an understanding

www.panstanford.com
186 Wireless Body Area Network Implementations for Ambulatory Health Monitoring
of the physiology, pathology, etc., and to build a device accordingly.

From the physician’s perspective, the device might be too complex
to use, the data might be too verbose to analyze, or the physician
might not trust the data compared to the existing standards. From
the patients’ perspective, the device might be too cumbersome or
irritating to wear, the battery might die too quickly, the device might
require too much maintenance, or they might not see sufficient value
in the device to comply with the doctor’s instructions. There are
many other stakeholders that must also be considered through the
design process: nurses or other caregivers, insurance companies or
other sources of reimbursement. There is a great book that discusses
the design process for medical devices in detail [1], and is highly
recommended.
Once the high-level application, set of sensing modalities, data
storage, and wireless telemetry mechanism are determined, the
requirements of the individual components need to be flushed
out. It isn’t sufficient to “record ECG”; rather, the application will
result in requirements that need to be satisfied by the design. For
example, many ECG circuits will bandwidth limit the signal to 40 Hz,
and thus only require a sampling rate around 80 Hz (based on
Nyquist–Shannon sampling theorem). However, if the data is to be
used to measure heart-rate variability, then a sampling rate of 1 kHz
is required to have the required resolution. It is easy to miss this step
and to start designing a circuit based on a number of assumptions
which might not be valid — only to find out months later and have
to spend considerably more time and effort trying to fix the design
or work around the flaws. This is a lesson the author has learned the
hard way — spend the time understanding the application and the
requirements, and the interdependence of requirements between
the various components.
To help understand the design process, we will give examples
involving an ambulatory health monitor developed at the Berkeley
Institute of Design, named the Tricorder (Fig. 7.1). The original goal
for the Tricorder was a small and highly integrated device, which
records a number of parameters that can be used to assess the
wellness-state of the user. The data was intended to be used by
health care providers to better monitor the health state of the user,
to help in the treatment of any chronic condition, and to provide
Existing WBAN Implementations 187
Figure 7.1. The Berkeley Tricorder. See also Color Insert.
advanced notice of any possible health problems. Based on our

application, we developed the following high-level design goals:
• A means of data storage for at least 24 h and a mechanism

for offloading said data.
• Remote telemetry for real-time data viewing.
• A comfortable means of wearing the device.
• As many relevant sensing modalities as possible in a single
highly integrated form factor.
Based on these design goals, we decided on Bluetooth for the

wireless telemetry, microSD card for the data storage, and a number
of sensing modalities, detailed in Table 7.1. Examples involving the
Tricorder will be presented in the various sections of this chapter.
7.2 Existing WBAN Implementations
There are a large number of existing WBAN (wireless body area

network) implementations for an array of applications. Any attempt
at categorizing implementations is artificial, and the distinctions
Table 7.1. Device parameter summary
Modality Sampling rate Bits/sample Comment

ECG 256 Hz 12 2-Stage HPF; RFI filer; single channel
EMG 256 Hz 12 1-Stage HPF; RFI filter
Respiration 256 Hz 12 Phase; 12 Mag. 4-Electrode configuration; 350 μA
(bioimpedance) current at 50 kHz
Acceleration (3-Axis) 256 Hz 8/Axis
Blood oxygenation 256 Hz 15 Reflective forehead sensor; 2-stage
amp with DC-offset subtraction
between different deployments can be very different from the

perspective of two different users. We will attempt to group together
various common characteristics between some of the different
implementations used in the noninvasive health/medical space
based on my experiences and research, and present it as a starting
point.
Other sections in the chapter are dedicated to the implementa-
tion of sensing modalities at a low level — this section will focus on
the differences at a higher level: the hardware design paradigm, the
firmware implementation, and mechanism by which data is handled.
7.2.1 Hardware Paradigms

There is a big tradeoff between implementations that are designed
and built from the ground up, and designs which use OEM (original
equipment manufacturer) components to accelerate development.
Fully custom designs will generally take longer to develop, test,
and debug — but they will provide the smallest form factor,
lowest power consumption, and best fit for your application. These
implementations might have higher up front costs associated with
all the required development tools, but will cost lower to produce in
volume.
Examples of highly integrated circuits include a non-contact
ECG electrode with an integrated amplifier [2]. In this case, the
electrode form factor requires a highly integrated device, which
is not possible with OEM options. The same is true with another
group that has developed a button-sized wireless pulse oximeter
[3]. Even greater levels of integration can be achieved by fabricating
Existing WBAN Implementations 189
an application-specific integrated circuit (ASIC) containing all the

required components for a WBAN system on a single chip [4], or just
a portion [5].
A majority of WBAN devices utilize at least one OEM component
— ranging from small PCBs that implement the wireless link to fully
integrated sensing devices such as a pulse oximeter, which provides
the blood oxygen saturation percentage data over a serial stream [6].
The use of OEM components can greatly accelerate the design cycle;
however, each module often implements redundant components
such as the power stage and microcontroller’s which consume more
power, take up more space, might not provide an ideal form factor,
and have higher costs.
The most common OEM modules are ZigBee-based motes.
Motes1 are a class of devices that are conform to the 802.15.4
communications standard and allow for a data transfer and relay
between motes in a mesh configuration. Motes can include an
ADC (analog-to-digital converter) and serial ports to interface with
custom sensing boards, such as the MICAz mote from Crossbow.
The Telos moteiv also includes temperature, humidity, and light
intensity sensors. Their popularity is based on their low levels of
power consumption, a well-established networking interface, and
the ability to relay data between motes.
There are many mote-based implementations, such as the group
at the University of Alabama, which has a built a custom ECG module
to interface with a Telos Mote [7, 8].
Another group has proposed a system that utilizes a number of
OEM modules, including a Nonin ipod pulse oximeter, a belt-based
respiration rate sensor, and MICA based motes [6]. A joint effort
between Stanford University and NASA has resulted in the Lifeguard
[9], a platform that utilizes a number of OEM components such as the
Nonin Xpod pulse oximeter and the Accutracker II blood pressure
monitor.
7.2.2 Firmware
There are two main camps in terms of how the firmware operates
— either as a single application, or as an operating system that
1 Dr. Kris Pister, inventor of motes, defines a mote as a single dust particle
runs the application. As with the hardware example, writing a

stand-alone application provides the greatest flexibility in how to
access the hardware and utilize its resources, whereas implementing
an operating system, such as TinyOS2 , can help accelerate the
development cycle and provide a large set of built-in tools but
introduces overhead into the application and imposes some limits
in terms of how the application is structured.
TinyOS is the primary operating system running in most of
the motes to handle network communications, and is used by a
number of projects. The CodeBlue group at Harvard University have
developed a number of sensors that use TinyOS motes for data
acquisition and telemetry [10] and evaluated their use for triage
[11].
A group at the Imperial College London has utilized a TinyOS-
based sensor module developed at the University of California,
Berkeley to accelerate the development cycle of a number of
biosensors. These include an ECG module and a pulse oximeter
module, each of which can relay data directly to a PDA [12].
7.2.3 The Data

Once data is collected, it can be stored, processed, transmitted,
or any combination thereof. For data storage, the predominant
mechanism is by the use of a secure digital card, especially the
microSD form factor, which is 15 mm × 11 mm × 1 mm. These
devices require standard Serial Peripheral Interface (SPI), present
on most microcontrollers. Their use is further promoted by the
availability of numerous royalty-free interface libraries and the
availability of multi-gigabyte cards. This is especially important for
devices such as the Berkeley Tricorder, which can generate up to 823
MB of data in 24 h [13]. It should be noted that during a write cycle,
SD cards use considerable bursts of current, which can introduce
noise to your system. For the Berkeley Tricorder, we found this noise
was coupled onto the LED current, which produced noise from the
pulse oximeter signal (visible in Fig. 7.15). A redesigned current
source is expected to correct this.
2 http://www.tinyos.net/
Signal Acquisition 191
Processing of data can involve digital filtration to remove

signal noise, or calculating the percentage oxygen saturation from
the output of the pulse oximeter photodiode and lookup tables.
Processing of data consumes limited CPU cycles and increases the
power consumption of the microcontroller. Hence, as much data
processing as possible should be performed on more powerful
devices (i.e., a cell phone or a personal computer).
In terms of telemetry, there are a large number of possible
implementations that depend on: the desired telemetry range,
requirements of data relaying, the operating frequency, and the
requried bandwidth. As a general rule of thumb, the technology that
meets all the requiremets while minimizing the power consumption
and PCB real-estate will provide the best option.
7.3 Signal Acquisition
The number of possible sensing modalities is very large — a

discussion of all the possible modalities is beyond the scope of this
chapter. Instead, we will discuss some common strategies used in
interfacing analog and digital components, with examples in the
modalities used in the medical space.
The strategy for any sensing modality is to start with a
characterization of the signal of interest in the context of how that
data is to be used. The parameters of interest include the frequency
bandwidth, the required dynamic range, and source impedance. It
is always critical to start with signal source and design the rest of
the system from that point; otherwise you might find yourself in the
position of being forced to redesigning portions of your circuit at the
expense of extra time and cost.
In addition to designing the circuit based on the signal of interest,
it must also be designed to be immune to noise coupled from
the power supply, from radio-frequency interference (RFI), or from
other components on the circuit board while adhering to various
safety requirements.
A similar strategy exists for sensing modalities that do not
require any analog interface, such as an accelerometer that
communicates through a digital interface. In this situation, it’s
important to understand data’s application and to design around

those requirements. For example, an accelerometer that is used for
fall detection might require a high sampling rate, but not many bits
of resolution, whereas an accelerometer used to measure cardiac
output (i.e., balistocardiograph) would require significantly higher
resolution but a lower sampling rate.
7.3.1 Frequency Bandwidth of Interest

An understanding of the frequency bandwidth ( fmin − fmax ) of
interest (FOI) for any particular sensing modality is perhaps the
single most important characteristic to understand. For analog-
based signals, it will determine how the front-end interface is
designed, will influence the choice of the amplifier, and determine
the sampling requirements for the ADC. The FOIs are best
determined by doing research on the modality of interest, especially
for any biologically originating signals as most of them have been
well characterized.
The FOI is often derived by an examination of the power
spectrum of the modality of interest. For example, consider the
power spectrum of an ECG signal sampled at 256 Hz (Fig. 7.2).
By 40 Hz, we can see a 15 dB drop in the power of the signal,
−15
which means that the importance of the 40 Hz frequency is 10 10 =
Figure 7.2. Power spectrum of ECG signal. The peak at 120 Hz is a

harmonic of the 60 Hz power-line interference. See also Color Insert.
Figure 7.3. Biopotential signal transduction pathway
3.1% that of the dominant frequency. The importance continues

−40
to drop down to −40 dB, or 10 10 = 0.01%. Depending on the
application, this data can help determine what range of frequencies
are of interest.
7.3.2 Measuring Surface Biopotentials

There are a number of sensing modalities that are based on
detecting and amplifying biopotentials from the skin. These include
the electromyogram (EMG) for detecting the activity of muscle
fibers, the electrocardiogram (ECG) for detecting the electrical
activity of the heart, the electroencephalogram (EEG) for detecting
the electrical activity of the brain, the electroretinogram (ERG) for
measuring the activity associated with eye movement, among oth-
ers. All these modalities operate under the same basic mechanism
(Fig. 7.3).
7.3.2.1 The electrode

The first site of signal transduction occurs at the skin-electrode
interface, where ionic potentials are converted into electrical
potentials through reduction/oxidation reactions that follow the
basic form
C C n+ + ne− (7.1)
A m− A + me− (7.2)
where C and A represent the cation and the anion, respectively,
and m and n represent their respective valencies. These reactions
are reversible for most electrode/electrolyte combinations, but not
all. As a result, it is wise to not be too creative with electrode
Figure 7.4. Silver-silver chloride electrode redox reaction.
choices. There are many electrode/electrolyte combinations that

have various advantages and disadvantages, but we will focus on
the most commonly used one: the silver-silver chloride (Ag-AgCl)
electrode. Among other benefits, the electrode chemistry will help
minimize the effects of motion artifacts [14].
The Ag-AgCl electrode consists of a layer of silver surrounded
by a very thin layer of silver chloride. A gel containing a high
concentration of Cl− ions and saturated with AgCl is often used in
conjunction with the Ag-AgCl electrodes (Fig. 7.4).
The interface of the Ag-AgCl electrode, the electrolyte, and
the skin form a equilibrium, which is governed by the following
equations:
Ag Ag+ + e− (7.3)
Ag+ + Cl− AgCl (7.4)

Metallic sliver will give up electrons and spontaneously combine
with free chloride ions to form silver chloride, which due to its
limited solubility, will precipitate out of solution adding to the silver
chloride layer. These reactions occur constantly and reversibly, but
the concentrations of the various constituents remain the same
averaged over time resulting in no net current. However, if the
potential of the tissue at the location of the electrode changes, then
the reactions will be driven in one direction or another causing
a net flow of electrons into or out of the electrode, which can be
measured.
It is also possible to measure the electrical potential differences

using capacitative electrodes that do not have a galvanic connection
[15]. These electrodes can be fabricated with a thin layer of
nonconducting dielectric material and offer benefits, including being
able to sense potential differences though clothes, but suffer from a
very high sensitivity to external noise and an attenuation in the low-
frequency components of the signal.
7.3.2.2 Filtering
As mentioned in Section 7.3.1, different biological processes exhibit
electrical activity at different frequencies, and should be filtered
accordingly. Beyond filtering for unwanted noise, a low-pass filter is
always required to avoid aliasing, while a high-pass filter is required
to bias the signal. Furthermore, it is always a wise idea to add an
RFI filter [16] to prevent output offset errors due to RF rectification
inside the amplifier.
7.3.2.3 Amplifier
Picking an optimal operational amplifier (OpAmp) for the applica-
tion is a critical step in the design in a biopotential measurement
device. Care must be taken that the amplifier does not contribute
additional noise to the system, that its input impedance is
sufficiently greater than the source impedance of the signal to
prevent distorting the signal, and that it can provide sufficient gain
at the frequencies of interest. A number of other issues exist, such as
verifying that the amplifier is able to operate at the supply voltage
of the PCB or that the quiescent current consumption is sufficiently
low, but these issues are more obvious and do not require as much
attention.
As most biosignals are based on the difference in potential
from two different sites, the use of an instrumentation amplifier
(InAmp) is highly recommended. An InAmp is designed to provide
balanced, high input-impedance (typically 109 ) inputs with low
bias currents, and a low impedance output with low DC-offset error.
An InAmp can be constructed with two or three OpAmps; however,
using a monolithic single-package InAmp has the advantage of
Figure 7.5. Texas Instruments INA331 Instrumentation Amplifier Gain-

Frequency Bode Plot.
having highly matching passive components that are often trimmed

to provide lower errors and very high common-mode rejection.
Additionally, a monolithic InAmp will use less PCB space to help
reduce the WBAN size. There are a number of free online references
that help explain OpAmps [17] and InAmps [18].
The first step in picking the right amplifier is to see if it’s able to
provide a flat gain response over the range of frequencies of interest
(Section 7.3.1) by examining the Bode plot, or the gain-frequency
response of the amplifier on a log-log plot (Fig. 7.5).
Although most passive components can generate noise, referred
to as thermal noise, the noise from active components (i.e.,
OpAmps/InAmps, transistors, etc.) usually dominates the signal.
There are a number of noise sources in OpAmps and InAmps, but the
main ones to be concerned with are input bias currents and input-
referenced voltage noise. For biopotentials, the source impedance
is often very high (upward of 105 ), where the bias-current noise
will dominate. Only by understanding the characteristics of the
signal requiring amplification can you determine what amplifier
requirements exist for your implementation.
7.3.2.4 Analog Digital Converter and Microcontroller

The number of low power microcontrollers (MCU) is ever increasing
with certain applications allowing a device to remain operational
for years with a single battery. Most contemporary MCUs include
ADC and digital to analog converter (DAC), which greatly simplify
the design; however, do not feel restricted to using the provided
peripherals if they do not satisfy your design requirements.
For the design of the Berkeley Tricorder, we required an MCU
that provided us with multiple fast, high-precision ADC inputs for
data acquisition as well as a DAC for our SpO2 implementation. We
needed multiple serial (UART, SPI, I2 C) interfaces, a DSP for digital
filtering, and low power consumption for extended battery life. We
found the Texas Instruments MSP430 MCU a perfect fit for our
needs. Our current design utilizes the MSP430F2618, which is fast
(16 Mhz), and has large RAM (8 kb). The built-in ADC is capable of
200 k samples per second with a 12-bit resolution and is multiplexed
across eight channels.
7.3.3 Electrocardiograph
The electrocardiograph is a device that measures the electrical
activity of the heart. The heart can be thought of as four distinct
pumps, which are synchronized to beat in a specific order. Blood
from the body first fills the right atrium, which is essentially a low-
pressure pump that fills the right ventricle. The right ventricle is
a higher pressure pump that pumps blood into the lungs. When
blood returns from the lungs, it similarly fills the left atrium, which
fills the left ventricle whose job is to pump blood at high pressure
throughout the body.
The timing of these pumps is coordinated by means of the heart
tissue itself. A region known as the sino-atrial node, known as the
pacemake of the heart, initiates an electrical depolarization that
causes muscle tissue nearby to start contracting. A healthy heart is
able to direct the wave of electrical depolarization in such a way
to cause a highly orchestrated series of contractions resulting in a
synchronized pumping action.
Figure 7.6. ECG waveform. See also Color Insert.
When the heart initiates a contraction, a potential difference

forms between different regions of the heart and follows the wave
of depolarization as it progresses though the heart. The potential
difference can be treated like a dipole that moves through a volume
conductor, generating potential differences on the surface of the
body. This potential difference is what the ECG detects, and the
placement of the electrodes provides different views of the heart’s
electrical activity (Fig. 7.6).
A damaged heart, as is the case after a heart attack, can
result in problems with the propagation of the wave of electrical
depolarization resulting in a number of chronic conditions. By
utilizing an ECG to detect how the electrical activity of a heart has
been altered, a cardiologist can determine how the heart has been
damaged.
There are many additional uses for an ECG. The simplest use is to
measure the heart rate, which is performed by measuring the main
left-ventricular contraction, which appears as the strong positive
deflection known the R-peak.
ECGs are described by either the number of channels or the
number of electrodes. A channel refers to the number of different
differential pairs, whereas the number of electrodes simply refers
to the number of electrodes. The same electrode can be used as

a reference for multiple differential pairs, but when people speak
about 5- or 12-electrode ECGs, standard known configurations are
used. At a minimum, two electrodes are required to generate an ECG
trace, but 3-electrode, 5-electrode, and 12-electrode configurations
can provide additional views of the heart’s electrical activity to
aid in monitoring and diagnosis. In this section, we will detail the
implementation of a single-channel ECG, which can be extrapolated
out to configurations requiring additional channels.
The range of frequencies of interest for an ECG depends on the
application. Often, the distinction is made between monitoring-ECG
and diagnostic-ECG. Monitoring-ECG is used for routine examination
of the ECG waveform, and is more heavily filtered to remove noise
artifacts with a pass-band of 0.5 Hz to 40 Hz. Diagnostic-ECG is
filtered less to provide a more detailed waveform for diagnosing
cardiac problems, and typically ranges from 0.05–150 Hz. You can
see the relative contributions at the various frequencies in the power
spectrum in Fig. 7.2. The peak at 120 Hz results from the first
harmonic of the 60 Hz power-line noise; the 60 Hz peak is not
noticeable due to the high signal-to-noise ratio at that frequency.
The heart is a large muscle capable of generating a strong
signal on the surface of the skin, in the range of several millivolts
in amplitude, which makes it a fairly easy signal to detect. For a
two-electrode configuration, placing an electrode on either side of
the chest will work; however, it will be useful to understand the logic
behind some standard ECG configurations in determining where to
place the electrodes.
The first electrode configuration proposed was by Einthoven
(Fig. 7.7), and uses two channels and three electrodes. The
electrodes are placed on the left arm, right arm, and the left leg,
and forms a triangle with the heart in the center. The differential
voltage between any two electrode locations defines a view from
the lead defined by electrodes, denoted as I, II, and III. Given the
symmetry of the triangle, only two differential pairs are required to
be measured, and the third lead can be determined from the sum or
difference of the two differential measurements [19]. For example,
if both positive electrodes are placed on the right foot (V F ), and the
negative lead from one channel is placed on the left arm (V L ) and the
Figure 7.7. Einthoven ECG lead locations (from wikiCommons). See also
Color Insert.
other is placed on the right arm (V R ), then you have the following
relationships:
VII = V F − V R (7.5)
VIII = V F − V L (7.6)
It can then be shown that the difference of equation 7.6 from 7.5 will
yield V I .
VII − VIII = (V F − V R ) − (V F − V L )
VII − VIII = V L − V R = VI (7.7)
There are a number of other electrode configurations worth
considering [20–22], but it’s important to note that differential
surface potentials can be arithmetically combined to form different
electrical views as long as more than one channel is being used.
If the electrodes are close to any other large muscles, such as the
pectorals, then they can also pick up the electrical activity associated
with the muscle activity. It is possible to build a device known as an
electromyogram (EMG) to measure the activity of arbitrary muscles,
which will be discussed in Section 7.3.4.
After deciding where the electrode, are to be placed, leads must
be attached connecting the electrodes to the amplification circuitry.
The leads should be kept as close to one another, and ideally
be twisted together to prevent picking up electromagnetic noise

differentially. The instrumentation amplifier can do a very good job
of rejecting common-mode noise, which appears on both wires;
however, the further the wires are from each other, the greater the
probability that noise can be coupled onto just one wire, which will
show up as a differential signal and amplified by the InAmp.
7.3.3.1 Berkeley Tricorder

The Berkeley Tricorder was designed to have a monitoring-ECG
rather than a diagnostic-ECG as our use-case required any detected
abnormalities to be escalated to a cardiologist for a workup. The
design utilizes a second-order high-pass filter biased at the mid-
supply voltage followed by radio-frequency interference filter and
an instrumentation amplifier with a gain of 1000. A second-order
high pass filter was required to minimize the DC bias drift resulting
from motion artifacts while maintaining a low −3 dB corner
frequency (0.79 Hz). A low-pass filter was not required as the
instrumentation amplifier has a natural roll-off starting at 200 Hz
at a gain of 1000. Although this frequency is higher than required, it
simplified the design to use the built-in low-pass filter than to add
an additional one.
The
output of the instrumentation amplifier is biased to mid-
rail 3.32 V = 1.65 V , and the ADC has a positive/negative reference
voltage of 3.3V/0V, respectively. Thus, the output of the amplifier can
swing ±1.65 V before saturating.
For a gainof 1000, this translates
to an input voltage of ±1.65
1000
V
= ±1.65 mV from the skin surface
potential.
7.3.4 Electromyogram
The focus of this section will be on surface EMG measurements.
Although transcutaneous (“needle”) EMG electrodes are useful
for a number of applications [23], it is not likely that a WBAN
implementation will have this requirement and the safety concerns
associated with placing electrodes under the skin is beyond the
scope of this chapter.
A lot of the same methodologies apply to EMG as it does to

ECG, after all, the ECG is a specialized case where the electrical
activity of the myocardium is being measured. As with the ECG,
the EMG measures surface potentials generated by muscle fibers
during contraction. There are a large number of muscles that can be
measured, all with different applications ranging from kinesiology
studies [25] of motion to stress detection [26].
Unlike ECG, the EMG signal has higher frequency components
as noted in its power spectrum (Fig. 7.8) and thus require high
sampling rates. It is typically accepted that the EMG signal has a
bandwidth of 500 Hz, thus minimum a sampling rate of 1 kHz is
typically used due the Nyquist sampling theorem. Given the shape
of the power spectrum, a bandpass filter between 5–10 Hz and
500 Hz is typical. The band-pass filter is created by placing a low-
pass filter of 500 Hz in series with a high-pass filter at 5–10 Hz.
However, if only the envelope of the signal is required, rather than
the individual peaks, the signal can be rectified, low-pass filtered at
50 Hz, and sampled at 100 Hz [27] (Fig. 7.9). EMG surface potentials
can range from 0–6 mV peak to peak, depending on the muscle being
Figure 7.8. EMG power spectrum (redrawn from [24]).

Figure 7.9. (a) Raw EMG signal, (b) rectified EMG signal, (c) low pass
filtered.
investigated. Hence, the amplifier should provide sufficient gain to

utilize the full dynamic range of the ADC without saturating.
For a more detailed examination of the requirements associated
with EMG measurements, the reader is directed to reference [24].
7.3.4.1 The Berkeley Tricorder

Our initial interest in an EMG sensor was to measure back tension
by measuring muscle activity as a proxy for stress. For this, we did
not require a very fast input stage and reused the bulk of the ECG
design. Given that the bandwidth of the EMG signal is in the range of
20–500 Hz, we could be more aggressive with the HPF. A single stage
with a a −3 dB of 7.9 Hz was chosen and found to be effective. The
amplifier we chose demonstrates a gain roll-off starting at 300 Hz,
and allows us to sample the signal at a slower rate. This is sufficient
to measure a general level of muscle activity. However, for a more
diagnostically relevant EMG implementation, we are considering
migrating to a faster amplifier and higher sample rates.
7.3.5 Pulse Oximetry (SpO2 )

Blood oxygenation is a critical parameter that can help diagnose
conditions of pulmonary distress, hypovolemia [28], vascular
perfusion, among other. Blood oxygenation has a secondary benefit
in that it can be used to compute pulse transit time (PTT), or the
amount of time between the ejection of blood from the heart to the
pulse arrival time at a peripheral location. It has been shown that
there is a direct correlation between PTT and arterial blood pressure
[29].
Pulse oximetry is a method by which the variation in the

absorption of light by tissue is used to determine the percentage
saturation of blood hemoglobin by oxygen. The modern technique
was invented by Takuo Aoyagi, and works based on the following
two principles. The first involves the macromolecule responsible for
oxygen transport, hemoglobin. The iron containing macromolecule
found inside blood cells binds to oxygen molecules and helps
transport them at much higher concentrations than possible with
freely dissolved oxygen in the blood. Deoxygenated hemoglobin (Hb)
absorbs red light more than oxygenated hemoglobin (HbO2 ), giving
HbO2 a brighter red color. The opposite is true for infrared light,
where HbO2 absorbs infrared light more than Hb (Fig. 7.10). This
means that by measuring the ratio of absorption of red versus
infrared light, the percentage oxygenation of the hemoglobin can
be determined in samples of blood. However, there are many other
types of tissues found in the body that also absorb varying levels of
red and infrared light, so the simple technique cannot be applied to
blood in the body.
The second principle, which was the key to Takuo’s discovery,
was that it is possible to measure the variation in amplitude of
light being absorbed due to the pulsatile nature of blood. As the
Figure 7.10. Hemoglobin light absorption curves.

heart beats, a pressure wave travels through the arteries to the

capillaries, which change in volume as they fill and empty with
blood. By examining the ratio of absorption for just the pulsatile
portion of the signal, you are able to compare red and infrared light
absorption by the portion of blood that is changing, allowing for a
direct measurement of a component of blood rather than that of any
other tissue components.
The implementation of this technique is simple: measure the
ratio of the pulsatile portion of red to infrared light as it passes
and is absorbed by tissue, multiplied by a calibration value. Both
these steps are performed by shining alternating red and infrared
light from closely mounted (ideally on the same die) LEDs through
tissue (i.e., finger, ear) and measure intensity of transmitted light by
a PIN photodiode. A PIN photodiode is sensitive light detector. The
resulting waveform will resemble Fig. 7.11. Next, in software, the
amplitude of the waveform is measured for both light frequencies
(AC R E D , AC I R ) as well as the DC bias of the waveforms (D C R E D ,
D C I R ). The ratio AC RE D
AC I R
needs to be corrected for variations in light
intensity from the differed LEDs, the difference in absorption of the
different frequencies of light by the tissue, as well as the varying
sensitivity of the photodiode
to the different frequencies of light. The
DC I R
ratio of DC biases D C R E D takes into account all of these variables,
Figure 7.11. Light absorption through tissue by infrared (top curve) and
red (bottom curve) light.
and the product of the AC RE D

and the correction factor DDCCRIERD will
AC I R
result in a ratio R = DDCCRIERD AC RE D
AC I R
which directly corresponds to
the percentage saturation of hemoglobin by oxygen
(SpO
2 ).
AC R E D
log10 DC RE D
Note that most literature defines R as . However, as
AC I R
log10 DC I R
the ratio R is an arbitrary number used in conjunction with an

empirically derived mapping, there is no need to take the logarithms.
This simplified explanation of the process does not explain the
underlying theory; for more information, the reader is directed to
reference [30].
As contemporary amplifiers are very fast compared to the rate
of the heart beat, it is possible to only have the LEDs on for a very
short period of time, take a measurement, then shut off the LEDs to
conserve power for some period of time. An additional light intensity
measurement should be taken while the LEDs are off in order to
subtract out any background light from both the infrared and red
light signals. The duration of the LEDs on time to the off time is
referred to as the duty cycle, and the lower the duty cycle, the less
power the circuit consumes, which is an important consideration for
a battery-powered device.
Although a theoretical correlation between R and the percentage
SpO2 exists, sufficient deviation exists to necessitate use of an
empirically derived mapping of R and percentage SpO2 . The
empirical mapping is made by simultaneous measurements of R
through pulse oximetry and blood oxygen though samples drawn
from a subject while the subject is being deprived oxygen. Based on
these measurements, a conversion factor that maps R values to SpO2
percentages can be determined, which is valid for the device and
LED/photodiode configuration. A new empirically derived mapping
will have to be made if the circuit or any of the components change.
Also, the mapping is only valid down to approximately 80% blood
saturation as depriving oxygen to the test subjects below that
point is dangerous. This is the case with all commercially available
monitors. However, given that there are number of commercially
available and calibrated pulse oximeters, it is sufficient to use one
of these devices to derive an R-SpO2 mapping, rather than drawing
and analyzing blood from test subjects.
The description thus far has been for a transmissive pulse

oximeter that works by measuring the levels of light absorbed by
blood as light passes through tissue. There exists another type
referred to as a reflective pulse oximeter that works by measuring
the levels of light absorbed by blood as the light travels through
tissue and backscatters back to the same side of the tissue where
the LEDs are. The most prominent placement for such a sensor is the
forehead, but other locations have also been evaluated in reference
[31].
There are a number of tradeoffs that should be considered early
in the design process:
• The SNR is proportional to the volume of blood in the

tissue being measured, thus; tissues that suffer from poor
blood flow or vasoconstriction due to cold temperatures
will result in a poor signal.
• The greater the SNR, the further the LEDs and photodiode
are from each other, as the light travels through and is
affected by more tissue, but additional LED current is
required (and LEDs that can handle higher currents) which
decreases battery life.
• The faster the amplifier and ADC circuitry, the shorter
the duty cycle of the LEDs, which results in lower power
consumption at the expense of additional noise due to
higher cutoff frequency for the low-pass filters
There are many different SpO2 circuit implemetnations, such as this

one by Texas Instruments [32]. However, the basic building blocks
consist of noise-free current source to drive either a RED or IR LED,
a PIN photodiode to pick up the light transmitted, a transimpedance
amplifier to convert the current from the photodiode to a voltage,
and a DC subtraction and amplification stage to remove the DC bias
and further amplify the signal.

In order to derive blood oxygen saturation, determine pulse arrival
time, and keep costs low, we needed to record the SpO2 waveform
rather than to use an OEM device, which only computes the oxygen
Figure 7.12. Block diagram of pulse oximeter.
saturation percentage. Being able to visually inspect the SpO2

waveform is also important to the physician as a figure of merit for
the SpO2 calculation. Poor perfusion or poor SpO2 placement can
lead to erroneous oxygenation.
Our design (Fig. 7.12) called for a circuit capable of illuminating
one of the two LEDs with a constant current and measuring
the amount of light picked up from a nearby PIN photodiode.
We decided to do our initial development and testing using a
commercial reflective SpO2 sensor manufactured by Nellcor (Max-
Fast, Nellcor/Tyco, Pleasanton, CA). Nellcor produces a series of
disposable and reusable sensors, which utilize a DB-9 connector
with identical pinouts allowing for the ability to quickly test different
devices and configurations.
The output stage of our device consists of a constant-current H-
Bridge LED driver. The LEDs on the Nellcor sensor have their anode
and cathodes connected. By driving one side high and the other low,
one can choose which LED to illuminate. It is critical to drive the
LEDs at a constant current to avoid coupling supply noise onto the
received signal.
The input stage consists of a transimpedance amplifier (TIA)
connected to a PIN photodiode in the sensor followed by a
differential amplifier used to subtract out the dc-offset of the signal
and add additional gain. The offset subtraction signal is generated

from the MSP430’s on-board DAC. A feedback loop keeps the SpO2
signal from saturating by adjusting the DAC output. The DAC value is
also stored and used for the blood oxygenation computation.
The system also measures background levels by repeating the
measurements with both LEDs are turned off. These values are also
used in the blood oxygenation calculation. All the data is then stored
and/or transmitted for further processing to extract the relevant
features.
7.3.6 Respiration
Monitoring respiration is important in a large number of cases. It’s
critical to monitor respiration in patients with cardiopulmonary
issues such as congestive heart failure and patients who are on
medication that suppresses breathing. It’s also useful in diagnosing
a number of conditions such as systemic inflammatory response
syndrome (SIRS).

Our original attempt at respiration rate detection relied on acoustic
pickup with a microphone, but we found bioimpedance to be a much
more reliable indicator of respiration rate. Bioimpedance also allows
us to measure relative breathing tidal volumes and to detect coughs.
Bioimpedance can also be used for a number of other diagnostic
measurements beyond respiration such as measuring cardiac output
[33] and body fat [34], although we have not investigated these
applications.
A high frequency sinusoidal signal (50 kHz) is used so that the
effects of the skin impedance are minimized through capacitative
coupling. We chose a 350 μA current at 50 kHz in order to provide
us with a good signal, while being significantly lower than the
maximum safe current as defined by IEC60601-1-2005. Using a pair
of electrodes connected to a high impedance differential amplifier,
we can accurately measure the voltage difference between the
electrodes, which is directly related to the impedance between those
points.
We originally planned on building a fully analog impedance mea-

surement system; however, we were able to significantly reduce the
system complexity by utilizing Analog Device’s AD5933 impedance
analysis IC. The AD5933 has previously been shown effective in
bioimpedance measurements [35]. Our output stage consisted of a
voltage-to-current circuit followed by a 3.3 nF DC blocking capacitor.
The input stage utilizes the ECG electrodes, passes the signal through
an RFI filter to a programmable differential amplifier which feeds
the amplified signal back to the AD5933. Although the AD5933
should be very effective at filtering out any high-frequency noise, we
felt it would be prudent to add an RFI filter to the input stage —
given its importance in ECG measurements. The IC then computes
the real and imaginary components of the signal, which we convert
to a magnitude and phase. The magnitude of the impedance is
proportional to the chest volume, and breathing can easily be seen
as a variation of the impedance (Fig. 7.15, bottom trace). From this
signal, we can determine respiration rate and approximate tidal
volume.
7.3.7 Accelerometry
Most health monitors incorporate an accelerometer. With this, one
can determine body orientation, activity levels, and perform fall
detection. With additional processing, it is possible to deduce what
activity the subject is engaded in [36, 37–39]. Furthermore, it is very
easy to implement — hence its prevalence.

Our first implementation utilized a 3-axis analog accelerometer,
which was connected to the MSP430 by means of three analog
input lines. Although this approach worked well, it used too many
ADC channels of the MSP430. We later switched to the LIS302DL
manufactured by STmicroelectronics. Not only does it provide a
digital serial interface, but it also has built-in hardware fall detection.
Although initially designed to detect the fall of a portable electronic
device, the IC can be used to detect if the patient falls down without
taxing the microcontroller.
Wireless Interface 211
7.4 Wireless Interface
The number of wireless options is ever changing and evolving with

no clear dominant player at this point. One must consider all the
requirements based on the end application to determine the right
technology. Issues to consider include transmission range, data rate,
interoperability, and operating frequency. It is also insufficient to
simply present a table with the comparison of the various features
as many of the values will differ based on the manufacturer and
operating environment.
We will discuss the requirements for the Berkeley Tricorder and
the rational behind the choice of its wireless technology, followed by
a discussion on some of the issues that are important to consider in
making a decision for the wireless data link. Additionally, Table 7.2 at
the end of this section provides a general summary for a number of
wireless technologies. The complied data should not be considered
concrete as there are many different implementations of the various
technologies that it should only be considered a starting point.
7.4.1 The Berkeley Tricorder

Telemetry is an important requirement for a medical device as it
provides a means for the diagnostic data to be viewed in real-time, or
to be transmitted to caregiver or a more powerful system to perform
computational analysis of the data. We chose Bluetooth to satisfy our
telemetry needs for a number of reasons.
Table 7.2. Comparison of wireless technology standards
ZigBee Bluetooth ANT Wi-Fi Wi-Fi
IEEE Standard 802.15.4 802.15.1 Propriatary 802.11b 802.11g

Frequency 868 Mhz/915 Mhz/ 2.4 Ghz 2.4 Ghz 2.4 Ghz 2.4 Ghz
2.4 Ghz
Max Rate 20/40/250 Kbps 1−24 Mbps 1 Mbps 11 Mbps 54 Mbps
Over Air
Max Achievable 10/20/125 Kbps − 20 Kbps 6.3 Mbps 31.4 Mbps
Rate
Range (meters) 10−75 (1500 for Pro) 1−100 10 10−100 10−100
Current (mA) 1−50 1−35 1−22 100−350 100−350
• High level of penetration in consumer devices such as cell

phones and laptops.
• Standardized profiles (serial, audio, object transfer, dial-
up networking, etc) requiring no additional development
effort.
• High-speed wireless (v1.2 Bluetooth is rated at 1 Mbit/s).
• Lower power consumption per bit than competing devices.
• Single IC implementation.
• Multiple built-in interfaces peripherals to help simplify
device design such as an ADC/DAC/USB interface
Rather than using a monolithic serial to Bluetooth module, we have

incorporated the CSR plug’n’go BlueCore3 chipset directly. Not only
does this give us full access to many peripheral interfaces, but it
also significantly reduces the device cost. The BlueCore3 IC provides
audio input/output, a USB interface, 16 general-purpose I/O lines,
which can be used for I2 C or SPI communication, ADC, DAC, and a
serial interface. The firmware can be configured to provide many
different profiles including object exchange (OBEX) to facilitate bulk
data transfer, and more importantly, dial-up networking (DUN) to
allow telemetry without the need for any custom phone software.
Our current implementation utilizes Bluetooth as a serial device and
interfaces with the MSP430 at up to 1 Mbps.
7.4.2 Power Consumption

Of primary importance to many WBAN projects is the power
consumption of the wireless data link, which in most cases is the
dominant power consumer for the device. As we stated earlier,
it’s almost impossible to put together a table that’s capable of
comparing, apples to apples, the power consumption of the various
radio technologies. This is because it’s impossible to separate
the radio technology from an implementation of that technology.
Consider a radio transceiver that generates the RF signal. There
are many transceivers to choose from which all have different
power consumption characteristics, even if they operate at the
same frequency, have identical output power, and modulation.
For example, both Texas Instruments CC2500 IC and Nordic
Semiconductor’s nRF2401A IC operate at 2.4 Ghz and can support

Gaussian Frequency-Shift Keying (GFSK) modulation, which is used
for Bluetooth. However, the nRF2401A datasheet specifies 13 mA
at 0 dBm (1 mW) output power while the CC2500 specifies
21.2 mA. Even this comparison is unfair as there are many other
differences between the feature set of the components, but it demon-
strate the difficulty in comparing power consumption between
technologies.
Newer implementations of a protocol can vary the power
consumption of a device — depending on the feature set. For
example, Bluetooth version 4.0 defines Bluetooth low energy
technology, which trades off bandwidth for a decrease in power
consumption (50% to 99% reduction based on use case) while
doubling the maximum range.
The duty cycle of the radio will have a huge impact in regards
to power consumption. If your wireless technology allows for a
maximum bandwidth of 1 Mbps, but your application only requires
a 10 Kbps stream to be transmitted, then you should be able to shut
down the radio 99% of the time to save on power consumption.
This is just a very rough approximation, and it’s never so so clear
cut. There is additional time required for the radio to exit low
power, it needs to power on at regular intervals to see if some other
transmitter is sending it data, it needs to deal with the protocol
overhead, and if there is a problem with the data link, then the radio
will need to retransmit data.
7.4.3 Data Range and Transmit Power

The range of a wireless data link is dependent on the power of the
transmitter, the quality and orientation of the antennas (TX & RX),
the frequency of the transmitter, and the surrounding material. Let’s
first discuss how the power from the radio is sent to the antenna,
through space, and to the receiver and the various points of power
loss along the path.
The transmit power plays a very large role in the range of a
wireless link, and how much of this signal makes it to the receiver
will determine the range of the link. RF power is measured in dBm
which decibel value in reference to 1 mW of transmit power, and is
defined as
10(x/10)
P = (7.8)
1000
where P is the transmit power in watts and x is the dBm. Note that
0 dBm is equal to 1 mW, and the power doubles for every 3 dBm,
or is halved for every −3 dBm. The RF transceiver will generate
this power, and it will be sent to an antenna; however, impedance
mismatches between the transmitter, the PCB trace(s) (or coax
cable(s)), and the antenna can reduce the amount of transmitted
power.
The type of antenna and its radiation pattern will also play a role
in how efficiently the power is transmitted. The simplest antenna is
formed on the PCB by a specially designed trace pattern. These are
the simplest and cheapest solutions, but not necessarily the best.
The use of a chip antenna is a good compromise with generally
improved performance while utilizing minimal board space at a low
cost.
The radiation pattern of the antenna determines how the
antenna can focus energy in a particular direction. The ability
of an antenna to direct its energy is measured in dBi, or the
decibels of radiated power for a particular radiation geometry as
compared to an isotropic antenna, which radiates equally in all
directions. Antennas exist that can focus the transmission energy
giving significant gain in one orientation, but at the expense of
being significantly weaker in other orientations. If the body-worn
RF device is to be consistently oriented in reference to a remote
receiver, then the transmit range can be extended by using a
directional antenna with a high dBi. However, for most applications,
the relative antenna orientations cannot be guaranteed and thus
an omnidirectional antenna will be preferred to maximize transmit
power independent of the subject’s orientation.
Once the RF power hits the antenna and is transmitted in free
space, the RF energy degrades with the square of the distance
between the transmitter and receiver. Furthermore, the ability for
the receiving antenna to pick up a signal also degrades with the
square of the frequency. The loss attributed to these two factors
determines the free-space path loss (FSPL) and is defined as the
ratio of transmit power (Pt ) to receive power (Pr ). For an isotropic
antenna,
Pt = 4π d 2 S (7.9)
and
Sλ2
Pr = (7.10)
4π
where S is power per unit area, λ is the wavelength, and d is the
distance between the transmitter and receiver. Considering the ratio
of Eqs. 7.9 and 7.10

Pt 4π Sd 2 2d
2
d 2
FSPL = = = 16π 2 = 4π
Pr Sλ2 /4π λ λ
Substituting in λ = c/ f , where f is the frequency and c is the speed
of light,
2
4π
FSPL = df (7.11)
c
Expressed in dB,
2
4π
FSPL(d B) = 10 log df (7.12)
c
For a 2.4 Ghz transmitter with a receiver 10 m away, the free
space path loss will be in the order of 10 log((4π/3×108 )(10×2.4×
109 )) ≈ 30 dB.
However, the losses will be greatly increased if there is no clear
line of sight between the transmitter and receiver. This loss is
dependent on both the frequency as well as the blocking material,
such as the building walls or the human body. Consider 2.4 Ghz
radiation — it is used by microwave ovens to heat water in foods.
Likewise, given that the human body consists mostly of water,
the energy of a 2.4 Ghz RF link (about 1,000,000 weaker than a
microwave) is absorbed and heavily attenuated by the human body.
This is a huge problem with many of the technologies that operate
at 2.4 Ghz, such as certain ZigBee and Wi-Fi implementations, and
Bluetooth. If the body-worn device is on the front of the subject’s
chest, and the receiver unit is located behind the subject, severe
signal degradation will occur.
7.4.4 Data Rate

The maximum data rate is determined by the operating frequency
and the modulation technique of the technology in question. This
(often sited) theoretic upper limit does not include the protocol
overhead and assumes a high quality RF link. One should consider
the maximum required data rate for the application at hand,
determine the location of any bottlenecks in the system, and verify
that the chosen technology can satisfy that data link given real
values, not the ones found in datasheets.
For example, consider the case of two ZigBee-based sensors,
each sampling with a resolution of 12 bits at 4,000 samples per
second (sps) with the data being received by a PDA mounted with
a ZigBee receiver. If every sample is to be sent as a 16-bit word, then
each sensor will generate 64,000 bits per second of data (bps). This
means that the receiver must be able to handle 128 Kbps. ZigBee
is rated to handle 250 Kbps according to the spec sheet, but this
doesn’t include all the protocol overheads or any retransmissions
due to loss. A good rule of thumb for ZigBee is that you should
expect about half specified transmission rate with a good data link,
or 125 Kbps. In this case, a reduction in the sampling rate or some
other technique to lower the data rate would be required — or using
a different technology that can handle the higher data rates.
7.4.5 Human Safety

As discussed in Section 7.4.3, the human body is capable of
absorbing RF energy and converting it to heat. Although a radio
transmitter is orders of magnitude less powerful than a microwave
oven, it is important to consider how the energy can affect the
body or implants such as pacemakers, and so as to minimize any
potential health hazards. There are a number of different regulatory
bodies that have established safety guidelines for the use of RF
transmitters that should be referenced. In the United States, the
FCC guidelines were derived from the recommendations of the
National Council on Radiation Protection and Measurements (NCRP)
and the Institute of Electrical and Electronics Engineers (IEEE).
Many European countries use the guidelines established by the
International Commission on Non-Ionizing Radiation Protection

(ICNIRP). The NCRP, IEEE, and ICNIRP guidelines specify different
permissible energy levels based on the frequency and on whether
part of the body or the whole body is exposed.
The FCC’s guidelines can be found in OET bulletins 56 3 and
654 , and are highly recommended readings. As far as RF emissions
are concerned, the FCC differentiates between whole-body exposure
from partial-body exposure from a smaller device, such as a
cellphone or WBAN device. The FCC limits for partial-body exposure,
defined as a radiating source operating within 20 cm of the user, are
based on the specific absorption rate (SAR) of a particular frequency
of energy per unit mass of tissue. For the general public, 1.6 W/kg
averaged over 1 g of tissue are permissible, though less restrictive
limits are defined by the regulatory agencies in other countries.
For example, the CENELEC-imposed limit in the United Kingdom is
2 W/kg averaged over 10 g of tissue5 .
The shielding in pacemakers and other implantable devices
should prevent most interference from a WBAN device; however,
contemporary implantable devices are designed for bidirectional RF
communication and thus a risk interfering with the operation of the
device exists. As a result of a study performed in 1997 [38], the FCC
recommends that the transmitter should be at least 6 inches away
from the pacemaker.
7.4.6 Security
For most WBAN-type applications, especially if health-related infor-
mation is being gathered, then data security is a must. Depending
on the country the device is to operate in, various government
regulations apply, such as the Health Insurance Portability and
Accountability Act (HIPAA) in the United States. Violations of these
requirements can result in incarceration, so care must be taken in
the implementation of the authentication of peers and encryption of
data to be transmitted.
3 http://www.fcc.gov/oet/info/documents/bulletins/#56
4 http://www.fcc.gov/oet/info/documents/bulletins/#65
5 http://webstore.iec.ch/webstore/webstore.nsf/artnum/033746
Many of the wireless technologies support some level of

encryption, but as security expert Bruce Schneier often states,
it’s not the encryption that fails, but the implementation of the
encryption. Consider the WEP encryption used by many Wi-Fi
routers. The encryption used is RC4, which is still secure. However,
the implementation of the encryption is faulty, and as a result a WEP
password can be determined within minutes using standard tools.
In the case of WEP, this is because the encryption vectors are reused
often if there is sufficient traffic allowing for cryptographic attacks.
This is a fundamental flaw in WEP, and similar flaws might exist in
the implementation of the security of your wireless technology — or
in your particular implementation of a technology.
If you are required to use a particular technology and you need
to comply with regulations such as HIPPA, but the security provided
by your particular technology is insufficient, there is no reason
you can’t implement your own additional layer of authentication
or security. Bruce Schneier’s book Applied Cryptography is a great
resource in understanding the fundamentals of encryption and
authentication if you plan on implementing your own. Some
encryption schemes are simple to implement and can be done using
spare clock cycles in a microprocessor — others might require
more power computational power on a complex programmable logic
device (CPLD) or digital signal processor (DSP). However, care must
be taken if a multi-IC solution is used where a secret key might be
transmitted from one IC to another in an unencrypted fashion.
7.5 Batteries
In choosing what type of battery to use to power a WBAN device,

one must consider the battery size, capacity/power density, nominal
operating voltage, internal resistance, and safety. The right battery
for a WBAN implementation is the result of weighing all the
options relative to the design requirements and making an informed
decision.
Size is important in such that it helps determine the minimum
physical dimensions of the WBAN device — though this often
at the expense of capacity. The smallest battery options include
Batteries 219
coin-cells, which can be as small as 4.8 mm in diameter and

1.2 mm in height, and lithium-polymer ion (LiPo) batteries, which
can be manufactured to any specified size. Coin-cell batteries are
convenient that they can be mounted directly to the PCB; however,
it also means that the end user needs to have access to the
PCB to change the battery. And although rechargeable coin-cell
batteries are available, they suffer from higher level of self-discharge
and lower charge densities than other rechargeable technologies,
making them less practical.
Capacity is measured in ampere hours (Ah), or milliampere
hours (mAh) for smaller batteries. The runtime can be determined
by examining the discharge graph for the battery chemistry you
are interested in. If you are using a linear regulator, then you
must maintain the battery voltage above the sum of the operating
voltage of your device and the dropout voltage from the linear
regulator. Consider the case of a lithium-ion battery in Fig. 7.13. If
your microcontroller requires 3.3 V to operate at, and your linear
regulator has a 150 mV dropout voltage, then the minimum battery
voltage required is 3.45 V. The point at which the discharge curve
intersects with the minimum required voltage will indicate your
Figure 7.13. Discharge graph for various battery chemistries (adapted

with permission from mpoweruk.com).
runtime. For a single-cell lithium-ion battery, you can operate down

to 90% of your battery’s capacity, which means 10% of the capacity
is wasted. By lowering the required voltage, you can decrease this
wasted capacity.
If your device uses a switching mode power supply (SMPS), you
will be limited by whichever voltage is higher: the minimum safe
operating voltage of the battery or the minimum operating voltage of
the switching mode supply. However, an SMPS will introduce noise
onto your supply lines, and possibly onto any analog signals in your
device, unless great care is taken in its implementation.
The nominal voltage is the characteristic operating voltage of
the battery, and if the required voltage is greater than the nominal
voltage of a single cell, then putting additional cells in series will
increase the voltage. For example, a LiPo battery has a nominal
voltage of 3.7 V. Two batteries in series will produce a voltage of
7.4 V.
The discharge rate of batteries often normalized and presented
as the discharge rate in ampere hours divided by the capacity of
the battery in ampere hours and is presented as fractions of C . 1C
represents fully discharging a battery in one hour, 2C in half an
hour, and 0.5C in 2 hours. The discharge curve of a battery will be
depressed as the battery ages, at higher discharge currents, and at
lower operating temperatures. For example, in the case of Fig. 7.14,
operating a LiPo battery at 0◦ C will decrease its capacity by 17%
versus at 25◦ C.
The internal resistance of a battery limits the maximum available
current that the WBAN device will consume. Although the WBAN
device might be designed with a low average current requirement,
various components on the device can consume high current in
bursts, especially the wireless transmitter. For example, if your
wireless transceiver is a GSM cell module, then you can expect bursts
of power consumption in the range of 2 amps while your WBAN
device averages less than 100 mA of current. If the battery has a high
internal resistance, then the voltage provided by the battery can dip
below the required levels, causing the digital components on your
device to latch up, and/or result in erroneous data on the analog
side. The use of large-capacitance low effective serial resistance
(ESR) capacitors can help buffer the power usage and minimize
these effects.
Batteries 221
Figure 7.14. Discharge graph for lithium-ion at various temperatures and

currents.
Of the various battery options, LiPo batteries have very low

internal resistance and can provide many amps of current if required
making them an ideal choice. However, it is important to provide
a low-resistance path to the ICs that will be consuming the bursts
of power or else there will be little benefit in using a battery with
low internal resistance. This is often accomplished by the use of
dedicated supply and ground planes on the PCB. This technique will
also help minimize potential supply voltage differences in the circuit,
which can result in unexpected behaviour in analog circuits. Another
advantage to LiPo batteries is that they can be manufactured in to fit
any required shape, though the size is directly proportional to the
amount of charge the battery can store.
If the end user is expected to replace the batteries often, then the
use of a standard battery should be considered. In the United States,
they include sizes denoted as AAA, AA, or 9-Volt. These batteries are
easily found in both rechargeable and nonrechargeable flavors. The
main advantage in using these types of batteries is their availability.
In terms of safety, it is important to design the WBAN device
to prevent the battery, or the cables attached to the battery, from
contacting an unintentional part of the device. For example, if the
WBAN device includes ECG electrodes, and the battery accidentally

contacts the electrode leads, then sufficient current can flow through
the heart to cause fibrillation and kill the user. Furthermore, some
battery chemistries, such as LiPo, are flammable/explosive if the
enclosure is punctured and can result in sever burns. Thus, it is
important to provide an adequate enclosure for the battery to
guarantee that the user or some other component cannot damage
the battery.
7.6 Final Thoughts and the Berkeley Tricorder
The Berkeley Tricorder, currently in its 7th version, has been used
in a number of studies. These include two to evaluate its ability to
record ambulatory data, one on evaluating activities of daily living,
and to evaluate sports training. A sample of the data acquired from
the ambulatory study can be found in Fig. 7.15. The EMG signal was
derived from electrodes placed over the latissimus dorsi muscle on
the back. A reflective SpO2 sensor manufactured by Nellcor (Max-
Fast, Nellcor/Tyco, Pleasanton, CA) was placed on the forehead
as it minimized motion artifacts. The system performed well with
Figure 7.15. Sample data from the Berkeley Tricorder. See also Color
Insert.
References 223
Figure 7.16. ECG signal (lower plot) captured while subject in motion
(upper plot). See also Color Insert.
minimal motion artifacts. Fig. 7.16 shows the ECG signal while the
subject was walking.
Many lessons have been learned through the process of designing
the Berkely Tricorder, and this chaper has tried to convey some
of those learned lessons and expereiences. There is no cookbook
methodology of desinging a WBAN device; rather, the need to un-
derstand the problem and formulate a solution while understanding
the various trade-offs. This same theme has been reitterated in many
sections of this chapter and is perhaps the single most important
lesson this author has learned.
References
1. S. Zenios, J. Makower, P. Yock, T. J. Brinton, U. N. Kumar, L. Denend,

and T. M. Krummel (January 2009) Biodesign: The Process of Innovating
Medical Technologies, New York, Cambridge University Press.
2. Y. Chi, S. Deiss, and G. Cauwenberghs (2009) Non-contact low power
EEG/ECG electrode for high density wearable biopotential sensor
networks, in Proceedings of the 2009 Sixth International Workshop on
Wearable and Implantable Body Sensor Networks, 00, pp. 246–250.
3. Y. Mendelson, R. J. Duckworth, and G. Comtois (July 2006) A wearable

reflectance pulse oximeter for remote physiological monitoring in
Proceedings of the International Conference of IEEE Engineering in
Medicine and Biology Society 2006, 1, pp. 912–915.
4. T. H. Teo, G. K. Lim, D. S. David, K. H. Tan, P. K. Gopalakrishnan,
and R. Singh (2007) Ultralow power sensor node for wireless health
monitoring system in Proceedings of ISCAS 2007, pp. 2363–2366.
5. L. Wang, J. Y. Zhang, J. Huang, L. K. Chen, L. Yu, Y. T. Zhang, and G. Z. Yang
(April 2009) A field programmable analogue ASIC towards low power
processing-on-node BSNs, in Proceedings of the Sixth International
Workshop on Wearable and Implantable Body Sensor Networks, pp. 1–4
doi: 10.1109/P3644.54.
6. E. Teaw, G. Hou, M. Gouzman, K. W. Tang, A. Kesluk, M. Kane, and J. Farrell
(August 2005) A wireless health monitoring system, in Proceedings
of the 2005 IEEE International Conference on Information Acquisition,
pp. 247–252.
7. C. Otto, A. Milenkovic, C. Sanders, and E. Jovanov (January 2006) System
architecture of a wireless body area sensor network for ubiquitous
health monitoring, Journal of Mobile Multimedia 1(4), pp. 307–
326.
8. E. Jovanov, A. Milenkovic, C. Otto, and P. C. de Groen (March 2005) A
wireless body area network of intelligent motion sensors for computer
assisted physical rehabilitation, pp. 1–10.
9. K. Montgomery, C. Mundt, G. Thonier, A. Tellier, U. U. Barker, R. Ricks,
L. Giovangrandi, P. Davies, Y. Cagle, J. Swain, J. Hines, and G. Kovacs
(April 2004) Lifeguard — A personal physiological monitor for extreme
environments, pp. 1–4.
10. D. Malan, T. Fulford-Jones, M. Welsh, and S. Moulton (2004) CodeBlue:
An ad hoc sensor network infrastructure for emergency medical care,
in International Workshop on Wearable and Implantable Body Sensor
Networks, 5 pp.
11. T. Gao, D. Greenspan, and M. Welsh (December 2005) Improving patient
monitoring and tracking in emergency response, pp. 1–6.
12. B. P. Lo, S. Thiemjarus, R. King, and G. -Z. Yang (May 2009) Body
sensor network — A wireless sensor platform for pervasive healthcare
monitoring, pp. 77–80.
13. R. Naima and J. Canny (April 2009) The Berkeley tricorder: Ambulatory
health monitoring, 2009 Sixth International Workshop on Wearable and
Implantable Body Sensor Networks, pp. 53–58 doi: 10.1109/P3644.57.
References 225
14. R. WEBSTER, Medical instrumentation sm 3e, lavoisier.fr (January

1997) URL http:// www.lavoisier.fr/notice/frLWO6A2LARRWLKO.
html.
15. M. Oehler, V. Ling, K. Melhorn, and M. Schilling (2008) A multichannel
portable ECG system with capacitive sensors, Physiological Measure-
ment, 29, pp. 783–793.
16. C. Kitchin, L. Counts, and M. Gerstenhaber Reducing RFI rectification
errors in in-amp circuits, Analog Devices Inc, Application Note (AN-
671).
17. B. Carter and T. Brown (2001) Handbook of Operational Amplifier
Applications, Texas Instruments Application Report, SBOA092A.
18. C. Kitchin and L. Counts (January 2000) A designer’s guide
to instrumentation amplifiers, school.mech.uwa.edu.au URL http://
school.mech.uwa.edu.au/mechatronics/El-resources/Kitchin-Op-
Amplifiers%2520Guide.pdf.
19. F. Wilson, A. Macleod, and P. Barker (December 1931) The po-
tential variations produced by the heart beat at the apices of
Einthoven’s triangle, American Heart Journal, 7(2), pp. 207–211, URL
http://linkinghub.elsevier.com/retrieve/pii/S0002870331904110.
20. E. Goldberger (1942) A simple, indifferent, electrocardiographic elec-
trode of zero potential and a technique of obtaining augmented,
unipolar, extremity leads, American Heart Journal, 23(4), pp. 483–492.
21. H. Burger and J. V. Milaan (1948) Heart-vector and leads, British Heart
Journal, 10(4), pp. 229.
22. J. Malmivuo, R. Plonsey, and J. Cameron (January 1995)
Bioelectromagnetism: Principles and applications of bioelectric and
biomagnetic fields, Informa Pharma Science URL http://www.
informapharmascience.com/doi/abs/10.3109/03091909609009004.
23. G. Rau, E. Schulte, and C. Disselhorst-Klug (2004) From cell to
movement: To what answers does EMG really contribute? Journal of
Electromyography and Kinesiology, 14(5), pp. 611–617.
24. B. Gerdle, S. Karlsson, S. Day, and M. Djupsjobacka (1999) Acquisition,
processing and analysis of the surface electromyogram, in Modern
Techniques in Neuroscience (ed. U. Windhorst and H. Johansson), Berlin,
Springer Verlag, pp. 705–755.
25. J. Ahlgren (1967) Kinesiology of the mandible an EMG study, Acta
Odontologica, 25(6), pp. 593–612.
26. U. Lundberg, R. Kadefors, B. Melin, G. Palmerud, P. Hassmen, M.
Engstrom, and I. E. Dohns (1994) Psychophysiological stress and EMG
activity of the trapezius muscle, International Journal of Behavioral

Medicine, 1(4), pp. 354–370.
27. R. Merletti (1999) Standards for reporting EMG data, Journal of
Electromyography and Kinesiology, 9(1), pp. 3–4.
28. M. Shamir and C. Weissman (January 2003) Plethysmographic wave-
form variation as an indicator to hypovolemia, Anesthesia and Analgesia
URL http://www.anesthesia-analgesia.org/content/97/2/602.2.full.
29. L. Geddes, M. Voelz, and C. Babbs (January 2007) Pulse tran-
sit time as an indicator of arterial blood pressure, Psychophysi-
ology DOI: 10.1111/j.1469-8986.1981.tb01545.x, URL http://www3.
interscience.wiley.com/journal/119574157/abstract.
30. T. Rusch, R. Sankar, and J. Scharf (1996) Signal processing methods for
pulse oximetry, Computers in Biology and Medicine, 26(2), pp. 143–159.
31. A. Tobola and C. Douniama (June 2007) Evaluation of alternative deriva-
tion areas for plethysmography and pulse oximetry, in Proceedings
of SENSATION Second International Conference, Monitoring Sleep and
Sleepiness with New Sensors Within Medical and Industrial Applications
pp. 1–2.
32. V. Markandey (June 2009) Pulse oximeter implementation on the
tms320vc5505 DSP medical development kit (MDK), Texas
Instruments Application Report SPRAB37 pp. 1–28 URL http://
focus.ti.com/lit/an/sprab37/sprab37.pdf.
33. G. Cotter, Y. Moshkovitz, E. Kaluski, A. Cohen, and H. Miller
(January 2004) Accurate, noninvasive continuous monitoring of
cardiac output by whole-body electrical bioimpedance, Chest URL
http://www.ncbi.nlm.nih.gov/pubmed/15078756.
34. R. Sung, P. Lau, C. Yu, and P. Lam (January 2001) Measurement of body
fat using leg to leg bioimpedance, Archives of Disease in Childhood URL
http://www.fetalneonatal.com/content/85/3/263.full.
35. F. Seoane, J. Ferreira, and J. Sanch’ez (January 2008) An analog front-
end enables electrical impedance spectroscopy system on-chip for
biomedical applications, Physiological Measurement URL http://
iopscience.iop.org/0967-3334/29/6/S23/pdf/0967-3334 29 6 S23.
pdf.
36. U. Maurer, A. Smailagic, D. Siewiorek, and M. Deisher (2006) Ac-
tivity recognition and monitoring using multiple sensors on differ-
ent body positions, in Proceedings of the International Workshop
on Wearable and Implantable Body Sensor Networks (BSN’06) URL
http://www.ece.cmu.edu/˜bfrench/ewatch files/papers/ewatch%2520
papers/01612909.pdf.
References 227
37. S. Patterson, D. Krantz, L. Montgomery, P. Deuster, S. Hedges, and L.

Nebel (2007) Automated physical activity monitoring: validation and
comparison with physiological and self-report measures, Psychophysi-
ology, 30(3), pp. 296–305.
38. D. Hayes, P. Wang, and D. Reynolds (January 1997) Interference with
cardiac pacemakers by cellular telephones, The New England Journal
of Medicine URL http://content.nejm.org/cgi/content/abstract/
336/21/1473.
39. M. J. Mathie, B. G. Celler, N. H. Lovell, and A. C. F. Coster (2004)
Classification of basic daily movements using a triaxial accelerom-
eter, Medical and Biological Engineering and Computing, 42 (5),
679–687, DOI: 10.1007/BF0234755,URL: http://www.springerlink.
com/content/0140-0118/42/5/.
Chapter 8
Ambulatory Recording of Biopotential

Signals: Constraints and Challenges for
Analog Design
Refet Firat Yazicioglu,a Sunyoung Kim,a Tom Torfs,a

Julien Penders,b Buxi Singh Dilpreet,b Inaki Romero,b
and Chris Van Hoofa, b
a IMEC, Kapeldreef 75, Leuven, 3001, Belgium
b IMEC-nl/Holst Centre, High Tech Campus 48, 5656 AE Eindhoven,
The Netherlands
fi[email protected]
The monitoring of biopotential signals such as EEG, ECG, and

EMG is a common procedure in modern clinical practice. The
instruments that can monitor these signals are traditionally realized
by combining precision building blocks with powerful DSP modules.
The growing interest toward the improvement of patients’ quality of
life and the use of biopotential signals in nonmedical applications
such as entertainment, sports, and brain-computer interfaces re-
quires the implementation of miniaturized and wireless biopotential
acquisition systems with ultra-low power dissipation. This has
dramatically changed the way of developing instruments for the
extraction of biopotential signals, placing stringent constraints
on the design of analog front-end circuits that can be used in

www.panstanford.com
230 Ambulatory Recording of Biopotential Signals
ambulatory biopotential monitoring applications. In addition, the

ambulatory monitoring of patients has introduced new challenges
that can jeopardize the signal integrity.
This chapter will focus on the design of analog-integrated
circuits that can be used in ambulatory biopotential monitoring
applications. Several constraints in terms of power dissipation
and signal-to-noise ratio will be addressed. Later, an important
challenge in ambulatory biomedical signal monitoring, namely
motion artifacts, will be introduced and different strategies to tackle
this problem will be explained.
8.1 Introduction: The Need for Portable

Medical Electronics Systems
The need for portable medical electronics stems from the fact that
the current health care routine is very much centralized by heavily
focusing on the hospitalization of patients. This requires that people
adapt to this data-centric approach, which is indeed very efficient
for curative care delivery but, on the other hand, very expensive.
Keeping this in mind, demographical changes indicate that the
continuously aging population is leading to significant rise in chronic
diseases, resulting in ever-increasing health care costs as shown
in Fig. 8.1 [1]. Such increase in costs not only creates a significant
burden on the economical system, but also reduces the efficiency of
health care delivery.
Therefore, there is a strong interest to introduce the parts of
health care cycle into the daily routine of people. It is expected
that this will significantly reduce hospitalization costs and also
help in increasing the efficiency of health care delivery. The initial
interest is to introduce the practice of signal monitoring and analysis
into people’s daily routine through ambulatory and continuous
monitoring systems with miniature size and wearable form factor.
This can significantly reduce the hospitalization requirements
of people and assist health care professionals in data analysis,
significantly reducing the time and the cost of long-term patient
monitoring. In the long run, such monitoring systems will evolve to
Introduction: The Need for Portable Medical Electronics Systems 231
Figure 8.1. Change of total expenditure on health (percentage of gross

domestic product) over years. See also Color Insert.
smarter systems capable of performing reliable diagnostics based

on acquired medical signals, and even perform partial treatment
and/or suppression of the disorder through drug delivery and/or
electrical/visual stimulation systems.
The key for the realization of such a vision is a technology
platform that can enable the collection of data from patients and
communicate these to medical professionals in a reliable manner.
The e-Health project, “The use, in the health sector, of digital data
— transmitted, stored, and retrieved electronically — in support of
health care, both at the local site and at a distance.” as defined by
WHO, can be the platform for the support and the diagnostics of
patients. E-health is claiming to offer the potential to reduce medical
cost, enable personalized health care, deliver remote health services,
and increase the delivery efficiency in real-time. Therefore, the
gathering of fast, reliable, and continuous medical information from
patients lies in the center of the e-Health project further addressing
the necessity for the realization of miniature and smart systems for
medical signal monitoring.
Figure 8.2. Technology vision for future medical monitoring and diagnos-
tics systems [2]. People will be carrying their personalized sensor nodes,
collecting medical information from the patient and their surroundings,
analyze it, and communicate the results to medical professionals with
wireless communication.
This requirement is being addressed by body area networks

(BAN) [2], which consist of smart and miniaturized sensor nodes
collecting information from patients and its environment, process-
ing this information, and wirelessly communicating the results to
medical professionals. The main challenge behind the realization of
these sensor nodes is the fact that the available power budget for
performing these functionalities is strictly limited due to the small
form factor of these sensor nodes. Hence, research focuses on the
realization of power-efficient implementation for each and every
building block of this smart system.
Therefore, this chapter will focus on the analog front-end part
of these miniature sensor nodes and how to implement low-power
analog front-end circuits, which not only extract biomedical signals
with high signal quality, but also provide a method to monitor signal
integrity problems in ambulatory monitoring systems.
The outline of the chapter is as follows: Section 8.2 will
present the basics of biopotential signal acquisition. Section 8.3
will address constrains and challenges in analog circuit design
for the monitoring of biopotential signals in portable/ambulatory
biopotential recording systems. Section 8.4 will describe how
to design power efficient instrumentation amplifiers extracting
Basics of Biopotential Signal Acquisition 233
biopotential signals in a power-efficient manner. Last but not

the least, Section 8.5 will describe signal integrity problems in
ambulatory measurement systems.
8.2 Basics of Biopotential Signal Acquisition
Figure 8.3 shows the characteristics of biopotential signals as

presented in reference [3]. These signals can be grouped in two
categories. The signals in the first group are electrocardiogram
(ECG) and electromyogram (EMG) signals, which are due to
muscular activity, i.e., activity due to the cardiac muscles and the
skeletal muscles, respectively. On the other hand, the signals in
the second group are due to neural activity within the brain, and
the naming convention changes according to the measurement
invasiveness and the focus of the measurement. The action potential
(AP) measurements are invasive measurements and refer to
the signals’ single neurons. Similarly, local field potential (LFP)
measurements are also invasive, but the signal of interest is the
average activity of group of neurons. It should be noted that since
these two measurements are invasive, large biopotential signals
can be picked up. On the other hand, as the measurements move
toward noninvasive methods such as electrocorticogram (ECoG),
Figure 8.3. Frequency and amplitude characteristics of biopotential

signals — most commonly monitored signals in medical practice [3]. See
also Color Insert.
where the signals are measured on the surface of the cortex,

and electroencephalogram (EEG), where the signals are measured
above the skull, the signal amplitudes decrease significantly,
requiring higher performance analog building blocks for signal
acquisition.
The acquisition of biopotential signals requires two main
elements: 1) biopotential electrodes and 2) an instrumentation
amplifier. The purpose of the electrodes is to convert the ionic
current within the body into electronic current so that the
instrumentation amplifier can process this signal for amplification
[3]. The measurement of biopotential signals can be grouped
into three categories according to the electrode configurations
as:
• Type I: Bipolar measurement with third electrode biasing
• Type II: Bipolar measurement with lead biasing
• Type III: Unipolar measurement
Type I measurement configuration is intended for precision mea-
surements, where the common-mode signals need to be rejected
from the measurement. However, this measurement needs the use
of a third electrode for setting the DC potential of the patient. On
the other hand, Type II enables the bipolar measurement of signals
without using the third biasing electrode. This is generally the
preferred measurement method for applications requiring the use
of minimum number of electrodes, where the common-mode signals
need to be rejected. The third, and the last measurement method, is
the unipolar measurement method. This method is generally used,
where the signal amplitudes are sufficiently large, or the common-
mode signals are sufficient low, such that it is not required, by the
application, to reject the common-mode signals.
Similar to the measurement configuration, the design of an
instrumentation amplifier also depends on the application. For
applications where the signal levels are very small, together
with Type I measurement a high-performance instrumentation
amplifier is required. On the other hand, an instrumentation
amplifier with lower performance specifications may be sufficient
for measurements with larger signals enabling the implementation
of a lower power instrumentation amplifier.
Constrains and Challanges 235
Figure 8.4. Different measurement and biasing methods for biopotential

signal acquisition. See also Color Insert.
8.3 Constrains and Challanges
The previous section has given a brief introduction to the measure-

ment of biopotential signals. Of course, such an ideal model indicates
that the implementation of an analog front-end can be a straightfor-
ward process and that simple amplifier architectures can be used
to achieve very low-power analog front-end circuits. However, this
is unfortunately not the case for biopotential measurements. There
are significant aggressors and error sources that create challenges
and constrains for instrumentation amplifier design. This section
will summarize these constrains and challenges, which will enable
the analog designer to select the best amplifier topology to be used
in different applications of biopotential measurements.
The nonidealities in biopotential measurements can be due to:
i. Electrostatic Interference
ii. Biopotential Electrodes
iii. Instrumentation Amplifier
These three main groups of nonidealities are shown in Fig. 8.5

for a Type I measurement. The first nonideality comes from the
interference to human body appearing due to the electrostatic
coupling to surrounding mains, which results in a displacement
current flowing through the human body [4] leading to the
Figure 8.5. Measurement of biopotential signals (Type I measurement)

introducing the nonidealities from different sources. See also Color Insert.
generation of an AC potential over the body. This potential appears

as a common-mode signal to the instrumentation amplifier and
requires sufficiently large common-mode rejection ratio (CMRR) so
that we can reject the large common-mode signals and extract the
weak biopotential signals.
The second nonideality is due to the electrodes. An electrode
sitting on a skin, or on a tissue, results in complex impedance and
a polarization voltage between the signal source and the input of the
instrumentation amplifier [3]. The latter leads to a net DC input to
the instrumentation amplifier, which can be rejected by introducing
AC coupling prior to amplification. On the other hand, the prior,
combined with the finite input impedance of the instrumentation
amplifier, leads to the conversion of the common-mode signals into
differential mode according to:
VDM Z ELEC
= (8.1)
VCM Z IN
where VDM is the differential mode signal at the input of the
amplifier, VCM is the common-mode signal on the human body,
ZELEC is the mismatch of the impedances between two recording
electrodes, and ZIN is the input impedance of the instrumentation
Constrains and Challanges 237
amplifier. Therefore, it can be seen that even if we can implement

an instrumentation amplifier with infinitely large CMRR, the finite
input impedance and the mismatch of electrode impedance lead to
the conversion of the common-mode signal into differential mode.
This means that the affective CMRR is significantly reduced by the
finite input impedance of the amplifier.
Last but not the least, the nonidealities from the instrumentation
amplifier can affect the accuracy of the biopotential recordings.
Two major nonidealities with significant impact on the quality
of recordings are the noise and the finite input impedance of
the instrumentation amplifier. The latter, unless sufficiently large,
may load the electrodes, as well as, leading to the conversion of
common-mode signals into differential mode. The prior, on the other
hand, significantly reduces the signal-to-noise ratio of biopotential
recordings due to the fact that biopotential signals have very low
frequency behavior [5].
As a result of this discussion, we can divide the main performance
specifications of instrumentation amplifiers into four groups:
i. DC Headroom
ii. Common-Mode Rejection Ratio
iii. Input Impedance
iv. Noise Performance
It should be noted that the quantitative specifications for these
requirements will be defined by the application, and the choice
of the instrumentation amplifier should be able to meet these
specifications. A very good example on how application defines the
instrumentation amplifier specifications can be seen from different
brain activity measurement methods as shown in Fig. 8.6.
During EEG measurements, the electrodes are placed above the
skull. This leads to biopotential signals with very weak amplitudes,
which make them very sensitive to various aggressors. On the other
hand, due to the less strict material choice, the electrodes are more
stable leading to more relaxed DC headroom requirements. This
means that instrumentation amplifiers addressing EEG applications
should realize very high performance, i.e., high CMRR, very high
input impedance, and very low noise, whereas, the DC headroom
specifications are not very strict.
Figure 8.6. Different measurement methods of brain activity showing

how the application defines the instrumentation amplifier specifications.
On the other hand, if we have a look at more invasive measure-

ments, for instance ECoG measurements, the signal amplitudes are
larger, relaxing noise, CMRR, and input impedance specifications.
This means that circuits targeting these applications may be more
aggressive on low-power consumption rather than the noninvasive
measurements of biopotential signals.
Finally, if we have a look at more invasive measurements such
as AP measurements, the signal amplitudes even further increase,
which means that the noise and CMRR requirements can be more
relaxed. On the other hand, the size of the electrodes is significantly
reduced, which leads to a very large electrode impedance. In
addition to that, very strict biocompatibility requirements limit
the choice of electrode material. Hence, even larger polarization
voltages can be seen while recording biopotential signals from such
electrodes. Hence, instrumentation amplifiers implementing very
large input impedance and incorporating large DC headroom are
required.
Therefore, the conclusion is that the design of instrumentation
amplifiers for different biopotential recording applications targets
different specifications. The next section will review these different
instrumentation amplifier architectures and why they suit very well
for specific applications.
Design of Instrumentation Amplifiers for Biopotential Recordings 239
8.4 Design of Instrumentation Amplifiers for

Biopotential Recordings
The most common instrumentation amplifiers that are being used

for biopotential recordings can be grouped into two main categories:
i. Uncompensated Instrumentation Amplifiers

ii. Compensated Instrumentation Amplifiers
The name compensation represents the use of circuit techniques

that reduce the effects of flicker noise in CMOS transistors
and mismatch related errors. Therefore, the amplifiers utilizing
compensation are being used in applications requiring very high
performance instrumentation amplifiers. On the other hand, the
uncompensated amplifiers focus on aggressively improving the
power efficiency of the instrumentation amplifiers, which is one
of the most important criterions in implantable applications.
Therefore, this section will describe the most commonly used
instrumentation amplifiers for the extraction of biopotential signals.
Their key advantages and disadvantages will be reviewed.
8.4.1 Uncompensated Instrumentation Amplifiers

There are two main types of instrumentation amplifiers that do
not use any compensation for the amplifier nonidealities. These
instrumentation amplifiers differ in the sense that the first type
uses the ratio of resistors to define the gain of the instrumentation
amplifier, or namely resistive instrumentation amplifiers, whereas
the second type uses the ratio of capacitors, or so called capacitive
instrumentation amplifiers.
The typical architecture of the resistive instrumentation ampli-
fiers is shown in Fig. 8.7 [6]. The instrumentation amplifier consists
of an input transconductance stage and an output transimpedance
stage. The input stage converts the differential input voltage into
a current over the resistor R1 . This current is converted into a
voltage at the transimpedance stage. Hence, the voltage gain of the
architecture is defined by:
R2
AV = (8.2)
R1
Figure 8.7. Simplified architecture of the resistive instrumentation

amplifier topology [6] that does not make use of any compensation
technique for improving CMRR and reducing noise.
If we analyze the important characteristics of this instrumentation

amplifier topology, the critical performance specifications can be
summarized as:
i. DC Headroom: The lack of AC coupling significantly reduces

the DC headroom of this instrumentation amplifier. Additional
circuitry is required for the introduction of AC coupling.
ii. CMRR: It is defined by components mismatches. Especially, the
gain mismatch of the input buffers will lead to the conversion
of the common-mode voltage into differential current over the
input resistor R1 . As a result, the CMRR of this amplifier topology
will be in the moderate-low range.
iii. Input Impedance: One of the key advantages of this architecture
is its input impedance, which is solely defined by the parasitic
capacitances at the inputs of the transconductance stage. This
simply means that the input impedance of this architecture is the
maximum possible value for a given instrumentation amplifier
topology.
iv. Noise Performance: The input-referred noise of this instrumen-
tation amplifier architecture is defined by:
v 2R2
2
vIA = 2 × vBUF
2
+ v 2R1 + (8.3)
A 2v
This shows that the noise of the instrumentation amplifier is
defined by the noise of the input buffers together with the
resistors defining the transconductance of the input stage. It
Figure 8.8. Simplified architecture of the capacitive instrumentation

amplifier topology [8].
should be noted that the noise of the buffers will be dominated

by flicker noise, significantly reducing the signal-to-noise ratio
of the measurements at low frequencies, i.e., the range of
biopotential signals.
As an alternative to the resistive instrumentation amplifier, the

architecture of the capacitive instrumentation amplifier is shown
in Fig. 8.8 [7, 8]. It is similar to a differential resistive feedback
amplifier, but resistors are replaced with capacitors, hence the gain
is defined as:
c1
AV = (8.4)
c2
In such a configuration, the DC voltages at high impedance nodes are
defined by the resistors. Note that these resistors are implemented
using transistors operating in weak inversion, implementing very
large values on-chip by using a very small area [8]. The critical
performance specifications can be summarized as:
i. DC Headroom: This architecture is inherently AC coupled.

Supply level DC voltages can be rejected by this instrumentation
amplifier.
ii. CMRR: It is defined by the components mismatches. Hence, the
expected CMRR is also in the moderate-low range.
iii. Input Impedance: The input impedance is defined by the input

capacitances C1 and can be given as:
1
Z IN = (8.5)
j wC 1
Hence, the input impedance of this architecture is also very
large considering the fact that biopotential signals have very low
frequency behavior.
iv. Noise Performance: The total input referred noise of the
instrumentation amplifier is mainly dominated by the noise
of the core operational transconductance amplifier (OTA) as
represented in Eq. 8.6. The only nonideal effect that elevates
the instrumentation amplifier noise above the core OTA noise
level is the presence of the parasitic capacitance C p . The value
of this parasitic capacitance appears at the numerator of the
noise equation. Hence, the capacitors C1 and C2 should be sized
properly to minimize the effect of C p . In addition, the core OTA
also suffers from flicker noise. This significantly increases the
noise of the OTA for low-frequency applications.

C1 + C2 + Cp 2
2
vIA = × vOTA
2
(8.6)
c1
As a conclusion, if we compare the two uncompensated instrumen-
tation amplifier topologies, both of the architectures suffer from
large 1/f noise and limited CMRR, but the capacitive one presents
a clear advantage of having maximal DC headroom, which is an
important requirement for the extraction of AP and LFP signals.
On the other hand, neither of the instrumentation amplifiers
has sufficient performance for the recording of weaker biopotential
signals, where the 1/f noise and limited CMRR are important error
sources. Hence, compensated instrumentation amplifiers are used
for applications requiring very low noise levels.
8.4.2 Compensated Instrumentation Amplifiers

A well-known technique that reduces flicker noise and improves
the CMRR of amplifier is called chopper modulation (Fig. 8.9) [9].
This technique works on the principle that the input signal can
be modulated to a frequency location, where flicker noise and
Figure 8.9. Operating principle of chopper modulation. Signal aggressors

are modulated to out of the signal band.
other DC errors are not present, passed through the amplifier

and demodulated back. In such an implementation, square-wave
modulation is used, where the modulators are implemented using
cross-coupled switches toggling at alternating clock phases. These
modulators only modulate the differential input signal, but they are
transparent to common-mode signals. Therefore, not only the flicker
noise of CMOS transistors can be pushed out of the signal band, but
also the CMRR can be increased.
The only problem that prevents the use of chopper-modulated
instrumentation amplifiers for biopotential measurements is the
fact that this technique inherently leads to the implementation
of DC-coupled instrumentation amplifiers. A simple solution to
this problem can be the use of off-chip passives implementing
high-pass filter characteristics. However, this can lead to the use
of large number of external components, especially important
for multichannel EEG recording applications with high channel
count.
As an alternative to off-chip passives, a recently proposed tech-
nique utilizes a DC servo to implement AC coupling (Fig. 8.10) [10].
In this architecture, the input of the DC servo input is connected
to the output of the amplifier after the demodulator. It checks the
Figure 8.10. Implementation of an AC-coupled chopper-modulated in-

strumentation amplifier. The DC servo loop regulates the input of the core
amplifier to reject the differential input impedance to chopper-modulated
instrumentation amplifiers [10].
DC level of the output signal and regulates the input voltage of the
amplifier to reject the input DC voltage.
Similar to uncompensated instrumentation amplifiers, there
are two different instrumentation amplifier architectures, which
use chopper modulation. The first architecture is called resistive
compensated instrumentation amplifiers. Figure 8.11 shows the
actual implementation of an AC-coupled chopper modulated in-
strumentation amplifier relying on resistors as the gain elements
[10, 11].
Figure 8.11. The complete architecture of a compensated instrumenta-

tion amplifier using a DC servo loop for implementing high-pass filtering
characteristics [10, 11]. See also Color Insert.
The core of the instrumentation amplifier is an uncompensated

resistive instrumentation amplifier, whose simplified architecture is
shown in Fig. 8.7. The modulators and demodulators are located
at the input and output of the resistive instrumentation amplifier.
The input of the DC servo is sensing the DC level of the output and
subtracts it from the input of the amplifier. This subtraction takes
places at the terminals of the resistor R1 , which is actually based
on the principle shown in the inset figure. This principle simply
indicates that applying a voltage over a resistor leads to a current IR ,
flowing through R. However, the source of this current will depend
on the presence of other current sources forcing current trough R.
For instance, if a current source forces a current IF through resistor
R, the voltage source only needs to supply IR -IF through the resistor
R. This is the exact operating principle of DC servo loop presented
in Fig. 8.11. The input transconductance only supplies the rest of the
current due to IGM . DC servo ensures that IGM equals to the input DC
voltage divided by R, so that our transimpedance will not receive this
current to convert it into voltage.
The critical performance specifications can be summarized as
follows:
i. DC Headroom: The DC servo loop introduces AC coupling to
the chopper-modulated instrumentation amplifier. However, the
maximum current that can be supplied from the DC servo limits
the DC headroom to:
DC Headroom = IGM,max × R1 (8.7)
ii. CMRR: Chopper modulation eliminates the errors related to
component mismatches. Therefore, the CMRR of instrumenta-
tion amplifiers can reach more than 120 dB.
iii. Input Impedance: The input impedance of this architecture
is slightly reduced compared with the input impedance of an
uncompensated resistive instrumentation amplifier due to the
up modulation of the input signal.
iv. Noise Performance: The main advantage of chopper-modulated
architecture is the removal of flicker noise. Hence, this architec-
ture can implement very low noise at low frequencies.
The second compensated instrumentation amplifier architecture
uses capacitive instrumentation amplifier as the core and introduces
Figure 8.12. The complete architecture of a compensated capacitive

instrumentation amplifier using a DC servo loop for implementing high-pass
filtering characteristics [12].
chopper modulation. Figure 8.12 shows the architecture of the

instrumentation amplifier [12]. Similar to the resistive instrumen-
tation amplifier, a DC servo loop is used for the introduction of AC
coupling. Different from the resistive instrumentation amplifier, the
DC servo is implemented through capacitive coupling. Again, the
current through the input capacitor C1 can be seen as the addition
of DC and AC input signals. In order to prevent its amplification,
the current due to the DC input is subtracted through the DC servo
path from the amplification signal path leading to AC coupling
characteristics.
The basic specifications of this instrumentation amplifier can be
summarized as:
i. DC Headroom: The DC servo introduces AC coupling. The DC
headroom is again defined by the limitations of the DC servo as:
C DC VDD
DC Headroom = × (8.8)
C1 2
ii. CMRR: Chopper modulation increases the CMRR of the architec-
ture.
iii. Input Impedance: One of the main drawbacks of this architec-
ture is the input impedance. The modulation of the input signal
reduces the equivalent input impedance. This is due to the fact

that input impedance is defined by the capacitance C1 . Therefore,
the input impedance of the architecture can be given as:
1
Z IN = (8.9)
j (win c1 ± n × c1 )
where wi n is the frequency of the input signal and n is defined by
the harmonics of the square-wave modulation as:
4 ∝ 1 4 ∝
n= × k × wclk = wclk (8.10)
π k−1,3... k π k=1,3...
The first scaling factor within the summation formula comes

from the odd harmonics of a square wave, and the second
scaling factor comes from the increasing frequency of the odd
harmonics. These two factors cancel each other leading to a pure
summation of the clock frequency as the scaling factor of C1 . Note
that the effective range of C1 is defined by the resistance of the
chopping switches and the capacitance of the instrumentation
amplifier.
iv. Noise Performance: Chopper modulation eliminates the flicker
noise of the core amplifier. This means that the noise equation
of this architecture can be written as Eq. 8.6, with the key
improvement that the core amplifier noise is only defined by the
thermal noise.
8.4.3 Summary and Comparison of Instrumentation

Amplifier Topologies
In this section, we have presented four different instrumentation
amplifier architectures, which are commonly used in biopotential
measurement applications. These architectures are categorized
according to their gain elements as resistive or capacitive and
according to the use of compensation techniques.
It should be noted that the correct amplifier choice for a
given application should combine the lowest possible power with
sufficiently good performance. In addition to the discussion on
amplifier specifications, Fig. 8.13 compares the power efficiencies
of the instrumentation amplifiers presented in this section. It
is interesting to see that the best efficiency is achieved by
Figure 8.13. Comparison of different instrumentation amplifiers in terms

of their power efficiency. Y-axis indicates the current consumption of the
instrumentation amplifier and x-axis indicates the noise density of the
instrumentation amplifier. Dashed lines indicate the constant contours of
noise efficiency factor (NEF) lines [8]. The diameter of a circle indicates the
total integrated noise of the instrumentation amplifier. See also Color Insert.
capacitive uncompensated amplifiers. This is due to the fact that this

instrumentation amplifier topology uses only a single active circuit,
which is the core OTA. Unfortunately, the implementations using
the capacitive uncompensated instrumentation amplifier topology
achieve relatively large total integrated noise. On the other hand,
due to the presence of the servo loop, compensated amplifiers
have slightly worse efficiency, but they have very much lower
total integrated noise compared with capacitive uncompensated
instrumentation amplifiers. Finally, uncompensated resistive and
industry standard three opamp [13–15] instrumentation amplifiers
have simply much worse power efficiencies preventing their use in
ambulatory biopotential measurement applications.
It can be concluded from this discussion that capacitive uncom-
pensated instrumentation amplifiers can be the preferred choice for
applications that can live with relatively large circuit noise level.
Table 8.1. Comparison of different instrumentation amplifier architec-

tures
Figure 8.7 Figure 8.8 Figure 8.11 Figure 8.12

Resistive Capacitive Resistive Capacitive
uncompensated uncompensated compensated compensated
DC Headroom Limited Maximum Limited Limited
CMRR Low to Moderate Low to Moderate High High
Input Impedances High High High Moderate
Noise Performance Large 1/f noise Large 1/f noise Low Noise Low Noise
Last but not the least, Table 8.1 summarizes the properties of
instrumentation amplifiers presented in this section and presents
the comparison of their key characteristics. Not only due to their
bad power efficiency but also due to their lower performance
specifications, resistive uncompensated instrumentation amplifiers
are not the preferred choice for portable biopotential monitoring
systems. On other hand, the rest of other three instrumentation
amplifier topologies are very attractive in terms of their power
efficiencies as well as their performance specifications.
In addition to having the best power efficiency, uncompensated
capacitive amplifiers have very large input impedance and maximal
DC headroom. If we refer to Fig. 8.6, these characteristics match very
well with the requirements of AP and LFP measurements.
The compensated version of this amplifier, i.e., compensated
capacitive instrumentation amplifiers, additionally removes flicker
noise and improves the CMRR of the instrumentation amplifier. On
the other hand, the equivalent input impedance is reduced due to the
capacitive nature of the input impedance. Hence, this architecture
is generally used for ECoG applications where lower noise and
higher CMRR performance are expected from the instrumentation
amplifier, but lower input impedance values can be allowed by the
application due to the use of invasive electrodes with large surface
area.
Finally, resistive compensated instrumentation amplifiers have
very high input impedance, low noise, and high CMRR. Therefore,
their characteristics match very well with noninvasive biopotential
measurements, where a high performance instrumentation ampli-
fier is required.
8.5 Signal Integrity Problems in Ambulatory

Measurements
In addition to the noise sources such as circuit noise, interference to

measurement setup, and irrelevant biological signals that correlate
the biopotential measurements, another important noise source
in ambulatory measurements is motion artifact signals. These
potentials occur in the electrode cables, in the skin, and at the
electrode/electrolyte interface. While artifacts coming from cables
can be reduced by appropriate electrode cables, artifacts from the
skin and electrode/electrolyte interface are difficult to reduce by
design.
The main source of the motion artifacts, as the name implies,
is the motion present in ambulatory measurements. During ambu-
latory measurements, the subject/patient is continuously moving.
The relative movement of electrode and tissue leads to significant
voltage fluctuations called motion artifact signal. It is assumed that
the source of the motion artifact signal is the fluctuations in the
polarization generated at the electrode-tissue interface and within
the tissue [16].
These motion artifact signals may have large amplitudes, and
also they may have similar frequency spectrum as biopotential
signals. Hence, amplitude-based or frequency-based algorithms, or
the algorithms combining both methods, can significantly suffer
from these motion artifacts reducing the reliability of biopotential
measurements in ambulatory conditions. Therefore, there is a strong
need for the methods and circuits that can enable the differentiation
between the medically irrelevant motion artifact signals and the
biopotential signals.
8.5.1 Methods Focusing on Motion Artifact Reduction in

Biopotential Recordings
There have been several proposals in the literature trying to tackle
the motion artifact problem in ambulatory biopotential recordings.
Time synchronous averaging or median is one of them [16]. This
is a classic technique widely used for cleaning of signals like ECG.
However, these methods require the use of several ECG beats in
Signal Integrity Problems in Ambulatory Measurements 251
order to obtain one single clean beat, and, therefore, beat-to-beat

variations are removed.
Another technique is the use of fixed filters [17]. This technique
could give the information about beat-to-beat variations. However, it
does not adapt to the time-variant properties of noise and requires
some previous knowledge of the noise. Furthermore, the design of
linear phase band-pass filters is critical in preventing distortion of
the ECG. Wavelets have also interesting filtering properties [18].
Wavelet de-noising based on a decomposition of the signal using
wavelets that are highly adapted to the ECG gives a higher resolution
in lower frequencies and, therefore, could lead to better removal
of noise. However, wavelet processing is computationally heavy for
ultra-low power ambulatory biopotential recording systems.
An alternative technique called adaptive filtering was introduced
in the 60’s as a new tool for ECG de-noising [19]. Since then it
has become more and more popular, and several approaches have
been implemented. For noise removal, the input of an additional
reference signal is required, which should be highly correlated with
the noise. Several reference signals have been investigated with
successful results such as motion measured by accelerometers [20],
skin-electrode impedance [21], and skin stretch measured by optical
sensors [22].
Finally, independent component analysis (ICA), which contains
beat-to-beat information and requires no reference signal, aims to
reconstruct the ECG and motion artifact signals [23]. This method
is, however, also computationally expensive and requires more than
one ECG lead. Table 8.2 gives a summary of existing methods that are
trying to reduce the effect of motion artifact signals on biopotential
recording applications.
Among these methods, adaptive filtering is gaining more and
more attention since it focuses on removing motion artifacts
from biopotential signals without affecting ECG signal parameters.
Also another advantage of adaptive filtering is the fact that it is
computationally less heavy compared with ICA analysis.
8.5.2 Readout Circuits for Adaptive Filtering

Compared with other methods, adaptive filtering presents several
advantages. First of all it can be implemented using a single-lead
Table 8.2. Summary of methods found in the literature for artifact

reduction in biopotential recordings (ECG)
Advantages Disadvantages
Time Synchronous Low computational Nonstationary information removed,
Averaging complexity require fiducial point demarcation.
Filter Banks Beat-to-beat information No adaptation to time-varying noise
Adaptive Filtering Adapts to time-varying noise, Choice of reference signal is critical,
no previous knowledge of implying complexity in reliable noise
noise required measurement
Independent No reference signal required Multiple leads required.
Component Computationally heavy.
Analysis
biopotential measurement (unlike ICA analysis), it does not require

the history of the signal (unlike time synchronous averaging), and it
can successfully eliminate noise from biopotential recordings even
though the noise source looks exactly similar to the biopotential
signal of interest (unlike filter banks). Therefore, there has been a
specific attention on adaptive filtering over the last years. However,
a critical requirement for adaptive filters is the extraction of a signal
from the biopotential measurement system that is correlated with
the motion artifact but uncorrelated with ECG signals.
A possible approach to tackle this problem is to collect data
from other sensors that have maximum correlation with motion
artifact signals and minimal correlation with biopotential signals.
Several approaches has been tried in the literature, such as using
accelerometers to sense the subjects’ movement, electrode bend
sensors to detect the motion of electrodes, skin stretch sensors to
sense the movement of skin, and electrode impedance measurement
circuits to correlate electrode impedance measurement with the
motion artifact [22, 24–27].
Among these possibilities, the use of electrode-tissue impedance
as the reference signal input to adaptive filters is gaining even more
attention, since it does not require the use of an additional sensor
and also since the characteristics of the electrode-tissue interface is
actually responsible for motion artifact signals. Therefore, different
circuit topologies that focus on the simultaneous measurement
of biopotential signals with the electrode-tissue impedance have
Signal Integrity Problems in Ambulatory Measurements 253
Figure 8.14. A circuit topology proposed by [26] to monitor biopotential

signals simultaneously with the electrode-tissue impedance.
been proposed. Figure 8.14 shows the most common circuit

topology for the measurement of biopotential signals with the
electrode-tissue impedance [26]. A sinusoidal stimulation current
is injected differentially to the measurement electrodes. The AC
current frequency is selected such that it is out of the frequency
band of biopotential signals. Hence, the low-pass filtering of the
instrumentation amplifier output can extract biopotential signals,
whereas the demodulation of the output with the AC current
frequency can extract the electrode-tissue interface impedance.
Although this technique can be very effective for extracting the
required information, it requires the use of sinusoidal current
generation blocks, which can be very power consuming in CMOS
technology.
An alternative measurement method focuses on using the bias
electrode for extracting the electrode-tissue impedance simultane-
ously with biopotential signals [27], Fig. 8.15. Instead of using a
fixed bias voltage for the human body, an AC bias voltage is used.
Under ideal conditions, this bias voltage appears as a common-
mode signal to the instrumentation amplifier. However, as stated
in Eq. 8.1, the finite input impedance of the instrumentation
amplifier combined with the mismatch of the electrode impedances
converts the common-mode signal into differential mode. Hence, the
electrode-tissue impedance can be extracted using the formula:
|δ Z ELEC |
VIMP = × VBIAS (8.11)
Z IN
Figure 8.15. Another circuit topology proposed by [27] to monitor

biopotential signals simultaneously with the electrode-tissue impedance.
The method modulates the bias voltage of the patient to extract the
electrode-tissue impedance.
However, it can be seen from the measurement method that this

principle relies on the finite input impedance of the instrumentation
amplifier, which simply contradicts with the requirements from
biopotential measurement systems, i.e., as large as possible input
impedance.
Last but not the least, a recent circuit topology that focuses on
the measurement of electrode-tissue impedance with low-power
dissipation has been proposed by [28], Fig. 8.16. This circuit uses
square wave current stimulation rather than sinusoidal current
stimulation. This way an integrated circuit with much lower power
Figure 8.16. The implementation of electrode-tissue impedance mea-

surement circuit operating simultaneously with the biopotential measure-
ment circuit [28]. Instead of sinusoidal current sources and analog modula-
tors/demodulators, chopper-modulation technique has been employed. See
also Color Insert.
Conclusion 255
dissipation can be realized. This is due to the fact that the develop-
ment of sinusoidal currents and their demodulation only relies on
the use of chopper modulators leading to a significantly lower power
dissipation compared with the methods using sinusoidal currents
and analog modulators/demodulators.
8.6 Conclusion
As the cost of health care is increasing, the remote monitoring of

people is gaining more and more attention. This leads to a strong
urge for portable medical signal monitoring systems that can be
integrated into people’s daily lifestyle. It is expected from these
systems that they can collect data, analyze these, and communicate
the results to medical professionals.
One of the most important building blocks for this system is
the realization of the analog front-end circuit that can consume
minimal power and can still meet the challenging requirements
of medical monitoring systems. In this chapter, we have reviewed
the most commonly used instrumentation amplifier topologies
for different medical signal acquisition. It can be seen from this
review that there is not a single instrumentation amplifier topol-
ogy ideal for different biomedical signal acquisition applications;
rather, it is more important to consider the requirements of
the application and select the instrumentation amplifier topology
accordingly.
In addition to instrumentation amplifier design, an important
challenge in ambulatory medical signal measurement systems is
motion artifacts. These signals, though do not have any medical
relevance, appear together with biomedical signals and reduce the
reliability of signal analysis software. Several methods that address
this problem have been presented in this chapter. The advantages
and disadvantages of these different techniques have been reviewed.
In particular, adaptive filtering approach has been favored due to
its advantages over other methods. Last but not the least, the
implementation of analog circuits that can generate the reference
signal of an adaptive filter is described, and low-power techniques
for such circuits have been presented.
References
1. OECD health data 2009.

2. B. Gyselinckx, R. Vullers, C. Hoof, J. Ryckaert, R. Yazicioglu, P. Fiorini, and
V. Leonov (October 2006) Human++: Emerging Technology for Body
Area Networks, in 2006 IFIP International Conference on Very Large
Scale Integration, pp. 175–180.
3. J. G. Webster (1992) Medical Instrumentation: Application and Design,
2nd edn, Houghton Mifflin, Boston (Mass.).
4. J. C. Huhta, and J. G. Webster (March 1973) 60-Hz Interference
in Electrocardiography, IEEE Transactions on Biomedical Engineering,
20(2), pp. 91–101.
5. B. Razavi (2001) Design of Analog CMOS Integrated Circuits, McGraw-
Hill.
6. C. Toumazou, et al. (1989) Novel Current-Mode Instrumentation
Amplifier, IEEE Electronics Letters, 25(3), pp. 228–230.
7. R. Olsson, A. Gulari, and K. Wise (20–22 March 2003) A Fully Integrated
Band-Pass Amplifier for Extracellular Neural Recording, in Proceedings
of the 1st International IEEE EMBS Conference on Neural Engineering,
pp. 165–168.
8. R. Harrison, and C. Charles (June 2003) A Low-Power Low-Noise CMOS
Amplifier for Neural Recording Applications, IEEE Journal of Solid-State
Circuits, 38(6), pp. 958–965.
9. C. Enz, E. Vittoz, and F. Krummenacher (June 1987) A CMOS Chop-
per Amplifier, IEEE Journal of Solid-State Circuits, 22(3), pp. 335–
342.
10. R. F. Yazicioglu, P. Merken, R. Puers, and C. Van Hoof (6–9 February 2006)
√
A 60 μW 60 nV/ Hz Readout Front-End for Portable Biopotential
Acquisition Systems, IEEE International Solid-State Circuits Conference
Digest of Technical Papers, pp. 109–118.
11. R. F. Yazicioglu, P. Merken, R. Puers, and C. Van Hoof (3–7 February 2008)
A 200 μW Eight-Channel Acquisition ASIC for Ambulatory EEG Systems,
IEEE International Solid-State Circuits Conference Digest of Technical
Papers, pp. 164–165.
12. T. Denison, K. Consoer, A. Kelly, A. Hachenburg, and W. Santa
√
(2007) 2.2 μW 97nV/ Hz, Chopper Stabilized Instrumentation Am-
plifier for EEG Detection in Chronic Implants, IEEE International
Solid-State Circuits Conference Digest of Technical Papers, pp. 162–
163.
References 257
13. M. A. Smither, D. R. Pugh, and L. M. Woolard (September 1977) CMRR

Analysis of the 3-op-amp Instrumentation Amplifier, IET Electronics
Letters, 13, pp. 594.
14. A. C. Metting van Rijn, A. Peper, and C. A. Grimbergen (1990) High-
Quality Recording of Bioelectric Events; Part 1: Interference Reduction,
Theory and Practice, Medical and Biological Engineering and Computing,
28, pp. 389–397.
15. M. Burke, and D. Gleeson (February 2000) A Micropower Dry-Electrode
ECG Preamplifier, IEEE Transactions on Biomedical Engineering, 47(2),
pp. 155–162.
16. H. Tam, and J. G. Webster (1977) Minimizing Electrode Motion Artifact
by Skin Abrasion, IEEE Journal on Biomedical Engineering, 24, pp. 134–
139.
17. V. Afonso, W. Tompkins, T. Nguyen, K. Michler, and S. Luo (1996)
Comparing Stress ECG Enhancement Algorithms, IEEE Engineering in
Medicine and Biology Magazine, 15, pp. 37–44.
18. P. Augustyniak (2007) Separating Cardiac And Muscular ECG Com-
ponents Using Adaptive Modeling in Time-Frequency Domain, in
Proceedings of the WACBE World Congress on Bioengineering.
19. B. Widrow, J. Glover, J. R. J. McCool, J. Kaunitz, C. Williams, R. Hearn,
J. Zeidler, J. Eugene Dong, and R. Goodlin (1975) Adaptive Noise
Cancelling: Principles and Applications, IEEE Journal Proceedings, 63,
pp. 1692–1716.
20. D. Tong, K. Bartels, and K. Honeyager (2002) Adaptive Reduction of
Motion Artifact in the Electrocardiogram, in Proceedings of the 2nd Joint
EMBS/BMES Conference, 2, pp. 1403–1404.
21. P. H. Devlin, R. G. Mark, and J. W. Ketchum (1984) Detecting Electrode
Motion Noise in ECG Signals by Monitoring Electrode Impedance, in
Proceedings of Computers in Cardiology pp. 51–56, Los Angeles.
22. P. Hamilton, M. Curley, R. Aimi, and C. Sae-Hau (2000) Comparison
of Methods for Adaptive Removal of Motion Artifact, Computers in
Cardiology, 27, pp. 383–386.
23. M. Milanesi, N. Martini, N. Vanello, V. Positano, M. Santarelli, R. Paradiso,
D. De Rossi, and L. Landini (2006) Multichannel Techniques for Motion
Artifacts Removal From Electrocardiographic Signals, in Proceedings
of the 28th Annual International Conference of the IEEE Engineering in
Medicine and Biology Society, pp. 3391–3394.
24. P. T. Gibbs, L. B. Wood, and H. H. Asada (2005) Active Motion Artifact
Cancellation for Wearable Health Monitoring Sensors Using Collocated
MEMS Accelerometers, in Proceedings of the Society of Photo-Optical

Instrumentation Engineers.
25. P. S. Hamilton, and M. G. Curley (1997) Adaptive Removal of Motion
Artifact, in Proceedings of the 19th Annual International Conference of the
IEEE Engineering in Medicine and Biology Society, pp. 297–299.
26. J. Ottenbacher, M. Kirst, L. Jatobá, U. Großmann, and W. Stork (2007) An
Approach to Reliable Motion Artifact Detection for Mobile Long-Term
ECG Monitoring Systems Using Dry Electrodes, in Proceedings of the IV
Latin American Congress on Biomedical Engineering.
27. T. Degen, and T. Loeliger (August 2007) An Improved Method to
Continuously Monitor the Electrode-Skin Impedance During Bioelectric
Measurements, in proceedings of 29th Annual International Conference
of the IEEE Engineering in Medicine and Biology Society, pp. 6294–6297.
28. R. F. Yazicioglu, P. Merken, R. Puers, and C. Van Hoof (February 2010) A
30 μW Analog Signal Processor ASIC for Biomedical Signal Monitoring,
in Proceedings of IEEE International Solid-State Circuits Conference
Digest of Techical Papers, pp. 124–125.
Chapter 9
Network and Medium Access Control

Protocol Design for Wireless Body
Area Networks
Jamil Y. Khan
School of Electrical Engineering and Computer Science,
The University of Newcastle, Callaghan,
NSW, 2308 Australia
[email protected]
Wireless body area network (WBAN) is considered a special-

purpose sensor network designed to collect and transmit physi-
ological data within a short distance. Physiological sensors could
be either implanted or placed external to a human body. A
WBAN will generally operate either as a stand-alone personal area
network (PAN) or as an interconnected network in telemedicine
applications. The network configuration of a WBAN will depend
on medical applications. In a WBAN design, the MAC protocol
plays a very important role, which determines the QoS (Quality of
Service) performance, energy efficiency, and reliability of a network.
This chapter introduces the WBAN architecture, MAC protocols,
and network design techniques for medical applications. The
chapter initially introduces different network topologies followed
by fundamental description of MAC protocols and its classifications,

www.panstanford.com
260 Network and Medium Access Control Protocol Design for WBAN
and then reviews several WBAN-specific MAC protocols. Following

the MAC protocol discussions, the chapter introduces energy
management techniques and patient monitoring design techniques.
Some simulation results of a WBAN are presented to analyze the
performance of a CSMA/CA MAC-based WBAN networks.
9.1 Introduction
The WBAN is becoming a key component of future telemedicine-

based health care systems [1, 11]. Applications of WBAN in future
can bring a number of key benefits to the health care system.
Major benefits include collection of medical data of patients from
their home or workplaces, collection of transient medical data in
a cost-effective manner, patient’s mobility, reduction of health care
cost, improved patient care based on the availability of timely
information, etc. On some occasions, collection of transient data
for different diagnostic purposes could be difficult. In many cases,
patients need to be kept under observation in hospitals or clinics
for an extended period to detect certain medical conditions, which
is quite resource intensive. Applications of WBAN could reduce
the cost of patient observation and treatments because health care
professionals can access those data from anywhere anytime.
A WBAN is a special-purpose wireless sensor network with spe-
cific QoS requirements [21]. Key requirements of a WBAN are low
packet transmission delay, low to medium data transmission rate,
near zero packet loss, minimum delay jitter, and low energy usage.
Performance of a WBAN system will depend on a number of factors,
which include the hardware design, applications characteristics, and
network design.
In a patient monitoring system, data transmission reliability
and latency is extremely important. The reliability and latency
of a WBAN will largely depend on the design of physical (PHY)
and medium access control (MAC) layers. For optimum network
efficiency and data transmission reliability, the MAC layer needs
to be designed to suit specific needs of specialized applications.
Considering the importance of WBAN applications, an IEEE stan-
dard group (IEEE802.15.6 working group) is working on the
Introduction 261
development of a body area network architecture [22]. The new

standard will define the PHY and MAC layer management issues,
which could be used to develop a low cost, ultra-low power and
highly reliable wireless network. Reliability of a medical network
can be defined in terms of its main QoS parameters such as delay
profile, delay jitter, and information loss rate. Performance and
reliability of a WBAN will depend on the PHY and MAC layer
procedures as well as on the network topology used. Following
sections examine network architecture and MAC protocol design
issues for a WBAN-based patient monitoring system. Before moving
into the discussion of design issues, first we examine the network
and MAC protocol design requirements of a WBAN. Listed below are
some of the main requirements of a WBAN.
• A WBAN should be able to support a range of med-

ical applications, which includes acquiring data samples
from periodic and nonperiodic sources and transmitting
information to a service node within a maximum fixed
delay without any loss of critical information [2]. Also, the
system should be able to support exchange of nonmedical
and control information with various remotely controlled
appliances.
• A WBAN should be able to operate in a power-constrained
environment where power sources such as battery should
operate for a reasonably longer period of time. Power
savings for implantable nodes are more critical than other
nodes. Power savings can be achieved by combining PHY
and MAC layer procedures [3].
• A WBAN should be self-healing, secure, and reliable.
• A WBAN should support data rates between few tens of
kbs to several Mbps to host a range of applications such as
images and video clips.
• A WBAN should support QoS management features to
offer priority services. Particularly when a critical patient
is monitored, the system must guarantee the delivery of
critical information to a service node. For medical data, the
main QoS features will be the transmission delay and the
packet loss.
• A WBAN may incorporate a narrow transmission band

(400 MHz or 2.4 GHz) together with the UWB (ultra-wide
band) technology to cover different environments.
• A WBAN should operate and coexists with other network
devices operating in similar frequency bands. Also, a WBAN
should be able to operate in a heterogeneous networking
environment, where networks of different standards may
cooperate with each other to acquire information from
different sensors.
In this chapter, we concentrate on the network design techniques,

mainly concentrating on the MAC and energy management issues.
The MAC protocol implements a set of rules, which are used
to coordinate packet transmissions in a shared transmission
channel environment. Numerous MAC protocols have been designed
to support applications in different types of networks. In this
chapter, we review several MAC protocols that could be used for
WBAN applications. This chapter also introduces an interconnected
WBAN architecture, which can be used to support remote patient
monitoring applications. Rest of the chapter is organized as follows.
Section 9.2 introduces the network topology and configurations
used in wireless sensor networks. Section 9.3 introduces the basic
properties and classifications of MAC protocol. Section 9.4 discusses
scheduled and polling MAC protocols and their design features.
Section 9.5 discusses the random access protocols and their
design features. Section 9.6 presents a hybrid MAC protocol design
proposed by the IEEE 802.15.4 standard. Section 9.7 discusses the
energy management issues in a WBAN. Section 9.8 presents the
design of several patient monitoring networks. Section 9.9 presents
some simulation results of a ZigBee-based patient monitoring
system. Conclusions are presented in Section 9.10.
9.2 Network Topologies and Configurations
A communication network can be designed using a number

of different topologies that define the geometrical position of
communication nodes [12]. A network topology is generally selected
Network Topologies and Configurations 263
Figure 9.1. Network topologies: (a) centralized star topology; (b) distrib-
uted bus topology; (c) ring topology, and (d) fully connected mesh topology.
based on application scenarios. There are five standard topologies

generally used: star, bus, tree, ring, and mesh. Figure 9.1 shows
commonly used four network topologies. The star topology consists
of a centralized controller where all nodes communicate directly
with the central controller to exchange information. Information
between nodes is also exchanged via the central controller. The
star configuration is suitable for WBAN applications where a
central controller can collect and aggregate all physiological data
from different sensors. The bus structure supports communication
among various nodes in a decentralized manner. The topology has
been developed to support operation of random access protocols
with minimum protocol control overhead. Bus structure could be
used for wired body area network applications. The ring topology
is used to transmit packets in a contention free mode using a
token, which is a form of distributed polling network. In the ring
topology, nodes are connected in a serial closed-loop fashion where
a control packet, known as the token, schedules the transmission
of packets. A ring network may not be suitable in a WBAN due to
longer transmission delay and maintenance of token overhead. A
ring network is generally more suitable for high data rate networks
with high volume of traffic. The mesh topology is used to provide
full connectivity between all nodes. The topology provides many
redundant transmission links, thus offering higher reliability in a
large network. The mesh network topology is quite often used in

large sensor networks to provide full connectivity. For a WBAN
application, the mesh topology is not very useful because each body
area network will generally consists of limited number of nodes.
However, a mesh topology could be used in a large hospital network
for multi-patient monitoring systems.
The star topology in a wireless sensor network is organized as
a cluster tree topology where the star network controller becomes
the cluster head. A cluster-based network forms a PAN; in case of a
wireless clustered network, it is referred to as the WPAN (Wireless
PAN) where the cluster head becomes the PAN coordinator [14]. All
data communication in a cluster network goes through the cluster
head or in case of a PAN/WPAN, through the PAN coordinator.
The PAN coordinator is also responsible for resource allocation,
synchronization, allocation of a PAN ID. Each node in a cluster has a
unique cluster ID, which is controlled by the cluster head. The cluster
head/PAN coordinator is a more computationally powerful data
processing device, which collects information from all its sensor
nodes. In WPAN, nodes are generally classified into two types:
full functional device (FFD) and reduced functional device (RFD).
The FFD can operate in three modes serving as PAN coordinator,
a coordinator, or as a network node. The RFD can only act as a
simple node, which forward information to a coordinator or to a
PAN coordinator. RFDs can exchange information only through the
PAN coordinator. To operate a cluster-based WPAN, it is necessary
to activate a star configuration. In a cluster-based WPAN, a FFD is
activated first, which may establish its own network and become
the PAN coordinator. All devices in a PAN are identified by the PAN
identifier. Once the PAN identifier is chosen, the PAN coordinator
allows other nodes to join the network. WPAN network devices
generally do not implement the full seven-layer protocol stack;
rather, it implements only the lower two layers of the protocol stack,
i.e., the physical layer (PHY) and the data link layer (DLL).
9.3 Basics of Medium Access Control Protocols
A sensor node is designed using the standard OSI (open system in-
terconnection) model to exchange data using communication links.
Basics of Medium Access Control Protocols 265
Figure 9.2. Protocol stack (a) standard OSI model, (b) three-layer WPAN
model.
The basic seven-layer OSI model and the corresponding WPAN

protocol stacks are shown in Fig. 9.2. A WPAN device will only
implement two and half or three layers of the OSI model, which
includes PHY, mostly a sublayer of the DLL, and an application
layer [14]. The physical layer is responsible for the transmission
and reception of physical signals using the hardware resources. The
PHY layer is also responsible for conditioning of transmitted and
received signals suitable for the transmission channel. A WPAN will
generally only use the MAC sublayer of the DLL. The MAC protocols
are used in multiuser communication networks to transmit packets
in a contention-free mode. The MAC protocol allow users to transmit
packets in an interference free mode while maintaining high
network throughput and low delay, thus supporting QoS of different
types of traffic. MAC protocols can also control transmission
power of a node by optimizing packet transmission and scheduling
processes. MAC protocols have been evolving over the last few
decades to support different type of networks and applications.
MAC protocols can be classified into two broad categories: fixed
assignment- and dynamic assignment-based protocols as shown in
Fig. 9.3. Fixed assignment-based protocols use the basic properties
Figure 9.3. MAC protocol classifications.
(time, frequency, and code) of a transmission channel to divide

transmission resources and use the allocated resources to transmit
their information in an interference-free mode. Fixed allocation-
based MAC protocols generally do not alter allocated transmission
resources during the length of a data session. Fixed assignment
protocols can also be viewed as scheduled protocols. Some of the
commonly used fixed resource assignment-based protocols are:
FDMA (frequency-division multiple access), TDMA (time-division
multiple access), CDMA (code-division multiple access), and SDMA
(space-division multiple access) [13, 14]. For WBAN applications,
some of the systems use the TDMA technique to accommodate
multiple sensor nodes.
Dynamic assignment-based protocols are generally known as
ON demand-based protocols where transmission resources are
allocated to transmitters only when an actual transmission takes
place. ON demand protocols are more suitable for bursty non-
periodic traffic sources. The dynamic access protocols are again
divided into two subclasses: scheduled access protocol and random
access protocol. A scheduled access protocol allocates transmission
resources in a dynamic manner, but it coordinates transmission
either in a centralized or in a distributed manner to avoid any
interference or collisions. The random access protocols are also
ON demand-based transmission system where transmitters may
use some channel state information to transmit a packet. The
channel sensing mechanism is used to reduce the probability
of collisions. A random access protocol generally requires less
Basics of Medium Access Control Protocols 267
control signaling compared to the scheduled access protocols.

Transmission of control signaling has implications on a network.
Large control overheads could reduce the information transmission
capacity of a WBAN; also control signaling could increase the power
requirements of a node or a network.
One of the key issues of a WBAN MAC protocol design is
power management. Power requirements of a WBAN will be jointly
determined by PHY and MAC layers. The MAC protocol can also
control the physical layer of a node to control the sleep cycle of
a node, which can affect the energy requirements of a node. The
MAC protocol can determine sleep, wakeup, and listening periods
depending on the application. Most of the medical applications will
generate low duty cycle data. The MAC protocol can synchronize
various application duty cycle data and intelligently control the sleep
cycle of a transceiver [13, 4]. Further discussion on MAC energy
management is provided in Section 9.7.
QoS requirements of a WBAN protocol are simple and straight-
forward. Most of the nodes will transmit information in a periodic
manner using fixed size packets. It is necessary for a WBAN
that packets can be transmitted in a contention-free mode with
high guarantee of delivery and low packet transmission delay. For
WBAN applications, most cases packet retransmission needs to be
avoided because generally the sensor nodes will have little buffer
and processing capabilities. Also, retransmission of packets will
reduce the life of the WBAN energy source. Some of the specialized
WBAN nodes connected with a tiny camera may require higher
transmission bandwidth to transmit images. A WBAN can prioritize
its transmission by assigning different priority value to different
traffic sources. QoS requirements of WBAN applications have not
been strictly defined yet. With the further development of the IEEE
802.15.6 standard, it is expected that QoS profiles for medical data
will be developed for future applications.
9.3.1 WBAN Traffic Characteristics

Traffic sources in a WBAN are mostly medical sensors which
generate data depending on the properties of physiological signals.
Table 9.1 lists some of the common diagnostic medial signal
Table 9.1. Physiological signal characteristics
Parameter Value Frequency (Hz) Data Rate (bits/sec)1

ECG Signal (single lead) 0.5–4 mV 0.01–250 4,000
Respiratory rate 2–50 breaths 0.1–10 160
Blood pressure 10–400 mm Hg 0–50 800
EEG 3–300 μV 0.5–60 960
Body temperature 32–40◦ C 0–0.1 1.6
EMG (Electromyogram) 10 μV–15 mV 10–500 8,000
GSR (Galvanic Skin Reflex) 30 μV–3 mV 0.03–20 320
1
Assuming an 8 bit A/D converter is used.
properties. Data rate presented in the fourth column is calculated

based on the Nyquist frequency and an 8 bit/sample A/D converter.
The table shows that signal data rate and interarrival time vary quite
significantly. All the traffic generated by various sensors is periodic
in nature because of constant sampling rate. For patient monitoring
applications, it is also possible that sensors could transmit data on
demand basis, i.e., a sensor transmit data only when a sensor is
being queried by a medical device. In addition to following medical
data, a WBAN may also support image or video transmission from
either an implanted or an on-body camera. Data rate for imaging
devices will depend on application and image/video resolution and
frame size. A WBAN also needs to cater for irregular events such as
irregular heart beat when a patient may enter into a critical stage.
WBAN can be used in an ambulatory environment where a critical
patient could be monitored over a communication network. Hence,
it is apparent from this discussion that a WBAN generally needs to
cater for periodic data, but on occasions it may be necessary to cater
for nonperiodic bursty data. Besides these signals, a WBAN may
need to transmit/receive control signals from specialized medical
devices. In future for specialized applications, a WBAN will probably
consist of sensor nodes as well as diagnostic medical appliances.
9.4 Scheduled Protocols
Packet transmissions in a WBAN will be controlled by the MAC

protocol. A scheduled MAC protocol will allocate transmission
Scheduled Protocols 269
resources to all nodes in an orderly manner using either a fixed or

a dynamic channel assignment technique. In a WBAN transmission
resource is the radio channel. In communication networks many
other scheduled protocols are used in various systems; however,
for WBAN applications we focus on TDMA and polling protocols
because they are most likely to be used for medical applications. For
certain applications, fixed assignment protocols can be modified into
adaptive assignment protocol such as ATDMA (advanced TDMA)
protocol.
9.4.1 TDMA Protocol

Time-division multiple access (TDMA) is one of the simplest fixed
assignment protocols [14]. TDMA protocol allows multiple users
to share a transmission channel by dividing the transmission time
among users. The TDMA protocol operates using a frame and slot
structure where a frame consists of a numbers of time slots with a
fixed frame length. The time slots are allocated to each transceiver
that can transmit or receive information using their allocated time
slots. The TDMA frame is synchronous in nature, which repeats
at a constant rate. For a WBAN, TDMA links can be configured
either in a full duplex or in a half duplex manner. Duplex links
could be frequency-division duplex (FDD) or a time-division duplex
(TDD). The implementation of FDD links is more costly because
each node needs to use a duplex filter, whereas a TDD link can
be easily implemented using a software scheduler on a node. For
WBAN applications, most of the time traffic will be transmitted in
one direction, i.e., sensor to the coordinator; hence, either a full
duplex or a TDD link will be suitable. A TDMA-based WBAN can be
designed using both star and bus network topology. TDMA, being
a synchronous packet transmission system, can handle periodic
traffic more efficiently than nonperiodic traffic. A TDMA link can
provide fixed packet transmission delay, which is one of the key
requirements of a WBAN. Another benefit of the protocol is the
use of minimum number of overhead bits to support information
transmission due to the fixed channel allocation technique. The main
disadvantages of a TDMA-based system are scalability and efficient
support of nonperiodic traffic sources. The energy efficiency of
a TDMA-based system can be enhanced by synchronizing the

transmission frame/slots with the sleep cycle of a node [5]. Energy
efficiency issues are discussed in Section 9.7.
9.4.2 Polling Protocol

Polling is one of the dynamic scheduled access mechanisms that
can be used to support packet transmission in any size network
[6]. A polling network is a centralized network architecture where
a central network node controls the transmission of packets,
avoiding any collisions of packets. Unlike the TDMA protocol, polling
protocols generally do not require any fixed allocation of resources
but require higher level of signaling to transmit information.
Figure 9.4 shows a typical wireless polling network and its polling
sequence operating in star topology. This figure shows that all nodes
in the network are polled by the central controller in a round-robin
manner to receive data from these nodes. The controller sends a poll
message to a specific node, which in reply can send either a data
packet or a NACK (negative acknowledgement) specifying that it has
no data to send. To minimize packet transmission, the controller
could include the acknowledgement of the pervious transmission
with the current poll message. Since a poll message is a broadcast
message, all nodes in the network can receive the message. However,
Figure 9.4. A wireless polling network showing the polling sequence. See
also Color Insert.
Scheduled Protocols 271
the controller specifies in the poll message which node should reply.
A polling system is more flexible than a TDMA-based system due to
direct interaction between the controller and a node. In this case,
each node can be polled based on its transmission requirements. For
example, Table 9.1 shows that an ECG sample is produced after every
2 ms whereas an EMG sample is produced every 0.2 ms. If we assume
10 signal samples are transmitted in a packet, then the ECG node
needs to be polled after every 20 ms whereas the EMG node needs
to be polled every 2 ms. This flexibility is not available in a TDMA-
based system. Also, in polling networks, nodes can transmit different
size data packets depending on the application. These flexibilities
are achieved at the expense of extra signaling and the processing
capability of the central controller. Polling sequence and rate can be
tailored for different applications.
Packet transmission delay in a polling network will depend on
the polling cycle time [12]. As shown in Fig. 9.4, nodes are polled
in a sequential manner where a node can transmit a packet after it
has been polled. So the packet transmission delay will be the sum of
packet queuing time and the packet transmission time. In a WBAN,
propagation delay is extremely small, which can be neglected. In a
network if the controller is polling all the nodes at a constant rate,
then the polling cycle will be the sum of polling time and packet
transmission time. For example, if the poll and acknowledgement
packet transmission time is 1 ms and the data packet transmission
time is 2.5 ms, then each node in Fig. 9.4 will be polled after every
14 ms assuming that all nodes always have data to transmit when
polled. In this case, if the first node generates data in the middle of a
polling cycle, then it must wait 8 ms before a poll message is received
and then it will take another 2.5 ms to transmit the data packet
to the controller, incurring a 10.5 ms delay for the transmission.
For a WBAN application, if the nodes are polled by matching their
data generation rate, then the polling cycle length will vary due
to the number of active nodes in each polling cycle. A polling
system can be tailored to cater for data with different priorities
and data generation rates. A high priority node can be polled more
frequently compared to lower priority nodes. Polling networks are
easily scalable, particularly when the controller finds a new node it
can include the new node in its polling list or delete a node from
the list if a node does not respond to subsequent polls. The energy
requirements of a polling network can be variable depending on
the system design. Polling networks can be implemented using star,
bus, and ring topologies. For WBAN application, a star topology-
based polling network will be most suitable because the bus and ring
topologies introduce higher polling and message transmission delay.
9.5 Random Access Protocols
Random access protocols are generally used in distributed networks

without the presence of any central coordinator [12, 13, 14].
Nodes using random access protocols coordinate the transmission
of packets to minimize collisions. Nodes using this class of protocols
will use the transmission channels only when they have data
to transmit. Random access protocols are more complex than
scheduled access protocols; hence, careful approach is required
to design the protocol and the network. Random access protocols
are in use in local area networks for many years. This class of
protocol is also becoming popular for wireless sensor network
applications. The IEEE 802.15.x standard defines a number of
random access protocols for a number of short-range wireless
networks [7]. Bluetooth was the first short-range low-to-medium
data rate MAC protocol developed for WPAN applications. Later
the Bluetooth protocol was modified to develop the IEEE 802.1.5.1
standard. Subsequently, a low-power low data rate standard, the
IEEE 802.15.4 MAC protocol, was developed for the wireless sensor
network applications. In the IEEE 802.15.4 standard, which consists
of PHY and MAC layers, functionalities have been merged with
application layer protocols to develop the ZigBee standard. The
IEEE 802.15.x family of protocols is becoming one of the major
wireless sensor networking standards. The IEEE 8021.5.6 protocol
is currently under development to support WBAN applications.
It is most likely that the new WBAN standard will be based on
the CSMA/CA (carrier sense multiple access/collision avoidance)
standard [22].
One of the first random access protocols used in computer
networks is the ALOHA protocol. Since the development of the
Random Access Protocols 273
ALOHA protocol, many other random access protocols have been

developed by further enhancing the features of the protocol. A more
efficient random access protocol known as the CSMA protocol was
developed from the ALOHA/S-ALOHA (slotted ALOHA) protocols
[12]. Using the CSMA protocol, each transmitting node checks the
status of the transmission channel to decide whether to transmit
a packet or not. The philosophy of the CSMA protocol is listen
before transmit. The CSMA protocol can operate either in persistent
or in a nonpersistent mode. In the persistent mode, a node
constantly monitors the status of the transmission channel and
starts transmitting as soon as the channel becomes free. Using
the nonpersistent mode, a node constantly monitors the status
of a transmission channel. When the channel becomes free, the
node then can either immediately transmit or backoff for a certain
period of time to check the channel again. The decision whether
to transmit a packet or to defer a packet transmission depends on
the transmission probability value generated by a random number
generator at the MAC layer. The nonpersistent protocol has been
developed to minimize the probability of collisions by randomizing
packet transmission times. The persistent protocol may offer a low
transmission delay at the expense of higher level of packet collisions,
whereas the nonpersistent CSMA protocol offers network stability
at the expense of higher average packet transmission delays. A
compromised version of the CSMA protocol is quite widely used
and is known as the p-persistent protocol, where the probability
of a transmission or a deferment can be adjusted by selecting
appropriate value of p, which is the packet transmission probability
[15]. In a CSMA/CA network, an ACK packet is used to inform the
transmitter about the status of a transmission. A receiver transmits
an ACK packet when a packet is successfully received. A transmitter
may not receive an ACK packet if either the transmitted packet is
incorrectly received due to channel errors or packet transmission
is unsuccessful due to collisions. A transmitter can successfully
transmit a packet if all other transmitter remains silent for a
period of maximum propagation delay, tp , of the network. After
the propagation delay, all nodes of a network become aware of the
transmission and will not attempt to transmit any packet while the
current transmission continues.
Channel sensing and p persistent techniques are combined

to improve the performance of a network by trying to avoid
collisions. CSMA/CA-based protocols also use another collision
avoidance technique particularly to reduce the hidden user problem
[13]. A wireless terminal is considered to be a hidden terminal
if the terminal is unable to receive transmitted signal(s) from its
neighboring terminals because that particular terminal could in a
RF (radio frequency) shadow region compared to its neighboring
terminals. In such case, a hidden terminal could increase the packet-
collision level in a network. A hidden terminal may be unaware
of other transmission or transmission attempts and inadvertently
interfere with other transmissions. The CSMA/CA protocol uses
the RTS/CTS (ready to send/clear to send) signaling to reduce the
hidden terminal problem. However, for a WBAN application, the
hidden terminal will not be a problem due to the size of the network
and the proximity of terminals.
Random access protocols can offer several advantages for WBAN
applications. The first advantage is the support of periodic and
nonperiodic traffic with equal efficiencies. Using a random access
protocol, a transmitter can initiate a packet transmission on its own
only when a packet is generated at the node. Unlike the scheduled
access protocol, the transmitter does not need to wait for a time
slot or a poll message to initiate a transmission. In a WBAN when
periodic and nonperiodic traffic are mixed, the problem of collision
reduces due to randomization of packet generation rate. Second,
the main advantage of the random access protocol is the energy
consumption [16]. A node using the random access protocol can stay
in the sleep state most of the time, waking up only when transmitting
a packet. Another key advantage of a random access protocol is
the scalability. The networking size can easily be increased or
decreased without the need for any software or hardware upgrade.
The main disadvantage of the random access protocol is variable
packet delay due to collisions at the higher traffic load. However,
this issue is not a problem for WBAN applications if the volume of
traffic is kept relatively lower compared to the network transmission
capacity. Random access protocols have been widely accepted for
sensor network applications. Industrial standards such as ZigBee,
Bluetooth, etc are quite widely used in sensor network applications.
Hybrid MAC Protocol 275
Both ZigBee and Bluetooth standards are based on the CSMA/CA-

based MAC standards, which can be successfully used in WBAN
applications.
9.6 Hybrid MAC Protocol
Previous sections discussed scheduling and random access pro-

tocols, which are used in different wireless sensor networks. For
sensor network applications, the IEEE 802.15.4 standard defines
a hybrid packet transmission structure, which allows transmitters
to use both contention and guaranteed transmission modes [14,
17, 18]. The hybrid transmission structure has been developed to
efficiently support periodic, intermittent, and repetitive low latency
data. The IEEE 802.15.4 standard supports the hybrid transmission
mode using the super frame structure as shown in Fig. 9.5. The
super frame structure is controlled by the PAN coordinator. The
coordinator periodically sends beacons that identify the PAN. The
PAN is identified by a 16 bit PAN identifier. The beacon frame also
contains information, a list of outstanding frames, and other system-
related parameters. The time interval between beacon signals is
constant, and users can select a value which is a multiple of 15.38 ms.
The maximum beacon interval time could go up to 252 ms. Two
consecutive beacon signals form a super frame separated by 16
equally sized time slots as shown in Fig. 9.5. The super frame is
subdivided into active and inactive periods. All nodes, including
the coordinator, can turn off their transceivers and go into a sleep
Figure 9.5. IEEE 802.15.4 MAC super frame structure. See also Color
Insert.
state during the inactive period of a super frame. The active period
consists of 16 timeslots, where each timeslot is divided into two
groups. One set of slots forms the contention access period (CAP)
and the rest of the slots become part of the guaranteed time slots
(GTS). The length of the active and inactive periods is configurable,
hence allowing system design flexibility to accommodate different
types of applications. The CAP slots can be accessed using the
standard CSMA/CA procedure, whereas the GTS slots are allocated
by the coordinator when a request is made by a node. The flag
in the request packet indicates whether the slot will be used to
transmit or receive. Upon receiving a slot request, the coordinator
will allocate a slot when appropriate resources are available. GTS
slots are organized in the same way as the TDMA slots. When GTS
slots are used for transmission, a node can schedule its sleep cycle
and wake up just before the time slot starts and send the packet.
On the other hand, when a CAP slot is used, the node must wake up
in advance and perform carrier sensing or other collision avoidance
procedures before it can initiate a transmission.
The super frame structure based on the hybrid MAC protocols
could be well suited for WBAN applications where medical data with
different QoS requirements will be transmitted. Sensors of lower
priority data could be transmitted using CAP slots, whereas high
priority data could be transmitted using GTS slots. The super frame
structure is configurable where the active and inactive periods can
also be adjusted.
9.7 Energy Management in a WBAN
Energy management of nodes is one of the crucial requirements

of WBAN deployments [19]. WBAN’s could be deployed for patient
monitoring continuously for several days. In such deployments, it
is important that nodes are designed properly so that the battery
lasts for the duration of the deployment. Energy requirements of
a node will depend of a number of factors, which includes the
PHY layer and hardware design, MAC layer-controlled sleep cycle,
MAC packet scheduling, and the application requirements. Source
of energy of WBAN nodes will be a battery whose life cycle will
Energy Management in a WBAN 277
depend on the way a node will drain the energy. There are a number
of battery phenomena that could affect both charge and discharge
outcomes. Two important phenomena are the rate-capacity effect
and the recovery effect. The rate-capacity effect of a battery is the
idea that drawing a large, continuous current from a battery will
lead to it depleting faster than expected. In an ideal battery, it is
assumed that the voltage stays constant for the entire life of the
battery, and then drops to zero. In the real world, however, voltage
is not constant over the entire life of the battery; instead it drops
during the discharge. This drop in voltage during the discharge
varies in severity from battery to battery. In all cases though, it
leads to a perceived drop in battery capacity. Figure 9.6 shows the
typical rate capacity characteristics of a battery for a continuous
and an intermittent load situation [6]. An intermittent transmission
is a typically ON/OFF scenario when a transmitter sends data
for a brief period and then moves into an inactive state. When a
transmitter moves to an inactive state, the battery load will be low,
allowing the battery to recover. Figure 9.6 shows that an intermittent
application could prolong the life of a battery due to the recovery
effect. For a continuous discharge, the slope remains fairly constant
as expected. For an intermittent discharge, however, the battery is
able to recover some of the lost charge, resulting in a piecewise-
continuous discharge slope. This results in the battery having a
longer lifetime.
Figure 9.6. Battery charging and discharging phenomena.

As shown in Fig. 9.6, the battery lifetime of WBAN node can

be sufficiently extended by appropriately designing the packet
transmission schedule in a network. It is possible to optimize
the power requirements of a sensor node by using the following
techniques.
• Lower data transmission frequency by the use of data
aggregation technique.
• Reduction of proportional frame overhead by selecting
appropriate payload size.
• Appropriate use of power management algorithms to
coordinate the sleep cycle.
• Placement of nodes in a WBAN to avoid poor transmission
channel conditions.
To achieve significant advantages, each of the above techniques
needs to be carefully designed and integrated with the MAC protocol.
Lower the data transmission frequency can reduce the power
requirements of a sensor node. Average current consumed by a
sensor node is described by Eq. 9.1. The equation clearly shows that
transmission duration and average transmitter current requirement
will increase with more frequent transmission of data. In addition
to power requirement for data transmission, there are other power
requirements. For example, for a CSMA/CA-based network for
every packet transmission the node needs to listen to the channel
when the node acts as a receiver. On the basis of channel sensing
result, it either transmits or backoff for a certain duration. It is
necessary for reduced power consumption that during the backoff
period, the transmitter goes to the standby mode. The transmitter
could be switched ON by a timer for further channel sensing and
transmission decision-making. This intermittent ON/OFF process of
the transmitter will help to prolong the battery. Data aggregation
technique will lower frequency of data transmission as well as
reduce the frame overhead [10]. For example, EEG data samples
are generated at a rate of 60 Hz, i.e., samples are generated after
every 16.66 ms. If we use 8 or 12 bit A/D converter and transmit
each sample individually, then the payload size will be very small.
Instead of a single sample if a number of EEG samples are combined
and transmitted every second, then the payload size becomes
480 bits or 720 bits depending on the chosen A/D converter word
size. In this case the transmitter will use far more less power to
transmit without introducing unnecessary delays. Similarly the data
aggregation technique can be applied for a multichannel sensor
networks as described in Chapter 5. In case of a multichannel
WBAN, samples from different channels will be aggregated and
encapsulated in a single packet. The use of data aggregation
technique has another advantage for a random access network.
The data aggregation technique will reduce the number of packet
transmission attempts for a CSMA/CA network, thus reducing the
packet collision probability and increasing the QoS of a WBAN. In
a random access network such as ZigBee or Bluetooth network,
transmission of fewer large packets will reduce the collision level
as well as the transmission power requirements.
Iavg = Ttxon × Itxon + Trxon × Irxon + (1 − Ttxon − Trxon ) × Istb (9.1)
Ttxon is the transmission duration including packet transmission and

other time necessary to support the transmission.
Trxon is the packet reception time.
Itxon is the average current consumed by the transmitter.
Irxon is the average current consumed by the receiver.
Istb is the current consumed by the transceiver module in the
standby mode.
The coordination of sleep cycle is important for the node power

management. On the transmitter side, sleep cycle can be easily
managed because the transmitter blocks can be turned ON or OFF
by the MAC layer depending on a node’s transmission requirements.
The transmitter circuit will only be turned ON when the MAC
layer receives packet from its higher layer. Switching the receiver
block between ON/OFF modes is a critical issue. A node needs
to know when it is expecting a packet or a control packet. For a
TDMA or a poll-based system, the receiver can easily work out the
packet reception time. In this case, the receiver can be left in the
standby or sleep mode until the packet reception time. However,
using the CSMA/CA protocol the packet reception time is not fixed
until the transmitter and receiver work out the packet transmission
schedule. In case of WBAN applications, sensor nodes will be mostly
transmitting information packets. Sensor nodes will only receive

acknowledgement packets or occasional control packets. On the
other hand, the coordinator or the PAN coordinator will receive
information and control packets from different sensor nodes; hence,
it may be necessary for the coordinator the keep the receiver circuit
ON for the duration of its operation. A receiver can save more power
when receiving data from periodic sources. However, a WBAN node
should be able to receive data from different type of sources, i.e.,
periodic and nonperiodic sources.
Power transmission requirement of a node is also influenced
by transmission channel conditions. Although a WBAN will operate
within a small area, it is possible that the receive signal will be
corrupted when transmission loss increases due to the placement of
a node or a posture of a body. Most of the modern transmitters will
transmit at a minimum power level, which will allow the receiver
to correctly receive the transmitted information. For a multi-patient
monitoring system, it is important that mutual interference should
be minimum; hence, all transmitters must transmit at their optimum
power level. An optimum power level is defined as the minimum
power level at which a receiver can receive packets with zero
bit error rate (BER). The transmitter power level requirement
will increase with increasing transmission losses. A transmitter
can adjust its transmission power level based on the feedback
from the receiver, which informs the link condition. A receiver can
send a feedback using a positive ACK when a packet is received
correctly or a NACK (negative acknowledgement) in case a packet
is received incorrectly. For a WBAN, transmission of ACK or NACK
could introduce additional traffic load. It may be possible that ACK
or NACK mechanism can be used only for critical data leaving other
nodes to transmit data at a predetermined power level. In this
case, a receiver may occasionally send explicit instruction to the
transmitter to change its power level to maintain zero BER level.
To examine the effect of sensor node power management issues,
we use the CC2420 chip parameters for sample calculations. The
receiver module of the chip operates on a fixed power level, which
consumes 18.8 mA current. The transmitter power requirement
depends on the transmitted RF (radio frequency) signal power.
Transmitter current values are listed in Table 9.2 for different
Table 9.2. CC2420 IC current consumptions
RF signal Power Level (dBm) Transmitter Current (mA)

−25 8.5
−15 9.9
−10 11
−5 14
0 17.4
transmitted RF signal powers. The table shows that the chip can
operate at six different power levels. Strongest RF signal will
consume an average current of 17.4 mA, whereas the weakest RF
signal will consume 8.4 mA of current. The standby current of the
IC is 1 μA. The required RF signal power level will be determined
by the transmission loss value. The transmission loss value can
be calculated using the Eq. 9.2. For WBAN applications, additional
losses can add with the transmission loss. Major additional loss
component value is the tissue absorption particularly for implanted
nodes. The received signal strength should be at least equal to or
greater than the receiver sensitivity. The receiver sensitivity of the
CC2420 IC is -95 dBm. If a transmitter is generating a RF signal at
−25 dBm, then the total transmission loss should be lower than 70
dBm.
d
P L (d) dB = P L (d0 ) + 10γ log10 + Xσ, (9.2)
d0
where PL(d0 ) is the path loss at the reference distance, d0 is the
reference distance, and X σ is the shadowing variance.
Let’s now analyze the power requirements of a CSMA/CA-based
transmitter. The CSMA/CA protocol uses a series of procedures
before a node can transmit a packet. Figure 9.7 shows the
packet transmission sequence of the CSMA/CA protocol. When
the transmitter senses a channel and initially it finds busy, then
it waits until the channel is idle for the DIFS (distributed intra-
frame spacing) duration. After the DIFS waits the transmitter enters
Figure 9.7. Packet transmission sequence of the CSMA/CA protocol.

into a backoff period. Following the backoff, the transmitter could

transmit in a contention mode, i.e., transmit a packet with a risk
of collision. After the transmission of a packet, the terminal will
wait for a period known as SIFS (short IFS) [12]. After the SIFS, the
transmitter is expected to receive an ACK packet. Typical CSMA/CA
timing parameters are as follows. The DIFS length is 50 μs, minimum
back period is 140 μs, and the SIFS value is 10 μs. The packet
transmission time will depend on the packet size. The IEEE 802.15.4
CSMA/CA packet will consist of PHY layer header of 6 byte, MAC
layer header 5 byte assuming a 2 byte address field and a 2 byte
FCS (frame check sequence). Total transmission time of a packet
can be calculated using Eq. 9.3 neglecting the propagation delay.
For a 40 byte/320 bit payload, the packet length will be 51 byte,
and the packet transmission time of the CC2420 transmitter will
be 1.63 ms. Table 9.3 lists current consumption values of different
CSMA/CA parameters for different RF signal levels. Values listed
Table 9.3 is calculated based on the assumptions that a packet can
be transmitted in the first attempt; hence, these are best case values.
For example, when the channel is sensed and found to be busy, the
transmitter will backoff and resense the channel after the backoff
period. A node senses the channel after each backoff period until
it finds the channel free for a DIFS period when it could initiate a
packet transmission. In this case power transmission requirements
of packet can significantly increase. If we assume that a WBAN
node is using two 900 mAH AA batteries, then a node can transmit
30,706 packets at 0 dBm RF signal level. The same node can transmit
60,851 packets at −25 dBm RF signal level effectively doubling the
lifetime of a battery for the best transmission case condition. In a
Table 9.3. Current consumption values for different CSMA/CA parameters
Packet transmission
RF Signal (dBm) DIFS (μA) SIFS (nA) Backoff (nA) (μA) Total (μA)
0 0.94 0.01 0.14 28.36 29.31
−5 0.94 0.01 0.14 22.82 23.76
−10 0.94 0.01 0.14 17.93 18.87
−15 0.94 0.01 0.14 16.13 17.07
−25 0.94 0.01 0.14 13.85 14.79
Patient Monitoring Network Design 283
WBAN node, if one packet is transmitted per second, then the battery
can potentially last up to 17 h for the −25 dBm RF signal level.
Packet retransmissions can affect the life of a battery. However, in the
best transmission case, the battery can last slightly longer than the
above value due to the battery recharging phenomena as shown in
Fig. 9.6.
L
Tpacket = TDIFS + TBO + + T SIFS (9.3)
R
where TDIFS is the DIFS period, TBO is the backoff period, L is the
packet length in bits, R is the transmission rate in bits/s, and TSIFS is
the short interframe spacing.
Discussions in this section clearly show that the battery lifetime
of WBAN will depend on selected MAC parameters and traffic
characteristics. Data aggregation and packet scheduling technique
will significantly influence the power requirements of a node. For
a polling network node, transmission power requirements will
depend on the design of polling message and cycle.
9.8 Patient Monitoring Network Design
In this section, we will introduce the general rules and techniques

of WBAN design. As mentioned earlier, WBAN is a special-purpose,
low-to-medium rate sensor network with a typical operating
diameter of about 1 m. The number of nodes in a typical WBAN
is expected to be around 2 to 5 with each node connected to a
number of sensors. So the data rate requirements per node will
not be significantly high unless a node is connected to a camera
or supporting multichannel ECG data. Main QoS requirements of
a WBAN are low latency, no or very minimum packet loss, and
low power transmission to avoid any tissue heating or damage.
Two main networking requirements of a WBAN are low power
consumption and the network should be a scalable one. The size of
WBAN could range from a single-body network to an interconnected
multi-body network operating as a multi-hop network as shown in
Fig. 9.8. To design a WBAN, the following design issues need to be
considered.
Figure 9.8. A WBAN-based multi-patient monitoring system. Each

patient’s body forms a PAN and consists of sensor nodes and the PCU.
9.8.1 Transmission Capacity Requirements

It is necessary to develop the data profile of a network. Most of the
data will be generated from physiological sensors. It is important
to examine all physiological signal characteristics and calculate the
minimum and maximum transmission data requirements of each
signal. Data must be characterized in terms of their interarrival
rate, which is related to the sampling rate and the data burst size.
The designer should also consider the dynamic range of signals
and select the sampling rate and A/D converters appropriately.
Next, the aggregate data volume should also be considered. The
aggregate data rate is important for the selection of the networking
standard and the MAC protocol. The total aggregate data rate of a
network should not exceed 85–90% capacity of a scheduled access
network. In case of a random access network, the aggregate data
rate should be below 75% of the maximum capacity of the network.
Performance of a network could be significantly affected by the
aggregate data rate of a network. It is also important to work out

the peak-to-average data transmission rate of a network. In a WBAN,
most of the time data rate will be steady, but monitoring a critical
patient may occasionally generate peak traffic. If the peak-to-average
traffic rate is high, then occasional data loss can happen due to
congestions or buffer overflow.
9.8.2 PHY and MAC Layer Parameter Selection

It is important to select PHY and MAC parameters appropriately
to satisfy the requirements of a WBAN [8]. Although PHY layer
parameter selections are difficult because in most cases PHY
parameters are decided by the hardware design. Important PHY
layer parameters include the modulation and coding techniques that
can influence the performance of a sensor node. The modulation and
coding parameters determine the receiver sensitivity and effective
packet error rate (PER) based on the channel BER. Selection of
MAC parameters is very crucial in determining performance of a
network. Sections 9.4–9.7 have discussed various features of various
MAC protocols. MAC protocol selection will influence the power
consumption, QoS values, and scalability issues. Scalability could
be an important issue for the WBAN deployment. A WBAN should
offer flexibility to health care professionals so that they either
increase or decrease the number of nodes on a body without any
major change in software or hardware. It will be useful if the MAC
protocol can adaptively accommodate varying number of sensor
nodes. The TDMA protocol will not be very scalable unless software
modifications are done on the network controller side. Using a
polling network, it is possible to add or remove sensor nodes
without major changes [11]. From the scalability point of view, the
CSMA/CA protocol offers the best advantage. In case of CSMA/CA
protocol, since there are no central coordination necessary, any new
sensor node will read the broadcast message of its PAN coordinator
and will register with PAN coordinator for data exchange within a
WBAN.
Most of the sensor networking standards such as code blue,
Wibree, Mica2, etc. have selected the CSMA/CA MAC protocol to
develop medical applications [21]. As mentioned above, a CSMA/
CA-based system is adaptive in nature, requires minimum control

signaling, and supports required QoS for a WBAN. Also, the
CSMA/CA protocol can be optimized for different application
scenario by appropriate selecting transmission and priority para-
meters. The IEEE 802.15.4-based hybrid protocol could be a good
choice for future systems where low priority nodes can transmit
packet using the contention mode, whereas the GTS mode could be
used for high priority nodes. Hybrid protocol will also offer excellent
scalability feature.
9.8.3 Network Configuration

The network configuration of a WBAN depends on the deployment
scenario. A WBAN can be deployed either in a stand-alone mode or
in an interconnected mode [20]. A stand-alone mode could be used
for a single-patient monitoring system, whereas an interconnected
mode could be used for either monitoring multiple patients or
monitoring a patient remotely. A stand-alone mode will simply form
a PAN with a CCU (central coordination unit) as the PAN coordinator.
In this mode, the CCU will collect all sensor data and store them
for further processing [9]. The protocol architecture will be very
simple where application layer can directly communicate with the
MAC layer for transmitting and receiving information. In this mode,
we don’t need to consider mobility because it is most likely that all
nodes will be attached to a single human body and the CCU will also
be located either on the body or close to the body in a fixed location.
Multi-patient or remote monitoring system will require a
complex networking configuration where we need to construct a
multi-hop network as shown in Fig. 9.8. In this case, each body
will form a single PAN and all PANs need to be connected to a
PCU (patient coordinator unit — a wearable device that can be
carried, wearable by the patient). Each CCU will act as a router
for their PAN and transmit all its sensor data to the PCU. The PCU
could be connected to a hospital LAN (local area network) or to the
Internet. All CCU’s can transmit their packets in the broadcast mode
with PCU address in the MAC address field. The multi-hop network
needs to introduce some form of packet scheduler to minimize
the contention if the CSMA/CA protocol is used. In the multi-hop
networking scenario, all sensor nodes, CCUs, and PCU will transmit
packets in the contention mode; hence, it is necessary that some
form of power control and packet scheduling need to be introduced.
For example, sensor nodes in each PAN should transmit their packets
with a minimum power sufficient for their CCU to receive packets
with zero BER. If sensor nodes transmit at a higher power level than
necessary, then it may be possible packet transmissions in other
PANs will be affected. It is also necessary that CCUs should introduce
some form of pseudo synchronized packet scheduling so that all of
them should not try to send their packets at the same time. Each CCU
will individually adjust its transmitter’s power based on its PCU link
characteristics.
A multi-patient or a stand-alone patient monitoring system could
be converted to a remote patient monitoring system by introducing
the TCP/IP (transmission control protocol/Internet protocol) stacks
either at the CCU or at the PCU. The TCP/IP protocol stack will allow
these nodes to communicate with external networks. In this case no
modifications will be necessary for the sensor nodes because they
still will be sending their data within their PAN.
Some deployment scenario of WBANs may require the network
to support mobility. For example, mobility support is necessary
when a patient is monitored while they are carrying out their day-to-
day activities either at home or at workplace. Mobility can introduce
two additional problems: the variation of link characteristics and
routing data to the appropriate node. If a mobile patient is supported
within an indoor area, then it is most likely an access point will be
used within the home or work area, which can connect it with the
CCU to exchange data. In that case, we might use a Wi-Fi access
point to connect with the CCU. It is preferable to use a Wi-Fi access
point because it can support a larger cell size — up to 300 m link
distance and relatively cheaper to implement. In this case, we need
to develop a CCU with dual protocol stack. If we assume a ZigBee-
based PAN and a Wi-Fi-based access point, then CCU will act as a
router and it will have vertically split protocol stack supporting the
ZigBee protocol on the PAN side and Wi-Fi on the CCU to access point
link. Basically in this case, the CCU will be built using dual radio IEEE
802.15.4 and IEEE 802.11. If a WBAN requires mobility at a higher
speed for sports training applications, then wireless standards such
as GSM or UMTS or WiMAX should be used because Wi-Fi can only

support mobility up to 5 km/h.
9.9 Performance Analysis of a WBAN
In this section, we will analyze the performance of a ZigBee-

based patient monitoring system. Figure 9.8 shows a typical WBAN
network architecture for mult-patient monitoring applications. The
data from each sensor are transmitted to a PCU. The PCU aggregates
sensor data and transmit them to a CCU located at a short distance.
The PCU can gather data from both implant and external nodes.
The CCU acts as an intermediate network device that forward the
collected medical data to a patient database (DB) where remote
monitoring devices can retrieve patient’s data for health care
professionals. Each patient body forms a PAN where a PCU acts as
the PAN coordinator. In this WBAN application, the CCU acts as the
PAN coordinator for all PCUs, which can be considered a second-
tier network. The network operates in a multi-hop fashion where
each link transmits data using the CSMA/CA protocol. Relay nodes
such as the PCU and the CCU can be configured in two ways. One
of the configurations is that these nodes simply act as forwarding
node, i.e., it receives a packet and simply forwards the packet
to the next link. Other configuration is that these nodes receive
multiple packets from sensor nodes and then encapsulates multiple
packets and forwards a larger packet. The former approach can be
implemented with nodes with low-processing capability, whereas
the latter approach reduces the number of transmission access on
the forward link, thus reducing the probability of collisions for a
ZigBee-based system.
In this section, we present the performance of the above
WBAN for multi-patient monitoring applications using an OPNET
simulation model [23]. Using the simulation model, initially we
analyze the performance of the data aggregation technique. To
analyze the performance of data aggregation technique, we use a
single PAN model. Figure 9.9 shows the packet loss and packet
generation rate of the simulated WBAN for different payload sizes.
Result shows that with the increasing payload size, the throughput
Performance Analysis of a WBAN 289
Figure 9.9. WBAN packet loss and packet generation rate for different
data aggregation rate/payload size. See also Color Insert.
of WBAN increases due to fewer transmission attempts. The figure

shows that with single sample/payload (12 bits), the network
only managed to successfully transmit 120 packets, offering a
success rate of 40%. When the average payload size is increased
to 120 bits/packet, then the number of transmission attempts is
dropped to 50 packets/second, offering 100% throughput in the
network. The improvement in the efficiency can be achieved due to
lower contention level in the network, which resulted fewer packet
losses due to the packet dropping threshold. In this simulation, we
used five retransmission attempts as the packet dropping threshold,
which means if a packet is unable to successfully transmit a packet
in five successive attempts, then the packet is dropped from the
transmission queue. The delay profile of the WBAN for different
payload sizes is also investigated and presented in Fig. 9.10. The plot
shows that the average packet delay decreases with the increasing
payload size. Both results (Figs. 9.9 and 9.10) show the same
trend.
Next we investigate the performance of an interconnected WBAN
to analyze the performance of a multi-patient monitoring system.
In the proposed scenario, it is assumed that all patients’ sensor
Figure 9.10. WBAN delay profile for different payload sizes.
nodes transmit their data via the PCU and CCU to the database. In
this model, PCUs and CCUs are simply acting as a relay node, and
forwarding those data to the DB. In order to minimize the number
of packets, we use the data aggregation technique at the source
node. In this simulation, we connect all sensors on a body using
two sensor nodes transmitting data to the PCU. One of the sensor
nodes (sensor 1) aggregates ECG, body temperature, and blood
pH data and transmits these data after every 145 ms generating
7 packets/second. The other sensor node (sensor 2) aggregates
data from the blood flow, blood pressure, and respiratory rate
sensors, and transmits these data after every 485 ms generating
3 packets/second. This aggregated packet structure allows 73
data samples with a payload size of 880 bits/packet. Using the
aggregation technique, each PAN generates only 10 packets/second.
The effective data rate requirement of sensor 1 and sensor 2
becomes 6.05 kbps and 1.92 kbps, respectively.
The delay profile for a multi-patient networking scenario is
depicted in Fig. 9.11. The plot shows the average end-to-end delay,
which is the total delay over 3 hops (sensor→PCU, PCU→CCU, and
CCU→DB). It is observed that the delay increases with increasing
number of patients. In this case, the main reason for the increase
Conclusions 291
Figure 9.11. Delay profile of the multi-patient monitoring system.
in delay is caused by the contention or interference from different

PANs. For an operating ZigBee network, sensor nodes on each
patient’s body forms a working PAN with the PCU as the coordinator.
In a multi-hop networking, collisions can happen either within a
PAN or between PANs when simultaneous transmission goes on in
different PANs. Also, transmission from the PCU to CCU or from the
CCU to DB transmissions can be interfered by the transmissions
from other PANs if the IEEE 802.15.4 MAC is used on all the links.
In our simulation, all the wireless links used the CSMA/CA protocol.
The delay plot shows that the delay profile for up to six patients is
quite acceptable where the average packet delay is about 170 ms. A
monitoring system will be able to transmit large number of samples
within this 170 ms due to the aggregation technique used.
9.10 Conclusions
This chapter presented WBAN network design techniques for

various application scenarios. WBAN is currently considered a
developing technology, which needs to be standardized for clinical
applications. There are many networking techniques that need to
be appropriately developed to support medical applications. Since
the MAC protocol is one of the key components of a WBAN design,

it is necessary to select appropriate protocol and configure them to
achieve optimum performance. This chapter presented theoretical
discussions on MAC protocol and network design techniques,
which will be useful for medical, IT professionals for future
medical application development. Simulation results presented in
Section 9.9 can act as a guideline for WBAN design engineers.
Section 9.9 also demonstrates the importance of network simulation
for the development of an efficient WBAN architecture.
Acknowledgement
Author wishes to acknowledge contributions of Mr. Garrick Bulger

and Mr. Benjamin Harding. Mr. Bulger graduated with B.E. Electrical
Engineering from the University of Newcastle in 2007. Mr. Harding
graduated with B.E. Computer Engineering from the University of
Newcastle in 2009.
References
Journals
1. Soomro, A., and Cavalcanti, D. (2007) Opportunities and Challenges in
Using WPAN and WLAN Technologies in Medical Environments, IEEE
Communications Magazine, 45(2), pp. 114–122.
2. Golmie, N., Cypher, D., and Rebala, O. (2005) Performance analysis for
Low Rate Wireless Technologies for Medical Applications, Computer
Communications, 28, pp. 1266–1275.
3. Omeni, O., Wong, A. C. W., Burdett, A. J., and Tomazou, C. (2008) Energy
Efficient Medium Access Protocol for Wireless Body Area Networks,
IEEE Transactions on Biomedical Circuits and Systems, 2(4), pp. 251–
258.
4. Demirkol, I., Ersoy, C., and Boazici, F. A. (2006) MAC Protocols for
Wireless Sensor Networks: A Survey, IEEE Communications Magazine,
44(4), pp. 115–121.
5. Su, H., and Zhang, Xi. (2009) Battery Dynamics Driven TDMA MAC
Protocols for Wireless Body Area Monitoring Networks in Healthcare
References 293
Applications, IEEE Journal on Selected Areas in Communications, 27(4),

pp. 424–434.
6. Sharon, O., and Altman, E. (2001) An Efficient Polling MAC for Wireless
LANs, IEEE/ACM Transactions on Networking, 9(4), pp. 439–451.
7. Gutierrez, J. A., Naeve, M., Callaway Ed., Bourgeois, M., Mitter, V., and
Heile, B. (2001), IEEE802.15.4: A Developing Standard for Low Power
Low Cost Wireless Personal Area Network, IEEE Network, 15(5), pp.
12–19.
8. Ullah, S., Higgins, H., Islam, S. M. R., Khan, P., and Kwak, K. S. (2009) On
PHY and MAC Performance in Body Sensor Networks, EURASIP Journal
on Wireless Communications and Networking, 2009, pp. 1–7.
9. Yuce, M. R., Ng, P. C., and Khan, J. Y. (2008) Monitoring of Physiological
Parameters From Multiple Patients Using Wireless Sensor Network,
Journal of Medical Systems, 32(5), pp. 433–441.
10. Baek, S. J., Veciana, G. de, and Su, X. (2004) Minimizing Energy
Consumption in Large Scale Sensor Networks Through Distributed Data
Compression and Hierarchical Aggregation, IEEE Journal on Selected
Areas of Communications, 22(6), pp. 1130–1140.
Book
11. Khan, J. Y., and Yuce, M. R. (2010) Wireless Body Area Network
(WBAN) for Medical Applications, in New Developments in Biomedical
Engineering, (ed. Campolo, D.), In-Teh, Croatia, pp. 591–627.
12. Leon-Garcia, A., and Widjaja, I. (2004) Communication Networks:
Fundamental Concepts and Key Architecture, McGraw Hill, New York.
13. Karl, H., and Willig, A. (2005) Protocols and Architecture for Wireless
Sensor Networks, John Wiley, England.
14. Ganz, A., Ganz, Z., and Wongthavarawat, K. (2004) IEEE802.15,
in Multimedia Wireless Networks: Technologies, Standards and QoS,
Prentice Hall, New Jersey, pp. 165–185.
Proceedings
15. Chan-Soo , H., Kibeom, S., and Cioffi, J. M. (2006) Opportunistic p-
persistent CSMA in Wireless Networks, in Proceedings of the IEEE
International Communications Conference, pp. 183–188.
16. Timmons, N. F., and Scanlon, W. G. (2004) Analysis of the Performance of
IEEE 802.15.4 for Medical Sensor Body Area Networking, in Proceedings
of the IEEE SENCON, pp. 16–24.
17. Zhang, Y., and Dolman, G. (2009) A New Priority-Guaranteed MAC

Protocol for Emerging Body Area Networks, in Proceedings of the IEEE
International Conference on Wireless and Mobile Communications, pp.
140–145.
18. Li, C., Wang, L., Li, J., Zhen, B., Huan-Bang, Li., and Kohno, R. (2009)
Scalable and Robust Medium Access Control in Wireless Body Area
Networks, in Proceedings of the 20th IEEE International Symposium on
Personal, Indoor and Mobile Radio Communications 2009, pp. 2127–
2131.
19. Khan, J. Y., Yuce, M. R., and Harding, B. (2010) Battery Life Cycle
and Transmission Power Profile Analysis of a Wireless Body Area
Network With Implanted Nodes, in Proceedings of the 4th International
Symposium on Medical Information and Communication Technology.
20. Khan, J. Y., Yuce, M. R., and Karami, F. (2008) Performance Evaluation of
a Wireless Body Area Sensor Network for Remote Patient Monitoring,
in Proceedings of the 30th Annual International IEEE Engineering in
Medicine and Biology Society Conference.
Technical Report
21. Shnayder, V., Chen, B.-R., Lorincz, K., Fulford-Jones, T. R. F., and Welsh, M.
(2005) Sensor Networks for Medical Care, Technical Report TR-08-05,
Division of Engineering and Applied Sciences, Harvard University.
Website
22. http://www.ieee802.org/15/pub/TG6.html.
23. http://www.opnet.com.
Chapter 10
Power Management in Body Area

Networks for Health Care Applications
Vijay Sivaraman,a Ashay Dhamdhere,b and Alison Burdettb

a School of Electrical Engineering and Telecommunications,
The University of New South Wales, Sydney, NSW 2052, Australia.

b Toumaz Technology Limited, 115 Milton Park,
Abingdon, OX14 4RZ, United Kingdom.

{vijay, ashay}@unsw.edu.au; [email protected]
For an increasing number of people living with chronic medical

conditions, wearable wireless sensor devices can provide nonin-
trusive yet continuous physiological monitoring, allowing effective
clinical management without compromising quality of life. Truly
non-intrusive sensor devices will have light weight and small form
factor, placing fundamental constraints on the available energy,
which necessitates very careful energy management at all layers.
This chapter investigates the opportunities and challenges in
the use of dynamic radio transmit power control for prolonging
the lifetime of such energy-constrained, body-wearable sensor
devices.
We first present extensive empirical evidence that the wireless
link quality can change rapidly in body area networks (BAN), and
a fixed transmit power results in either wasted energy (when the
link is good) or low reliability (when the link is bad). We quantify

www.panstanford.com
296 Power Management in Body Area Networks for Health Care Applications
the potential gains of dynamic power control in body-worn devices

by benchmarking off-line the energy savings achievable for a given
level of reliability. We then propose a class of schemes feasible
for practical implementation that adapt transmit power in real
time based on feedback information from the receiver. We profile
their performance against the off-line benchmark and provide
guidelines on how the parameters can be tuned to achieve the
desired trade-off between energy savings and reliability within the
chosen operating environment. Finally, we implement and profile
our scheme on a MicaZ mote-based platform, and also report
preliminary results from the ultra-low-power integrated health care
monitoring platform we are developing at Toumaz Technology. Our
work sets the stage for a holistic approach to power management,
incorporating innovations across all layers of BAN design.
10.1 Introduction
Lifestyle changes combined with an aging population and poor diet

are contributing to an ever-increasing number of people living with
chronic medical conditions requiring ongoing clinical management.
This has resulted in a heavy burden on health care systems that are
primarily geared toward treating acute conditions. Wireless sensor
network technologies have the potential to offer large-scale and
cost-effective solutions to this problem. Outfitting patients with tiny,
wearable, vital-signs sensors would allow continuous monitoring
by caregivers in hospitals and aged-care facilities, and long-term
monitoring by individuals in their own homes.
To successfully deploy BANs that can perform long-term and
continuous health care monitoring, it is critical that the wearable
devices be small and lightweight, lest they would be too intrusive
on patient’s lifestyle. This places fundamental limitations on the
battery energy available to the device over its lifetime. Typical
prototype devices in use today, such as MicaZ motes [1] used in
Harvard’s CodeBlue [2] project, operate on a pair of AA batteries that
provide a few watt-hours (a few tens of kilo-joules) of energy. Truly
wearable health monitoring devices are emerging that have orders
of magnitude lower battery capacity — at Toumaz Technology we
Introduction 297
Figure 10.1. Toumaz Sensium TM digital Plaster. See also Color Insert.
are building a new generation of single-chip low-cost disposable

“digital plasters” (shown in Fig. 10.1) that provide nonintrusive
ultra-low power monitoring of ECG, temperature, blood glucose,
and oxygen levels. Our Sensium TM chip operates on a flexible
paper-thin printed battery [3] with a capacity of around 20 mW-h
(approximately 70 J). Such stringent energy constraints necessitate
very careful energy management.
Communication is the most energy-consuming operation that a
sensor node performs [4] and can be optimised at multiple layers
of the communication stack. At the physical layer, we at Toumaz
have innovated an ultra-low-power radio [5] suited to BANs: our
radio provides a proprietary 50 Kbps wireless link over a distance
of 2–10 m and consumes 2.7 mW at a transmit strength of −7 dBm
(compare this to the CC2420 radio [6] in MicaZ motes that consumes
22.5 mW for −7 dBm output). At the data-link layer, energy can
be saved by intelligent medium access control (MAC) protocols that
duty-cycle the radio, i.e., by turning the radio off whenever packet
transmission or receipt is not expected. Several such MAC protocols
have been developed in the literature (see [7] for a survey). The
B-MAC [8] protocol included in the TinyOS distribution provides
versatility to the application in controlling the duty-cycling of the
radio, while at Toumaz we have developed our proprietary MAC
protocol [9] suited to BANs. However, these MAC protocols only
control when the radio is switched on, they do not determine the
output power of the radio when it is on. The focus of this work is to
study the impact of transmission power (for any given transmission

schedule) in trading off reliability of the time-varying wireless
link for energy efficiency at the transmitting node. We note that
the ability to control the transmission power is available on most
platforms: the CC2420 radio in Crossbow’s MicaZ motes provides
32 transmission levels (ranging from −25 dBm to 0 dBm output)
selectable at run-time by configuring a register, while our Sensium TM
platform similarly supports eight levels (ranging from −23 dBm to
−7 dBm output).
First, we present extensive empirical trace data that profiles
the temporal fluctuations in the body area wireless channel. The
large variations show fixed transmit power to be suboptimal: when
link quality is poor, low transmit levels result in reduced reliability,
whereas when link quality is good, high transmit levels waste
energy. Furthermore, the rapid variations render existing schemes
(discussed in detail in Section 10.2), that adapt transmit power over
long time scales (hours and days), inappropriate for use in BANs.
Using trace data we compute off-line the “optimal” power control
scheme, i.e., one that minimizes energy usage subject to a given
lower bound on reliability. Though infeasible to realize in practise,
the optimal gives insight into the potential benefits and fundamental
limitations of adaptive power control in BANs, and also provides a
benchmark against which practical schemes can be compared.
Second, we develop a class of practical on-line schemes that
dynamically adapt transmission power based on receiver feedback.
These schemes are easy to implement and can be tuned for desired
trade-off between energy savings and communication reliability.
We show conservative, balanced, and aggressive adaptations of our
scheme that progressively achieve higher energy savings (14–30%)
in exchange for higher packet losses (up to 10%). We also provide
guidelines on identifying algorithm parameter settings appropriate
to application requirements and operating conditions.
Third, we present a real-time implementation of our power
control scheme on a MicaZ mote-based platform, demonstrating
that energy savings are achievable even with imperfect feedback
information. We also present preliminary observations of the effi-
cacy of our scheme in the ultra-low power platform for continuous
Related Work 299
health care monitoring being developed at Toumaz Technology. Our

work shows adaptive transmit power control as a low-cost way
of extending the battery-life of severely energy-constrained body
wearable devices and opens the doors to further optimizations
customised for specific deployment scenarios.
The rest of this chapter is organised as follows: Section 10.2
briefly discusses prior work in the area of adaptive power control.
Section 10.3 presents empirical observations on channel variability
in BANs and motivates dynamic power control. Optimal off-line
power control is explored in Section 10.4, while practical on-line
schemes are proposed and analyzed in Section 10.5. Section 10.6
describes our implementation and experiments on the MicaZ mote
and Toumaz Sensium TM platforms, while conclusions and directions
for future work are presented in Section 10.7.
10.2 Related Work
Transmit power control has been studied extensively in the

literature in several contexts with different objectives. A large
number of works, e.g., references [10–15], consider the IEEE 802.11
environment for wireless ad hoc networks and propose adjusting
transmit power or transmit rate for data packets based on several
factors such as wireless channel conditions (probed via RTS/CTS
messages), payload length, etc. In spite of the useful insights from
these works, there are significant differences between the IEEE
802.11 WLAN (wireless local area network) environment and a
BAN for health care monitoring such as: (a) WLAN devices send
sporadic data while BAN devices typically send data periodically,
(b) packet sizes are much smaller in BANs than in WLANs, and (c)
BAN devices operate on or very near the human body, which makes
radio propagation (and hence channel conditions) in BANs markedly
different from WLANs. These important differences merit study of
power control in the specific context of BANs, which to the best of
our knowledge has not been undertaken to date.
A large body of work in power control has also targeted multi-
hop networks with the objective of enhancing throughput [16],
increasing connectivity [17, 18] or reliability [19], and reducing

delays [20]. Joint routing, scheduling, and power control schemes
have also been proposed [21, 22]. Our work considers a single-hop
BAN where such issues do not arise. Moreover, the environment
presented by a BAN is much more dynamic than the psuedo-static
scenarios considered by these works.
More relevant to this work are existing power control schemes
that target energy savings in wireless sensor networks. The study
in refference [23] proposes two algorithms that adapt transmission
power by exchange of information among nodes and based on
signal attenuation, while reference [24] proposes a linear prediction
model for estimating the optimal transmission power based on
measured link quality. However, these studies have targeted static
deployments (such as for environmental or structural monitoring
applications) wherein variability in wireless link quality has been
shown empirically [25, 26] to be slow. In contrast, this work
considers wearable mobile devices for which the wireless link
quality can change significantly and rapidly since it is very
susceptible to position and orientation of the human body [27].
To the best of our knowledge, adaptive power control for body-
wearable devices has not been explored by other researchers
before.
Two recent papers make interesting observations that are
complementary to our work: the authors in reference [28]
investigate the number of different power levels that can be
effectively leveraged by power control algorithms in an indoor
WLAN environment. They show that, due to multipath and fading
effects, there is significant overlap between the RSSI distributions
for nearby power levels, making them practically indistinguishable
at the receiver, and claim that as few as four power levels may suffice
to make power control attractive. We leverage this fact when we
shift from the MicaZ (32 power levels) to the Sensium TM (8 power
levels). Lastly, the study in reference [29] points out that power
control by itself does not lead to significant energy savings unless
complemented by a good MAC protocol that has low duty-cycling
of the radio. We concur with this observation and note that the
MAC protocol used in the Sensium TM platform has very low duty
cycle.
The Case for Transmit Power Control in Body Area Networks 301
10.3 The Case for Transmit Power Control in

Body Area Networks
We begin by empirically profiling the temporal variations in the

quality of the wireless link between a body-worn device and
a fixed base-station, as a patient wearing the device performs
various activities. The patient was played by the first author. Our
experiments in this section use the MicaZ motes from Crossbow
Technologies [1], while some preliminary results that use the
Toumaz Sensium TM platform are presented in Section 10.6. In each
experiment the device was strapped around the patient’s chest,
simulating continuous monitoring of heartbeat and ECG. We also
conducted several experiments in which the device was strapped
around the patient’s arm (for monitoring blood pH and glucose); the
results were qualitatively similar and are omitted here due to lack of
space. The experiments were conducted indoors in an office space
containing 10 cubicles. The base-station was placed close to one side
of the room at an elevation (atop a shelf) to provide better line-of-
sight coverage across the office space.
The MicaZ mote operates in the 2.4 GHz frequency band and can
support a 250 Kbps data rate. It supports 32 RF output power levels,
controllable at run-time via a register; the output power (in dBm)
and corresponding energy consumption rate (in mW) for various
levels are shown in Table 10.1. Since our goal is to save energy at
the body-worn device (which typically has lower energy resources
than the base-station), our experiments involve emitting packets
periodically from the body-worn device at various power levels,
Table 10.1. Characteristics of the MicaZ CC2420 radio
Transmit level Output (dBm) Power (mW)

31 0 31.3
27 −1 29.7
23 −3 27.4
19 −5 25.0
15 −7 22.5
11 −10 20.2
7 −15 17.9
3 −25 15.3
and measuring the link quality at the receiver (base-station). Two

metrics of link quality are available on the mote platform: received
signal strength indicator (RSSI), which is computed internally in
the radio by averaging the signal power over eight symbol periods
of the incoming packet, and link quality indicator (LQI) metric,
which measures the chip error rate for the first eight symbols of
the incoming packet. For a static scenario, previous research [24,
30, 31] has observed LQI readings to suffer from early saturation
and be relatively less stable. We conducted experiments to verify
this in a BAN scenario. Figure 10.2a,b show the RSSI and LQI as
a function of transmit power level for various scenarios of patient
distance and orientation relative to the base-station. While the RSSI
seems to show a smooth increase with transmit power, the LQI
saturates early on for several scenarios and also seems to exhibit
a large variance (shown as error-bars), which makes the readings
less reliable. Figure 10.2c,d show the RSSI and LQI as a function of
Figure 10.2. Comparison of RSSI and LQI as indicators of channel quality

in a BAN.
The Case for Transmit Power Control in Body Area Networks 303
the packet receive ratio (PRR) observed across several experiments.

While the RSSI seems to provide a good estimate of packet loss rates
(e.g., RSSI of −90 dBm or larger always corresponds to PRR of 95%
or more), the LQI is seen to be a much weaker indicator of PRR due
to its high variance. Our subsequent study, therefore, uses RSSI as an
indicator of channel quality.
We profile, at different radio transmit levels, the changes in
link quality with time as patients perform their routine activities
involving resting, moving, turning, etc. To compare link quality
at different power levels, we should ideally take simultaneous
measurements at all power levels, which is infeasible. As an
approximation, we make the body-worn device transmit every
packet multiple times in quick succession at 16 different transmit
levels 31, 29, 27, . . . 1. The receiver (base-station) can thus record,
more-or-less simultaneously, the signal strength corresponding to
each transmit level. Measurements for three scenarios are described
next.
10.3.1 Normal Walk

This scenario has the patient walking back and forth in the room for
a few minutes at a normal walking pace; the patient stays between
1 and 8 m from the base-station at all times. The body device,
strapped onto the patient’s chest, generates a packet every second
(which is not entirely unrealistic for a heartbeat/ECG monitor) and
transmits it at 16 different output levels. The RSSI is recorded
at the base-station for each packet at each transmit level, and
plotted in Fig. 10.3a against time for four of the transmit levels.
For any fixed transmit power, the received signal strength fluctuates
widely: at fixed maximum transmit output (level 31 at 0 dBm), the
signal strength at the receiver changes from −64 dBm (at 34 s)
to −94 dBm (at 86 s — a change of 30 dBm under a minute.
There are nevertheless some discernible trends: for example, in the
interval 30–50 s, the received signal is consistently above −72 dBm
(at the maximum transmit level) due to the clear line-of-sight
presented by the patient walking toward the base-station, while the
subsequent interval 50–70 s exhibits RSSI below −75 dBm (again
at the maximum transmit level) due to the patient turning and
Figure 10.3. RSSI vs. time for various patient scenarios. See also Color
Insert.
blocking the line-of-sight with his body. The question of whether

these patterns present opportunities for energy savings by adapting
transmit power will be tackled in Section 10.4.
10.3.2 Slow Walk

In this scenario, we consider a slow-moving person (such as an
elderly or a handicapped person with restricted mobility) who takes
an exaggeratedly long time (over 6 min) to walk a distance of
3 m and back. As before, packets are transmitted every second at
several power levels, and the received signal strength at the base-
station is recorded and depicted in Fig. 10.3b. The trend in the
plot is very evident: the RSSI is fairly low for the first half, when
the patient’s body blocks the line-of-sight between the body-worn
device and the base-station, and then rises to a perceptibly higher
value in the second part of the experiment when the patient is
walking facing the base-station (barring the last few seconds when
Optimal Off-Line Transmit Power Control 305
the patient turns again). This scenario depicts the shortcomings of

fixed transmit power: a low transmit level would result in weak
signals (and packet loss) during the first half, while a high transmit
level would unnecessarily waste energy in the latter half. Such a
scenario, therefore, presents opportunities for saving energy by
reducing transmit power adaptively when the good channel persists.
10.3.3 Resting
In this scenario, the patient sits down to rest for approximately
20 min on a chair at a distance of about 6 m from the base-
station. Fig. 10.3c plots the RSSI over the entire period, at several
transmit levels. The wireless link is found to be fairly stable when
the patient is at rest (in spite of a few other people moving around
at several points in the experiment). This is in some sense an
“ideal” environment with tremendous potential for energy savings,
particularly with patients who are resting for a major part of the
day. These energy savings would be unattainable if the transmit level
were fixed, since a fixed setting would have to cater to the worst-case
scenario of a poor channel.
Having gained an understanding of the wireless channel under
various patient activity scenarios, the next section quantifies the
potential benefits of adaptive transmit power control.
10.4 Optimal Off-Line Transmit Power Control
To quantify the potential benefits of adaptive transmit power

control, we compute what the “optimal” transmission level might be
for each of the scenarios considered before. We define the “optimal”
as the lowest required transmit power level (as a function of time) to
achieve a minimum target RSSI. Based on our studies in Section 10.3
relating packet loss with RSSI for the MicaZ motes in a BAN setting,
we choose a conservative target RSSI of −85 dBm. The computation
of the optimal transmission level defined thus is done off-line, i.e.,
using the traces shown in the previous section. For each scenario, at
each time instant, we know the RSSI for each transmit power level,
and we can, therefore, identify the lowest transmit power at which
the signal strength at the receiver is no lower than the threshold
of −85 dBm (if all received signal strengths are below the lower
threshold we set the transmit level to be the maximum). We note
that such a scheme is not implementable in practice, since it would
require the transmitter to have instantaneous knowledge of the RSSI
at the receiver for each choice of transmit power level, which is
infeasible given that the channel varies with time.
The optimal transmit levels, and their associated RSSI values, for
each of the three scenarios are depicted as a function of time in
Fig. 10.4. Subplot (a) shows, for the normal walk scenario, that the
optimal changes rapidly to track the rapid fluctuations in channel
quality, thereby maintaining a fairly stable RSSI as shown in sub-
plot (b): for example, in the time interval 50–75 sec, the optimal
transmit level fluctuates multiple times between a high of 29 and
a low of 9. Based on the energy draw for each transmit power level
(shown in Table 10.1), we can compute the energy savings of optimal
power control to be around 34% as compared to using the maximum
transmit power. However, as the rapid fluctuations in the optimal
level indicates, a practical scheme is unlikely to be able to predict
the current optimal transmit level based on prior channel quality.
The optimal transmit power for a slow walk in Fig. 10.4c shows
high sensitivity to body orientation, even when the motion is very
slow. The rapid changes during the first 200 s arise from minor
variations in the patient’s body orientation while blocking the line-
of-sight between the body-worn device and the base-station (indeed
a few packets are lost even at the highest transmit power). But
when the patient turns (at approximately 200 s), there is a clear
line-of-sight, and the wireless link is relatively stable permitting the
optimal transmit power to remain low for a considerable length of
time (more than 2 min). This indicates that if the body orientation is
favorable, periods of slow activity could be capitalized by a transmit
control scheme to save energy without compromising realiability.
When the patient is resting, the link is fairly stable and
the optimal transmit power level is near-constant as shown in
Fig. 10.4e, which permits an energy savings of over 38% compared
to maximum transmit power. It would seem the quiescent wireless
channel in this case gives ample opportunity for practical schemes
to reduce transmit power without sacrificing reliability. The design
of such schemes is discussed next.
Practical On-Line Transmit Power Control 307
Figure 10.4. Optimal transmit power and associated RSSI for a normal
walk, slow walk, and resting position. See also Color Insert.
10.5 Practical On-Line Transmit Power Control
The optimal transmit power control scheme above was performed

off-line and required the sender to have a priori knowledge of the
link quality at the receiver, which is infeasible in reality. This section
develops practical algorithms that are then benchmarked against the

optimal.
A predictive approach to designing on-line schemes for power
control would require a wireless channel model for BANs. Modeling
the propagation of electromagnetic waves around the human body,
e.g., via “creeping waves” [32], is fairly complex as it needs to account
for the permittivity and conductivity of the different layers of bone
and tissue in the human anatomy. Additionally, the model has to
contend with changes in orientation of the human body, mobility of
the patient, and other spatio-temporal aspects (such as room layout,
people in the vicinity, etc.). We believe such predictive models are
too complex to implement on energy-constrained wearable devices
and do not pursue them in our work.
We focus instead on reactive schemes that adjust transmit
power based on feedback from the receiver (inspired by the way
TCP adjusts its transmission rate in reaction to congestion in the
Internet). Specifically, the base-station measures the RSSI for each
received packet and feeds it back to the body-worn device in
the acknowledgment packet. In this section, we assume that the
feedback information is perfect (i.e., acknowledgment packets are
never lost); this assumption will be relaxed in Section 10.6.1.
10.5.1 A Simple and Flexible Class of Schemes

At its core, any reactive power control scheme must ramp up
transmit power when the channel quality deteriorates (so as to avoid
packet loss) and decrease transmit power when the channel quality
improves (in order to save energy). We propose a general class of
schemes that allow these operations to be tuned via appropriate
parameter settings.
Algorithm 10.1 depicts our class of schemes and is characterized
by four parameters: αu , αd , TL , and TH . The scheme maintains a
running average R̄ of the RSSI, computed by exponentially weighted
averaging (steps 2 and 4) of each newly obtained sample. The
averaging weight αu for a sample representing an improving channel
can, in general, be different from the weight αd used for a sample
representing a deteriorating channel; this gives flexibility to a
scheme in reacting differently to a perceived increase or decrease
Algorithm 10.1 A class of power control schemes

Require: R {RSSI from the current sample}
Require: R̄ {Average RSSI}
1: if R ≤ R̄ then
2: R̄ ← αd R + (1 − αd ) R̄
3: else {R > R̄}
4: R̄ ← αu R + (1 − αu ) R̄
5: end if
6: if R̄ < TL then
7: Double the transmit power
8: else if R̄ > TH then
9: Reduce the transmit power by a constant
10: else {TL ≤ R̄ ≤ TH }
11: No action is required
12: end if
in channel quality, thereby placing different emphasis on energy

savings versus packet loss.
The scheme increases or reduces transmit power by comparing
the running RSSI average R̄ to lower and upper thresholds TL
and TH . Based on our previous analysis of RSSI and packet loss
(Fig. 10.2c), we believe a lower threshold TL = −85 dbm is
appropriate for the MicaZ platform. If R̄ falls below this threshold
(step 6), the transmit power is immediately doubled (step 7) to
avoid imminent packet loss in the deteriorating channel. If, on the
other hand, the average RSSI exceeds an upper threshold TH (step
8), the transmit power is reduced by a small amount (step 9). Our
experimental traces indicate that TH = −80 dBm is appropriate
for the MicaZ platform; lower values make the target RSSI range
[TL , TH ] too narrow, while larger values lead to higher transmit
power (and hence higher energy usage) than required. Readers may
note the similarity of our scheme to TCP in that the transmit power
increase is multiplicative while the decrease is additive, and thus has
a clear bias toward reacting faster to a deteriorating channel than an
improving one.
The class of schemes outlined in algorithm 10.1 is fairly easy
to implement in platforms with very limited CPU and memory
resources. It is also very flexible, and the parameters αu and

αd can, in particular, be tuned for appropriate trade-off between
energy efficiency and reliability. The performance of the schemes for
specific parameter settings is discussed next.
10.5.2 Example Adaptations of the General Scheme

A conservative approach to energy savings might be warranted in
applications where data loss is critical. A relatively high value αd =
0.8 could be used so that a low RSSI value triggers a rapid ramp-
up in transmit power, while keeping αu = 0.2 low so that transmit
power is reduced cautiously when good channel conditions prevail.
Applications in which energy is at a high premium and data loss is
not as critical may adopt an aggressive strategy with a high αu = 0.8
that reacts quickly to improvements in channel conditions while
setting αd = 0.2 so that a transient bad channel is ignored. A
balanced approach that equally values energy savings and packet
loss may place equal emphasis on whether the channel is getting
better or worse by setting αu = αd = α = 0.8.
We tested the efficacy of the conservative, aggressive, and
balanced schemes as described above on the trace data for the
three scenarios described earlier. The transmitter is assumed to
know the RSSI for each transmitted packet via feedback from the
receiver; it performs the exponential averaging using the parameters
listed for each scheme, and chooses an appropriate transmit power
level for the subsequent packet transmission. Figure 10.5 shows
the transmit power level and the corresponding RSSI for each of
the three schemes for each scenario. The average power draw per
packet, as well as packet loss rates, for each of the schemes under
the different scenarios is summarized in Table 10.2.
For normal walk, Fig. 10.5a shows that all schemes exhibit
considerable fluctuation in their transmit power, since the channel
varies quite rapidly. The conservative scheme yields good link
reliability (only 1.4% loss) but has high energy usage (30.58%
above optimal), while the aggressive scheme yields good energy
savings (5.73% within optimal) by sacrificing link quality (9.6%
packet loss). As one might expect, the energy usage and packet loss
Figure 10.5. Transmit power and RSSI under conservative, balanced, and
aggressive schemes for various scenarios.
under the balanced scheme are between those of the conservative

and aggressive.
For the slow walk scenario, Fig. 10.5 shows the aggressive
scheme to be fairly stable in the interval 210–350 s when the
patient is walking slowly facing the base-station, whereas the
conservative scheme shows some fluctuations in that region. Again,
Table 10.2. Power draw averaged across packets, and loss rate, for various
power control schemes
Normal Walk Slow Walk Resting

power loss power loss power loss
Scheme (mW) (%) (mW) (%) (mW) (%)
Maximum 31.32 0 31.32 1.2 31.32 0
Optimal 20.60 0 20.74 1.2 22.35 0
Conservative 26.90 1.4 25.52 2.18 24.36 0.08
Balanced 25.15 3.85 24.11 3.85 24.22 0.16
Aggressive 21.78 9.6 21.96 7.28 23.74 0.32
the conservative scheme uses 16.21% more energy than the

aggressive scheme, but has a packet loss rate much lower than
the aggressive scheme, showing that energy and reliability can be
traded-off in our schemes by choosing parameters appropriately. As
before, the balanced scheme lies between the other two in both its
metrics.
The transmit level under the three schemes for the resting
scenario, shown in Fig. 10.5e, is fairly stable, barring some rogue
glitches in the link quality (e.g., at 905 s) that the conservative
scheme over reacts to. The energy savings, as well as the packet
loss ratios, are comparable under all three schemes, indicating that
under quiescent channel conditions the parameter settings do not
influence the algorithm performance significantly.
10.5.3 Tuning the Parameters

In this section, we undertake a more detailed study of the impact
of the algorithm parameters αu and αd on the performance of
our class of schemes. We conducted several experiments in which
the patient is fairly active, since the parameters have a larger
impact on energy and loss under such scenarios. Figure 10.6
plots the energy consumption (left column) and packet loss (right
column) of our scheme for various parameters settings for a
representative scenario. Figures 10.6a,b show that for any fixed αd ,
the energy consumption falls monotonically with αu : this is because
a larger αu makes the scheme react faster to an improving channel,
Figure 10.6. Energy consumption and packet loss for various parameter
settings.
thereby reducing transmit power to save energy. When αu is held

constant and αd increases, the scheme reacts more quickly to a
degrading channel, and the packet loss rate diminishes (Fig. 10.6d)
at the expense of increased energy consumption (Fig. 10.6c). The
plots show clearly that increasing αu pulls the algorithm in the
direction favoring energy savings, while increasing αd pulls it toward
reliability — these opposite trends justifying our decision to devise a
scheme with two separate α values. For a balanced scheme that uses
αu = αd = α, Fig. 10.6e,f show that a scheme that is very reactive

to both good and bad channel conditions (i.e., high α) improves loss
performance as expected, but incurs a penalty in terms of slightly
increased energy usage.
An interesting question concerns the optimization of the para-
meters for a given operating scenario. For the trace considered in
this section, we found that (αu = 1.0, αd = 0.3) yields maximum
energy savings when the loss budget is high at 15%, implying that
the ramp-down in transmit power should be immediate while a
ramp-up happens slowly over several poor RSSI samples. When the
loss margin is stringent at 5%, energy savings are maximized for
(αu = 0.8, αd = 1.0), i.e., the scheme should react immediately to
a degrading channel and relatively slowly to an improving channel.
Though the optimal settings obtained here do not transfer directly
to other scenarios, they provide insight which could conceivably be
used by a system to adjust the parameters at run-time based on
application requirements and operating conditions.
10.6 Prototyping and Experimentation
This section reports on a real-time implementation of our power

control schemes on a MicaZ mote-based testbed, as well as
preliminary experiments on the Toumaz Sensium TM platform.
10.6.1 MicaZ Mote Platform

The previous section evaluated the performance of the transmit
power control schemes using trace data wherein the sender (body-
worn device) is assumed to know the RSSI at the receiver (base-
station) for each of the packets it has thus far transmitted. In
a real operational setting, such feedback information would be
contained in acknowledgment packets (from the receiver to the
sender), which may also experience loss. This section undertakes a
real-time implementation of our scheme on body-worn MicaZ motes
to evaluate the efficacy of our power control scheme under imperfect
feedback. We note that the base-station is assumed to have abundant
energy reserves and does not implement power control, always
using the highest power level for transmitting acknowledgments.
Prototyping and Experimentation 315
Figure 10.7. Transmit power and associated RSSI for real-time implemen-
tation of balanced scheme on the MicaZ platform.
The implementation performs exactly the steps shown in

algorithm 10.1, with the additional step that if an acknowledgment
is not received for a transmitted packet, the RSSI for that sample
is taken to be −100 dBm (thereby signaling a bad channel to the
algorithm). We present results with balanced parameter setting
αu = αd = α = 0.8 for a scenario where the patient undertakes
a mix of walking and resting. Figure 10.7a shows the transmit power
level under our scheme and Fig. 10.7b the corresponding RSSI. Our
scheme performs quite well, yielding average energy consumption
rate per packet of 20.23 mW (a 35.4% savings in energy compared
to using maximum transmit power), and packet loss rate of 3.8%.
Unfortunately, there is no way to benchmark this result (recall that
computing the optimal requires a trace that includes the RSSI for
all power levels at all times), but a careful look at the plots will
show regions where the RSSI is high and yet the transmit power is
not reduced, for example in the interval (410, 420) s. It was found
that several acknowledgment packets were lost in this interval (the
overall acknowledgment loss rate for this scenario was 11.82%),
and the absence of feedback information forced our scheme to use
a higher transmit power level than would have been necessary
with perfect feedback. High acknowledgment loss rates arising from
asymmetric link qualities can have a negative impact on the efficacy
of feedback-based power control schemes and merit deeper study in
future work.
10.6.2 Toumaz Sensium TM Platform

A major goal of this project is to evaluate the costs and benefits of
adaptive power control in the real-world continuous health monitor-
ing platform being developed by Toumaz Technology. Unfortunately,
the digital plaster (which includes the Sensium TM chip, printed
battery, etched antenna, and water-protective covering) is still in the
packaging process; so we present here preliminary results obtained
from using a Sensium TM chip mounted on a development board
strapped to the patient’s chest, as preliminary indicators on the
feasibility and benefits of adaptive transmit power control in the
Sensium TM platform.
The hardware-optimized design of the Sensium TM presented
several constraints in our experimentation: (i) the radio has only
eight transmit power levels (unlike 32 in the MicaZ radio), and
the energy savings are upper-bounded at 35% due to the limited
transmission power range (Table 10.3a), (ii) the RSSI readout is
only 3 bits (compared to 8 bits in the MicaZ), which gives only
a coarse estimate of the RSSI (Table 10.3b), (iii) a sleep time of
at least 1 s is imposed between successive packet transmissions
(for energy efficiency), which restricts our ability to sample the
channel at various transmit power levels without substantial change
in patient position/orientation, (iv) the RSSI feedback from receiver
to sender requires software modification of the acknowledgment
packets, which introduces a 1 s lag in the feedback. These limitations
not withstanding, we believe there is great value in experimenting
Table 10.3. Sensium TM Radio transmit and receive characteristics (a)

Transmit Characteristics (b) Receive Characteristics
Tx level output (dBm) power (mW)
7 −6 2.8 7 > −35

6 −7 2.7 6 −46 to −35
5 −9 2.6 5 −52 to −46
4 −10 2.5 4 −58 to −52
3 −12 2.4 3 −64 to −58
2 −15 2.2 2 −70 to −64
1 −18 2.0 1 −76 to −70
0 −22 1.8 0 < −76
Prototyping and Experimentation 317
Figure 10.8. RSSI vs. time using the Sensium TM platform. See also Color
Insert.
with this real-world wearable device, and these limitations can be

addressed in subsequent revisions of the Sensium TM hardware.
As with the motes, we strap the Sensium TM to the patient’s chest
and make it transmit a packet every second, cycling through the
eight available power levels, while the RSSI is recorded at the base-
station. A 10 min extract of the recorded RSSI at various transmit
power levels is shown in Fig. 10.8; in this scenario, the patient
is walking back and forth in the interval 30–80 s, during which
period the RSSI fluctuates rapidly, while in the intervals 280–380
s and 420–470 s the patient is stationary (facing away and toward
the base-station, respectively) and the RSSI is stable. This confirms
that the body area wireless channel presents significant temporal
fluctuations in the 862–870 MHz frequency range (much like the
2.4 GHz range of the MicaZ radio), and fixed transmit power is
suboptimal.
For the above trace, we compute off-line the optimal transmis-
sion power schedule for desired RSSI level of 3 (corresponding
to the range −64 to −58 dBm). Figure 10.9 shows the optimal
transmit power level (at the sender) and the corresponding RSSI
level (at the receiver) and clearly depicts that the transmit power
level can be reduced during quiescent periods such as 420–470 s,
while preserving reliability during periods when channel conditions
are poor. For this scenario, optimal transmit power control uses
14.63% less energy than fixed maximum transmit power, which
is a significant saving for this severely energy-constrained device.
Figure 10.9. Optimal transmit power and associated RSSI for the
Sensium TM .
Further results from our testing on the Sensium TM platform are

reported in our subsequent work [33].
10.7 Conclusions and Future Work
This chapter outlines the potential benefits and limitations of

adaptive radio transmit power control as a means of saving
precious energy in body-wearable sensor devices used for medical
monitoring. We experimentally profiled the radio channel quality
under different scenarios of patient activity and showed that fixed
transmit power either wastes energy or sacrifices reliability. We
then quantified the theoretical benefits of adaptive transmit power
control and showed that across different scenarios it can save nearly
35% energy without compromising reliability. We then developed
a general class of practical power control schemes suitable for
BANs, and showed specific instances that save 14–30% energy
(as compared to using maximum transmit power) in exchange
for 1–10% packet losses. We demonstrated that the adjustment
of parameters allow our schemes to achieve different trade-offs
between energy savings and reliability, making them suitable across
diverse applications in different operating conditions. Finally, we
demonstrated that a real-time implementation of our scheme on
the MicaZ mote-based platform is effective even in the presence
of imperfect feedback information and presented preliminary
experimental results indicating the potential for saving precious
References 319
energy in Toumaz’s real-world platform for continuous health care

monitoring.
Our work on dynamic power control in BANs, summarized in
this chapter from our publications [33–35], can be extended in
several ways. There is much further study required in exploring its
potential for specific health monitoring environments (e.g., critical
care in hospitals, aged care, athlete monitoring, etc.) which have
different characteristics in terms of patient mobility, periodicity, and
criticality of collected data. There is also scope for more extensive
experimentation with truly wearable health monitoring devices
used by real patients.
References
1. Crossbow-Technologies, Mica2 and MicaZ motes, URL http://www.

xbow.com.
2. V. Schnayder et al. (2005) Sensor Networks for Medical Care, Technical
Report TR-08-05, Division of Engineering and Applied Science, Harvard
University.
3. Power-Paper, Power Patch Platform, URL http://www.powerpaper.com.
4. D. Culler, D. Estrin, and M. Srivastava (August 2004) Overview of Sensor
Networks, IEEE Computers, 37(8), pp. 41–49.
5. A. Wong, G. Kathiresan, T. Chan, O. Eljamaly, and A. Burdett (Sep-
tember 2007) A 1 V Wireless Transceiver for an Ultra Low Power
SoC for Biotelemetry Applications, in ESSDERC/ESSCIRC, Munich,
Germany.
6. Chipcon, CC2420: 2.4 GHz IEEE 802.15.4 / ZigBee-ready RF Transceiver,
URL http://www.chipcon.com.
7. K. Langendoen and G. Halkes (2005) Energy-Efficient Medium Access
Control, Embedded Systems Handbook, CRC Press.
8. J. Polastre, J. Hill, and D. Culler (November 2004) Versatile Low Power
Media Access for Wireless Sensor Networks, in ACM SenSys, Baltimore,
MD, pp. 95–107.
9. O. Omeni, O. Eljamaly, and A. Burdett (August 2007) Energy Efficient
Medium Access Protocol for Wireless Medical Body Area Sensor
Networks, in Proceedings of the IEEE-EMBS Symposium on Medical
Devices and Biosensors, Cambridge, UK, pp. 29–32.
10. S.-L. Wu, Y.-C. Tseng, and J.-P. Sheu (September 2000) Intelligent
Medium Access for Mobile Ad Hoc Networks With Busytones and Power
Control, IEEE Journal on Selected Areas in Communications, 18(9),
pp. 1647–1657.
11. J.-P. Ebert, B. Stremmel, E. Wiederhold, and A. Wolisz (September
2000) An Energy-Efficient Power Control Approach for WLANs,
Journal of Communications and Network, 2(3), pp. 197–
206.
12. J. Monks, V. Bharghavan, and W. Hwu (April 2001) A Power Controlled
Multiple Access Protocol for Wireless Packet Networks, in IEEE Infocom,
Alaska, pp. 219–228.
13. J. Pavon and S. Choi (May 2003) Link Adaptation Strategy for IEEE
802.11 WLAN via Received Signal Strength Measurement, in IEEE ICC,
Anchorage, AK, pp. 1108–1113.
14. E. -S. Jung and N. Vaidya (January 2005) A Power Controlled
MAC Protocol for Ad Hoc Networks, Wireless Networks, 11(1–2),
pp. 55–66.
15. D. Qiao, S. Choi, and K. G. Shin (October 2007) Interference Analysis and
Transmit Power Control in IEEE 802.11a/h Wireless LANs, IEEE/ACM
Transactions on Networking, 15(5), pp. 1007–1020.
16. T. ElBatt, S. Krishnamurthy, D. Connors, and S. Dao (June 2000)
Power Management for Throughput Enhancement in Wireless Ad Hoc
Networks, in IEEE ICC, New Orleans, LA, pp. 1506–1513.
17. R. Ramanathan and R. Hain (March 2000) Topology Control of Multihop
Wireless Networks Using Transmit Power Adjustment, in IEEE Infocom,
Tel-Aviv, Israel, pp. 404–413.
18. M. Kubisch, H. Karl, A. Wolisz, et al. (March 2003) Distributed Algo-
rithms for Transmission Power Control in Wireless Sensor Networks,
in IEEE WCNC, New Orleans, LA.
19. D. Son, B. Krishnamachari, and J. Heidemann (October 2004) Ex-
perimental Study of the Effects of Transmission Power Control and
Blacklisting in Wireless Sensor Networks, in IEEE SECON, Santa Clara,
CA, pp. 289–298.
20. O. Chipara, Z. He, G. Xing, Q. Chen, X. Wang, C. Lu, J. Stankovic, and
T. Abdelzaher (June 2006) Real-Time Power-Aware Routing in Sensor
Networks, in IEEE IWQoS, New Haven, CT, pp. 83–92.
21. G. Xing, C. Lu, Y. Zhang, Q. Huang, and R. Pless (May 2005) Minimum
Power Configuration in Wireless Sensor Networks, in ACM MobiHoc,
Urbana-Champaign, IL, USA, pp. 390–401.
References 321
22. I. C. Paschalidis, W. Lai, and D. Starobinski (February 2007) Asymp-

totically Optimal Transmission Policies for Large-Scale Low-Power
Wireless Sensor Networks, IEEE/ACM Transactions on Networking,
15(1), pp. 105–118.
23. L. Correia et al. (October 2005) Transmission Power Control in
MAC Protocols for Wireless Sensor Networks, in ACM/IEEE MSWiM,
Montreal, Canada, pp. 282–289.
24. S. Lin, J. Zhang, G. Zhou, L. Gu, T. He, and J. Stankovic (November
2006) ATPC: Adaptive Transmission Power Control for Wireless Sensor
Networks, in ACM SenSys, Boulder, CO, pp. 223–236.
25. G. Zhou, T. He, S. Krishnamurthy, and J. Stankovic (June 2004) Impact
of Radio Irregularity on Wireless Sensor Networks, in ACM MobiSys,
Boston, MA, pp. 125–138.
26. J. Zhao and R. Govindan (November 2003) Understanding Packet
Delivery Performance in Dense Wireless Sensor Networks, in ACM
SenSys, Los Angeles, CA.
27. P. S. Hall and Y. Hao (2006) Antennas and Propagation for Body-Centric
Wireless Communications, Artech House.
28. V. Shrivastava, D. Agrawal, A. Mishra, S. Banerjee, and T. Nadeem
(October 2007) Understanding the Limitations of Transmit Power
Control for Indoor WLANs, in ACM/Usenix Internet Measurement
Conference (IMC), San Diego, CA, pp. 351–364.
29. J. Jeong, D. Culler, and J.-H. Oh (June 2007) Empirical Analysis of
Transmission Power Control Algorithms for Wireless Sensor Networks,
in International Conference on Networked Sensing Systems (INSS’07),
Kanazawa, Japan, pp. 27–34.
30. K. Srinivasan, P. Dutta, A. Tavakoli, and P. Lewis (November 2006)
Understanding the Causes of Packet Delivery Success and Failure in
Dense Wireless Sensor Networks, in ACM SenSys, Boulder, CO, pp. 419–
420.
31. K. Srinivasan and P. Lewis (May 2006) RSSI Is Under Appreciated, in
Workshop on Embedded Networked Sensors (EmNets), Boston, MA.
32. J. Ryckaert, P. D. Doncker, R. Meys, A. de Le Hoye, and S. Donnay (April
2004) Channel Model for Wireless Communications Around Human
Body, Electronics Letters, 40(9), pp. 543–544.
33. A. Dhamdhere, V. Sivaraman, and A. Burdett (December 2008) Experi-
ments in Adaptive Power Control for Truly Wearable Biomedical Sensor
Devices, in International Workshop on Adaptation in Wireless Sensor
Networks (AWSN), Sydney, Australia.
34. S. Xiao, A. Dhamdhere, V. Sivaraman, and A. Burdett (January 2009)

Transmission Power Control in Body Area Sensor Networks for Health-
care Monitoring, IEEE Journal on Selected Areas in Communications,
27(1), pp. 37–48.
35. A. Dhamdhere, V. Sivaraman, V. Mathur, and S. Xiao (December 2008)
Algorithms for Transmission Power Control in Biomedical Wireless
Sensor Networks, in IEEE Workshop on Wireless Network Algorithms
(WiNA), Yilan, Taiwan.
Chapter 11
Channel Modeling of Narrowband

Body-Centric Wireless Communication
Systems
Simon L. Cotton and William G. Scanlon

ECIT Institute, Queen’s University Belfast,
Queen’s Island, Belfast, United Kingdom
The modeling of wireless channels in body-centric communication

systems is of paramount importance when designing antennas,
protocols, and wireless transceiver hardware. In this chapter, we
briefly introduce the concept of on-body, off -body, and body-to-body
communication channels. We then focus specifically on the modeling
of narrowband on-body communication channels, which are an
integral part of today’s wireless body area networks (WBANs).
As well as identifying the factors responsible for shaping fading
characteristics in narrowband WBAN systems, we discuss a number
of the models commonly used to describe the first- and second-order
characteristics of amplitude distribution. The Akaike Information
Criterion is introduced as a method of ranking competing channel
models, and a worked example is provided to step the reader
through the modeling process, as well as providing information on
how to simulate the channels. The chapter then concludes with

www.panstanford.com
324 Channel Modeling of Narrowband Body-Centric Wireless Communication Systems
comments on the validity of the analysis presented, especially in the

case of non-isotropic signal reception.
11.1 Introduction to Body-Centric Communications
Recent commercial interest in body-centric communications, wire-

less systems that are designed to operate in close proximity to
the human body, has meant that antennas and propagation have
become important focal points for research. Study in these areas
will help to develop new and innovative solutions and systems
geared toward applications in the consumer, medical, and military
sectors. For example, requirements for specialist mobile health care
have helped to fuel a revolution in body sensor network (BSN)
technology, where networks of miniaturized wearable sensors
with wireless functionality are attached to the human body and
configured to monitor life vital signs such as body temperature,
electrocardiography, and motor activity. Other recent popular uses
of WBAN technology include data exchange between body-worn
cellular devices and wireless headsets, and recreational sports
monitoring.
At present, there are two general areas of body-centric com-
munication channel research, broadly categorized as on-body and
off-body channels. In on-body channels, such as those found in
WBANs, internetworked devices positioned on the human body
will typically use over-the-body surface propagation channels to
communicate as shown in Fig. 11.1. At ultra-high frequency (UHF)
and microwave frequencies, these wireless links are often formed
using on-body surface waves and signal components reflected,
diffracted, and scattered from other body parts. It should be noted
that some links may also utilize free-space propagation (e.g., waist
to wrist, etc.) as well as environmental multipath propagation
whereby signal transmissions from an on-body node are returned
toward the body from the local surroundings. Depending on network
topology, some WBAN applications may employ a single body-
worn master node for central processing of data and for onward
transmission of this information across other wireless networks
(e.g., Wi-Fi, Bluetooth, and cellular) using off-body communications
Introduction to Body-Centric Communications 325
Figure 11.1. Overview of on-body, off-body, and body-to-body

communications. See also Color Insert.
(Fig. 11.1). An emerging type of body-centric communications,

which also utilizes off-body communications, is body-to-body
networking. Body-to-body channels, such as those found in body-to-
body networks (BBNs), occur when a wireless device situated on one
person is communicating with a wireless device situated on another
person.
Much of the current research focus for WBAN applications is
centered on the unlicensed industrial, scientific, and medical (ISM)
bands at 868 MHz, 915 MHz, and 2.45 GHz. In this chapter, we
will focus on narrowband on-body channels at these frequencies.
The interested reader is directed to the following references
and those contained therein for channel studies on ultra-wideband
(UWB) WBANs [1], the use of diversity in WBANs [2–4], and
multiple-input multiple-output WBAN systems [5]. Those wishing to
learn more about off-body and body-to-body channel characteristics
are directed to the following references for narrowband off-body
[6], UWB off-body [7], narrowband and diversity in body-to-body

[8] and wideband body-to-body [9]. In the future, millimeter-
wave frequencies, such as the world-wide unlicensed spectrum
allocations at 60 GHz, will undoubtedly be adopted for high-
bandwidth applications. While the propagation characteristics in
these bands are difficult with severe blocking by almost all building
materials, furniture, and other objects, these same effects can
be used to the advantage of the system designer. For example,
the limited propagation at millimeter-wave frequencies can be
used to mitigate against potential signal-to-interference problems
associated with high densities of co-located WBAN users [10]. Other
advantages of millimeter-wave operation include the opportunities
for device miniaturization and the capacity benefits associated with
short-range spatial spectrum reuse.
11.2 Channel Modeling for Wireless Body Area Networks
Wearable wireless devices for WBANs are required to be compact,

lightweight, robust, unobtrusive to the user, and usually feature
antennas mounted conformal to or in extremely close proximity
to the body surface. These requirements must be met while still
maintaining a high level of performance, reliability, and efficiency.
Clearly, this will introduce design challenges at all layers in the
protocol stack, but physical layer characteristics are known to be
the limiting factor in the performance of body-centric wireless
communication systems.
A key discriminator between body-centric systems and tra-
ditional mobile communications is the unique time- and space-
varying propagation characteristics encountered. This is in part
due to the human body itself, which is an extremely complex
operating environment. Body-worn wireless devices are prone to
antenna-body interaction effects, which include near-field coupling,
radiation pattern distortion, and shifts in antenna impedance, which
may degrade the efficiency of the body-worn system and reduce
signal reliability [11]. Because the body acts as its own frame of
reference, body-centric communication channels are susceptible
to time-varying, received signal characteristics, which may range
Channel Modeling for Wireless Body Area Networks 327
from quasi-periodic to completely stochastic depending on the

person’s current physical state. The electromagnetic problem is
made even more difficult to analyze as signal reception in body-
centric systems may range from isotropic to strongly nonisotropic
[12], depending on the degree of body shadowing, attenuation of
the on-body creeping wave [13] (if present), and the nature of
the local surroundings. Within cluttered multipath environments,
inhomogeneous, irregularly spaced dielectric structures, including
nearby pedestrians, also cause the signal to undergo reflection,
diffraction, scattering, and absorption in varying degrees, creating
a spatially distributed electromagnetic field.
To further complicate the analysis of WBAN systems, the
magnitude of each of these factors will also be dependent on the
geometry of the on-body communication link, i.e., where nodes
are positioned on the body relative to one another. This effect is
demonstrated by the time series shown in Fig. 11.2, which shows
the signal received by nodes positioned on the right ankle and elbow
(as shown in Fig. 11.1) while the user was performing walking
movements in an anechoic environment. While it may be possible
Figure 11.2. Comparison of waist-to-ankle and waist-to-elbow on-body

channels at 2.45 GHz while the user was walking within an anechoic
environment. The measurement scenario and setup are described in
reference [12].
Figure 11.3. Comparison of waist-to-right head on-body channels at

2.45 GHz while the user was stationary and then walking within
(a) an outdoor environment and (b) an open office area. The measurement
scenario and setup are described in reference [15].
to identify a degree of channel repetition related to the user’s body

movements in each of the traces, it is quite clear that the distribution
of the received signal in both amplitude and time is dissimilar.
As discussed above, there can also be a considerable contrast in
wireless channel characteristics for the same on-body link over
different user states and environments. This is highlighted in
Fig. 11.3 for a signal received by a right-head positioned node from
a transmitter at the waist. Here the user is stationary in an outdoor
environment, and the received signal is relatively constant, although
a slight periodic variation due to respiration can be observed. When
the user moves into an office area environment, the channel is
again relatively stationary. However, the mean signal level becomes
elevated in this case by around 10 dB. For these scenarios (i.e., when
the person is relatively stationary), it has been proposed that the
lognormal distribution may be used to describe the path gain [14].
Another important effect of environmental multipath for on-body
channels is the alteration of the channel’s autocorrelation function.
As shown in reference [15], this effect is particularly prominent
for scenarios when the user is mobile. For anechoic environments,
many links may display identifiable quasi-periodic trends. However,
within rich multipath environments, the signal contributions from

off-body directions become important parts of the on-body channel.
Signal components returned to the body act to raise the signal level
(as shown in Fig. 11.3) and reduce the periodicity superimposed
on the channel by the user’s movements. Furthermore, as noted
in reference [16], the variation in mean signal level experienced in
different environments, and when the WBAN user performs a range
of very different exercises, means that on-body channels will be
expected to exhibit nonstationary channel statistics.
These characteristics mean that, except in the simplest of cases,
it will be very difficult to develop generic analytical models that ac-
curately model propagation characteristics in body-centric systems.
Simulation, even on today’s multiprocessor machines, is computa-
tionally intensive although parallelization of finite-difference time-
domain (FDTD) computations [17] may help improve computation
times. Given the large number of variable parameters found in body-
centric communications, it is often easier to analyze these channels
using statistics, where given a particular scenario, a probability level
is assigned to a channel output parameter such as the received
signal. To further simplify the analysis, it is also useful to investigate
environments with specific user actions separately (e.g., standing
in hallway, walking in open office area, running outdoors, etc.) to
create a set of baseline channel characteristics for WBAN operation.
In most cases, this will allow the on-body channel’s statistics to be
considered as wide sense stationary, that is the first and second
moments do not vary with respect to time. At present, the general
consensus from the literature is that there is no unified model
for received signal amplitude distribution in WBANs; rather, it is
accepted that particular models perform best for certain links and
scenarios. A range of statistical models have been used to describe
signal variation in WBANs; these include Rayleigh, Rice, Nakagami,
Weibull, and lognormal, which will be discussed in detail in the next
section.
11.2.1 Statistical Distribution of the Fading

Signal in WBANs
In this chapter, we restrict ourselves to the discussion of flat
fading on-body channels, often referred to as ”narrowband”
on-body channels. A wireless signal will undergo flat fading if

the channel has constant gain and linear phase response over
a bandwidth Bc greater than the bandwidth of the transmitted
signal Bs , i.e., Bs Bc . Signal variation in narrowband body-centric
communication channels is often reported using a combination of
first-order statistics, such as probability density functions (PDFs)
and cumulative distribution functions (CDFs), and second-order
statistics, such as level crossing rates (LCRs) and average fade
duration (AFDs). Second-order statistics are of particular interest
when designing mobile radio systems as they give a direct indication
to the rate of change, with respect to time, of the wireless
channel. Among their uses is the design of error correcting codes,
optimization of interleaver size, and system throughput analysis as
well as channel modeling.
11.2.1.1 Rayleigh and Rice distributions

There are a wide range of probability densities that have been
used to describe the distribution of signal fading in WBANs. Among
these are the popular Rayleigh and Rice fading models. Under
Rayleigh fading, the received signal is viewed as the resultant of
a large number of scattered signal components, each with random
amplitude and uniform phase. The envelope R of a fading signal that
follows a Rayleigh distribution may be expressed in terms of the in-
phase, X , and quadrature, Y , components of the fading signal as

R = X2 + Y2 (11.1)
where X and Y are mutually independent
Gaussian
processes with
E (X ) = E (Y ) = 0, and E X 2 = E Y 2 = σ 2 . The probability
density function for this random variable may be written as

r r2
p R (r) = 2 exp − 2 , r ≥ 0 (11.2)
σ 2σ

where E R 2 = 2σ 2 is the average envelope power. Rayleigh fading
has been reported for on-body links where communications occur
around the human torso at both 915 MHz and 2.45 GHz [18]. The
authors explain that diffracting components that travel distances
greater than 35 cm around the back of the body become significantly
attenuated. Instead, the received signal is made up of reflected
components, which follow a Rayleigh distribution. In body-centric

channels, where a dominant signal component (e.g., strong on-body
reflection or dominant creeping wave component) exists, it may
be more appropriate to model the received signal distribution as
following a Rice distribution. In this instance, X and Y are again
assumed
2 to bemutually
independent Gaussian processes such that
E X = E Y 2
= σ 2 with nonzero means E (X ) = μ X and
E (Y ) = μY , respectively. The PDF of the Rice distribution may be
written as
r r2 + s2 rs
p R (r) = 2 exp − I0 , r≥0 (11.3)
σ 2σ 2 σ2
where s 2 = μ2X + μ2Y is the power contained within the dominant
component and E R 2 = s 2 + 2σ 2 . In the absence of a dominant
component (i.e., s = 0), the Rice PDF becomes equivalent to the
Rayleigh PDF given in Eq. 11.2. Another important figure, used to
characterize the degree of fading in a wireless channel, is the Rice
factor, k, which is defined as the ratio of the square of the dominant
component to the scattered power, i.e., k = s 2 /2σ 2 . When k → 0,
and hence the dominant component s decreases, the fading becomes
closer to Rayleigh fading, and as k → ∞, the channel no longer
exhibits fading. Ricean fading for on-body communications has been
widely reported in the literature (868 MHz [19], 915 MHz [18], and
2.45 GHz [16, 18]). In reference [18], Ricean fading was reported
for on-body link distances less than 25 cm. It was also observed
that as the separation distance of the on-body link increases, the
k parameter decreases, due to greater attenuation of the creeping
wave component. The authors of reference [16] report that the
magnitude of the k factor also changes with time. In this study,
on-body channel characteristics were monitored over a wide range
of everyday environments, as a test subject performed various
activities such as driving a car, walking outdoors and indoors, and
sitting.
11.2.1.2 Nakagami distribution

The PDF of a Nakagami fading signal envelope is given by

2mmr 2m−1 mr 2
p R (r) = exp − , r≥0 (11.4)
(m) m

where m = E 2 (r 2 )/var(r 2 ) and = E r 2 . A physical model
for Nakagami fading has been proposed in reference [20], which
describes the received signal as the sum of random vectors all with
random amplitude and phase. It has also been shown that Nakagami
random variables may be generated as the sum of m independent
and identically distributed Rayleigh components with mean power
/m [21]. The Nakagami-m distribution is extremely versatile and
holds a number of important relationships with the Rayleigh and
Rice distributions. When m = 1, Eq. 11.4 becomes equivalent to
the Rayleigh PDF given in Eq. 11.2. For m > 1, the Rice k factor
can be approximated by the Nakagami-m distribution through the
following relationship [22]:
√
m2 − m
k= √ , m > 1. (11.5)
m − m2 − m
Moreover, for high values of the Nakagami-m parameter (m 1),
the Nakagami-m distribution can be used to approximate the
lognormal distribution [23]. Another attractive feature of the
Nakagami-m distribution is that it can be used to model fading
scenarios that are worse than Rayleigh, i.e., when m < 1. When
m = 1/2, the Nakagami distribution is equivalent to the semi-
Gaussian distribution.
In reference [24], for WBAN channel measurements made at
2.45 GHz, it was observed that when the test subject was stationary,
Nakagami-m parameters were always much larger than 1, irrespec-
tive of the operating environment. Under anechoic environmental
conditions, when the user became mobile, m parameters were
observed to decrease due to an increase in the diffuse contributions
from the body and interruption of the on-body creeping wave
component if present. In a similar fashion to Rice k factors reported
in reference [18], the Nakagami-m parameter was typically observed
to decrease with increasing separation distance between on-body
antennas. Fading in on-body channels was also observed to increase
when the user moved into a multipath environment, showing that
the local surroundings are an important consideration for on-body
systems. For on-body links operating at 4.5 GHz [25], Nakagami
fading has also been found to occur for scenarios when the test
subject performed walking movements in an anechoic chamber.
11.2.1.3 Weibull distribution

In a Weibull fading channel, the received signal is obtained as a
nonlinear function of the modulus of multipath components, where
the nonlinearity is expressed in terms of the power parameter α.
Yacoub [26] has shown that a Weibull-distributed envelope may
be obtained from two independent zero mean Gaussian random
variables with identical variance such that R α = X 2 + Y 2 . Following
from this, the PDF of a Weibull-distributed envelope may be written
as [26]
α
αr α−1 r
p R (r) = exp − (11.6)
λ λ
where λ = E (R α ). In a similar fashion to both the Rice and
Nakagami distributions, the Rayleigh distribution may be obtained
from the Weibull distribution by setting α = 2 and the negative
exponential distribution by setting α = 1 [27]. Weibull fading has
been reported for on-body measurements made at 2.36 GHz while
the test subject was walking and running [28] and at 4.5 GHz for
walking and standing up/sitting down motions [25]. In reference
[25], it is suggested that the Weibull model should be used for on-
body links that experience severe fading such as the shoulder and
ankle, while the user was walking, and the entire trunk action, while
the user was repeating standing up/sitting down motions.
11.2.1.4 Lognormal distribution

The lognormal PDF for an envelope R with mean μ and standard
deviation σ may be expressed by

1 [ln (r) − μ]2
p R (r) = √ exp − (11.7)
rσ 2π 2σ
where r > 0. The lognormal distribution has traditionally
been associated with shadow fading in wireless channels. It has
also been reported both for body shadowing [14] and recently
observed in small-scale fading [23]. Lognormal random variables
can be viewed as the result of a number of multiplicative factors
that become additive under logarithmic transformation. A physical
interpretation of lognormal fading in on-body channels has been
proposed in reference [29], where the received signal is the result

of diffraction, reflection, energy absorption, and antenna losses. The
occurrence of lognormal fading has generally been observed in on-
body channels in which the user is stationary [14, 15, 25] but has
also been reported for dynamic scenarios [24].
11.2.2 Higher Order Statistics

While the distribution of received signal amplitude is of interest,
quite often it is necessary to model not only the first-order statistics
of a wireless channel, but also the second-order characteristics to
understand how the received signal is distributed in time and how
often it crosses a particular threshold level.
11.2.2.1 Level crossing rate and average fade duration

The level crossing rate and average fade duration are two important
second-order statistics for fading channels. They give a direct
indication as to the rate of change of a wireless channel through
their relationship with the maximum Doppler frequency, fm . Both
the LCR and AFD are of great importance in the design of mobile
radio systems and the analysis of their performance [30]. The LCR,
N R , at a specified threshold level, R, is defined as the expected rate
at which a signal crosses R in a positive direction [31] and is given
by
∞
N R (r) = ṙ p Ṙ,R (ṙ, r) d ṙ (11.8)
0
where ṙ is the time derivative and pṘ,R (ṙ, r) is the joint probability
density of Ṙ and R.
For the case of isotropic scattering, and based on the knowledge
that the Rayleigh [31], Rice [32], Nakagami [33], Weibull [34], and
lognormal (after logarithmic transformation) [23] fading envelopes
have Gaussian time derivatives, and their envelopes and respective
slopes are independent, closed form expressions for the √ LCR of
these fading signals have been derived. Defining ρ = r/r , as
an example, Eq. 11.9 shows the form of the LCR equation for a
Parameter Estimation and Model Selection 335
Nakagami-m-distributed received signal envelope as derived in [33]

√ mm−(1/2) 2m−1
N R (r) = 2π fm ρ exp −mρ 2 (11.9)
(m)
where fm is the maximum Doppler frequencies and (•) is the
Gamma function. In a similar fashion, the average fade duration,
TR , is defined as the average length of time spent below a specified
threshold level
F R (r)
TR (r) = (11.10)
NR
where F R (r) is the cumulative distribution function of r. The
corresponding AFD equation for a Nakagami-m-distributed received
envelope may be written as [33]

m, mρ 2
TR (r) = √ (11.11)
2π fm mm−(1/2) ρ 2m−1 exp −mρ 2
where (•, •) is the incomplete Gamma function. It should be noted
that as signal reception in WBANs may often be nonisotropic [12],
Eqs. 11.9 and 11.11 and those contained in [31], [32], [34], and
[23], which were derived under the assumption of an isotropic
scattered signal contribution, will only provide an approximation of
the LCR and AFD in on-body channels in which the signal reception
is nonisotropic.
11.3 Parameter Estimation and Model Selection
Having acquired on-body channel measurements, an important step

in the data analysis procedure is the selection of the appropriate
fading model to represent the data. Before this is possible, it is
necessary to find parameter estimates for the candidate models.
The method typically employed is that of maximum likelihood
estimation (MLE).
11.3.1 Maximum Likelihood Estimation

MLE operates by determining the parameter estimates for a given
model that maximize the probability of the sample data. For
the Rayleigh distribution given in Eq. 11.2, if r1 , r2 , . . . , rn are
independent observations of the received signal, then the joint

probability of the observations is

n
L (σ |R) ≡ p R (R|σ ) = p R (ri |σ ) (11.12)
i =1
where L (•) represents the likelihood function, which is the
probability of obtaining σ , given the received envelope R. The
posteriori probability of a Rayleigh-distributed envelope may be
expressed in the form
n
n
−2n r2
L (σ |R) = (ri ) σ exp − i
2
(11.13)
i =1 i =1
2σ
The parameter σ for which the likelihood function is a maximum is
found by first taking the logarithm of Eq. 11.13, which is
n n 2
ri
ln [L (σ |R)] = ln (ri ) − 2n ln (σ ) − (11.14)
i =1 i =1
2σ 2
Now calculation has become easier as repeated multiplication has
been converted to addition. To obtain a maximum, the partial
differential of Eq. 11.14 is taken with respect to σ and the result
set equal to 0. With some rearrangement, the maximum likelihood
estimator of σ is given by

n 2
ri
σ̂ = (11.15)
i =1
2n
The log-likelihood reaches a maximum at the same point as the

original function and the nature of the stationary point may be
determined from its second derivative. As σ is a single parameter, the
Hessian (matrix of partial second derivatives) is scalar. The Fisher
information, I (σ ), is now
⎡ n ⎤
3 ri2
⎢ 2n ⎥
I (σ ) = −E ⎢ ⎥
i =1
⎣σ2 − σ4 ⎦ (11.16)
Maximum likelihood estimators for the lognormal distribution can

be found using a similar process, while estimators for the Rice
[35], Nakagami-m [36], and Weibull [37] distributions can be found
therein.
11.3.2 Akaike Information Criterion

The Akaike information criterion [38] (AIC) is a measure of the
goodness of fit of a chosen statistical model, which takes into
account the number of model parameters (i.e., complexity). Using
the AIC, competing models may be ranked according to their
respective scores and the model with the lowest AIC index selected
as the most likely candidate. The second-order AIC, often denoted
by AICc , which corrects for small sample size bias, may be written as
[39]
2P (P + 1)
A I C c = −2 ln L θ̂|data + 2P + (11.17)
(n − P − 1)
where ln[L (θ̂|data)] is the maximized log-likelihood for the pa-
rameters θ, given the data set and model under test, P is the
number of adjustable parameters available in the chosen model,
and n is the sample size. As noted in reference [29], the AICc has
a strong theoretical motivation as it provides an estimate of the
Kullback–Leibler information loss and aims to find the model that
will minimize the difference between the true and hypothesized
distributions. Two other metrics associated with the AICc which may
be used to interpret its output, are the delta Akaike and Akaike
weight. The delta Akaike of a model i , denoted by
i , is simply a
measure of each model relative to the best model, i.e.,

i = A I C c,i − min (A I C c ) (11.18)
Interpretation of the delta Akaike is straightforward; the optimum
model from the set of candidate models has a delta Akaike of 0.
Burnham and Anderson [39] state that a
i < 2 suggests that
substantial evidence for the adoption of a model exists, values
between 3 and 7 indicate that the model has considerably less
support, and a
i > 10 suggests that the model is very unlikely. The
Akaike weight of a model i , denoted by wi , represents the ratio of the
delta Akaike to the whole set of R candidate models and is calculated
as
R
wi = exp −
2i exp −
2r (11.19)
r=1
The Akaike weights indicate the probability that a model is the

best among the set of candidate models based upon the highest
weight on a scale of 0 to 1. Furthermore, the Akaike weights can

be compared with best model ( j ) to determine the evidence ratio,
w j /wi , i.e., the degree to which one model is better than another.
As a concluding remark on the AICc , it should be noted that its use
to select the appropriate fading model should be approached with a
relative degree of caution. The AICc can only select the best model
from a set of candidate models; therefore, if the set consists of a
selection of poor fitting models, the AICc can only select the best
candidate from those models. Therefore, it is recommended that the
selected model and estimated parameters are also validated visually,
to ensure that the model adequately describes the data at hand.
11.3.3 Worked Example

11.3.3.1 Model selection
As an example of the narrowband channel modeling process, we
now present a statistical analysis of the received signal at the
antenna positions shown in Fig. 11.1 from a transmitter mounted on
the user’s left waist. In this example, the user was mobile in both
an anechoic chamber and a reverberation chamber. The antennas
used were 5 mm higher mode patches [40] mounted parallel to the
body surface; these were designed to exploit the on-body creeping
wave mode of propagation, therefore, maximizing on-body coupling.
The received signal measurements were obtained using a vector
network analyzer configured to record the forward path gain (S21 )
at 2.45 GHz with a sample rate of 200 Hz. Full details of the channel
measurement campaign are provided in reference [12]. For the
analysis of fading characteristics, it is quite common to remove the
global or local mean depending on the scenario being analyzed. In
this example, we remove the global mean from the received signal
and report this value separately so that the reader may generate
the time correlated random variables associated with the fading
distribution using the simulation methods described at the end of
this section and subsequently modify the mean value of the envelope
to match the system scenario being considered.
We consider as our candidate models for these scenarios the
distributions described in Section 11.2.1. Quite often, the cumulative
Figure 11.4. A selection of empirical CDF plots and maximum likelihood

estimated theoretical plots while the user was mobile in the anechoic
chamber. See also Color Insert.
x
distribution function (CDF) F (x) = −∞ p R (r) dr of the received
signal is plotted on log-log scales to expand the tail of the CDF as it
is the deepest fades that cause the greatest degradation in wireless
system performance. A selection of the empirical CDFs for the ankle,
head, knee, and waist as the user was mobile in an anechoic chamber
are shown in Fig. 11.4 alongside the theoretical plots. All parameter
estimates for the competing models were obtained using maximum
likelihood estimation on data sets of 4000 samples per link and
are given in Table 11.1, including those not shown in Fig. 11.4. The
estimated fading parameters for each of the models show significant
variation between each of the on-body links. As expected, and
reported elsewhere [24, 29], fading appears to increase the greater
the on-body separation distance between nodes. Table 11.2 shows
the Akaike weights calculated using Eq. 11.19 for all positions. We
can see for this scenario, the Weibull and Rice models both account
for three on-body channels and the Nakagami distribution just
one. The Rayleigh model, which has only one adjustable parameter,
accounted for none of the channels as did the lognormal model.
In the reverberation chamber, the high level of environmental
multipath means that the fading observed in the on-body links
Table 11.1. MLE parameter estimates user mobile anechoic chamber
(σ ) Rice Nak-m Weibull Lognormal Mean

Position Rayl (s, σ ) (m, ) (λ, α) (μ, σ ) fm (Hz) (dBmV)
Ankle 0.78 0.74, 0.58 1.17, 1.23 1.1281, 2.20 −0.14, 0.59 5.20 −14.96
Chest 0.72 0.99, 0.15 10.76, 1.02 1.06, 7.37 −0.01, 0.16 5.82 1.21
Elbow 0.78 0.81, 0.53 1.06, 1.21 1.12, 2.26 −0.17, 0.68 3.20 −9.76
Head 0.82 0.00, 0.82 0.78, 1.35 1.12, 1.72 −0.23, 0.77 5.00 −24.40
Knee 0.79 0.73, 0.59 1.14, 1.23 1.13, 2.20 −0.14, 0.58 7.14 −9.13
Waist 0.73 0.96, 0.27 3.59, 1.07 1.10, 4.44 −0.04, 0.30 2.90 0.10
Wrist 0.83 0.00, 0.83 0.82, 1.37 1.13, 1.73 −0.21, 0.69 3.00 −10.64
Table 11.2. Akaike weights user mobile anechoic chamber
Position Weibull Nakagami Rice Rayleigh Lognormal
Ankle 0.67 0.32 0.01 0.00 0.00

Chest 0.00 0.00 1.00 0.00 0.00
Elbow 0.00 0.00 1.00 0.00 0.00
Head 0.00 1.00 0.00 0.00 0.00
Knee 0.07 0.00 0.93 0.00 0.00
Waist 1.00 0.00 0.00 0.00 0.00
Wrist 1.00 0.00 0.00 0.00 0.00
significantly increases. Table 11.3 shows that the fading parameters

for all of the distributions that contain Rayleigh as a special case (i.e.,
Rice, Nakagami, and Weibull) are approaching this limit. Observing
the empirical CDF plots for the head and waist in particular, we
see that there is very little difference in the Rice, Nakagami, and
Table 11.3. MLE parameter estimates user mobile reverberation chamber
Rice Nak-m Weibull Lognormal Mean

Position Rayl (σ ) (s, σ ) (m, ) (λ, α) (μ, σ ) fm (Hz) (dBmV)
Ankle 0.82 0.00, 0.82 0.87, 1.34 1.13, 1.80 −0.19, 0.67 5.90 5.16
Chest 0.80 0.52, 0.70 1.02, 1.27 1.13, 2.03 −0.16, 0.63 4.87 7.76
Elbow 0.79 .66, 0.64 1.05, 1.25 1.13, 2.09 −0.16, 0.63 5.25 8.98
Head 0.79 0.71, 0.61 1.04, 1.25 1.13, 2.10 −0.17, 0.66 5.07 7.71
Knee 0.79 0.67, 0.63 1.15, 1.24 1.13, 2.17 −0.14, 0.57 4.90 8.60
Waist 0.79 0.59, 0.68 1.01, 1.26 1.13, 2.04 −0.17, 0.65 5.44 7.71
Wrist 0.81 0.00, 0.81 0.92, 1.30 1.13, 1.89 −0.19, 0.68 7.00 7.56

estimated theoretical plots while the user was mobile in the reverberation
chamber. See also Color Insert.
Weibull plots. For this scenario, the Rice distribution was favored
on four occasions, the Weibull distribution twice, and Nakagami just
once. Once again, the Rayleigh and lognormal distributions were
not selected. When compared with the results from the anechoic
chamber, it reinforces that when a WBAN moves into an area
with a considerable level of environmental multipath, the local
surroundings become an important part of the overall channel
model.
Table 11.4. Akaike weights user mobile reverberation chamber
Position Weibull Nakagami Rice Rayleigh Lognormal

Ankle 1.00 0.00 0.00 0.00 0.00
Chest 0.15 0.11 0.53 0.00 0.21
Elbow 0.01 0.00 0.99 0.00 0.00
Head 0.00 0.00 1.00 0.00 0.00
Knee 0.01 0.99 0.00 0.00 0.00
Waist 0.05 0.02 0.88 0.00 0.05
Wrist 0.64 0.36 0.00 0.00 0.00
11.3.3.2 Level crossing rate

Continuing with the example and omitting the poorer fitting
Rayleigh and lognormal models for clarity, Figs. 11.6 and 11.7 show
the empirical LCR equations for the elbow-positioned antennas
while the user was mobile in the anechoic and reverberation
chambers, respectively. For both of these scenarios, the Rice
distribution was chosen by the AICc to be the most likely candidate
model for the fading distribution of the channel. The estimated
maximum Doppler frequencies for these channels were obtained by
minimizing the sum of the squared error between the empirical LCR
and the Rice LCR model [22]. Results for the remaining channels
in the anechoic and reverberation chambers were obtained using
a similar method using the LCR of the AICc favored model and are
given in Tables 11.1 and 11.3, respectively. The estimates of the
maximum Doppler frequency for these scenarios are in the range
2.90 to 7.14 Hz, which suggests relatively slowly changing channels.
Figure 11.6. Comparison of empirical LCR with theoretical Rice,

Nakagami and Weibull LCR plots for elbow-positioned antenna while the
user was mobile in anechoic chamber.
Figure 11.7. Comparison of empirical LCR with theoretical Rice,

Nakagami and Weibull LCR plots for elbow-positioned antenna while the
user was mobile in reverberation chamber.
11.3.3.3 Simulation of the received signal envelope

Having obtained an estimate of the maximum Doppler frequency, it
becomes possible to straightforwardly simulate Rayleigh and Ricean
on-body fading channels using the rayleighchan and riceanchan ob-
jects available in the communications toolbox of MATLAB. Nakagami
[41]1 , Weibull [42], and lognormal [23] envelopes may be generated
by adopting a rank matching approach. As outlined in reference
[41], where autocorrelated Nakagami random variables (RVs) are
produced from time-correlated Rayleigh variates, in this method a
Rayleigh envelope with the required autocorrelation properties is
generated (e.g., using the rayleighchan object in MATLAB). Statistical
rank is then exploited by rearranging the Rayleigh envelope using
order statistics (i.e., smallest to largest) while keeping note of the
1 Also describes a method for simulating Nakagami fading under nonisotropic
scattering conditions.
original index of the data prior to rearrangement. At the same time,

a randomly generated Nakagami, Weibull, or lognormal sequence
is also sorted into an ordered vector. Now the Nakagami, Weibull,
or lognormal sequence is rearranged so that its autocorrelation
structure matches closely that of the original (unordered) Rayleigh
reference sequence using the indexes stored earlier.
11.4 Conclusions
Due to the large number of variable parameters associated with

the design of body-centric communication systems, it is necessary
to make preliminary channel measurements for some well-defined
scenarios, which will form the boundary conditions for normal
system operations. In this chapter, we have shown how narrowband
on-body channels may be modeled using some of the most
commonly encountered fading distributions in wireless communi-
cations. Although there is no universal model that describes all on-
body channels for every possible scenario, it is clear that probability
density functions such as Rice, Nakagami, and Weibull may be used
to adequately describe the channel’s amplitude distribution.
One key requirement for future body-centric channel models is
an understanding of the directivity of the channel (i.e., the distrib-
ution of the angle of arrival). While directional signal reception will
have a negligible effect upon first-order characteristics, depending
on the degree of nonisotropic signal reception, it may have an
appreciable effect upon second-order statistics such as the LCR
and AFD. Therefore, models for these statistics, which were derived
under the assumption of isotropic signal reception, may only be used
as an approximation in body-centric channels when distribution of
signal arrival deviates strongly from the uniform case.
References
1. A. Sani, et al. (January 2010) Experimental characterization of UWB

on-body radio channel in indoor environment considering different
antennas, IEEE Transactions on Antennas and Propagation, 58(1),
pp. 238–241.
References 345
2. S. L. Cotton and W. G. Scanlon (January 2009) Characterization and

modeling of on-body spatial diversity within indoor environments at
868 MHz, IEEE Transactions on Wireless Communications, 8(1), pp. 176–
185.
3. I. Khan and P. S. Hall (January 2009) Multiple antenna reception at
5.8 and 10 GHz for body-centric wireless communication channels,
IEEE Transactions on Antennas and Propagation, 57(1), pp. 248–
255.
4. A. Serra, P. Nepa, G. Manara, and P. S. Hall (2007) Diversity measure-
ments for on-body communication systems, IEEE Antennas Wireless
Propagation Letters, 6, pp. 361–363.
5. I. Khan and P. S. Hall (January 2010) Experimental evaluation of
MIMO capacity and correlation for narrowband body-centric wireless
channels, IEEE Transactions on Antennas and Propagation, 58(1),
pp. 195–202.
6. S. L. Cotton and W. G. Scanlon (2007) Characterization and modeling of
the indoor radio channel at 868 MHz for a mobile body-worn wireless
personal area network, IEEE Antennas and Wireless Propagation Letters,
6, pp. 51–55.
7. A. A. Goulianos, T. W. C. Brown, B. G. Evans, and S. Stavrou (July
2009) Wideband power modeling and time dispersion analysis for UWB
indoor off-body communications, IEEE Transactions on Antennas and
Propagation, 57(7), pp. 2162–2171.
8. S. L. Cotton and W. G. Scanlon (April 2009) Channel characterization for
single and multiple antenna wearable systems used for indoor body-to-
body communications, IEEE Transactions on Antennas and Propagation,
Special Issue on Antennas and Propagation for Body-Centric Wireless
Communications, 57(4), pp. 980–990.
9. Y. Wang, I. B. Bonev, J. O. Nielsen, I. Z. Kovacs, and G. F. Pedersen (April
2009) Characterization of the indoor multiantenna body-to-body radio
channel, IEEE Transactions on Antennas and Propagation, Special Issue
on Antennas and Propagation on Body-Centric Wireless Communications,
57(4), pp. 972–979.
10. S. L. Cotton, W. G. Scanlon, and P. S. Hall (September 2010) A simulated
study of co-channel inter-BAN interference at 2.45 GHz and 60 GHz, in
European Wireless Technology Conference, Paris, pp. 61–64.
11. W. G. Scanlon and N. E. Evans (April 2001) Numerical analysis of
body-worn UHF antenna systems, IEE Electronics and Communication
Engineering Journal, 13(2), pp. 53–64.
12. S. L. Cotton, W. G. Scanlon, and G. A. Conway (2009) Autocorrelation

of signal fading in wireless body area networks, in 2nd IET Seminar on
Antennas and Propagation for Body-Centric Wireless Communications,
London, pp. 1–5.
13. J. Ryckaert, P. de Doncker, R. Meys, A. de Le Hoye, and S. Donnay (April
2004) Channel model for wireless communication around human body,
Electronics Letters, 40(9), pp. 543–544.
14. A. Alomainy, et al. (January 2007) Statistical analysis and performance
evaluation for on-body radio propagation with microstrip patch
antennas, IEEE Transactions on Antennas and Propagation, 55(1),
pp. 245–248.
15. S. L. Cotton, G. A. Conway, and W. G. Scanlon (April 2009) A time-
domain approach to the analysis and modeling of on-body propagation
characteristics using synchronized measurements at 2.45 GHz, IEEE
Transactions on Antennas and Propagation, 57(4), pp. 943–955.
16. Y. I. Nechayev, Z. H. Hu, and P. S. Hall (2009) Short-term and long-term
fading of on-body transmission channels at 2.45 GHz, in Loughborough
Antennas and Propagation Conference, Loughborough, pp. 657–660.
17. G. Z. Yang, A. Sani, S. L. Lee, Y. Zhao, and Y. Hao (2009) A simulation
environment for subject-specific radio channel modeling in wireless
body sensor networks, in 6th International Workshop on Wearable and
Implantable Body Sensor Networks, pp. 23–28.
18. A. Fort, C. Desset, P. Wambacq, and L. V. Biesen (December 2007)
Indoor body-area channel model for narrowband communications, IET
Microwaves, Antennas and Propagation, 1(6), pp. 1197–1203.
19. S. L. Cotton and W. G. Scanlon (2006) A statistical analysis of indoor
multipath fading for a narrowband wireless body area network, in 17th
IEEE International Symposium on Personal, Indoor and Mobile Radio
Communications, pp. 1–5.
20. W. R. Braun and U. Dersch (1991) A physical mobile radio channel
model, IEEE Transactions on Vehicular Technology, 40(2), pp. 472–482.
21. J. C. S. S. Filho and M. D. Yacoub (December 2009) On the second-
order statistics of Nakagami fading simulators, IEEE Transactions on
Communications, 57(12), pp. 3543–3546.
22. G. L. Stüber (2002) Principles of Mobile Communication, 2nd edn, Kluwer
Academic Publishers, New York.
23. S. L. Cotton and W. G. Scanlon (2007) Higher order statistics for
lognormal small-scale fading in mobile radio channels, IEEE Antennas
and Wireless Propagation Letters, 6, pp. 540–543.
References 347
24. S. L. Cotton and W. G. Scanlon (January 2009) An experimental

investigation into the influence of user state and environment on
fading characteristics in wireless body area networks at 2.45 GHz, IEEE
Transactions on Wireless Communications, 8(1), pp. 6–12.
25. M. Kim and J. I. Takada (2009) Statistical model for 4.5 GHz narrowband
on-body propagation channel with specific actions, IEEE Antennas and
Wireless Propagation Letters, 8, pp. 1250–1254.
26. M. D. Yacoub, D. B. da Costa, U. S. Dias, and G. Fraidenraich (2005)
Joint statistics for two correlated Weibull variates, IEEE Antennas and
Wireless Propagation Letters, 4, pp. 129–132.
27. M. D. Yacoub (January 2007) The α-μ Distribution: A Physical Fading
Model for the Stacy Distribution, IEEE Transactions on Vehicular
Technology, 56(1), pp. 27–34.
28. D. B. Smith, et al. (November 2009) Temporal correlation of dynamic
on-body area radio channel, Electronics Letters, 45(25), pp. 1212–
1213.
29. A. Fort, C. Desset, P. de Doncker, P. Wambacq, and L. van Biesen (June
2006) An ultra-wideband body area propagation channel model—from
statistics to implementation, IEEE Transactions on Microwave Theory
and Techniques, 54(4), pp. 1820–1826.
30. N. Youssef, T. Munakata, and M. Takeda (September 1996) Fade
statistics in Nakagami fading environments, in Proceedings of the
IEEE 4th International Symposium on Spread Spectrum Techniques and
Applications, pp. 1244–1247.
31. W. C. Jakes (1974) Microwave Mobile Communications, Wiley, New York.
32. S. O. Rice (January 1948) Statistical properties of a sine wave plus
random noise, Bell System Technical Journal, 27, pp. 109–157.
33. M. D. Yacoub, J. E. V. Bautista, and L. Guerra de Rezende Guedes (May
1999) On higher order statistics of the Nakagami-m distribution, IEEE
Transactions on Vehicular Technology, 48(3), pp. 790–794.
34. N. C. Sagias, D. A. Zogas, G. K. Karagiannidis, and G. S. Tombras (June
2004) Channel capacity and second-order statistics in Weibull fading,
IEEE Communications Letters, 8(6), pp. 377–379.
35. J. Sijbers, A. J. den Dekker, P. Scheunders, and D. van Dyck (1998)
Maximum likelihood estimation of Ricean distribution parameters, IEEE
Transactions on Medical Imaging, 17, pp. 357–361.
36. J. Cheng and N. Beaulieu (2001) Maximum likelihood-based estimation
of the Nakagami-m parameter, IEEE Communications Letters, 5, pp. 101–
103.
37. A. C. Cohen (November 1965) Maximum likelihood estimation in the

Weibull distribution based on complete and on censored samples,
Technometrics, 7(4), pp. 579–588.
38. H. Akaike (1974) A new look at the statistical model identification, IEEE
Transactions on Automatic Control, 19(6), pp. 716–723.
39. K. P. Burnham and D. R. Anderson (2002) Model Selection and
Multimodel Inference: A Practical Information-Theoretic Approach, 2nd
edn, Springer-Verlag, New York.
40. G. A. Conway and W. G. Scanlon (April 2009) Antennas for over-body
surface communication at 2.45 GHz, IEEE Transactions on Antennas and
Propagation, 57(4), pp. 844–855.
41. J. C. S. S. Filho, M. D. Yacoub, and G. Fraidenraich (March 2007) A simple
accurate method for generating autocorrelated Nakagami-m envelope
sequences, IEEE Communications Letters, 11, pp. 231–233.
42. D. B. Smith, et al. (2009) A simulator for the dynamic on-body area
propagation channel, in Antennas and Propagation Society International
Symposium, Charleston, pp. 1–5.
Chapter 12
Antenna Design and Propagation for

WBAN Applications
Tharaka Dissanayake
School of Electrical Engineering and Computer Science,
University of Newcastle, Callaghan, NSW2308, Australia
[email protected]
An essential part of wireless body area networks (WBANs), like for

any wireless system, is the transducer that converts electrical signals
into electromagnetic waves and vice-versa. This has been done by
antennas for medium-long range communications or inductive coils
for very short range communication. Usually the communication in
WBAN is either short range or medium range. The on-body and
implanted sensor nodes are essentially small devices that are, at
most, of the size of a standard mobile phone. There are matured
enough technologies to design low-profile antennas for mobile
wireless devices. The transformation of those technologies into
WBAN applications will be discussed. One of the main concerns
of antennas in WBAN is the interaction of near field electric
and magnetic fields with the human body. This phenomenon is
responsible for significant changes in antenna characteristics, and
researchers have found methods to mitigate these effects for on-
body antennas. Some of these methods, especially the antenna

www.panstanford.com
350 Antenna Design and Propagation for WBAN Applications
shielding, will be discussed. It is even more challenging to design in-

body (implanted) antennas with surrounding dissipative body tissue
at high frequencies. Recent research into the efficiency enhancement
of implanted antennas will be discussed in this chapter. Widely used
classes of antennas in wideband and narrowband WBAN, including
volume conduction antennas, will be presented as examples. The
commonly used design and experimental methods of WBAN antenna
evaluation will be discussed. A brief overview of antenna theory
and modern simulation techniques will be incorporated, where
necessary, for the readers who are not familiar with the material.
12.1 Introduction
Antennas can be considered reciprocal devices that convert currents

into field and fields into current. While reciprocity of antennas
does not have the same meaning for wideband operation, all
characterizing parameters are defined for both transmitting and
receiving antennas. For the sake of those who are not familiar with
the terminology used in antenna design, let us first define several
core definitions of antenna engineering. It is not by any means a
complete list but the most commonly found in design literature.
12.1.1 Antenna Gain

Gain of the antenna is defined very much different to the gain of
an amplifier. In the context of an amplifier or any other active or
passive circuit component, gain relates the input currents, voltages,
or power to the output currents, voltages, or power. This is exactly
the case in antennas; however, the difference is that the reference
is not an input parameter, but the properties of a hypothetical
isotropic antenna. The concept of gain of an antenna, essentially a
passive device, is defined as the radiation intensity (in watt/square
meter) in a given direction I (θ, φ)compared to the intensity I0 of an
isotropic antenna driven by the same power source. Hence,

I (θ, φ)
gain(dBi) = 10 log (12.1)
I0
Introduction 351
Note that I0 is not a function of elevation and azimuth angles θ and

φ, respectively. Some engineers use dipole antenna as the reference
antenna. The difference can be noted in the units of the gain. For
the isotropic reference, the unit is dBi. More often than not gain
is specified as a single value, which is the maximum out of all
directions.
12.1.2 Return Loss

One of the first things an antenna designer attempts to minimize is
the reflection of power at the antenna-transmission line interface.
Typical feed chain consists of a source, transmission line with
characteristic impedance matched to internal impedance of the
source and the antenna. Matching networks may be employed
to minimize power reflections at each interface. The reflection
coefficient is defined as
Z L − Zo
= (12.2)
Z L + Zo
where Z L is the input impedance of the antenna and Z o is the
characteristic impedance of the transmission line, which is often
matched to the source. The value of Z o is chosen as 50 in most
prototyping experiments. Nevertheless, actual output impedance of
a sensor node radio front-end can be different. In such situations, a
matching network may be employed. The return loss is now defined
as
RL = −20 log10 || (12.3)
which is a positive value. Hence, higher return loss means less
reflection. In generic designs, RL is kept above 10 dB; however, in
high power base station applications, the targeted return loss is
above 15 dB. For power limited WBAN sensor nodes, higher values
of RL are highly desirable.
12.1.3 Efficiency
Efficiency of an antenna is defined as the ratio between input power
and the radiated power. This ratio is determined by the losses
incurred in the antenna system. Useful power of an antenna is the
amount that is actually radiated to the far-field region. All the other
losses are
• Mismatch losses due to impedance differences between the
transmission line and the antenna
• Losses due to finite resistance of the conductive elements of
the antenna
• Losses due to dielectric loss tangent of substrate, super-
strate, or filling material (in case of dielectric filled horn
antennas)
• Near-field losses due to coupling between surrounding
conductive or dielectric material. Absorption losses of body
tissue, device casings, and radomes are some examples for
this category
In WBAN, the most prominent is the radiation absorption by the
tissue. As we shall see later, proximity of lossy material to an antenna
can severely degrade its efficiency. Without partial or full insulation,
especially for antennas on implanted devices, sustainability of
radiation is nearly impossible.
12.1.4 Reciprocity
The concept of reciprocity is often applied to antenna characteriza-
tion. In general, all of the aforementioned parameters are valid for
both transmit and receive cases. Nevertheless, this is not the case
for ultra-wideband (UWB) antennas analyzed in the time domain.
For wideband operation, antennas are better characterized using
transfer functions or impulse responses. Transmit impulse response
ht of the antenna is defined as
μ
E (θ, φ, t) = I (t) × ht (θ, φ, t) (12.4)
4πr
where E is the electric field and I is the excitation current. The
receive impulse response is defined as
Vrec (t) = E (t) × hr (θ, φ, t) (12.5)
Consequently, the reciprocity in the time domain is
1 ∂
ht (θ, φ, t) = hr (θ, φ, t) (12.6)
2 ∂t
Miniaturized Antennas 353
Note that it is still possible to characterize UWB antennas with

radiation patterns at several frequencies inside the operating band,
where the reciprocity is valid. This is what we may find in many
publications. The main reason is the unavailability of resources
to measure antennas in time domain. However, transfer function
approach is the most effective in optimizing radiation in different
directions, such as pulse shaping [1, 2]. It is also convenient to link
antenna impulse response with the channel impulse response to
model propagation scenarios [3].
12.2 Miniaturized Antennas
On-body or in-body devices, which are essentially the nodes of a

WBAN, are no more than the size of a mobile phone. The size
of a wireless endoscope can be even smaller, of the size of a
size 000 capsule. This leaves the antenna designer with very little
physical volume to work with. Fortunately, the problem has been
in discussion over the last decade among engineers, who have
developed antennas for mobile phones. The knowledge can well be
used in designing antennas for all types of WBAN devices, especially
those for biomedical applications [4–6].
12.2.1 Planar Inverted-F Antennas

A planar inverted-F antenna (PIFA) is essentially an evolution of
the microstrip patch antenna. Additionally, PIFA has a shorting
pin, which reduces the size of the microstrip patch antenna by
half (Fig. 12.1). This miniaturization makes it very popular among
small antenna designers. The location of feed with respect to
shorting pin/strip can be used to tune the antenna. However, PIFA is
relatively narrowband and various techniques have been employed
to overcome this deficiency. One such technique is capacitive loading
[7]. In addition, various other techniques employed in microstrip
antenna bandwidth enhancement such as thick, low permittivity
substrates with capacitive feeds can also be employed in PIFA
design. Miniaturizing techniques such as meandering are also
employed in advanced designs of PIFA antennas (Fig. 12.2) [8].
Figure 12.1. Planar inverted-F antenna configuration.
Figure 12.2. Meandered PIFA antenna.

Miniaturized Antennas 355
12.2.2 Planar Monopoles and Dipoles

Planar monopoles discussed here are the printed version of the
quarter-wave monopole antenna or the half-wave dipole antenna.
Such antennas have gained popularity as compact, integrated
components in mobile handheld devices. The ease of fabrication
dominates the list of advantages in planar antennas. However,
the tuning of quarter-wave printed monopole antennas depends on
the supporting ground plane. As shown in reference [9], the finite
size of the ground plane reduces the bandwidth and the frequencies
of operation. Several modifications have been suggested such as
adding additional strips to the ground plane. One of the challenges in
integrating printed monopoles or dipole antennas in WBAN devices
is retaining the compactness while manipulating the ground plane
to tune the antenna. One such technique is the use of asymmetric
feed location and the feed gap [9]. Being able to freely place the feed
point without losing mechanical rigidity is an advantage of printed
antennas.
The design challenge unique to the printed monopoles or dipoles
in WBAN is the antenna detuning due to the proximity of body
tissue. This has been shown in reference [5] using a planar compact
capsule-shaped antenna in the proximity of a tissue sample. The
change of matching is obvious in Fig. 12.3. When coupled to a
conducting body such as a tissue sample, currents are induced in a
larger volume. As a result, the effective size of the antenna rises by
several wavelengths. This results in lower resonant frequencies. At
first glance, this is helpful in antenna miniaturization. Nevertheless,
it must be noted that tissue is a poor conductor, which brings down
the radiation efficiency of the antenna. Furthermore, properties of
the tissue are spatial and temporal variables. Therefore, usually
it is not recommended that the presence of tissue is used as a
miniaturization technique.
The coupling effect is prominent with the strong electric fields at
the feed gap of dipole or monopole antennas, as shown in Fig. 12.4.
Therefore, it is important when designing and placing planar dipole
and monopole antennas in WBAN applications, the feed gap stays
clear of conductive and dielectric parasitic elements or there is
sufficient shielding [10], either conductive or dielectric.
Figure 12.3. Change of measured return loss characteristics due to the

presence of tissue sample [5].
Figure 12.4. Feed gap electric fields.

Implanted Antennas 357
12.2.3 Planar Slot Antennas

The slot antennas are also known as magnetic antennas. The loop
antennas fall in the same general category [11]. These antennas have
the advantage of loosely coupling to nearby conductive structures.
The selection of slot antennas not only permits greater freedom in
tuning antennas for packaged devices, but also delivers unchanged
pattern stability as discussed in reference [12]. The dielectric
loaded antenna presented by us, which will be discussed in the
next section, was a compact wide-slot as well. The annular slot
antenna presented in reference [3] proves the same since the
shape of the radiated pulses remained unchanged with the distance
between the antenna and the body. The power of the received
pulse changed proving that the closer the antennas to the body, the
lower the efficiency. However, this effect is weak in slot antennas
compared with monopoles or dipoles. Use of dielectric [13] or
conductive shielding can further improve the antenna performance
as shown in reference [10]. Shielding also reduces the radiation
absorption by the tissue. Absorption is proportional to the electric
field components inside the tissue. It changes with the magnitude
and the direction of the near-field patterns [14]. Although it is
possible to design to minimize the absorption by manipulating the
near field, total shielding is an easier, somewhat straightforward,
solution. There is, however, one main disadvantage in adding a
reflector to otherwise omnidirectional antenna such as a wide slot.
The confinement of the fields between the shield and the antenna
(or return currents in another perspective) results in high Q, and
in case of UWB antennas as shown in reference [10], it causes
the reduction of the matched bandwidth of the antenna. That is
why in narrowband WBAN applications, PIFA with ground plane is
essentially the favorite as mentioned in Section 12.2.1.
12.3 Implanted Antennas
We have seen that on-body antenna design can closely follow the
principles of small antenna design for wireless handsets near lossy
dielectric tissue. In contrast, the implanted antennas are totally
surrounded by lossy dielectrics. This makes the antenna design,
especially sustaining the required efficiency, very difficult. This

challenge has been addressed in both narrowband and wideband
communication by many research groups, including that of the
author.
A highly cited theoretical analysis of implanted antennas can
be found in reference [15]. A dipole inside a head model has
been analyzed using finite difference time domain (FDTD) and
dyadic Green’s function (DGF) methods. The FDTD simulation
has used a detailed body model with 30 different tissue types,
whereas the DGF uses spherical model with six concentric layers
of different tissue types. The FDTD simulation used a detailed
human phantom model and, hence, delivered more accurate and
realistic results. Such simulations are possible with most of the
commercial electromagnetic simulation software as they all provide
detailed full-body phantom models, thanks to Visible Human Project
sponsored by US National Library of Medicine. However, computing
resource availability may limit the simulation to only a portion of the
phantom. The microstrip and PIFA antennas presented in reference
[15], with 0.16% and 0.25% efficiency, respectively, have been
simulated with simplified phantom models. It shows that the use
of simplified models does not compromise the accuracy of return
loss characteristics. After all, only the near-field electromagnetic
interactions decide the return loss characteristics of an antenna.
Therefore, it is possible to first simulate the antenna itself in a
simplified model and then use the resulting electric and magnetic
fields with a phantom model to calculate far-field electric fields. This
approach allows a fine mesh for the small antenna and a coarse mesh
for the large phantom model [16] when using FDTD method. Such
multilevel meshing techniques are built-in for commercial software
packages such as Microwave Studio by CST.
Although sophisticated modeling of antennas with phantom
models is possible with modern electromagnetic solvers [17, 18],
the precise modeling of the changing properties of the body
tissue is nearly impossible. There is no guarantee that tissue
properties remain the same from person to person or time to time.
Physiological conditions can change the water content of tissue and,
hence, the conductivity and permittivity. Typically, there is high
contrast between the electromagnetic properties of adjacent tissue
layers. For example, the relative permittivity of muscle is 64 at

434 MHz, whereas that of fat is 5 at the same frequency. One cannot
expect consistency in placing implanted devices in practice. As such
implanted device may end up in a fat layer or muscle. Dimensions
of the antenna have to be significantly different between these two
situations.
One of the methods to address this design challenge is making
the antenna less sensitive to surrounding tissue properties. Note
that almost all implanted antennas have some type of superstrate,
which prevents it from touching the lossy tissue [15, 18–20].
The permittivity of the superstrate has been chosen similar to
that of the substrate or in some cases determined by the type of
biocompatible material used. We have shown that superstrate can
be made to act as a shield between the antenna and the surrounding
tissue properties. In other words, the permittivity of the superstrate
can be selected that the sensitivity of the return loss to tissue
properties can be minimized [6, 21].
12.3.1 Dielectric Loaded Matching of Implanted Antennas

Dielectric loading of antennas in dissipative medium is a technique
that has been used for underground and underwater antenna
applications [22, 23]. The objective was to completely shield the
currents on the antenna from touching the conductive medium. The
experiments in reference [24] have been conducted in salt water,
which is very similar to an antenna in a tissue simulating medium.
Recent work on insulated dipoles in dissipative medium [25] has
brought to light some interesting results, which we have explored
in reference [21].
It can be shown that, using expressions for power reflection
coefficient at the insulator and tissue medium interface, if the
dielectric constant of the insulator is sufficiently high and remains
close to an average tissue dielectric constant, the antenna matching
can be achieved over a wide range of tissue properties. The preferred
value for dielectric constant of the insulator in the experiments
conducted in reference [6] is 50. Figure 12.5 shows the antenna
configuration. The antenna has the shape of a size 000 capsule.
It has two distinctive regions: the grounded coplanar waveguide
Figure 12.5. Insulator loaded printed slot antenna configuration. The

shaded area is metalization.
and the wide slot, printed on a Rogers TMM10i substrate. Rogers

provide a range of laminates that have very low dielectric losses. For
antenna applications, it is important that high frequency laminates
are chosen. Usually designers use FR4 laminates, which are widely
used in printed circuit board design. But the significant dielectric
losses cannot be tolerated, where high losses in propagation through
tissue prevail. The TMM10i has the highest dielectric constant in
the range. Furthermore, high dielectric constant of the substrate
reduces the guided wavelength at a given frequency. Dimensions
of the antenna and the transmission lines are determined by the
guided wavelength. For example, half-wave printed dipole means the
antenna length is actually half of the guided wavelength as opposed
to free space wavelength.
By selecting high dielectric constant substrate and even higher
dielectric constant insulator, we have achieved very high level of
compactness. Thus the antenna can be planted inside a capsule-
shaped enclosure. Note that the portion of the antenna with
Figure 12.6. The loaded antenna inside a tissue simulating cylinder in a

CST simulation. See also Color Insert.
the coplanar waveguide can indeed support other electronic

components. It is essential that antenna designers must consider
the effects of other components of the implanted device. We have
achieved this by introducing a bulk conductor to the upper half of
the antenna, and SMA connector in experiments. Figure 12.6 shows
the antenna in a tissue-simulating cylinder with its insulator loading.
The properties of the tissue cylinder have been changed to match
various tissue types. The results are shown in Fig. 12.7. It confirms
that the antenna is matched well within the expected bandwidth off
1 GHz, suitable for UWB applications.
12.3.1.1 Biocompatibility of dielectric loaded antenna

Although it is found that dielectric constant of the insulator has to be
close to 50, matching the permittivity of the biocompatible material
of the device casing to that value is a manufacturing challenge. As far
as we are aware, the engineers developing biocompatible material
have very little concern about the electromagnetic properties
of those. Therefore, it is expected that the antenna with its
insulator has a casing of biocompatible material such as silicone.
Properties of this material definitely affect the antenna performance.
The appendices of this chapter include a set of MATLAB codes
Figure 12.7. Return loss characteristics of the antenna shown in Figs. 12.6
and 12.5 for various tissue types.
for calculating the modified reflection coefficient after adding a

thin shell of low dielectric constant biocompatible material as a
modification to the problem analyzed in reference [25], as shown
in Fig. 12.8.
As an example, Table 12.1 shows the reflection coefficient at
the insulator-shell interface (note that the reflected wave inside
the insulator affects the antenna return loss characteristics) for
insulator dielectric constant of 50, shell dielectric constant of 10,
frequency of 3 GHz, and insulator radius of 5 mm. The shell thickness
Table 12.1. Reflection coefficient at the insulator-biocompatible shell

interface for fixed electromagnetic properties of the insulator shell. The
surrounding media is muscle tissue.
Reflection coefficient (dB) Shell thickness (mm)

−30 0
−37 0.1
−19 1
−5 10
Volume Conduction Antennas 363
Figure 12.8. Insulated Hertzian dipole with a dielectric shell in a

dissipative medium.
is varied from 0 mm to 10 mm. The implant is assumed to be inside

muscle tissue. The frequency-dependent tissue properties have been
determined using fourth-order Cole–Cole equation (tissue prop.m).
It is obvious from the results in the above table that with
thinner biocompatible shells, the relatively small dielectric constant
of the biocompatible material does not make much difference to
the reflection coefficient. Note that insulator and the biocompatible
material were assumed lossless.
12.4 Volume Conduction Antennas
The antennas we have discussed so far are physically separate

from the human body. Any currents induced on body tissue due to
the antenna’s operation, although unavoidable, are responsible for
significant losses in the system operation. Thus we take measures to
minimize these currents. All tissue types have increasing conductiv-
ity with frequency. At microwave frequencies, this phenomenon is a
problem in terms of propagation and antenna matching. However,
Figure 12.9. The concept of x-antenna.
there is a class of antennas suitable for WBAN communication

that uses tissue conductance for its operation. They are known as
volume conduction antennas [26–29]. In this technology, tissue is
in direct contact with the feed probe such that the former acts
as an antenna or a transmission line. Systems based on volume
conduction principle use internal electrodes and skin electrodes.
Although signal propagation (i.e., currents) is completely inside the
tissue with nonlinear characteristics, at macro level the reciprocity
theory holds [26].
The x-antenna presented in reference [29] (Fig. 12.9) has shown
promising characteristics as a volume conduction antenna. At 1 kHz
for an input power of 1.3 μW, 0.05 mV signal has been generated
on an electrode 15 cm away. Further improvement of this method
can deliver not only volume conduction, but also surface conduction
communication at higher frequencies.
12.5 Summary
In this chapter, we have discussed antenna elements that are suitable

for WBAN. The main focus is on designing miniaturized antennas
and taking the effect of human body into consideration. The most
popular antennas that have been proposed for WBAN devices are
PIFA, planar monopoles and dipoles, printed slot antennas, and
volume conduction antennas. Variants of these antennas can be
used in either in-body or on-body devices. The newly introduced
Appendix A 365
technique of dielectric loading to impedance match the implanted

antennas has been discussed. As a further improvement to make
the suggested method industrially viable, addition of thin shell
of biocompatible material is considered. A set of MATLAB codes
for theoretical analysis of such modification is provided here.
Readers may modify these codes to optimize antennas for different
implanted applications. For example, it is possible to calculate the
optimum thickness of the superstrate layers of a dielectric-loaded
PIFA by breaking the antenna currents into dipole-like elements and
determining optimum distances from each of them to the insulator-
tissue interface.
Appendix A
A.1 Function Calculating the Reflection Coefficient

%This is the reflection coefficient at the interface
%between the first insulator and biocompatible material.
%r1 is the inner radius of the thin shell and r2 is the
%outer radius of the thin shell. X111 and X514 are
%defined in X111.m and X514.m, respectively. k1,k2, and
%k3 are calculated using function k.m
function gamma=gamma(k1,k2,k3,r1,r2)
x1=x111(-k1,r1);
x2=x111(k1,r1);
x3=x111(-k2,r1);
x4=x111(k2,r1);
x5=x514(-k1,r1);
x6=x514(k1,r1);
x7=x514(-k2,r1);
x8=x514(k2,r1);
x9=x111(-k2,r2);
x10=x111(k2,r2);
x11=x111(-k3,r2);
x12=x514(-k2,r2);
x13=x514(k2,r2);
x14=x514(-k3,r2);
p1=x3*(x11*x13-x14*x10)/(x13*x9-x12*x10)+x4*
(x11*x12-x14*x9)/(x10*x12-x13*x9);
p2=x7*(x11*x13-x14*x10)/(x13*x9-x12*x10)+x8*
(x11*x12-x14*x9)/(x10*x12-x13*x9);
gamma=(x1*p2-p1*x5)/(x6*p1-x2*p2);
end
A.2 X111 and X514

function x111=x14(k,r)
x111=deli(k,r)*k^2;
end
function x514=x514(k,r)
x514=deli(k,r)/r+k^2*del(k,r);
end
A.3 Function calculating K vectors

%This function generates the k vectors using material
%properties and frequency. Frequency is an input
%variable to this function. On top of that, in case of
%biological tissue epsr and sigma are functions
%of frequency and they will be calculated by the
%function cole4.m. Usually mur=1.
function k=k(epsr,mur,sigma,freq)
eps0=8.85e-12;
mu0=4*pi*1e-7;
eps=epsr*eps0;
mu=mur*mu0;
Appendix B 367
omega=2*pi*freq;
kr=omega*sqrt(eps*mu/2)*sqrt(sqrt(1+sigma^2/
eps^2/omega^2)+1);
ki=omega*sqrt(eps*mu/2)*sqrt(sqrt(1+sigma^2/
eps^2/omega^2)-1);
k=kr-i*ki;
end
Appendix B: Calculating Frequency-Dependent Tissue

Properties
A.4 Cole–Cole function

%This function delivers relative permittivity and
%conductivity of biological tissue at a given
%frequency, when all the parameters for 4th order
%Cole-Cole equation are given. Cole-Cole parameters
%for 24 tissue types are given in Table 3.1 of Peter S.
%Hall and Yang Hao, "Antennas and Propagation for
%Body-centric Wireless Communications",Artec House,
%2006,.
function [epsr,sigma]=cole4(f,einf,sig,alpha1,alpha2,
alpha3,alpha4,de1,de2,de3,de4,tau1,tau2,tau3,tau4)
eps0=8.85e-12;
ehat1=de1/(1+(j*2*pi*f*tau1*1e-12).^(1-alpha1));
ehat=einf+(ehat1+ehat2+ehat3+ehat4)+sig/
(i*2*pi*f*eps0);
epsr=real(ehat);
sigma=-2*pi*f*imag(ehat)*eps0;
end
A.5 Calculating the properties

%This function delivers epsilon and sigma of a given
%tissue based on Cole-Cole equation. There is a text
%file "tissue.txt", which has all the Cole-Cole
%parameters for 24 tissue types (as given in Table 3.1
%of "Antennas and Propagation for Body-centric Wireless
%Communications" by Hall and Hao). The function reads
%from the text file and returns the permittivity and
%conductivity at a given frequency.
function [epsr,sigma]=tissue_prop(tissue_type,f)
switch tissue_type
case ’Aorta’
row=1;
case ’Bladder’
row=2;
case ’Blood’
row=3;
case ’Bone(cancellous)’
row=4;
case ’Bone(cortical)’
row=5;
case ’Brain’
row=6;
case ’Breast’
row=7;
case ’Cartilage’
row=8;
case ’Spinal Fluid’
row=9;
case ’Cornea’
row=10;
case ’Eye Tissue’
Appendix B 369
row=11;
case ’Fat’
row=12;
case ’Bladder’
row=13;
case ’Heart’
row=14;
case ’Kidney’
row=15;
case ’Liver’
row=16;
case ’Lung’
row=17;
case ’Muscle’
row=18;
case ’Skin(dry)’
row=19;
case ’Skin(wet)’
row=20;
case ’Small Intestine’
row=21;
case ’Stomach’
row=22;
case ’Testis’
row=23;
case ’Tongue’
row=24;
otherwise
disp(’Tissue Not in Library’);
end
M=dlmread(’tissue.txt’);
einf=M(row,1);
sig=M(row,2);
alpha1=M(row,3);
alpha2=M(row,4);
alpha3=M(row,5);
alpha4=M(row,6);
de1=M(row,7);
de2=M(row,8);
de3=M(row,9);
de4=M(row,10);
tau1=M(row,11);
tau2=M(row,12);
tau3=M(row,13);
tau4=M(row,14);
[epsr,sigma]=cole4(f,einf,sig,alpha1,alpha2,alpha3,
alpha4,de1,de2,de3,de4,tau1,tau2,tau3,tau4);
end
A.6 Function for optimization

%\\A Hertzian dipole antenna embedded in tissue like
%material is surrounded by a pherical insulator and an
%outer layer of thin sperical shell of bio-compatible
%material.
%\\This function calculates the reflection coefficient
%at the inner surface of the thin shell.
%\\in_tissue (string) is the type of tissue surrounding
%the implanted antenna, selected from a set of tissue
%types as defined in the function tissue_prop(), lines
%9 to 58.
%\\sigma1 and sigma2 are conductivity of the insulator
%and the shell, may be considered as zero.
%\\2<epsr1<100 and 2<epsr2<10 are relative permittivity
%of the insulator and the shell, respectively.
%\\100e6<f<3e9 (Hz) is the frequency.
%\\0.1<rl<5 is the radius of the insulator.
%\\0.1e-3<thickness<5e-3 (m) is the shell thickness
function ref=reflection(x)
in_tissue=’Muscle’;
sigma1=0;
sigma2=0;
epsr1=x(1);
References 371
epsr2=x(2);
f=x(3)*1e9;
r1=x(4)*1e-3;
thickness=x(5)*1e-3;
[epsr3,sigma3]=tissue_prop(in_tissue,f);
k1=k(epsr1,1,sigma1,f);
ref=10*log(abs(gamma(k1,k2,k3,r1,r1+thickness)));
end
References
1. Dissanayake T., and Esselle, K. P. (2006) Waveform Optimization of UWB

Radio Systems Over Range of Directions, Electronic Letters, 42, pp. 384–
385.
2. Dissanayake T., and Esselle, K. P. (2006) Correlation Based Pattern
Stability and a Figure of Merit for UWB Antennas, IEEE Transactions on
Antennas and Propagation, 54, pp. 3184–3191.
3. Klemm, M., and Troester, G. (2006) Textile UWB Antennas for Wireless
Body Area Networks, IEEE Transactions on Antennas and Propagation,
54, pp. 3192–3197.
4. Furse, C. M. (2009) Biomedical Telemetry: Today’s Opportunities and
Challenges, in IEEE International Workshop on Antenna Technology:
Small Antennas and Novel Metamaterials, Santa Monica, CA, pp. 389–
392.
5. Dissanayake, T., Yuce, M. R., and Ho, C. (2009) Design and Evalu-
ation of a Compact Antenna for Implant-to-Air UWB Communica-
tion, IEEE Antennas and Wireless Propagation Letters, 8, pp. 153–
156.
6. Dissanayake, T., Esselle, K. P., and Yuce, M. R. (2009) Dielectric
Loaded Impedance Matching for Wideband Implanted Antennas, IEEE
Transactions on Micorwave Theory and Techniques, 57, pp. 2480–
2487.
7. Corbett, R. W., and Ross, D. M. (1996) A Capacitively Loaded PIFA
for Compact PCS Handsets, in IEEE Antennas and Propagation Society
International Symposium, pp. 742–745.
8. Kim, J., and Rahmat-Samii, Y. (2006) Implanted Antennas in Medical

Wireless Communications, 1st edn, Morgan and Claypool.
9. Ammann, M. J., and John, M. (2007) Some Techniques to Improve Small
Ground Plane Printed Monopole Performance, 2007 IEEE Antennas and
Propagation Society International Symposium, 1–12, pp. 2583–2586.
10. Klemm, M., Kocs, I. Z., Pederson, G., and Troster, G. (December
2005) Novel Small-Size Directional Antenna for UWB WBAN/WPAN
Applications, IEEE Transactions on Antennas and Propagation, 53,
pp. 3884–3896.
11. Schantz, H. G. (2003) UWB Magnetic Antennas, IEEE International
Syemposium on Antennas and Propagation, 3, pp. 604–607.
12. Dissanayake, T., and Esselle, K. P. (April 2008) UWB Performance of
Compact L-Shaped Wide Slot Antennas, IEEE Transactions on Antennas
and Propagation, 56, pp. 1183–1187.
13. Pelosi, M., Franek, O., Knudsen, M. B., and Pedersen, G. F. (2009)
Influence of Dielectric Loading on PIFA Antennas in Close Proximity to
User’s Body, Electronics Letters, 45, pp. 246–247.
14. Kivekas, O., Lehtiniemi, T., and Vainikainen, P. (5 November 2004)
On the General Energy-Absorption Mechanism in the Human Tissue,
Microwave and Optical Technology Letters, 43, pp. 195–201.
15. Kim, J., and Rahmat-Samii, Y. (2004) Implanted Antennas Inside
a Human Body: Simulations, Designs, and Characterizations, IEEE
Transactions on Micorwave Theory and Techniques, 52, pp. 1934–
1943.
16. Chirva, L. C., Hammond, P. A., Roy, S., and Cumming, D. R. S. (April
2003) Electromagnetic Radiation From Ingested Sources in the Human
Intestine Between 150 MHz and 1.2 GHz, IEEE Transactions on
Biomedical Engineering, 50, pp. 484–492.
17. Izdebski, P. M., Rajagopalan, H., and Rahmat-Samii, Y. (April 2009)
Conformal Ingestible Capsule Antenna: A Novel Chandelier Meandered
Design, IEEE Transactions on Antennas and Propagation, 57, pp. 900–
909.
18. Xia, W., Saito, K., Takahashi, M., and Ito, K. (April 2009) Performance of
an Implanted Cavity Slot Antenna Embedded in the Human Arm, IEEE
Transactions on Antennas and Propagation, 57, pp. 894–899.
19. Houzen, T., Takahashi, M., Saito, K., and Ito, K. (2008) Implanted Planar
Inverted-F Antenna for Cardiac Pacemaker System, in IEEE International
Workshop on Antenna Technology (iWAT): Small Antennas and Novel
Metamaterials, Chiba, Japan, pp. 346–349.
References 373
20. Warty, R., Tofighi, M. R., Kawoos, U., and Rosen, A. (2008) Characteri-
zation of Implantable Antennas for Intracranial Pressure Monitoring:
Reflection by and Transmission Through a Scalp Phantom, IEEE
Transactions on Micorwave Theory and Techniques, 56, pp. 2366–2376.
21. Dissanayake, T., Esselle, K. P., and Yuce, M. R. (2009) UWB Antenna Im-
pedence Matching in Biomedical Implants, in 3rd European Conference
on Antennas and Propagation, Germany, pp. 3523–3526.
22. Wu, T. T., King, R. W. P., and Giri, D. V. (July 1973) The Insulated Dipole
Antenna in a Relatively Dense Medium, Radio Science, 8, pp. 669–709.
23. Jacobi, J. H., Larsen, L. E., and Hast, C. T. (1979) Water-Immersed
Microwave Antennas and Their Application to Microwave Interrogation
of Biological Targets, IEEE Transactions on Micorwave Theory and
Techniques, 27, pp. 70–78.
24. Siegel, M., and King, R. W. P. (1971) Radiation From Linear Antennas in a
Dissipative Half-Space, IEEE Transactions on Antennas and Propagation,
19, pp. 477–185.
25. Tai, C. T., and Collin, R. E. (October 2000) Radiation of a Hertzian Dipole
Immersed in a Dissipative Medium, IEEE Transactions on Antennas and
Propagation, 48, pp. 1501–1506.
26. Sun, M., Liu, Q., Liang, W., Wessel, B. L., Roche, P. A., Mickle, M.,
and Scalabassi, R. J. (2003) Application of the Reciprocity Theorem
to Volume Conduction Based Data Communication Systems Between
Implantable Devices and Computers, in IEEE 25th Annual International
Conference of the Engineering in Medicine and Biology Society, pp. 3352–
3355.
27. Yao, N., Lee, H. N., Scalabassi, R. J., and Sun, M. (2006) Low Power Digital
Communication in Implantable Devices Using Volume Conduction of
Biological Tissues, in IEEE International Conference of the Engineering
and Biological Society, pp. 6249–6252.
28. Wessel, B. L., Roche, P., Sun, M., and Scalabassi, R. J. (2004) Optimization
of an Implantable Volume Conduction Antenna, in the IEEE 26th Annual
International Conference of the Engineering in Medicine and Biology
Society, 6, pp. 4111–4114.
29. Sun, M., Wessel, B. L., Liang, W., Roche, P., Liu, Q., Mickle, M., and
Scalabassi, R. J. (2003) A Volume Conduction Antenna for Implantable
Devices, in the IEEE 25th Annual International Conference of the
Engineering in Medicine and Biology Society, pp. 3356–3359.
Chapter 13
Coexistence Issues with Wireless Body

Area Networks
Axel Sikora
University of Applied Sciences Offenburg,
Badstrasse 24, D77652 Offenburg, Germany
axel.sikora@hs-offenburg.de
13.1 Introduction
Coexistence issues in wireless systems are known to be critical.

The wireless channel is a common, i.e., shared media, where an
exclusive assignment can be performed only within the limits of
time, frequency, or code multiplexing. The issues may be especially
unpredictable when license-free frequency bands for industrial,
scientific, and medical applications (ISM- or license-exempt bands)
are used, because cooperation with extrinsic and intrinsic system
influences must be supported. While doing that, the ISM systems
have to observe the overall rules of the regulations for the specific
ISM band.
The impacts of coexistence issues are extensively discussed for
hard real-time systems from industrial automation applications.
But they might also have a major impact on wireless body area
networks (WBAN), where they might endanger latency (jitter) due

www.panstanford.com
376 Coexistence Issues with Wireless Body Area Networks
to increased packet loss or might put additional load on the energy

budget, caused by prolonged channel observation times and the
need for retransmissions. Both aspects are especially troublesome
for low-power WBANs, which might be fed from batteries or from
energy converters, and for safety-relevant (medical) systems.
First, Section 13.2 gives an overview on the coexistence issues
in low-power, low-cost wireless systems in ISM bands. Special
attention is paid to interferers with high-duty cycle and with
unpredictable behavior. Their impact on the communication is
described is Section 13.3.
Based on these characteristics, Section 13.4 presents the cur-
rently available approaches to overcome coexistence issues. Those
concentrate on those relevant to WBANs and include listen-before-
talk behavior, channel monitoring and prediction techniques, back-
off strategies, reservation schemes, (slow) frequency adaptation
and (fast) frequency hopping, and redundancy with regard to
frequencies, antennas, and coding. An outlook on wide-band and
ultra wide-band (UWB) systems will be given.
In a third part, a selection of solutions will be presented. These
will include, but won’t be limited to developments from the author’s
institute and shall give an overview on the state of the art.
This part will also discuss the attempts to overcome the
coexistence issue from the side of standardization.
The fourth and last part will give an outlook on future research
and development directions.
It should be mentioned that this relatively short chapter can give
only a limited view on the issue. No exhaustive view on the manifold
literature seems to be feasible.
13.2 Analysis of Interferers
13.2.1 Classification
The coexistence issue for wireless ISM systems includes all kinds
of interferences. It is the effect of an emission from one (subject)
system, called the interferer, to a second (object) system, which is
potentially affected, called the interfered station. Interferers might
be classified into three groups:
Analysis of Interferers 377
• Intrinsic interference from wireless systems running the

same protocol in the same or neighbored frequency band
• Extrinsic interference from wireless systems running a
different protocol in the same or neighbored frequency
band
• Extrinsic interference from (micro) electronic systems with
electromagnetic or RF-emission (EMI, RFI)
In addition, we have to distinguish between the interference to a

system (i.e., the impact to our system) and the interference to other
systems.
13.2.2 Regulation Issues

It is anticipated within the whole chapter, that the predominant
number of WBAN systems are and will be operated in the
license-free frequency bands for industrial, scientific, and medical
applications. Those are denominated ISM or license-exempt bands.
As for all electronic devices, the use of these frequency bands is
regulated from the regional regulation body. For Europe, the rules
are elaborated and harmonized by ETSI and then put into national
laws for each single country. The Americas and Australia follow the
rules of Federal Commission of Communication (FCC). In Japan, the
Ministry of Telecommunications (MKK) regulates the Telec Radio
Regulations, the ARIB for Southern Pacific regions and so on.
Those rules might limit the behavior on the physical and on the
data link layer:
• On the physical layer, the regulation impacts the frequency

behavior. The typical parameters here to describe the
frequency behavior are
– The spectral mask
– The effective radiated power
– The peak power density
– The frequency range(s)
– The transmitter’s and receiver’s spurious emissions
This behavior has consequences on the possible modulation
schemes, the achievable data rates, the channel spacing, and
alike. Regulations on the physical layer might also include

regulation (transmit power control, TPC) or dynamic
frequency selection (DFS). The latter is sometimes also
known as Dynamic Channel Selection (DCS) or Adaptive
Frequency Agility (AFA).
• On the data link layer, key elements of the medium access
are regulated. This might include a maximum duty cycle per
basic time unit and access pertinence (listen-before-talk,
LBT).
The conformance to the rules depends on the frequency band and

on the region. It might be certified by a self-declaration (e.g., ETSI,
2.4 GHz ISM band) or by a certified body (e.g., FCC, 2.4 GHz ISM
band).
Albeit all these rules, no prediction can be made about the
density of devices within the range of the system, and thus no
statement can be made on the overall load on the channel.
13.2.3 Intrinsic Interference

The first group of interferers consists of stations that follow the
same communication protocol as the interfered station. This group
is characterized by the following behaviors:
• All nodes use the same frequency characteristics, i.e., the

same bandwidth and the same modulation scheme.
• The nodes may work in the same frequency band; however,
they might work in the same or in different channels. If
the intrinsic interferers work on the same channel as the
interfered stations, they might exchange information in
order to perform avoidance strategies.
• All nodes follow the same channel access mechanisms, i.e.,
LBT or back off strategies.
• All nodes may come with similar traffic characteristics, with
regard to traffic load, traffic cycles.
For completeness sake, it is highlighted that passive interferers

are disregarded within this overview. As passive interferers we
Analysis of Interferers 379
understand echoes due to multipath propagation. In this case, one

station does not hear only the direct signal from the sender, but also
the delayed signals that are reflected at surrounding objects.
These multipath effects are disregarded as data rates for WBANs
tend to be low and distances to be small.
Of course, the specialitiy of WBAN networks of course lies in the
attenuation through the human body.
13.2.4 Extrinsic Interference of RF-Stations within the

Same Frequency Band
Extrinsic interference from wireless systems within the same
frequency band is a common challenge for ISM-based systems.
Interferers of this group can be characterized by the following
behavior:
• The nodes are operated within the same frequency band.

The transmission might occur such that the interferer might
have a full or a partial overlap of the frequency band of the
interfered station. This overlap might be timely constant or
variable.
• The frequency characteristics of the interferer might be
different from the interfered station. This leads to the
fact that — even if the different stations work in the
same frequencies — there is no possibility to exchange
information and thus to cooperate.
• The traffic characteristics might be completely different. So
predictions of the future behavior are not possible. Analog
systems mostly come with an extremely high duty cycle.
Digital systems tend to be much more event-driven.
• If the interferer’s performance is reduced by the WBAN’s
activity, this might still worse the problem, as retransmis-
sions and frequency agility further reduce the available
bandwidth.
Figure 13.1 shows some of the most widespread systems in the ISM
band from 2.400 GHz to 2.4835 GHz (as defined by ETSI).
Figure 13.1. Frequency characteristics of some of the systems working in

the 2.4 GHz ISM band [29].
13.2.5 Extrinsic Interference of Other Systems within the

Same Frequency Band
All microelectronic devices, might they be dedicated for commu-
nication or not, are subject of electromagnetic interference (EMI).
The characteristics are also described by regulation and should be
below the so-called noise level. Practically, in all cases no limitations
can be given for the timely behavior of the interferer, i.e., it may be
always on. In no case, cooperation via a communication channel can
be supported.
Probably, the most prominent in-band interferers for ISM band
communication are microwave ovens, which are operated in the
2.4 GHz band in the whole world. For completeness sake, it should
be mentioned that this frequency was chosen in the fifties as a
reasonable compromise for heating food, as the lowest resonance
frequency of liquid water is around 18 GHz, of gaseous water above
22 GHz.
Figure 13.2 shows the spectral characteristics of a random choice
of commercially available microwave oven for home use.
Effect on Transmission 381
Figure 13.2. Frequency characteristics of commercially available micro

ovens for home use [6]. See also Color Insert.
13.3 Effect on Transmission
13.3.1 Fundamentals
The required bit error rate (BER) for the communication between
two nodes translates into a certain packet error rate (PER) or
frame error rate (FER), which depends on the packet length. In
order to achieve this BER, a certain signal-to-noise-ratio (SNR) is
required, which mainly depends on the modulation scheme and
on the processing gain that can be achieved, e.g., through coding
or frequency spreading. In this sense, interferers contribute to a
noise level, which degrades the SNR and thus increases the BER and
consequently the PER.
The impact on the actual transmission shows a strong depen-
dency from the characteristics of the input filter of the receiver.
This is normally described by the selectivity, which describes the
maximum power level for an intrinsic interferer working at the same
channel or at neighbored channels.
The carrier-to-co-channel interference ratio (C/I ratio) or simply
co-channel interference (CCI) describes the actual or required
power ratio of the wanted carrier signal (Pc ) with regard to the
unwanted interfering signal (Pi ), which is transmitted on the same
channel.

Pc
C C I = 20 log (13.1)
Pi
The adjacent channel selectivity (ACS) describes the same ratio with
regard to an interfering signal on a neighbor channel. The first ACS
describes the ratio for the directly neighbored channel, the second
ACS the ratio for the second next neighbor channel, and so on.
13.3.2 Simulation of a Dense Sensor Network (Intrinsic

Interference)
The power level of the emitted signals reduces along Frijs Free Space
Equation [7], as long as line-of-sight connection in the sense of
undisturbed Fresnel zones is maintained.

λ α
P R = PT · ·GR ·GT (13.2)
4π d
In Eq. 13.2, PT and P R are the power level at the transmitter and the
receiver, respectively, G T and G R describe the antenna gain at the
transmitter and the receiver, respectively, λ is the wavelength, d is
the distance between transmitter and receiver and is the coefficient
which describes the attenuation. Typical real-life values are between
2 and 6 [8]. The theoretical limit is 2 and is very often used in
simple model calculations for the so-called free space loss (FSL).
This square dependency is predominantly calculated in logarithmic
scale:
F S L = P R /PT = 20 · log(λ/4π d) + log (G R ) + log (G T ) (13.3)
Doubling d, leads to the well-known reduction of FSL by – 6 dB.

Based on this square propagation law, we can easily calculate the
C/I ratio under the assumption that all parameters (G T , G R , PT , and
λ) are identical for the two senders:
C I = 20 log(P R,c /P R,i ) = 20 log(di /dc ) (13.4)
If CI is required to be larger than 10 dB (e.g., for a commercially
available single chip radio transceiver at 2.4 GHz and a data rate of
250 Kbps [34]), then di ≥ 100,5 dc = 3.16 dc , i.e., distance between
interferer and receiver (denominated di ) must be more than three
times larger than the distance between the wanted (channel) sender
and the receiver (denominated dc ).
If the first adjacent channel sensitivity is regarded, which is
−20 dB, then the ratio shrinks to di ≥ 10−0.1 dc = 0.1 dc , i.e., the
wanted channel can be 10 times farer apart from the receiver than
the interferer.
Let S1 and S2 be two senders communicating with two
receivers R1 and R2 , respectively. Then, we can illustrate the spatial
characteristics as shown in Fig. 13.3. This figure can be read as
follows: Let us first assume that the CIi must be higher than 6 dB and
both station S1 and S2 transmit on the same channel. Left of the solid
Figure 13.3. Spatial distribution of 6 dB and −6 dB C/I and ACS lines.

line, the CI1 is above 6 dB, i.e., the receiver R1 must be left of the line
in order to successfully receive data from S1 . Respectively, the CI2 is
above 6 dB right of the dotted line, i.e., the receiver R1 must be left
of the line in order to successfully receive data from S1 .
The same kind of figure can also be used for adjacent channel
sensitivity. If we assume that both stations S1 and S2 transmit on the
neighbor channel and that AC Si shall be higher than −6 dB, then
receiver R1 must be left of the dotted line in order to successfully
receive packets from S1 and R2 must be right of the solid line.
In this case, we consequently have an overlap space, where both
communication paths work.
As a result, we achieve a so-called space division multiple access
(SDMA), which is discussed in Section 13.5.1. A notion often used
in this context is the so-called spatial capacity, which is originally
defined as bits/sec/cubic-meter [28]. In more recent sources, it is
defined as bits/sec/square-meter, since the relevant coverage area
is usually two dimensional rather than three dimensional [41].
It should be highlighted at this point that the characteristics only
depend on the ratio of the distances and of C /I , but neither on
the absolute distances, nor on the absolute output powers. This is
shown in Fig. 13.3 for a two-dimensional case, where the axes show
the relative distances between Section 13.5.1 both stations S1 and
S2 in two directions. Increasing the output power, as it is proposed
in various cases, does not help to solve the intrinsic coexistence
issue. Increasing the output power even aggravates the situation,
as the spatial capacity degrades. Intrinsic interference is extensively
examined in the literature for mobile applications (GSM, UMTS) [2].
13.3.3 Measurement of Real Packet Losses due to

Extrinsic Interference
Many analytical models, network simulations, and measurements in
literature describe extrinsic interference behavior between WLAN
and short-range wireless networks for local, home, body, or personal
area networks (LAN, HAN, BAN, PAN). It is especially for the 2.4 GHz
band, where a multitude of different technologies and products are
available, and where SRWN are the potentially endangered, due to
100%
80%
frame loss rate
60%
40%
20%
0%
2425 2430 2435 2440 2445 2450 2455
IEEE802.11b mean frequency [MHz]
Figure 13.4. Loss of IEEE802.15.4 frames with a high activity IEEE802.11

DSSS at varying center frequencies; the 802.15.4 channel kept constant at
2440 MHz [10].
the fact that their output power may be significantly lower than that
of the competing systems, e.g., WLAN.
Some representative results shall be shown here:
• Reference [18] gives a fundamental analytical estimation on

the coexistence impact of an IEEE 802.15.4 network on IEEE
802.11b devices.
• Reference [35] describes lab-based measurements, where
interfering signals are directly injected into the RF path.
• Reference [10] reports on real-world measurements, where
IEEE802.11b- and IEEE802.15.4-stations coexist and inter-
fere. Worst case packet loss rate for IEEE802.15.4 goes up
as 92% (Fig. 13.4).
13.3.4 Effects of Coexistence Problems

Needless to say that packet loss is not a problem per se and of course
does not necessarily lead to data loss, as within the course of reliable
communication, lost packets can be retransmitted. However, these
retransmissions may lead to an increased delay, may introduce delay
variations (jitter), and may add to the energy consumption.
Figure 13.5. Delay of Bluetooth point-to-point UDP traffic under the

influence of interference [33].
The delay and the jitter due to packet loss add to the overall
system delay and jitter. There are multiple analyses on the jitter
behavior of wireless communication under interference conditions.
A very detailed analysis for WLAN and Bluetooth behavior can be
found at in reference [33]. Figures 13.5 and 13.6 show the results for
delay and jitter for UDP traffic in a point-to-point topology. In case of
single coexistence with one interfering station (time 30 s to 60 s),
Figure 13.6. Delay of Bluetooth point-to-point UDP traffic under the

influence of interference [33].
Countermeasures — An Overview 387
the average delay increases by 26%, whereas in double coexistence

with two interfering stations (time 60 s to 90 s) by 49%.
13.4 Countermeasures — An Overview
13.4.1 Safety Aspects

It is one of the basic requirements for networking to provide safe
operation [9]. “Safety is the condition of being protected against
physical, social, spiritual, financial, political, emotional, occupa-
tional, psychological, educational, or other types or consequences of
failure, damage, error, accidents, harm, or any other event . . . . Safety
can take the form of being protected from the event or from exposure
to the event [51].”
It is especially the second sentence, which describes the two
main directions that
• The system either avoids to be exposed to the event. Ex-
posure is mainly avoided by planning and/or coordination
with other systems.
• Or the system attempts to be protected against the
event. Protection can be achieved by redundancy and/or
adaptivity.
13.4.2 Classification
These two directions of countermeasures can be performed on the
various levels of (wireless) networking (physical, data link, network,
and partially transport layer). They can be achieved through the
following activities:
• Company Policies: This might include prohibition of some
wireless products on campus, e.g., Bluetooth or 802.11b.
Obviously, applications are moving away from this practice,
as they are not suitable for changing topologies.
• Regulation Bodies: The rules from regulation bodies might
include basic coexistence rules, such as LBT or TPC (Section
13.2.2)
• Standards Bodies: The most prominent example of
a standard-based approach was offered by the legacy
IEEE802.2 workgroup with regard to the coexistence be-

tween IEEE802.11 (WLAN) and IEEE802.15.1 (Bluetooth)
[43].
• Technical Innovations: These look for new solutions with
regard to physical and data link layer protocols or with
regard to system level (driver) solutions.
With regard to the collaboration with other protocols, one might also
distinguish collaborative approaches, which take into the account
the specific behavior or other protocols, and non-collaborative
approaches, which don’t.
The countermeasures on the physical layer (Section 13.5) are
around the technologies to split up one medium into different
channels, e.g., with space, frequency, or code division multiple access
(SDMA, FDMA, CDMA). In contrast to that the countermeasures on
the data link layer (Section 13.6) are built around the variations
of time division multiple access (TDMA), which allows multiple
stations use one channel.
13.5 Countermeasures on Physical Layer
13.5.1 Channel Classification and Selection

One of the basic countermeasures is the space division multiple
access (SDMA) approach, already described in Section 13.3.2. Thus,
the space division might be relevant not only for channel reuse, i.e.,
reuse of the same channel, but also for neighbor channel selection.
In order to perform the channel selection, manual frequency
planning has been performed for decades, e.g., for radio and
television broadcast services and for mobile (GSM) services. Of
course, this manual planning has no practical importance for mobile
services, such as WBANs. Automatic channel selection requires
autonomous channel classification of the stations. For practical use,
two flavors can be found [1]:
• Active classification can be done during the course of nor-

mal communication, or the devices can exchange dummy
packets with the specific goal of building a classification list.
Countermeasures on Physical Layer 389
• Passive classification is accomplished by listening to chan-

nels. Most of today’s single-chip transceivers come with two
options of passive channel supervision:
– Energy detection (ED) is an estimate of the received
signal power within the bandwidth of the channel. No
attempt is made to identify or decode signals on the
channel [44].
– Carrier sense reports a busy medium only upon
the detection of a signal compliant with the same
protocol with the same modulation and spreading
characteristics of the PHY that is currently in use by
the device. This signal may be above or below the ED
threshold [44].
In the case of active classification, it is mostly the actual channel,

which is classified, whereas in the case of passive classification,
the stations may scan all available channels and thus determine a
spectral activity map.
Two basic characteristics add complexity to this classification:
• As the wireless signal is spatially distributed, the obser-

vation of one station has only local significance. It cannot
describe the quality at any other station or any other
localization. As this has only small relevance for small-scale
networks, its importance raises/becomes crucial with the
size of the network. It is especially critical for spatially
distributed multi-hop network, where different parts suffer
from different interfering influence. This case is illustrated
in Fig. 13.7. The three interferers I1 , I2 , and I3 have an
impact on only a subset of the stations Sx y . In addition, the
optimum case for one station, e.g., S31 , might be prohibitive
for another station, e.g., S24 . Some solutions to this problem
are proposed in reference [11].
• If nodes are extensively using power-down modes, they
might not be informed about a change in frequency —
and thus have to re-register. This adds additional jitter to
the transport characteristics [30]. Furthermore, channel
observation is impossible in power-down phases, which
Figure 13.7. Illustration of spatial distribution in a wireless mesh

network.
may cause complete communication failure when turning

back to active mode.
• In addition, the classification has only limited timely
relevance, too. If the channel classification is performed
periodically, the channel selection can be adapted. This
is performed in legacy telecommunications systems, as in
DECT [46] or GSM [2].
However, many of the simpler SRWN protocols perform the channel

classification only at network or connection start-up [44].
In real-life wireless protocols, adaptive channel selection is
merely used as addition to basic services. For IEEE802.11, only the
European h-extension provides the required service primitives, such
as measurement request, measurement report, supported channels,
and channel switch announcement (Fig. 13.8) [48].
ZigBee networks, since the 2007 specification [45], allow a
slow dynamic adaptation of the used channel, which is denomi-
nated frequency agility. It provides a method for the network to
change channels in the event of interference. In order to avoid
the aforementioned problems of spatial distribution, it is the
network coordinator that detects potential interference on a channel
and directs the devices on the network to change to a better
channel.
IEEE 802.11 STA IEEE 802.11 STA

SME MLME MLME SME
Decision
to switch
channel
Channel Switch
MLME-CHANNEL Announcement frame MLME-CHANNEL
SWITCH.req (multiple) SWITCH.ind
Decision
to follow
switch
MLME-CHANNEL
SWITCH.rsp
Channel switch Channel switch

via PLME via PLME
MLME-CHANNEL MLME-CHANNEL
SWITCH.cfm SWITCH.cmf
Channel Channel
switch switch
complete complete
Figure 13.8. Channel switch decision in IEEE802.11h [48].
Frequency agility differs from frequency hopping with regard to

the time constants of frequency changes and the adaptivity to real-
world impacts.
Some vendor-specific WLAN management tools run an add-
on to IEEE802.11-WLAN-Access Points, e.g., [37]. Bluetooth V1.2
[47] enables the frequency hopping (Section 13.5.2) to be adaptive
(AFHSS).
13.5.2 Frequency Hopping

Frequency hopping spread spectrum (FHSS) is the simplest spread
spectrum technique, which helps to counteract against frequency
specific interference on a statistical basis. FHSS uses M different
carrier frequencies that are modulated by the source signal. At one
moment, the signal modulates one carrier frequency; at the next
moment, the signal modulates another carrier frequency. In the long
run, all M frequencies are used. The frequency pattern must be
agreed between the synchronized sender and receiver.
Table 13.1. Probability of l = 3 subsequent collisions assuming n pairs of

nodes using m channels in a random selection. m = 58 corresponds to one,
m = 36 to two WLAN interferers [9]
p
n m = 80 m = 58 m = 36
2 1,95E-06 5,13E-06 2,14E-05
3 7,81E-06 2,05E-05 8,57E-05
4 1,95E-05 5,13E-05 2,14E-04
5 3,91E-05 1,03E-04 4,29E-04
10 3,22E-04 8,46E-04 0,0035367
20 0,0025977 0,0068168 0,0285093
30 0,0087794 0,0546389 0,0963579
40 0,0208207 0,0546379 0,2285273
50 0,0406748 0,1067401 0,4464668
The stability against constant-frequency interferers is only

limited to those time slots when the frequencies do not overlap
(as discussed in Section 13.3.1). This also allows running several
FHSS systems in parallel. However, due to non-orthogonality, the
theoretical aggregate gross rate of M times of the gross rate of one
channel is unachievable.
Table 13.1 describes the collision probability of multiple fre-
quency hopping system within the same range (intrinsic interfer-
ence) in the presence of WLAN systems (extrinsic interference).
In reference [13], an information-theoretic approach for analysis
of various multiple access schemes is given considering various
propagation factors such as path loss, shadowing, and fading.
Figure 13.9. Frequency Hopping Spread Spectrum System.

However, this contribution mainly concentrates on outdoor radio

channels.
Reference [14] evaluates the possible capacities of FHSS systems
under indoor conditions and indicates that the number of piconets
supported in an area of 20×20 m2 varies from 15 for a fully loaded
system to around 50 for a 20% load system.
Whereas FHSS systems are robust against non-hopping systems,
those non-hopping systems may show a severe degradation. This
problem is especially well studied for the coexistence between
IEEE802.11 WLAN and Bluetooth (BT), which has been in the
focus of IEEE802.15.2 work group [43] [36]. In this case, 22 out
of the 79 Bluetooth frequencies overlap with one WLAN channel.
In addition, WLAN packet length may be twice as long as the BT
hopping time (1.293 ms for an IEEE802.11 packet with 1514 bytes
at 11 MBit/s compared to 625 μs BT time constant). Furthermore,
overlap times (Section 13.6.3) must be taken into account. The
results [15] show the severe impact on WLAN and the relative
moderate impact on BT. These results are widely validated also in
recent publications, e.g., reference [16].
In general, frequency hopping is a very popular approach, as it
is simple in implementation, uses narrowband transmission, and
thus enables low power during operation, and delivers relatively
good robustness. The major disadvantage of FHSS systems is
the need of synchronization, which has to be sustained also
during power-down periods. Thus, FHSS needs either a relatively
long re-synchronization period after a power-down period, or a
low-power timer, which is sufficiently precise to keep the node
synchronized.
Bluetooth is certainly the FHSS with highest prevalence. So
three of the recent developments of Bluetooth shall be mentioned
here:
• In order to reduce the overall energy consumption due

to synchronization times, the Bluetooth low energy tech-
nology reduces the number of synchronization channels
to four. This approach is also proposed for many other
frequency hopping or frequency agile systems, e.g., refer-
ences [9] and [45].
• The statistical approach of plain FHSS can be extended

through adaptivity, as described in Bluetooth V1.2 (adaptive
FHSS, AFHSS). The adaptivity helps to blacklist a subset
of frequencies. The rules, which lead to the blacklisting,
are not standardized, but is obvious that — as active
classification is used — it is targeted towards static
interferers, e.g., IEEE802.11 WLAN.
It should be mentioned that the blacklisting of some
channels within the course of AFHSS
– Is mainly helpful for the robustness of the WLAN
transmission, as the probability that Bluetooth packets
interfere with the WLAN frames on their static
frequency is significantly reduced
– Decreases the potential spatial capacity of Bluetooth
• The third approach is on the driver-side in order to
coordinate the channel access of the different media.
The most prominent example for this is certainly the
cooperation between IEEE802.11 WLAN and Bluetooth.
This cooperation on driver level has already been proposed
in reference [43]. It came to life early, but only within
proprietary solutions [42].
Many other protocol developments targeted towards real time and

safe operation chose the FHSS as well. The most prominent examples
are Wireless HART [49], a protocol for larger-scale multi-hop
networks, where frequency patterns are adaptively agreed between
pairs of communication partners and where synchronization is
maintained through high-quality timers, and WISA [12] for short-
latency transmission.
Proprietary protocols include [9] and allow up to six retransmis-
sions on three different frequencies within 5 ms using low-cost and
ultra-low energy (battery powered) transceiver and microcontroller.
13.5.3 Frequency Spreading and Code Division Multiple

Access
The direct sequence spread spectrum (DSSS) technique also
expands the bandwidth of the original signal, but the process is
Figure 13.10. Direct Sequence Spread Spectrum (DSSS) System.
different. In DSSS, we replace each data bit with n bits using a

spreading code. In other words, each bit is assigned a code of n
bits, called chips, where the chip rate is n times that of the data
bit.
At the receiver, the received signal is de-spread with the same
spreading code. Due to the autocorrelation of the transmitted
PN sequence with the PN sequence that the receiver believes
the transmitter is using, DSSS achieves a so-called process gain,
which lowers the required SNR. Thus the impact of uncorrelated
interferers can be reduced.
The longer the PN sequence is, the better the achievable
autocorrelation might be. As a result, the process gain increases.
The process gain may be so large that code division multiple access
(CDMA) can be used. However, as also the required bandwidth
increases, the most protocols targeted for WBANs come with only
relatively short PN sequences, e.g., between 4 (e.g., IEEE802.15.4
[44]) and 11 (e.g., IEEE802.11 [48]). The achievable process gain
mainly helps to suppress deep fading in multipath environments, but
does significantly improve coexistence robustness, as can be seen in
Section 13.3.3.
The only exception to this, which is currently visible, is UWB.
13.5.4 The Promise of Ultra-Wide-Band

UWB technologies have found their place for manifold radar- and
measurement applications, as well as for military communication,
mainly since the eighties of the last century. It is only in this decade
that UWB gained importance for communication in consumer and
automation applications [3].
The basic idea of UWB transmission is the usage of an extremely
broad bandwidth, so that the transmission power is distributed
Figure 13.11. Spatial capacity comparison between IEEE 802.11, Blue-

tooth, and UWB [31].
over this broad spectrum. Thus the intensity of UWB-radio systems,

i.e., the power per frequency (W/MHz) is correspondingly low.
The demodulation and decoding of the received signal is achieved
through an appropriate process gain. Hence UWB devices appear as
noise to most primary users. Reference [7] gives a good overview on
the challenges of UWB systems, which come with a lot of strategic
advantages for sensor-actuator networks [20, 21], including the
coexistence robustness. Some detect-and-avoid (DAA) strategies for
UWB are discussed in reference [22]. An excellent analysis on the
coexistence behavior of UWB systems can be found in reference [32].
UWB also promises to achieve an extraordinary spatial capacity,
as shown in Fig. 13.11.
13.6 Countermeasures on Data Link Layer
13.6.1 Basic Medium Access Control

As introduced in Section 13.4, the countermeasures are built around
the variations of time division multiple access (TDMA), where
stations share a common physical channel. Thus the whole discipline
discusses about the best moment of time for the stations to access
Countermeasures on Data Link Layer 397
the medium. Thus the problem is commonly denominated medium

access control (MAC). For MAC algorithms, we classify centralized
and de-centralized architectures. In de-centralized MAC algorithms,
the coordination function is distributed among several peer stations,
i.e., there exists no master node to synchronize the medium access.
For body area networks, both approaches have certain relevance.
Mainly very small and simple WBANs follow a centralized approach,
for example with one master data collector. For more complex
and spatially distributed WBANs, more and more de-centralized
algorithms are applied.
13.6.2 Centralized Approach

Centralized architectures, where a master device assigns time slots
to its slaves, allow deterministic, i.e., hard real-time behavior, as long
as the physical channel (Section 13.5) is available.
If the assignment works properly, no collisions should oc-
cur. However, centralized approaches come with two major dis-
advantages, which restrict their use in general sensor WBAN
architectures.
• All slave stations must remain synchronized with the

master, which in the general case requires precision timers
and regular activity.
• In case that the synchronization is performed within the
communication channel, the topology is limited to star or
hierarchical star, i.e., tree topologies. This is, however, not
necessarily a major restriction for WBANs. It can be over-
come by synchronization methods for changing topologies.
The most prominent example is certainly IEEE1588 [50],
where major efforts are invested into the topology setup of
a tree.
Excellent overviews of synchronization protocols for WSNs can be

found in references [24], [25], and [26]. In addition, it is a relatively
common practice to use side-channel synchronization, e.g., with the
help of terrestrial wide area clock signals (DCF77 [38], WWV [39] or
alike, or spatial synchronization signals, i.e., GPS [40].
In addition, special care must be taken for the registration of the

slave at the master station. It is either registered via a (manual) side
channel or during a predefined period of time, when de-centralized
access is enabled. This second approach directly leads to hybrid
methods.
13.6.3 Duty Cycle Management

A very easy approach to set up a decentralized medium access,
which, however, is still commonly used in many wired and wireless
systems, is the restriction of the duty cycle. If the load on the
channel is low enough, the probability that two packets collide may
be small enough to have no significant impact on the transmission
characteristics. Duty cycle may be limited by regulation (Section
13.2.2). It can be as low as 1% (e.g., 868.6–868.7 MHz, CEPT
ERC/REC 70-03, Annex 7) or even 0.1% (e.g., 869.25–869.3 MHz,
CEPT ERC/REC 70-03, Annex 7). The regulation details describe the
exact conditions, i.e., the related time. For example, for the above-
mentioned CEPT ERC/REC 70-03, it is 1 h.
Duty cycle management is still used in many fields. It is simple
and needs only (unidirectional) transmitters. However, it is of course
only generally suitable for low-performance connectivity.
13.6.4 Channel Sensing Methods

Many regulation rules prescribe listen-before-talk (LBT), which
is sometimes also called carrier sense (CS). In this case, the
(bidirectional) transceiver has to supervise the channel for a certain
amount of time before starting the transmission. LBT helps to avoid
collision with an already active transmission and is, therefore, very
helpful in reducing the collision probability.
However, two aspects restrict its use for wireless networks:
• LBT does not give a guarantee that a collision cannot

occur. If two stations start their LBT simultaneously, then
both stations can start their transmission without breaking
the rules, but causing a collision. For a real system with
propagation times and turn around times (required time
to switch a half-duplex transceiver from receiver into
transmitter mode), we are talking about time intervals for
Figure 13.12. Timing behavior of channel sensing for LBT strategy,

including turn around and propagation times (translated from [5]).
simultaneity. This problem is well known since decades,

as it already occurred in the wired coax- or hub-based
Ethernet. It is a matter of collision avoidance strategies
like persistency to further reduce the collision probability.
The timing behavior at one transceiver is illustrated in
Fig. 13.12.
• The other restriction is caused by the spatial distribution of
wireless communication and which is commonly described
as hidden station problem. If it is assumed that in Fig.
13.7 two stations, e.g., Si 1 and Si 3 , do not hear each other,
but have overlapping ranges, so that the intermediate
station Si 2 can hear both. Applying LBT at Si 1 and Si 3
does not remedy a collision at Si 2 . This problem can only
be solved with additional countermeasures, e.g., RTS-CTS-
mechanisms.
As already described in Section 13.5.1 there are different methods
of channel sensing. The impact of different channel sensing
methods is, for example, evaluated in reference [17] with regard
to the impact on IEEE802.15.4 performance under IEEE802.11b
interference.
13.6.5 Persistency and Collision Avoidance

It is also the hidden station problem, which rules out the possibility
to use the collision detection (CD) approach, as it is applied in
IEEE802.3 (wired Ethernet). For this, it is necessary that a sender

can make reasonable assumptions about the collision (coexistence)
situation at the receiver at different displacement. The CD approach
has the major advantage that a transmission can be stopped at a very
stage of the transmission.
Due to the hidden station problem, it is only after the end of a
transmission and a missing acknowledgment (ACK) that a station
learns about a lost packet. This might lead to the situation that in a
busy network, multiple stations have to perform retransmissions. If
all of them perform LBT directly after the medium was relinquished
and use the same time constants, this necessarily leads to a collision.
Furthermore, the collision probability may be increased in case
of repeated broadcast packets. If, for example, S21 in Fig. 13.7 sends
out a broadcast packet, which is then forwarded by all directly
neighbored stations (S11 , S12 , S22 , S31 , and S32 ), this will lead to
collisions despite the usage of LBT.
This problem can be improved if the stations reduce greed to
access the channel, which is described by the notion of persistency. A
persistent (or 1-persistent) transmits its data that is has to send with
a probability of 1 whenever it finds the channel idle. If the channel is
busy, the station waits until it becomes idle (greedy approach).
Non persistent behavior differs from 1-persistent, when the
channel is found busy. In this case, the station deliberately waits
a random period of time before sensing the channel again. This
approach was first used in the legacy Aloha system [23] for
retransmissions. In Aloha and in many other systems (including
CSMA/CD Ethernet), a so-called binary back off algorithm is applied.
If the transmission is organized in time slots, we can define a
probability p that a station transmits its data that is has to send.
With a probability of q = 1 − p, a p-persistent station defers the
transmission until the next slot and the same procedure iterates. If
the channel has become busy meanwhile, the station waits a random
time and starts again. If the channel is busy when first sensing it,
the station waits until the next slot and repeats the procedure. The
impact of the variation of p in the presence of a FHSS interferer is
investigated in reference [16].
As an example, the back off mechanism of IEEE802.15.4 is shown
in Fig. 13.14. Each IEEE802.15.4-compatible device shall maintain
Figure 13.13. CSMA/CA mechanism including back off algorithm for

unslotted operation in IEEE802.15.4 [44].
three variables for each transmission attempt: NB, CW, and BE. NB is
the number of times the CSMA-CA algorithm was required to back
off while attempting the current transmission; this value shall be
initialized to zero before each new transmission attempt. CW is the
contention window length, defining the number of back off periods
that need to be clear of channel activity before the transmission
can commence; this value shall be initialized to two before each
transmission attempt and reset to two each time the channel is
assessed to be busy. The CW variable is only used for slotted
Figure 13.14. Possible delay of data frames for EnOcean Radio Protocol
[own calculations].
CSMA-CA. BE is the back off exponent, which is related to how many

back off periods a device shall wait before attempting to assess a
channel. BE shall be initialized to the value of macMinBE. In slotted
systems with the received BLE subfield set to one, this value shall
be initialized to the lesser of two and the value of macMinBE. Note
that if macMinBE is set to zero, collision avoidance will be disabled
during the first iteration of this algorithm.
Although the receiver of the device is enabled during the channel
assessment portion of this algorithm, the device may discard any
frames received during this time.
For unslotted CSMA-CA, the MAC sublayer shall initialize NB and
BE and then proceed directly to step 2. The MAC sublayer shall delay
for a random number of complete back off periods in the range 0
to 2BE – 1 (step 2) and then request that the PHY perform a CCA
(step 3), which shall start immediately.
If the channel is assessed to be busy (step 4), the MAC sublayer
shall increment both NB and BE by one, ensuring that BE shall be
no more than macMaxBE. If the value of NB is less than or equal to
macMaxCSMABackoffs, the CSMA-CA algorithm shall return to step 2.
If the value of NB is greater than macMaxCSMABackoffs, the CSMA-
CA algorithm shall terminate with a channel access failure status.
It should be mentioned that persistency is not limited to use

with channel sensing. Reference [27], for example, uses a random-
based back off mechanism for unidirectional transmitter in order
to keep the collision probability low. The usable time slots for
this system are shown in Fig. 13.14. The upper part shows the
timing of the three transmission attempts of a standard module,
where the first two attempts are performed at fixed times. For the
third attempt, one of four possible time slots is selected through
a random process with equal probabilities for all of the four time
slots.
The lower part of Fig. 13.14 shows the delay of a repeated frame
under the condition that the repeater receives the first frame of a
sender. The time slots for the repeated packets are also selected on
random processes.
In addition, back off times can also be used to prioritize the
traffic. This approach is mostly handled with regard to the so-called
inter-frame spaces (IFS), which anyhow have to be regarded due
to the physical characteristics of the channel (propagation time,
echoes) and the transceivers (turn-around time). It is applied in
many wireless systems in order to keep data packets and their
respective acknowledgments together. In addition, it can be used to
allow management or other low-latency traffic prioritized access to
the channel.
IEEE802.11 defines five different IFSs to provide priority levels
for access to the wireless media [48]: short IFS (SIFS), PCF IFS
(PIFS), DCF IFS (DIFS), arbitration IFS (AIFS), which is used by the
quality of service facility, and the extended IFS (EIFS). Figure 13.15
Figure 13.15. Some IFS relationships in IEEE802.11 [48].

shows some of these relationships in IEEE802.11, where the short

IFS give a higher priority to the stations or the messages.
13.6.6 Medium Reservation Methods

The persistency and collision avoidance methods do not have any
impact in case of hidden stations. This problem can be improved
by the usage of ad hoc medium reservation methods. The basic and
maybe mostly used mechanism is the so-called RTS-CTS handshake,
as it is defined in IEEE802.11 [48, 4].
The handshake of Fig. 13.16 begins when the sender sends a
short request-to-send (RTS) packet; the receiver then replies with
a clear-to-send (CTS) packet.
• Only afterward, the sender begins the actual data transmis-

sion.
• A node further away from the sender may not receive the
RTS during carrier sense; it receives, however, the CTS
which prompts it to back off.
• If a collision occurs during RTS, the packet is deliberately
made small that the collision doesn’t make much damage.
Figure 13.16. Medium reservation mechanism with RTS-CTS

management frames, as used in IEEE802.11 [48], after [4].
Conclusions 405
13.7 Conclusions
Coexistence remains one of the major concerns and challenges of

license-exempt bands, as they are used for WBANs. A variety of
approaches has been developed, as the avoidance of coexistence
impact is subject to a conflict of objectives. Therefore, no approach
can be found, which is optimum for all cases, but the selection has to
be performed on a case-by-case basis.
References
Book reference
1. R. K. Morrow (2004) Wireless Network Coexistence, McGraw Hill.
2. J. Eberspächer and H.-J. Vogel (1997) GSM Global System for Mobile
Communication (Vogel).
3. I. Oppermann, M. Hämäläinen, and J. Iinatti (2004) UWB — Theory and
Applications, John Wiley and Sons.
4. A. Sikora (2003) Wireless Personal and Local Area Networks, John Wiley
and Sons.
5. G. Kupris and A. Sikora (2008), ZigBee – Datenfunk mit IEEE802.15.4 und
ZigBee, Franzis-Verlag Poing,
Journal reference
6. C. Borean (2007) Analysis of ZigBee Interferences in WBANs, in First
European ZigBee Developers’ Conference, June 07, Munich, Germany.
7. H. T. Friis (1946) A Note on a Simple Transmission Formula, in
Proceedings of the IRE, 34.
8. U. Walter (September 2005) Behavior of IEEE 802.15.4 Standards in
the Various Frequency Bands, in Design and Elektronik Developer Forum,
Berlin.
9. A. Sikora and D. Lill (2008) Wireless Technologies for Safe Automation –
Insights in Protocol Development, in 13th IEEE International Conference
on Emerging Technologies and Factory Automation (ETFA), Hamburg.
10. A. Sikora and V. Groza (2005) Coexistence of IEEE802.15.4 With
Other Systems in the 2.4 GHz ISM Band, in IEEE Instrumentation and
Measurement Technology Conference (I2 MTC), Ottawa.
11. A. Sikora (2005) Adaptive Coexistence of Meshed ZigBee Networks and

Other Systems in the Same ISM Band, in Wireless Technologies Kongress,
Mannheim.
12. G. Scheible, J. Endresen, D. Dzung, and J.-E. Frey (February 2007)
Unplugged but Connected: Design and Implementation of
a Truly Wireless Real-Time Sensor/Actuator Interface, IEEE Spectrum,
44(2).
13. D. Verhulst, M. Mouly, J. Szpirglas (July 1984) Slow Frequency Hopping
Multiple Access for Digital Radio Telephone, IEEE Journal on Selected
Areas in Communication, SAC-2.
14. A. Kumar and R. Gupta (2001) Capacity Evaluation of Frequency
Hopping Based Ad-Hoc Systems, in Proceedings of the 2001 ACM
SIGMETRICS International Conference on Measurement and Modeling of
Computer Systems.
15. N. Golmie, R. E. V. Dyck, A. Soltanian, A. Tonnerre, and O. Rébala
(2003) Interference Evaluation of Bluetooth and IEEE 802.11b Systems,
Wireless Networks, 9, pp. 201–211.
16. I. Ashraf, K. Voulgaris, A. Gkelias, M. Dohler and A. H. Aghvami
(November 2009) Impact of Interfering Bluetooth Piconets on a
Collocated p-Persistent CSMA-Based WLAN, IEEE Transactions on
Vehicular Technology, pp. 4962–4975.
17. S. Y. Shin and H. S. Park (June 2008) On the Impact of Channel
Sensing Methods to IEEE802.15.4 Performances Under IEEE802.11b
Interference, Journal of Communications and Networks, 10(2).
18. I. Howitt and J. A. Gutierrez (2003) IEEE 802.15.4 Low Rate Wireless
Personal Area Network Coexistence Issues, in IEEE Wireless Communi-
cations and Networking Conference (WCNC).
19. C. -C. Chong, F. Watanabe, and H. Inamura (2006) Potential of UWB
Technology for the Next Generation Wireless Communications, in 2006
IEEE Ninth International Symposium on Spread Spectrum Techniques and
Applications.
20. G. Dolmans, O. Rousseaux, L. Huang, T. Fu, B. Gyselinkx, S. D’Amico,
A. Baschirotto, J. Ryckaert, and B. van Poucke (2007) UWB Ra-
dio Transceivers for Ultra-Low Power and Low Data Rate Com-
munications, in IEEE International Conference on Ultra-Wideband
(ICUWB).
21. J. Zhang, P. Orlik, Z. Sahinoglu, A. Molisch, and P. Kinney (March
2009) UWB Systems for Wireless Sensor Networks, Mitsubishi Electric
Research Laboratories Report TR2009-007.
References 407
22. S. M. Mishra and R. W. Brodersen (2007) Cognitive Technology for

Improving Ultra-Wideband (UWB) Coexistence, in IEEE International
Conference on Ultra-Wideband (ICUWB).
23. N. Abramson (1970) The ALOHA System — Another Alternative for
Computer Communications, in the AFIPS Conference Proceedings, 37,
FJCC 1970, S. 695–702.
24. F. Sivrikaya and B. Yener (July/August 2004) Time Synchronization in
Sensor Networks: A Survey, IEEE Network, pp. 45–50.
25. S. Johannessen (April 2004) Time Synchronization in a Local Area
Network, IEEE Control Systems Magazine, pp. 61–69.
26. B. M. Sadler and A. Swami (2006) Synchronization in Sensor Networks:
An Overview, in MILCOM 2006.
27. F. Schmidt, G. Scholl, A. Anders, H.-J. Körber, and H. Wattar, RF-
Embedding of Energy-Autonomous Sensors and Actuators Into Wireless
Sensor Networks, in Multifunctional Structures/Integration of Sensors
and Antennas Meeting Proceedings, RTO-MP-AVT-141, Paper 3, Neuilly-
sur-Seine, France, pp. 3.1–3.14
28. C. Paillard and J. Wight (2006) Inside UWB Design: A Tutorial,
CommsDesign, 29.3.2006.
29. C. Borean (2007) Analysis of ZigBee Interference WBANs, in First
European ZigBee Developers’ Conference, Munich, Germany.
30. G. Kupris (2008) Channel Agility and ZigBee/WLAN (IEEE 802.11 g and
n) Coexistence, in Second European ZigBee Developers’ Conference.
31. J. Foerster, E. Green, S. Somayazulu, and D. Leeper (2001) Ultra-
Wideband Technology for Short- or Medium-Range Wireless Commu-
nications, Intel Technology Journal, Q2.
32. M. Hämäläinen (2006) Single-Band UWB Systems: Analysis and
Measurements of Coexistence With Selected Existing Radio
Systems, Faculty of Technology, University of Oulu, available at
http://herkules.oulu.fi/isbn9514280644/isbn9514280644.pdf.
33. A. Miaoudakis, D. Stratakis, V. Zacharopoulos, and E. Antonidakis (June
2009) Co-existence Performance Evaluation of Wireless Computer
Networks in a Typical Office Environment, ComSIS, 6(1), pp. 169–
184.
Data Sheets and Application Notes

34. Nordic Semiconductor ASA, Product Specification Single Chip 2.4 GHz
Transceiver nRF2401, Rev. 1.1, June 2004.
35. R. Rodriguez, MC1319x Coexistence, Freescale Semiconductor, Ap-

plication Note, Document Number: AN2935, Rev. 1.2, 07/2005,
http://www.freescale.com/files/rf if/doc/app note/AN2935.pdf.
36. M. B. Shoemake (February 2001) Wi-Fi (IEEE 802.11b) and Blue-
tooth Coexistence Issues and Solutions for the 2.4 GHz ISM Band,
Texas Instruments, version 1.1, available at focus.ti.com/pdfs/vf/
bband/coexistence.pdf.
Websites
37. http://www.cisco.com/en/US/products/sw/cscowork/ps3915/.
38. http://www.ptb.de/en/org/4/44/442/ index.htm.
39. http://tf.nist.gov/timefreq/stations/wwv.html.
40. http://tycho.usno.navy.mil/gps.html.
41. http://www2.rohde-schwarz.com/en/technologies/Wireless
Connectivity/UWB/information/.
42. http://www.ti.com/bluetooth80211.
Standards and Protocols

43. IEEE
802.15.2-2003 IEEE Recommended Practice for Telecommunications
and Information Exchange Between Systems — Local and Metropolitan
Area Networks Specific Requirements — Part 15.2: Coexistence of
Wireless Personal Area Networks With Other Wireless Devices
Operating in Unlicensed Frequency Band, http://www.
ieee802.org/15/pub/TG2.html.
44. IEEE Std 802.15.4TM -2006: Wireless Medium Access Control (MAC) and
Physical Layer (PHY) Specifications for Low-Rate Wireless Personal
Area Networks (WPANs), http://www.ieee802.org/15/pub/TG4.html.
45. ZigBee
46. Digital Enhanced Cordless Telecommunications (DECT); Common
Interface (CI); Part 3: Medium Access Control (MAC) layer, EN 300 175–
3 V1.4.2 (1999–06).
47. https://www.bluetooth.org/foundry/adopters/document/
Bluetooth Core Specification v1.2
48. IEEE Standard for Information technology — Telecommunications and
information exchange between systems — Local and metropolitan
References 409
area networks —Specific requirements; Part 11: Wireless LAN Medium

Access Control (MAC) and Physical Layer (PHY) specifications.
49. HART Field Communications Protocol Specification, Revision 7.0,
August 2007.
50. IEEE P1588TM /D1-E, “Draft Standard for a Precision Clock Synchroniza-
tion Protocol for Networked Measurement and Control Systems”, 18 Mar
2007.
51. IEC 61508 “Functional safety of electrical/electronic/programmable
electronic safety-related systems (E/E/PES)”
Chapter 14
Implanted Wireless Communication

Making a Real Difference
Henry Higgins
Zarlink Semiconductor, Castlegate Business Park,
Portskewett, Caldicot, NP26 5YW , UK
henry m [email protected]
14.1 Introduction
This chapter outlines the requirement for in-body communication

with descriptions of body properties, design, testing, and ultimate
manufacture of a product for widespread use.
Since the first cardiac pacemaker was implanted in Arne
Larsson’s chest in 1958, it has been essential to know what implants
are doing and change the operation in the best interest of the patient.
Fitting an implant and hoping for the best is not a way forward.
The implant can be in a unique position to sense what is going on
within the body without the need for invasive surgery. Operating on
a patient increases the risk of infection, can be expensive, distresses
him or her, and may not show what is really happening. The
implant can become almost like part of the body and can feedback
information when he or she is going about their normal day-to-day
business. The patient does not need to be sedated or anesthetized,
and the data gathering can go on almost anywhere.

www.panstanford.com
412 Implanted Wireless Communication Making a Real Difference
14.2 Why In-body Communication?
The implant may not be operating to its best, so changes in vital

parameters may be necessary. This can best be done with a wireless
link. It hardly needs to be said that wires passing through the skin
are a source of infection, can result in injury (if pulled or knocked),
and are often uncomfortable for the patient.
The communication system is an addition to the implant and is
there to make it work better and provide feedback to the medical
team.
14.3 Applications
Typical applications for a communication system are closely linked

to the implant applications. The pacemaker has been mentioned
above; other applications include the life-saving cardiac defibrillator,
stimulation of muscles can bring movement back to limbs paralyzed
by injury or disease, control of bowel or bladder incontinence can
mean freedom for thousands of otherwise very distressed patients.
The cochlear implant has brought hearing to many whom the
conventional hearing aid will not help.
Many other applications are under development, such as nerve
and muscle stimulation. Imaging the gastrointestinal tract has
proved to be a valuable diagnostic tool where a camera pill transmits
pictures of the esophagus, stomach, small and large bowel. These
images are not possible to obtain even with a MRI scanner [1].
14.4 MICS and ISM Bands
There are several things to consider when choosing a frequency

band, or bands, for implant communication. Tissue penetration is
important as well as approved worldwide and accepted for implant
use.
The band accepted worldwide for implant use is the Medical
Implant Communication System (MICS) band, which is based at
403 MHz [2]. This band may be widened if the approval bodies
agree. This band has the advantage of good propagation through
RFID Techniques 413
the body, and it can be used for two-way communication. There is

a restriction on the effective transmit power of 25 μW (−16 dBm);
this limit applies outside of the body. The band is split up into 300
kHz channels with limits on out-of-channel interference. Use of this
band within the prescribed limits does not require a license.
The Industrial Scientific and Medical Bands (ISM) are several in
nature [3]. This band can be used, for example, to transmit data from
a camera pill working its way through the digestive system, as this
application is not considered to be an “implant.” An ISM frequency
can be used to wake up a MICS band circuit but not for two-way
data transfer. ISM bands can also be used between non-implanted
hearing aid. The permitted power levels and frequency bands vary
from country to country, so it is essential to check on any national
regulation that may preclude operation. One solution to this is to
have country variants of the system. An important consideration
when using ISM bands is that they are not universal; for example,
in the United States there is a 915 MHz band, but in Europe
the equivalent is 868 MHz with no overlap. If one of these bands
were to be used, a variant would need to be made for each block
or the implant would need to be capable of being reprogrammed in
situ to change frequency. If one of these bands were to be used, then
there may need to be country variants — hard wired or software
controlled.
Typically, the range of operation, from patient to base-station,
will be 3 m or less. If a body-worn data gathering system is used,
the range could be even less.
As regulations will change with time, it is prudent to check on the
current status.
A 27 MHz band has been used for control of a capsule within
the gastrointestinal tract, but the antenna will be very much smaller
than 5 m, which would be the case for a dipole in free space and,
therefore, will be of poor efficiency.
14.5 RFID Techniques
Communication with electromagnetic induction similar to Radio

Frequency Identification(RFID) has been successfully used for years,
where only a small data transfer is needed. Wherever practical, this

approach can be useful due to the big advantage of not needing a
battery or for not draining the existing battery. RFID communication
cannot, however, be initiated by the implant but needs to be started
externally.
RFID is useful where an implant may be in place for a very long
time without recourse to a battery change or where the data transfer
is small. Applications for this include cranial pressure sensing, which
is performed at regular intervals, or interrogating the status of an
aortic arterial aneurism (AAA) repair [4].
14.6 Propagation Through the Body, Changes in Body

Shape and Posture
Body tissue is variable in its electrical properties and also varies

with frequency as Table 14.1 shows. The relative dielectric constant,
εr , affects the wavelength of an RF signal; in free space, εr = 1 and
the wavelength at 403 MHz is 74.4 cm, but when traveling through
muscle (εr = 58), this will be reduced to 9.76 cm. The conductivity,
σ , will affect how the tissue absorbs the signal. For air, σ is almost
zero. The characteristic impedance, Z r , is an issue when a signal
goes from one medium to another as there will be a reflection at
the boundary as shown in Eq. 14.1. For air, Z r = 377 , sometimes
denoted Z o .
Za − Zr
= (14.1)
Za + Zr
Table 14.1. Dielectric constant, conductivity, and characteristic

impedance of muscle and fat
Muscle Fat
Frequency εr 6 Zr εr 6 Zr
100 66.2 0.73 31.6 12.7 0.07 92.4
400 58.0 0.82 43.7 11.6 0.08 108
900 56.0 0.97 48.2 11.3 0.11 111
MHz S.m−1 S.m−1
Source: William Scanlon Queens University, Belfast.

Antennas 415
where is the reflection coefficient, or how much of the signal

power is reflected at the boundary,
Z a is the characteristic impedance of first medium .
Z r is the characteristic impedance of second medium .
Much has been written about propagation of RF (radio frequency)
signals through a human body. The depth of the implant is
important, as is the posture. Not only the amplitude but also the
polarization of the signal will be a function of the size, shape, and
posture of the body and implant depth. It should be borne in mind
that a surgeon will fit an implant where it is best for the patient
and the function and not necessarily for the best RF propagation.
As a patient lives with an implant, so he/she may put on weight or
exercise and loose fatty tissue so the propagation will continue to
change. Posture will also change the propagation as Johansson has
shown [5].
Calculation of body propagation can be made, but this will have
limitations as not every body is the same, a good reference is [6].
14.7 Antennas
The implant antenna is a very important part of the link. Because

of the wavelength of a 403 MHz signal and that it is immersed in a
weakly conductive medium, any antenna will appear poor compared
to a full size dipole in air. The wavelength of an RF signal is found
from Eq. 14.2.
λ = 300/ f m (14.2)
where f is frequency in MHz.
This is reduced by being immersed in a dielectric material such
as muscle as Eq. 14.3
√
λm = λ/ (εr )m (14.3)
where λm is the wavelength in medium.
Figure 14.1 shows the wavelength vs frequency for operation in
air and in muscle (εr = 58).
There are several options for antenna, but the configuration will
depend on the implant size and where it is to be located. A patch
Figure 14.1. Wavelength in air and typical muscle vs frequency. Note: Log
scales. See also Color Insert.
antenna is a flat structure that can be made as part of the implant

case or can be attached to a convenient surface. It needs a layer of
dielectric between the active surface and the 0 V (“ground”) plane.
The size of the patch is usually defined by the available flat surface.
At 403 MHz, even in a body, any patch antenna will be electrically
small and so it will have low gain. A typical example is shown in
Fig. 14.2, which has been used for both 403 MHz and 2.45 GHz.
Figure 14.2. Silver on alumina patch antenna mounted on test implant

case. See also Color Insert.
Antennas 417
The impedance in air is almost entirely reactive but when measured

in contact with skin and body tissue, the impedance becomes more
resistive. It is essential that any antenna be measured when in
contact with a body or immersed in body phantom liquid. The point
of measurement needs to be clearly defined at either the antenna
input or at the end of a length of coupling cable.
Small antennas can be made to be electrically larger by
meandering the structure [7] or employing a fractal [8] pattern.
A loop antenna will tend to be inductive and a monopole will
be capacitive. Any antenna will be electrically small even within
the body. Typically the antenna designer has to make the best of
the volume allocated. A good theoretical reference for antennas is
Kraus [9].
The measurement of the impedance of an electrically small
antenna can be challenging as a network analyzer may not give the
most accurate result, so other methods should be employed. Details
of the measurement techniques and the hardware used can be found
in Yang [10]. Fujimoto has written extensively on the design and use
of small antennas [11].
Three-dimensional simulation software can be used in the design
process where it is possible to simulate an antenna within a medium
that resembles a typical human body. This will enable the effect
of changes in body tissue electrical parameters to be estimated.
Simulation of a 3D structure can take several hours and is dependent
on good initial design. Software is available from Agilent (ADS), CST,
and Ansoft.
14.7.1 Use of Smith Chart in Coupling Network Design

In the design of a coupling network between the antenna and the
transmitter or receiver, calculation can be difficult; so a graphical
technique can be used. This can be done either by using a software
package or by hand using a preprinted chart, pencil, and a calculator.
Philip Smith designed the chart, thus bearing his name. More
detail can be found in his book [12]. The concept is to give a
graphical representation of any complex impedance showing clearly
the resistive and reactive elements. There are two basic versions of
the Smith Chart: impedance and admittance; these can be overlaid
to make a powerful graphical tool. Detailed use of the Smith Chart

in the design of RF networks can be found Lee [13], so it will not be
repeated here.
The impedance chart comprises circles of constant resistance
with their center on a horizontal line; the outer circle has 0
resistance, with smaller circles indicating progressively higher
resistance. The reactive (imaginary) part is represented by partial
circles with inductance above the center line and capacitive below.
The center line represents pure resistive impedance.
Often a Smith Chart is shown normalized where the impedance is
divided by the characteristic impedance (50 for example) and the
admittance multiplied by the characteristic impedance. Figure 14.3
shows a typical normalized impedance only chart available from the
University of Berne [14].
As well as a visual representation of lumped impedance and
admittance, the chart also represents adding transmission line.
Various versions of the chart are available from Internet sources.
To design a network, the antenna impedance must be known
and plotted on the chart. The destination impedance must be
defined. Then by adding series and parallel reactive components,
including the impedance of the IC and possibly line lengths, the
Figure 14.3. Blank normalized impedance Smith Chart (University of

Berne).
Antennas 419
target impedance at the transmitter/receiver integrated circuit (IC)

can be reached.
It is useful to design a network using a paper Smith Chart and
calculator as it gives the designer a better feel of what difference
adding a component will make. The next stage is to design the PCB
tracking for the coupling network. It is suggested the PCB comprise
signal layer on top and a 0 V (“ground”) plane such that it is possible
to have transmission line of known impedance. Using surface-mount
components, it is possible to make this very small so the track effect
will be minimal at 403 MHz. The coupling to the antenna should
also be known or the impedance measured where it enters the PCB
such as the end of a short length coax. When the above are known
and the paper initial design completed, simulation is recommended,
which will include the PCB track and dielectric material with real
models of the passive components; these are often available from
the manufacturers or can be derived from the published data on
self-resonant frequency, tolerance, and Q. Simulation of the entire
network should be close to, but not the same as, the initial design.
Simulation can be run to see the effect of component tolerance and
potential yield given the nominal values.
Note: When operating within the body, there is no “ground” and the
whole implant, along with monitor and stimulation wires, will be
floating.
14.7.2 Design of Antenna Coupling Networks

Assuming an available transceiver is to be used, most of the
RF system hardware design has already been done leaving the
PCB layout, software, and antenna coupling network. The antenna
coupling network is important as it needs to present the transmitter
and receiver with the correct impedance, conjugately match the
antenna and provide filtering. The coupling network may also be
required to separate a 403 MHz data signal from a 2.45 GHz wake up
and prevent each branch from loading each other. A typical example
of a communication IC that uses 2.45 GHz for wake up and 403 MHz
communication is the Zarlink ZL70101 [15].
The impedance required by a transmitter may be different to
that required by a receiver. For optimum power, the transmitter
impedance is found from Eq. 14.4
(0.7071 · Vpp )2
Rl = (14.4)
Po
where Rl is the optimum load resistance, Po is the required power
output, and Vpp is the maximum voltage swing of the transmitter
output.
The receive impedance may be optimized for noise as well as
gain. This may require a more complex network and possibly include
a switch to present different impedances for transmit and receive.
Often an optimum value is specified by the IC manufacturer.
Below are examples of the design of coupling networks:
14.7.2.1 Design example 1

Consider a simple example at 403 MHz, where the antenna
impedance is 10 + j20 ; normalized, becomes 0.2 + j0.4 , is
required to be transformed to an impedance of 200 , normalized
4 . See Fig. 14.4. Adding a series 10 nH transforms the impedance
to 10 + j48 , normalized 0.2 + j0.96 . Adding a parallel
capacitance of 8.3 pF will produce an impedance of 200 + j0. Part
of that 8.3 pF could be the tuning capacitance of the IC. See Fig. 14.5.
As has been seen, the only part of the IC input impedance that is
considered is the capacitance, as a transmitter or receiver will be
required to work into a resistive load that is often in the 100–500
range, and the real part of the IC impedance is ignored.
14.7.2.2 Design example 2 (SAW filter)

In many countries, the emergency services use a communication
network called Tetra. One of the Tetra bands is 308 to 400 MHz,
which is very close to the MICS band. The transmit powers are high
with fixed base-stations closely spaced to provide wide coverage.
This high power signal could overwhelm the sensitive input of a
MICS implanted or base-station receiver as the front-end filtering
may not be able to provide enough selectivity. Additional filtering to
the RF path can be provided with a SAW filter centered on the MICS
band. These filters have a narrow-pass band that severely attenuates
Antennas 421
Figure 14.4. Simple matching antenna uses chart normalized to 50 ω

(Software University of Berne).
Figure 14.5. Matching schematic from Fig. 14.4. See also Color Insert.
out of band signals with relatively low attenuation of the in-band

signal.
These filters are available in small size, which can be included
in an implanted design. They can be single sided at both ports or
single sided on one port and double sided on the other. Typically, the
impedance of the ports is nominally 50 on one side and 200
on the other. The impedances SAW filters tend to be capacitive. This
reactance needs to be conjugately matched as part of the coupling
network design.
The filter can be put into the receive path or in a common
transmit and receive paths as part of the antenna coupling network.
A typical example is the SF16-0403C5UB02 from AVX/Kyocera. A

SAW filter in a common transmit and receive path will not only
reduce out-of-band signals overloading the receiver but will also
attenuate any out-of-band signals in the transmitted signal.
If a differential SAW filter is used with a single-ended RF IC, then
a balun needs to be designed. A balun will transform a differential
signal to single sided using capacitors and inductors. The first step
is to conjugately match the SAW differential output such that it
presents a non reactive impedance of known value. Next design the
balun transforming the differential SAW resistance to the required
impedance at the RF IC.
Consider the example of a nominal 200 SAW (AVX/Kyocera
SF16-0403C5UB02) coupling to a 400 impedance at an IC input.
Consider also that the IC input has a 2 pF internal capacitance. Each
differential port of the SAW has an impedance of 35 + 5.7 pF. This
is shown on the normalized Smith Chart of Fig. 14.6 along with the
locus of the SAW input impedance. Adding a 36 nH inductor to each
differential output port to 0 V will conjugately match the capacitive
element and result in an impedance that is almost a pure resistance
of 183 on each port giving a differential impedance of 366 .
A balun to convert to 400 can be designed using the circuit of
Fig. 14.7 and Eqs. 5 and 6.
√
L = (R l · RS /2)/ω (14.5)
where Rs is single-sided impedance, 400 in this example; Rl is
differential impedance, 360 in this example.
At 403 MHz, L = 100 nH
Also
√
C = 1/[ω · (Rl · RS /2)] (14.6)
This gives C = 1.5 pF.
The conjugate matching and impedance transformation can
be rationalized by adding the parallel inductor/inductor and
inductor/capacitor combinations to a single inductor on each SAW
port to 0 V.
The output of the balun is a pure 400 resistance, but the IC
input has a capacitance of 2 pF. A 77 nH inductor to 0 V from the
Antennas 423
Figure 14.6. SAW filter input and (differential) output; 400 to 420 MHz;
Ansoft SV software. See also Color Insert.
Figure 14.7. Inductor/capacitor balun.
IC pin will conjugate match this capacitance. The path from the
differential ports of the SAW to the IC now looks like Fig. 14.8.
The single-ended port now needs to be matched to the antenna
remembering that the SAW has a capacitive impedance. For example,
if the SAW filter were to match to an antenna of impedance 10 +
j20 (normalized 0.2 + j0.4), then the impedance of the SAW needs
to be transformed to produce a conjugate match of the antenna.
Figure 14.8. Complete differential SAW to IC port network; includes balun

and tuning.
The conjugate match has the same real part but the negative of the
imaginary part. In this example, the SAW impedance needs to be
transformed to 10 – j20 (normalized 0.2 – j0.4). Adding the balun
and coupling network to the differential output changes the single-
sided input to 40 – j6.2 (normalized 0.8 – j0.124); this is almost a
pure resistance as Fig. 14.9 shows. Adding a parallel 15 pF capacitor
to 0 V transforms the impedance to 10 – j17.5 (normalized 0.2 –
j0.35). Adding a series 150 pF capacitance will produce the required
20 – j20 conjugate match of the antenna (Fig. 14.9).
SAW filters are also available with both single-ended input and
output such as the AVX/Kyocera SF16-0403C5UU01.
The schematic is shown in Fig. 14.10.
If a 2.45 GHz wake-up is employed using a single antenna, then
the signals need to be split as close to the antenna as possible. This
can be done with an inductor in the 403 MHz path and a small
capacitor in the 2.45 GHz path.
An alternative approach can be found in Yang [10].
When the design is complete and the PCB is laid out, the
whole circuit, including tracking, should be simulated using realistic
models of the passive components and the SAW filter. Real
components will have a specified value, but there will be a tolerance
and the parameters will change with frequency that may change the
response from that predicted.
The above examples used pure capacitors and inductors
with no PCB allowed for. Real passive components will have
Antennas 425
Figure 14.9. Single-sided SAW to antenna matching; normalized plot;

Ansoft software.
Figure 14.10. Single-sided SAW input to antenna coupling network.
parasitic elements such that a capacitor will have a typical

equivalent circuit for a 10 pF surface-mount capacitor is shown in
Fig. 14.11.
This will result in an impedance that will change from capacitive
to inductive (and vice versa for an inductor); a typical 10 pF surface-
mount capacitor impedance and series resistance plot is shown in
Fig. 14.12 (courtesy AVX).
Figure 14.11. Typical capacitor equivalent circuit.
Figure 14.12. Impedance and series resistance for 10 pF 0402 capacitor

(courtesy AVX). See also Color Insert.
14.7.2.3 Use of simulation for antennas and design of

coupling networks
There are several software packages that will simulate a PCB with
passive components. These include ADS from Agilent and Microwave
Office from AWR and free software such as AppCad from Agilent,
SmithCalc from the University of Berne, and Ansoft Student Version
(SV). To give accurate results, the PCB and dielectric constant
should be precisely defined. The models for the passive compo-
nents must include the parasitic preferably using manufacturers’
models.
Antennas 427
14.7.3 Physical Body Simulator

Before any live trials can be considered, it is essential that the
communication system be proven in an environment that resembles
the electrical properties of a human body. The body phantom is
a useful tool as a battery-powered implant can be placed in it at
various positions and depths. This enables the effectiveness of the
antenna in a body medium to be assessed and the reliability of the
link transferring data.
Testing an implant in dead tissue, such as meat from a butcher,
does not have the same electrical properties as the living does.
A body phantom comprises a 76 cm high, 30 cm diameter
Perspex tank filled with a liquid having electrical properties close
to that of living tissue. There are recipes for the liquid from
Wojeck [16] that cover various frequencies as Table 14.2 shows. The
implant must be water tight, battery powered, and suspended in
the liquid attached to a PTFE (polytetrafluoroethylene) (or similar
nonconductive) support.
A system will be used in a doctor’s office where the signal will
be reflected off of various surfaces causing multipath fading. To test
implant to base station link without the complication of reflections,
an anechoic chamber can be used. This is a chamber that has walls,
floor, and ceiling covered with absorbent material, thus eliminating
stray reflections. Using a body phantom in an anechoic chamber will
show the link can work and the ideal data rate for given.
14.7.4 Body Simulator Measurements and Sample Results

Testing in an anechoic chamber will require an implant and base
station as well as measurement antennas and other equipment.
Table 14.2. Body phantom liquid (* HEC = hydroxyl ethyl cellulose)
% by weight
Frequency GHz 0.1 to 1.0 1.5 to 2.5
Water 52.4 45.3
Suger 45.0 54.3
NaCl 1.5 0.0
HEC* 1.1 0.4
Figure 14.13. Power measured from implant in tank of body phantom

liquid vs depth (transmit power = −16 dBm).
Measurement can be made with the implant and measurement

antenna at various polarizations. Test software has to be written,
which will enable the implant to transmit a continuous wave signal
for a period of time that will enable measurements to be made.
Also it must be possible to set it into the receive mode record and
transmit back the RSSI (received signal strength indicator) level
measured when the base station is transmitting a CW signal.
Using a body simulator tank, and a battery-powered implant with
a patch antenna, measurements were made of signal levels and error
rate. Signal levels were measured using a calibrated log–periodic
antenna and spectrum analyzer.
Figure 14.13 shows how the signal from the implant varied with
depth. Here “depth” is the distance from the front of the tank to the
implant antenna.
Figure 14.13 shows the measured power level increases with
depth up to 6 cm after which it declines. This is thought to be due
to the body tissue acting as a parasitic antenna, which improves
propagation. After 6 cm, the body phantom attenuates the signal as it
is weakly conductive. Measurements made of data rate also confirm
this result.
14.7.5 The Role of Automatic Antenna Tuning

When implanted, an antenna will present a different impedance
from when it was in air. The implanted environment will not be
predictable and may change with time, so it is essential that the
Error Correction 429
implanted system be capable of automatically tuning the antenna

coupling network for optimum performance without external
intervention. This feature may be built into the RF IC along with a
routine that will activate on wake up. If this feature does not exist,
then external varactor diodes may need to be added and a routine
written that will optimize both the receive signal and transmit
power.
14.8 Implant Power Constraints and Battery

Considerations
For long-term use, battery drain must be kept to a minimum as

the communication system is in addition to the primary implant
function. Typically, a pacemaker will be expected to last over
seven years before requiring replacement, so the overhead that the
communication system adds must be minimal. With implants having
more features and pressure to make them smaller, the battery size
will also be constrained. A smaller battery will hold less charge, so
it is necessary to only power up the RF system when it is to be
used.
For a new application, it should be remembered that the voltage
of a primary (nonrechargeable) battery will decline gradually,
whereas a secondary (rechargeable) will show a much steeper drop
when a critical point is reached. The RF communication system can
feed back the battery status.
14.9 Error Correction
When packets of data are transmitted, error can be caused by noise,

interference, or multipath fading. To recover from this, the system,
if the receiving end detects an error, can invoke error correction.
This involves an overhead on the data transmitted but may enable
a corrupt packet of data to be corrected. If the error is such that
correction does not work, then the receiving end may request the
whole packet be re-sent. Using such techniques, the data integrity is
maintained and the final error rate is minimized.
14.10 RF Circuit Hardware Options
RF transmitters will need supply current related to the output power

— that is fairly obvious. For a receiver to operate on a low-level
signal, it will require careful design for gain and low noise. Typically,
the more sensitive a receiver is, the higher the supply current will
need to be. Receivers can take as much current as transmitters.
There are several ICs available that operate both in the MICS and
some of the ISM bands. To choose an IC for implant use, first consider
the following:
• Required data rate.

• Whether the implant is going to be on permanently or
intermittently
• Available supply voltage and current.
• Available volume
• Space for antenna
• Whether the IC is available in implant quality
• Whether wakeup is needed?
Several ICs have sleep mode options, which will reduce the average
current drain significantly. Waking up the IC from sleep mode can
be done either by transmitted signal in the band of use or by a
separate ISM band. Table 14.3 shows several ICs that cover the MICS
band. Most ICs will have an SPI (serial peripheral interface) to an
application processor.
Other manufacturers include Lime, Maxim, and Zlinx.
When choosing an IC, it is necessary to be sure what the
additional components are as they can significantly increase the
required volume.
Typically, the RFIC will be controlled by a separate microcon-
troller, which may be from a different manufacturer or may be the
controller for the primary function. This will involve setting and
reading register values, which can be a considerable task; so the
use of an evaluation board is recommended. These are available
from most RFIC manufacturers and will enable basic functions to
be run and allow access to individual registers for more advanced
operation.
Base Station 431
Table 14.3. Example values
Maximum Sleep Wake On Minimum Maximum

IC date rate current mode current frequency frequency
Melexis 115 50 Software 6.1(Rx) 15(Tx) 27 930
TH7122
RFM TRC105 200 100 Software 2.7(Rx) 16(Tx) 300 610
Semtech 200 100 Software 3(Rx) 15(Tx) 863 and 902 870 928
SX1211
TICC1101 500 9800 Wake on radio, 15(Tx0dBM 315 915
poll every second and Rxmode)
Ti Toumaz 50 None 4.5 868 and 902 870 928
CC981H
(Sensium)
Zarlink 800 100 2.45GHz 5 402 and 433.5 405 434.4
ZL70101
kb.s nA mA MHz MHz
Refer to data sheet for more details.
Some RFICs also include a processor that could be used to

control the primary function if it is adequate for the purpose.
Microcontrollers will also take current, which must be included in
the power budget. As with RFICs, some have low power modes,
which can significantly reduce current drain. The TI MSP430 series
is considered to be among the lowest power products currently
available.
14.11 Base Station
When designing an implant, it is also necessary to design the

external base station to control the communication and handle the
data. Often it is possible to use the same IC for both the implant and
the base station. Power and space are not so much of a problem
as it is for the implant; so enhancement of the ICs performance is
possible by adding a pre-amplifier in the receive path and using a
better antenna system.
In a real environment, RF signals will reflect off of walls,
furniture, equipment, etc. Also polarization will depend on the
implant position and patient posture. This can result in constructive
interference, which will improve the RF link but can also result in
Figure 14.14. Typical base station block diagram.
destructive interference, which could break the link. This can be

countered by employing an array of antennas to provide spatial
and polarization diversity. The antennas can be selected with an RF
switch. This will increase the size of the base station, but additional
antennas could be integrated into the enclosure.
A typical base station will have an RF IC, microcontroller,
pre-amplifier, antennas, antenna switch, and a USB interface to
a PC as shown in Fig. 14.14. A USB interface can also be used
to provide power. This includes a separate wake-up transmitter
working on an ISM band. The wake transmitter will need to have
the capability of operating over different frequency limits and power
levels depending on the country of operation. Frequency hopping
may be required for the wake transmitter, which will require
reprogramming of the frequency source (usually a crystal oscillator,
phase locked loop (PLL) and voltage-controlled oscillator (VCO)),
allowing time for it to settle before transmission. Care must be taken
to be sure that there is no transmission outside of the limits when
the frequency is changed. This can be done by switching off the
output stage during the frequency hops.
14.11.1 Link Budget

A simple link from base station to implant is shown in Fig. 14.15,
where the base station antenna has a gain of 0 dB, the range is 3 m,
the body loss is assumed to be 16 dB, and the implant antenna has
a gain of −20 dB. The implant antenna gain is for a poor antenna at
403 MHz; better gains are possible.
Base Station 433
Figure 14.15. Link from base station to implant (simple example).
For the base station to implant path, the radiated power is 25 μW

(−16 dBm). The path loss in air (assuming no reflected signal path)
is found from Eq. 14.7 (Friis equation).
Ps = G t (λ/(4π R))2 (14.7)
where Ps is signal at skin surface, λ is wavelength in meters, G t is
antenna gain, and R is range in meters.
The signal at the implant receiver is found from adding the
transmit power antenna and path gains (losses) as Eq. 14.8.
Pi = −16 − 34 − 16 − 20 dBm
= −86 dBm (or 2.5 pW) (14.8)
where Pi is the signal delivered by the implant antenna
The implant to base station path has the advantage that the
25 μW limit is outside of the body; so the transmit power can be
higher as long as the SAR (specific absorption rate) limit is not
reached, and the battery can supply the current. In this example, the
transmit power is 1 mW (0 dBm); with antenna and body losses, the
power at the body surface is −36 dBm; this is within the capability of
several ICs. Figure 14.16 shows a simple example. Adding the gains
(losses) gives a base station signal of −70 dBm, or 100 pW, which is
much higher than the base station to implant signal.
In a real environment, there will be stray reflections that can
interfere constructively or destructively at the receiver. The signal
at the receiver must be factored into a signal-to-noise ratio to access
the viability of the link. There may also be background noise sources
that will deteriorate the signal quality, but if the link does not
work with these very simple calculations, it will not work in a real
situation. See also reference [17].
Figure 14.16. Implant to base station link (simple example).
14.12 Environment
The human body is an environment of an almost constant temper-

ature and mostly free of mechanical shock (this is not the case for
cochlear implants in children where normal rough and tumble can
be the cause of significant shock). The implant will need to survive
the distribution process that may involve rough treatment and large
temperature swings.
An important consideration for the design of an implant is the
materials that will be exposed to body liquids that can attack many
materials. It is not adequate to cover a metal surface with a plastic
coating as this may become porous and enable leaking of toxic
materials. The metals that are considered safe for implant use are
platinum, platinum/iridium, and an implant grade of stainless steel.
Metals such as silver and copper are toxic and gold will be attacked
by body liquids.
14.13 Manufacture
It is not too difficult to design and build a single implant for

experimental use by hand adjustment of the parameters. An
experimental implant may not need to operate for a lengthy time,
and it may not matter if it fails. An implant that can be fitted to
patients is required to work consistently, does not need physical
intervention, and has to be of a very high standard. It must be
possible for an implant to be purchased against a know specification.
Conclusions 435
To take the concept to a product that can be fitted to patients

requires a design that can be reproduced reliably in significant
volumes.
For an RF system, the design requires to be thoroughly tested in
a simulated environment and may undergo live testing. The design
should not require components to be selected at manufacture.
Components used must come from a source that will supply to
implant grade — not every manufacturer will do this. It may be
necessary to change a design to enable implant grade components
to be used. The printed circuit board (PCB) is an essential part
of the product, and it should not be taken for granted. For
consistent product manufacture, the assembly of components onto
the PCB should be by machine and not by hand; even the best
assembly workers will, occasionally, fit a component into the
wrong location. This will result either in time consuming rework
or products being scrapped. This adds to the overall cost of
manufacture.
The final testing of the assembly should be to check for the part
was manufactured correctly. Final testing should not be to verify the
design as this should be well established along with the choice of
components. Following manufacture, there may be a need for burn
in testing where the product is powered for a period of time, possibly
at an elevated temperature. This is to make sure that any component
weakness is revealed.
14.14 Conclusions
Adding a wireless link to a human body implant can greatly improve

the diagnosis of diseases and improve the quality of life of many
patients. It offers the possibility of long-term monitoring without
the need for a long and expensive stay in hospital. By monitoring
the performance of the body and the implant, therapeutic regimes
can be optimized without the need for surgical intervention. In
the future for this technology will use even lower power devices
with ever smaller component sizes including integrating passive
components into the PCB. The PCB technology is also critical to size
reduction.
Implant technology will make even more significant improve-

ments to the lives of otherwise very ill patients.
Notation
dB: a log power ratio, 10 log10 (P1 /P2 ).

dBm: a power ratio referenced to 1 mW, 10 log10 (P1 /1 mW).
Rx: receiver.
Tx: transmitter
λ: wavelength in meters
References
1. Sivard, A., et al. (November 2004) Challenge of Designing In-body

Communications, EE Times, http://www.eetimes.com/design/
embedded/4025029/The-challenge-of-designing-in-body-communica-
tions.
2. ETSI EN3018390, Electromagnetic Compatibility and Radio Spectrum
Matters (ERM); Radio equipment in the frequency range 402 MHz to
405 MHz for Ultra Low Power Active Medical Implants and Accessories;
Part 1. 2006. 06921 Sophia-Antipolis, Cedex, France.
3. Loy, M., et Al. (May 2005) ISM Band and Short Range Device Regulatory
Compliance Overview, Texas Instruments Application Report SWRA048.
Dallas, Texas.
4. Finkenzeler, K (2003) RFID Handbook, 2nd edn, Wiley. Munchen,
Germany.
5. Johansson, A. J. (2002) Wave Propagation From Medical implants —
Influence of Body Shape on Radiation, Department of Electroscience,
University of Lund, Lund, Sweden.
6. Rahmat-Samii, Y., and Kim, J. (2006) Implanted Antennas in Medical
Wireless Communications, 1st edn, Morgan and Claypool. California USA.
7. Warnagris, T. J., and Minardo, T. J. (December 1998) Performance of a
Menadered Line as an Electrically Small Transmitting Antenna, IEEE
Transactions on Antennas and Propagation, 46(12), pp 1797–1801.
www.ieee.org
8. Felber, P. (2001) Fractal Antennas, revised edn, Illionis Institute of
Technology, Revised January 16. www.iit.edu
References 437
9. Kraus, J. D. (1988) Antennas, 2nd edn, McGraw Hill. Columbus OH, USA.
10. Yang, G. Z. (2006) Body Sensor Networks, Springer, pp 117–143. London,
UK.
11. Fujimoto, K., et al. (1987) Small Antennas, Research Studies Press (John
Wiley). Bognor Regis, UK.
12. Smith, P. (1969) Electronic Applications of the Smith Chart, McGraw Hill.
Columbus OH, USA.
13. Lee, R. F., and Chen, W. (1997) Advances in Microstrip Printed Antennas,
Wiley, pp. 35–70. Columbus OH, USA.
14. Dellsperger, S., University of Applied Sciences, Berne,
[email protected].
15. Zarlink ZL70101 Data Sheet: http://www.zarlink.com/zarlink/zweb-
zl70101-datasheet-dec09.pdf.
16. Wojclk, J., et al. Tissue Recipe Calibration Requirements, SSI/DRB-TP-
D01-003, Spectrum Sciences Institute RF Dosemetry Research Board,
51 Spectrum Way, Nepean, Ontario, K2R 1E6 Canada.
17. Planning For Medical Implant Communication Systems (MICS) and
Related Devices, SPP 6/03, October 2003. Australian Communications
Authority, Canberra, Australia. www.acma.gov.au.
Chapter 15
Wireless Power and Data Telemetry for

Wearable and Implantable Electronics
Zhi Yang, Yu Han, Linh Hoang, YiKai Lo, Kuanfu Chen,

Jian Lao, Mingcui Zhou, and Wentai Liu
University of California, Santa Cruz, CA, USA
[email protected]
The recently reported power and data telemetry systems in the

literature will allow wireless communication with tissue and by
doing so enable implantable/portable microelectronic devices to
treat presently incurable diseases, analyze biologic signals, and
enhance the performance of surgical tools. While telemetry design
is, in general, guided by several basic themes — heat dissipation,
data communication, and integration and miniaturization — this
chapter addresses the above themes through system-level modeling,
analysis of critical blocks, and design examples based on dual-band
scheme [1].
15.1 Introduction
Inductive links are popularly used in newer generations of bio-

medical microelectronic systems for providing power and data.
Supported by telemetry, these systems are more appealing in terms

www.panstanford.com
440 Wireless Power and Data Telemetry for Wearable and Implantable Electronics
of portability, safety, and system integration in comparison to their

counterparts, those powered by a battery and commanded by
signals through tether wires connected externally. Some important
examples of wearable and implantable electronics are cochlear
implant [2–4], retinal implant [1, 5–9], tissue imaging [10], brain-
computer interface [11–13], artificial limb [14], neural recording
[15–19], and health meters [20].
Functionally, a telemetry system consists of two links: a power
telemetry link to deliver energy for powering electronics and a
data telemetry link to communicate information between the object
and an external station. To deliver power and data wirelessly for
those wearable and implantable electronics, both links have two
components of electronics. Those located externally or physically
detached from the objects are referred to as primary side units,
e.g., external battery, power transmitter, signal processing station,
forward data transmitter, reverse data receiver, etc. Those located
under the skin (implanted electronics) or along with the objects
(wearable electronics) are typically referred to as secondary units,
including power receiver, forward data demodulator, stimulator
[21], signal sensor and processor [22], and reverse data transmitter,
etc. To accomplish the functions of powering and communicating
with secondary electronics, the telemetry system should achieve
high-power efficiency to avoid heating tissue, low bit-error-rate
(BER) to ensure reliable data communication, and small in size
and light in weight to target the highest level of system integration
and miniaturization. As will be shown in this chapter, the mentioned
design goals are inherently connected; thus careful design optimiza-
tion and trade-off analysis are important to achieve the optimal
performance given the application.
This chapter is organized as follows. In Section 15.2, analysis
and methodology to design highly efficient power telemetry link
are discussed first, followed by modeling and optimization of coils,
which are the discrete devices of the largest footprint size. Further,
the trade-off between power efficiency and coil size is elaborated
by correlating power efficiency, Q factor, coupling, and geometry
parameters analytically. In Section 15.3, a review/comparison of
several data communication schemes is presented based on power
dissipation, design complexity, data rate, BER, and communication
Power Telemetry 441
distance. In Section 15.4, a telemetry design example to support a

256-channel epiretinal stimulator array is given with test results. In
the appendix, detailed derivations and design guidelines related to
coils and power telemetry are presented.
15.2 Power Telemetry
In this section, critical components of a power telemetry link,

including coupled resonant amplifier, high Q inductor, and rectifier,
are introduced first, followed by detailed modeling, explanation, and
implementation results.
Resonant Amplifier: As shown in Fig. 15.1, a power telemetry link
used in biomedical applications involves two components: power
transmitter and power receiver. The power transmitter is typically
located outside the body, converting energy from a battery (Vb) to
radio frequency electromagnetic (EM) wave. The receiver harvests
energy from the EM wave, which further goes through rectification
and voltage boosting, powering the rest electronics at the secondary
side. For the convenience of discussion, the power transmitter
and receiver are conceptually simplified as two isolated resonant
tanks (small coupling coefficient between coils, small power load
at the secondary side). The primary resonant tank has a resonant
√
frequency at ω1 = 1/ L 1 C res1 , and the secondary resonant tank
√
has a resonant frequency at ω2 = 1/ L 2 C res2 . If ω1 = ω2 − ω0 and
the carrier frequency ω = ω0 , both primary and secondary tanks are
in resonance, achieving high power efficiency. Details on modeling
these two resonant frequencies will be presented in Section 15.2.1.
Figure 15.1. Conceptual diagrams of a power telemetry system.

Power Efficiency and Inductor Optimization: Power efficiency,

defined as the ratio of power harvested at the secondary side over
the power drained from the battery (Is Vs /Ib Vb in Fig. 15.1.), is
an important parameter to evaluate the performance of a power
telemetry link. It can be analytically approached using different
simplified models, and the results show that power efficiency is a
monotonously increasing function of the Q factors of coils and the
coupling between them. In case the coils are weakly coupled and the
power load is small, power efficiency is approximately proportional
to k2 Q 1 Q 2 (Q 1 and Q 2 are the Q factors of coils, and k is the
coupling coefficient between the coil pair), suggesting coils to be the
most critical components for power telemetry. Given the restrictions
on coil size and coil separation, the achievable coupling coefficient
is determined. Therefore, one of the emphases of this chapter is
on improving the power efficiency by increasing the coil Q factor.
Details on coil modeling and method to improve the Q at the carrier
frequency are given in Sections 15.2.2 and 15.2.3.
Power Rectification and Regulation: The induced voltage at the

secondary resonant tank is rectified and regulated to power the
rest of the electronics. For rectifier design, particularly on chip
implementation, diodes or diode-connected MOS transistors are
commonly used. Figures 15.2a, b illustrate full-wave rectifier (gener-
ate two voltage supply levels) and cascoded rectifier (generate four
voltage supply levels) topologies. Generally, diodes are implemented
using MOS transistors, and they face two challenges. First, as
shown in Fig. 15.2c, diode-connected PMOS transistor has VT + VOD
source to drain voltage drop (VT is the transistor turn on voltage,
VOD is transistor over-drive voltage). A large voltage drop across
the diode-connected transistor degenerates the power conversion
efficiency.
V0 V H − (VT + VOD )
ηrectifier = = , (15.1)
VH VH
where ηrectifier is the power conversion efficiency, V O is the rectified
DC voltage, and V H is the peak voltage in the secondary resonant
tank connected to the rectifier input.
Second, substrate has to be biased to the largest posi-
tive/negative voltage level for PMOS/NMOS transistors. Otherwise,
Power Telemetry 443
Figure 15.2. Rectifier topologies: (a) Full-wave rectifier providing Vdd and
Vss ; (b) Cascoded rectifier providing four voltage supply levels: Vdda , Vddb ,
Vssa , and Vssb , avoiding DC-DC voltage drop from Vddb /Vssb to Vdda/ Vssa (only
positive branches Vddb and Vdda are drawn); (c) PMOS implementation diode
for rectifying positive branches; its substrate has to be biased to the highest
voltage level to avoid leakage current; (d) a bias circuitry for the substrate
of PMOS diode. Vsub = max(VH sin ωt, Vo ).
large leakage current to substrate appears. This requirement is more

complicated in biomedical applications, since the voltage in the
resonant tank is AC and its amplitude appears to be the largest at the
secondary side. Consequently, special bias circuitry for the substrate
is required, e.g., floating well biasing, dynamic bulk switching
technique. An example of bias circuitry for PMOS diode substrate is
shown in Fig. 15.2d, where the substrate voltage fluctuates to track
the higher one of source and drain. More discussion on the choice of
rectifier structure is presented in Section 15.2.4.
15.2.1 Mega-Hz and Sub-Mega-Hz Power

A commonly used power telemetry architecture is shown in
Fig. 15.3a, where the transmitter is adopted from Class-E power
Figure 15.3. (a) Class-E amplifier based on power telemetry link,

including Class-E power amplifier, power receiver resonant tank, rectifier
and regulator; (b) Linearized circuit model for the power telemetry link.
amplifier, the receiver is a resonant tank followed by rectifier and

regulator. The corresponding linearized circuit model of the power
telemetry architecture is shown in Fig. 15.3b.
The Co-Resonance of Primary and Secondary Tanks: Given compo-

nents’ values and coil coupling coefficient, there are two resonant
frequencies satisfying the following constraints:
∂ I1 (ω) ∂ I2 (ω)
= 0|ω=ωres1 , = 0|ω=ωres2 (15.2)
∂ω ∂ω
where ωres1 and ωres2 are the resonant frequencies of primary
and secondary tanks after considering load reflection through coil
coupling. At ωres1 and ωres2 , currents through the coils (I1 , I2 ) reach
Power Telemetry 445
their maximum respectively. To improve power efficiency, both

primary and secondary resonant tanks are desirably to resonate at
the same frequency point.
After inclusion of coil coupling and power load, ωres1 and ωres2
deviate from the ideal form. Through straightforward analysis based
on linearized circuit model, V1 , V2 , I1 , and I2 meet the following
constraints:
I1 (R esr1 + 1/( j ωC res1 ) + j ωL 1 ) + I2 j ωM = Vi n , (15.3)

Resr2 + 1/j ωC res2
I1 j ωM + I2 j ωL 2 + R AC = 0 (15.4)
Resr2 + 1/j ωC res2 + R AC
I1 j ωL 1 + I2 j ωM = V1 , I1 j ωM + I2 j ωL 2 − V2 , (15.5)
where M is the mutual inductance between L1 and L2 .
Combine Eqs. 15.2–15.5, under the resonant condition, circuit
components’ values satisfy

ω2 M 2
ωL 1 + 1/ωC 1 + I M = 0,
j (ωL 2 − 1/ωC res2 ) + R esr2 + R AC .esr
(15.6)

ω M
2 2
ωL 2 + 1/ωC res2 + I M . (15.7)
j (ωL 1 − 1/ωC res1 ) + R esr1
where IM{.} stands for the imaginary part, R AC .esr is the converted
R AC to series resistance in the secondary resonant tank.
Based on Eqs. 15.6 and 15.7, the optimal power carrier frequency
ωC can be represented as a function of L 1 , L 2 , C res1 , and C res2 .

Resr2 +R AC .esr
C res1
− CRres2
esr1
ωC = . (15.8)
(R esr2 + R AC .E S R )L 1 − Resr1 L 2
Power Efficiency: Both the primary and secondary tanks are in
resonance is a necessary condition to achieve high efficiency. In
this subsection, focus is made to balance undesired primary and
secondary power losses.
The power drained from the external battery consists of
three parts: power dissipation in the primary resonant tank (P1 ),
power dissipation in the secondary resonant tank (P2 ), and power
consumed by R AC (Pd ). Among these three components, Pd is
the power delivered to support electronics, while P1 and P2 are

undesirable power losses, which should be minimized.
Assume the resonant conditions of Eqs. 15.6 and 15.7 are
satisfied, P2 is approximately computed as
P2 = I22 R2 = 0.5(Vr + VT )2 ( j ωC res2 + 1/R AC )2 ωL 2 /Q 2 (15.9)
where Vr is the rectified voltage, VT is the diode forward voltage

drop when it turns on, and Q 2 is the secondary coil Q factor.
P1 is the Ohm loss in the primary resonant tank (derived
considering the energy spent in the secondary side), represented as
P1 = I12 R1 = Vi2nd /(k2 ωL 2 Q 1 )

2(P2 + Pd )2
= (15.10)
(Vr + VT )2 ( j ωCres2 + 1/R AC ) k ωL 2 Q 1
2 2
where Vi nd is the induced voltage on the secondary side, Q 1 is

the quality factor of the primary resonant tank, and k is the
coupling coefficient between the primary and secondary coils (k =
√
M/ L 1 L 2 )
Fix the power requirement of electronics at the secondary side
(Pd = Pd0 , Vr = Vr0 ); the optimization goal is to reduce the undesired
power losses P1 and P2 .
arg min α P1 + β P2 | Pd =Pd0 ,Vr =Vr0 ,ω=ωc (15.11)
where α and β are weighting coefficients for power loss at the

primary side and the implant side, respectively.
When the carrier frequency and components’ values satisfy the
resonant conditions in Eqs. 1.6 and 1.7, the objective function α P1 +
β P2 is minimized to be

αβ 2α Pd
α P1 + β P2 ≈ 2Pd 2
+ 2 (15.12)
k Q1Q2 k Q1Q2
It is interesting to see that this minimum undesired power loss is

determined by Q and coupling coefficient alone. In the following
section, focus will be on boosting the coil Q factor at the carrier
frequency.
Power Telemetry 447
15.2.2 Inductor Q Boosting

When a coil operates at low frequency, its Q factor is defined
as Q = ωL/R, where ω is frequency, L is inductance, and R is
series resistance. As frequency increases, frequency-related effects,
including skin effect, proximity effect, and self-resonance modify
both L and R, dramatically degenerating the Q factor. According to
reference [23], a formula for predicting the Q factor is
2π f L 1 − f 2 /F sel
2
f
Q( f) = (15.13)
R 1 + f 2 / fh2
where f is frequency, fh is a parameter to quantify the impact from
proximity effect (skin effect), and fself is the coil’s self-resonant
frequency; fh and fself are expressed by geometry and physical
parameters, briefly
1 1
fh ∝ 1 √ , fsel f ∝ (15.14)
rs σ N 1 N s a ln(a/r1 ) C p,k (k − p)2
as shown in Fig. 15.4. rs , rt , and a are the radii of a strand, a turn, and
the coil loop (usually litz winding is used, where one turn contains
multiple strands; otherwise, rs = rt ). Ns and Nt are the number of
strands per turn and the number of turns of the coil winding. σ is the
metal conductivity, C p,k is the parasitic capacitance between turn p
and turn k whose expression is shown in the appendix.
Importantly, for any given coil, there is an optimal frequency,
f peak , that has the maximal Q
∂Q
= 0| f = f pwK (15.15)
∂f
Figure 15.4. Coil geometry illustration. Left: A multiple-turn, multiple-

layer coil winding. Middle: Cross section of one turn, single-strand wire.
Right: Cross section of one turn, multiple-strand wire (litz wire).
To maximize the power efficiency, f peak ’s of the coil pair should be

designed in accordance with the power carrier frequency. Based on
Eqs. 15.13–15.15, the analytical form of f peak is
2
1/ f peak ≈ 1/ fh2 + 3/ fsel
2
f (15.16)
Equation 15.16 represents a key design equation and a closed form
analytical solution for f peak . With this single equation of merit, the
maximum Q, as well as the maximum efficiency of the telemetry
system, can be determined. By changing the design parameters, one
may tune f peak close to the target frequency and maximize the power
efficiency of the telemetry.
Example a. Inner diameter: To increase the self-resonant frequency

of a coil and thus f peak , an effective and practical method is to
increase the separation between layers. As shown in Fig. 15.4., fixing
outer radius aout and reducing the inner radius ai n , an increased
layer separation can be achieved.
According to Eq. A.7, a coil’s total parasitic capacitance is

l
C sel f = C b (l − 1)m + cm (2i − 1) (m − 1) / Nt2 . (15.17)
2
i =1
where C b is the parasitic capacitance between two nearby turns in
the same layer and C m is the parasitic capacitance between different
layers, as shown in Fig. A.3 in the appendix.
For a tightly wound coil, the parasitic capacitance between two
nearby turns is
π/4
π D i r0
C b = C m = ε0 εr dθ (15.18)
0 ς + εr r0 (1 − cos θ)
For coils with spacing between layers, the parasitic capacitance is
⎧ π/4 π D i r0
⎪
⎨ cb = ε0 εr 0 dθ
ς + εr r0 (1 − cos θ) (15.19)
⎪ π D i r0
⎩ cm = ε0 εr π/4 dθ
0
ς + εr r0 (1 − cos θ) + 0.5εr h
where h is the separation between two layers. The calculation
results using Eqs. 15.17–15.19 under different layer separation is
shown in Table 15.1 (insulation layer thickness: ς = 3 um, dielectric
constant εr = 3), suggesting tightly wound coil is not necessarily the
best choice for achieving high Q .
Power Telemetry 449
Table 15.1. Comparison of coils with and without interlayer separation
Coil Din Cb Cm L C sel f fsel f f peak

h=0 2.76 cm 15 pF 15 pF 85 μH 17 pF 4.1 MHz 1.6 MHz
h=r0 2.66 cm 15 pF 1.0 pF 77 μH 1.5 pF 15 MHz 2.5 MHz
Example b. The number of strands: If the cross section of the

winding is fixed, reducing the number of strands is an effective way
to increase f peak and the Q at higher frequencies for two reasons.
First, fsel f increases due to the increased separation between nearby
turns. Second, fh increases since the total number of strands in the
winding decreases. Therefore, both f peak and the Q are larger at
higher frequencies.
As a numeric example, comparisons of several coils with different
geometry parameters are given in Table 15.2. Specifications of coils
under comparison: D out = 3.2 cm, D i n = 2.7 cm, b = 2.5 cm, t =
2.5 mm, ds = 50 um, Nt = 36, ς = 3 um, εr = 3, wire strand size =
American wire gauge (AWG) 44.
15.2.3 Rectifiers and Regulators

On-chip rectification is commonly implemented by diode-bridge
configured in a half-wave rectifier scheme for simplicity or a full-
wave scheme to acquire a smaller output voltage ripple and higher
conversion efficiency. However, the use of on-chip PN-junction diode
and simple diode-connected transistor in rectifier circuitry usually
leads to a voltage drop of ∼1 V (diode turn on voltage plus overdrive
voltage, as shown in Fig. 15.5a), resulting in a power efficiency
degradation, which is more severe in low-voltage applications.
Furthermore, the diode substrate has to be properly biased to avoid
leakage current through the parasitic BJT.
Table 15.2. Comparison of coils of different number of strands
Coil Ns RDC fsel f f peak Q peak
IV 3 9.5 ohm 26 MHz 8.5 MHz 198

V 7 4.1 ohm 19 MHz 4.2 Mhz 215
VI 15 1.9 ohm 14 MHz 2.1 MHz 222
VII 30 0.95 ohm 5.6 MHz 1.1 MHz 219
Figure 15.5. (a) Diode-connected transistor as a rectifier; (b) Synchro-

nized rectifier; (c) Floating well substrate biasing; (d) Dynamic switching
substrate biasing (Vi n is the input voltage to the rectifier, Vo is the output
voltage from the rectifier).
To enhance the conversion efficiency, active diode configuration

has been reported and its generic type is shown in Fig. 15.5b. The
basic principle of this circuitry is to mimic an ideal diode with zero
drop-output voltage. The transistor MP is used as a switch, and its
gate-control voltage is derived from the rail-to-rail output voltage
of a comparator and accordingly, a smaller forward voltage drop
across the diode can be achieved since it operates in triode region.
By carefully controlling the turn-on time of the transistor, reverse
current leakage that comprises power efficiency is prevented.
To avoid leakage current through the parasitic BJT, bulk can
be biased using floating well biasing (the substrate Vsub is biased
to V peak − VT , as shown in Fig. 15.5c), or dynamic bulk switch
technique (the substrate is biased to the larger one of source and
drain Vsub = max(Vsour , V Drai n ), as shown in Fig. 15.5d).
15.3 Data Telemetry
In applications involving stimulation, a data telemetry link transfers

commands and stimulation patterns from the primary station to
Data Telemetry 451
the secondary electronics. In some other applications involving

signal sensing and processing, a data telemetry link transfers those
recorded data at the secondary side to the primary station for
further processing. Depends on applications, a set of desired perfor-
mance metrics and constraints may vary, such as data bandwidth,
implementation complexity, power consumption, system size, and
reliability. In this section, several commonly used communication
schemes are discussed for different applications.
Amplitude-Shift Keying: Amplitude-shift keying (ASK) is one of
widely used modulation techniques for data communication for
wearable and implantable electronics [5, 24, 25]. Under ASK scheme,
binary data are transmitted through modulating the envelope of
carrier wave. One prominent advantage of ASK-based system is
that the circuit blocks for both modulation and demodulation are
relatively efficient in terms of design complexity and integration. In
reference [5], a neuro-stimulus chip with ASK telemetry units for
retinal prosthesis has been reported. The reported chip can support
a 10 by 10 electrode array, providing up to 250 Kbps data rate. In
reference [26], Z. Tang et al. proposed a load shift keying (a special
type of ASK) (LSK) modulator for data recording, called LSK circuit
configuration modulator (LSK-CCM). The system modulates the
secondary load with digitized recorded data, which induce voltage
fluctuation across the primary side inductor for data demodulation.
The design can provide 36 Kbps data with modulation index down
to 0.02.
Despite the efficiency in implementation, one of the challenges an
ASK-based data link faces is the susceptibility to the signal amplitude
fluctuation, which can be severe if the object is moving [27] and
produce erroneous bits.
Frequency-Shift Keying: Frequency shift keying (FSK) is another
commonly used modulation technique where carrier frequency
acts as the data-carrying entity [28–30]. Three FSK designs for
biomedical applications have been evaluated in reference [28],
where measurement results show that the differential FSK design
achieves relatively better performance: maximum 4 MHz data rate,
0.29 mm2 die area, and power consumption of 0.38 mW at 5/10 MHz
carrier frequencies.
Compared with ASK, FSK has decreased susceptibility to signal

amplitude fluctuation, achieving more robust data communication.
However, there are two major drawbacks. First, FSK signal spectrum
is relatively wide and cannot benefit from a high Q inductive-
capacitive tank at the receiver side. Consequently, it is more
inefficient to recover power from FSK signal compared with ASK
signal with small modulation index. Second, FSK receiver circuitry
is more difficult to design; for example, the receiver requires
synchronization circuitry, since the internal clock cannot be simply
derived from the carrier.
Phase-Shift Keying: The drawbacks of ASK and FSK derive from

fluctuations in signal magnitude and frequency spectrum. Such
fluctuations are suppressed in phase-shift keying (PSK). Instead of
using carrier amplitude and frequency to represent digital data, PSK
modulates carrier waves through changing their initial phases. A
binary PSK (BPSK) demodulator that incorporates a digital phase-
locked loop (PLL) has been reported in reference [31]. The design
can support 100 Kbps forward data rate at 13.56 MHz carrier
frequency. Their design has bidirectional telemetry units: BPSK-
based forward data transmission and multilevel LSK-based reverse
data transmission. In a different implementation of BPSK-based on
COSTAS loop [32], the measured data rate is 1.12 Mbps with 0.7 mW
power consumption.
Featuring stable signal amplitude and carrier frequency, how-
ever, BPSK has to use coherent demodulation method in the
receiver circuitry due to its sensitivity to phase information. A
phase synchronization circuit with high precision is particularly
important.
Differential Phase-Shift Keying: In some biomedical applications,

for example retinal prosthesis, power and data required to operate
the implant goes up to a few hundred mW and a few Mbps [33].
The power telemetry design must take the coil size, tissue magnetic
field exposure, and power efficiency into consideration, while the
data telemetry should provide a high data rate that can control
a large number of stimulation pixels at real time. In a single-
band approach, where power and data are transmitted through
Data Telemetry 453
the same carrier, high power efficiency and high data rate are two
contradicting specifications [5]. Dual-band telemetry is a better
approach [28, 34] because it transmits power and data at different
frequencies. However, the power-to-data interference in the dual-
band scheme can potentially corrupt the data demodulation. As a
result, it is important to incorporate a dual-band approach with a
power interference cancelation scheme.
A noncoherent differential phase-shift keying (DPSK) scheme,
where “1” is coded as a phase shift of 180◦ and “0” is coded as
no phase shift [35], can be used to reduce the power interference.
By assigning data rate equal to the power carrier frequency, two
consecutive symbols are under the same power interference, which
is eliminated by differentiation. The reported DPSK data link in
reference [1] is based on subsampling scheme [36], where sampling
frequency is lower than the data carrier frequency. The adoption
of the subsampling scheme is based on two reasons. First, a low
frequency operation reduces demodulator power. Second, it samples
the band-pass signal without down mixing, which involves the
phase-locked loop circuitry. In this sense, the DPSK technique can be
regarded as a noncoherent technique, and the DPSK receiver design
is much simpler than BPSK receiver since the phase information is
not precisely required. Measurement results in reference [1] show
that the DPSK data link can provide up to 2 Mbps data with total
power consumption of 6.2 mW/5 V, when tested under 12 dB
interference-to-signal ratio.
Ultra-Wideband Telemetry: Very recently, ultra-wideband (UWB)-

based data link has been reported in biomedical applications to
transmit data. A UWB design can offer low power consumption, low
signal power spectrum density, high data rate, immunity against
power interference and noise [37]. UWB radio is defined by FCC
to support high data rate with a fractional bandwidth fc greater
than 0.2 or a bandwidth of at least 500 MHz at all times of
transmission [37–39]. The fractional bandwidth is defined by the
following expression:
fH − fL
fc = 2 (15.20)
fH + fL
where f H and f L are the upper and lower frequencies measured at

the −10 dB emission points, respectively.
Instead of using continuous carrier waves as in common
communication systems, UWB generally uses very low duty cycle
impulses to transmit data. Such efficient transmission mode high-
lights UWB to transfer high data rate in a relatively large distance
(100 Mbps, several meters). Further, UWB pulse can be extremely
narrow, thus robust to multipath fading effects. Though UWB data
transmission features high data rate and low power consumption,
there still exist many challenges, such as antenna design for
wearable and implantable applications, high-peak power pulse
generation, and interference with other devices in the overlapping
spectrum. These challenges should be carefully considered before
implementation.
BER Performance Analysis: BER performance reflects the reliability

of data transmission. To compare the mentioned modulation
techniques, simulated BER curves based on theoretical analysis in
reference [35] are shown in Fig. 15.6. It is shown that PSK and DPSK
biomedical systems are superior to those of ASK and FSK systems in
terms of BER performance.
Figure 15.6. Simulated BER curves of different modulation schemes.

Design Example 455
15.4 Design Example
A 256-channel epiretinal prosthetic system is illustrated in Fig. 15.7.

Video images are captured by an external camera and encoded into
bitstreams by an external signal-processing unit. The bitstreams
are transferred to the implant. At the receiver side, data are
demodulated and distributed to control the 256-channel stimulator
at 60 frames per second.
In this epiretinal prosthetic system, the power and data
telemetry links are realized under dual-band architecture, where
power and data are transferred at two different frequency carriers.
The 2 MHz carrier power telemetry link, including a Class-E-based
power amplifier, inductively coupled coils, rectifiers, and regulators,
Figure 15.7. (a) Block diagrams of power and data telemetry in a 256-
channel epiretinal prosthetic system [33]. (b) Chip photo of the power and
data receivers.
Figure 15.8. Measured results. (a) and (b) Data telemetry. Trace 1:
the measured voltage across the transmitter coil; Trace 2: the measured
voltage across the receiver coil; Trace 3: the measured data receiver output.
(c) Power telemetry. Trace 1: the measured voltages across the power
transmitter coil; Trace 2: the measured voltage across the power receiver
coil (input to rectifier); Trace 3: rectifier output for powering secondary
electronics; (d) output of two stimulator drivers that are powered and
controlled by the power and data telemetry [33]. See also Color Insert.
generates in four supply voltages of ±12 V and ±1.8 V, in total

100 mW for the implant. The 22 MHz carrier data telemetry link
is implemented based on a DPSK modulation scheme, which can
tolerate power-to-data interference. The power regulators and the
DPSK data receiver are integrated on a chip in 0.18 μm 32 V CMOS
process and tested on the bench top. The bench-top test setup
includes: an NI data acquisition card (Model: NI USB-6259), used
Appendix 457
as the signal source to generate 2 Mbps image bitstream for the

full path demonstration; an FPGA (Model: Altera Flex10k), receives
bitstream from the NI card and produces the DPSK data; and a
2MHz clock for external power and data transmitters. Power and
data transmitters induce RF 2 MHz power carrier and modulated
22 MHz data carrier, which are inductively coupled to the implant
side. A power receiver that consists of a tapped coil, resonant tank,
rectifiers, and regulators harvests the RF energy and generates
DC voltages. A data receiver that consists of a data coil, a DPSK
demodulator, and an 8 MHz crystal oscillator decodes the 2 Mbps
image bitstream from the modulated 22 MHz signal and further
feeds to digital controllers. The 256-channel stimulator initiates
stimuli based on decoded bitstream accordingly.
Figures 15.8a,b show the measured results of the data telemetry
subsystem. Figure 15.8c illustrates the measured waveforms of the
power telemetry subsystem. Both power and data coil pair are
coaxial. Waveforms are measured at a coil separation of 2 cm. In
Fig. 15.8d, traces 1 and 2 present the distributed system clock
and the image bitstream on the global bus, while traces 3 and 4
show the output waveforms of two arbitrarily selected channels
(the stimulator array is powered by the power telemetry and
commanded by the data telemetry. Relec = 10 k in series with
C elec = 100 nF are used to mimic the electrode-tissue interface
in this test). The signal-to-interference ratio and BER of the DPSK
demodulator are measured as −11 dB and 10−5 [33], respectively.
Acknowledgement
The authors thank funding provided by NSF through BMES-ERC, and

DOE. The authors thank Dr. Guoxing Wang and Dr. Mehmet Yuce for
technical discussions.
Appendix
A.1 Equivalent AC Resistance

The resistance and power dissipation of any conductor carrying
DC current is usually calculated with the assumption of a uniform
current density distribution. Any nonuniform current distribution,

which is induced by the imposed magnetic field perpendicular to
the current flow, can significantly increase the power dissipation.
The term “skin effect” applies to nonuniformity in a single conductor,
caused by the magnetic field induced by itself. “Proximity effect” is a
nonuniformity caused by the magnetic fields of other conductors. In
this section, we give formula to calculate power loss from skin effect
and proximity effect for popular circular shape conductors.
P D C γ berγ bei γ − bei γ ber γ
P D C + PS K = (A.1)
2 (ber γ )2 + (bei γ )2
√
2 2πrs ber2 γ ber γ + bei 2 γ bei γ 2
PP R O =− H peak , (A.2)
σδ (berγ )2 + (bei γ )2
where PS K and P P R O are induced power dissipation from skin effect
and proximity effect, beri and beii are the real and imaginary parts
√
of a i th Bessel function, γ = j j μσ ωr (ω is the angular frequency
of the current), μ is the permeability of the conductor, σ is the
conductivity of the conductor, Hpeak is the magnitude of H field
imposed by nearby conductors.
In the case that the radius of the cylindrical conductor is
smaller or comparable to the skin depth, an approximation of
Bessel functions allows the calculation of the AC resistance of one
strand/turn
R S K = 0.021(rs /δ)4 R D C , (A.3)
π 3rs4 μ20 σ f 2 2
PPRO = H peak , (A.4)
2
where rs is the radius of an individual strand, R D C is the DC
√
resistance, and δ = 2/μσ ω is the skin depth.
Depending on the number of turns, and how tightly the coil
is wind, the contribution from skin effect and proximity effect to
the total power varies. For coils of multiple turns and strands
(Nt Ns 1), proximity effect contributes much more power than
skin effect does. Based on Eqs. A.3 and A.4, the AC resistance of a
coil can be represented as
R AC = R D C [1 + 0.125Nt Ns η β(πrs2 μ0 σ )2 ], (A.5)
Appendix 459
Figure A.1. Numeric values of η for coils with different cross sections.
b and t are the coil’s width and thickness.
where η is a parameter relates to coil winding cross section

geometry (shown in Fig. A.1), β is the area efficiency defined as
the ratio of the total conducting area over the cross section of the
winding.
A.2 Coil Model

As illustrated in Fig. A.2, a coil is modeled as a distributive RCL
network, where inductive and capacitive couplings between turns
Figure A.2. Distributive equivalent model of a coil. I L i represents the

current going through the inductive branch L i , Ri is the equivalent ESR of
the inductive branch L i , IC p,k denotes the mesh current through the parasitic
capacitance C p,k , Ie is the external driving current, and M p,k is the mutual
inductance between turn p and turn k.
( p and k) are modeled as mutual inductance M p,k and parasitic

capacitance C p,k .
The inductive and resistive elements in Fig. A.2 can be described
based on the former appendix and physics formulae
Ri = Ri.D C [1 + 0.125Nt Ns η β(πrs2 μ0 σ )2 ]
L i = 0.5 μ0 D i ln(D i /OD)

M p,k = (cos θ/l)dsds
where D i is the diameter of the conductor loop, OD is the diameter
of a single turn, θ is the angle of inclination between the two loop
elements ds and ds’, and l is the radius vector between them.
The calculation of parasitic capacitance between turns is
performed using the model shown in Fig. A.3. The turn-to-turn
parasitic capacitor is formed by the insulation layers of turns and
nonconductive air gap. Per unit angle, the turn-to- turn parasitic
capacitance value is
0.5C gap C i nsulate
Ct = . (A.6)
C gap + 0.5C i nsulate
Under the condition that the insulation layer is thin compared with
the turn radius, the parasitic capacitance between two turns is

π D i r0
C p,k = ε0 εr dθ, (A.7)
ς + εr r0 (1 − cos θ) + 0.5ε, h
Figure A.3. Turn-to-turn parasitic capacitance. (a) Illustration of the

parasitic capacitance in a coil winding. (b) Equivalent model to calculate the
turn-to-turn parasitic capacitance.
References 461
where θe is the effective angle between turn p and turn k, and h is

the separation of two turns. As a first order approximation, assume
that each turn except those on the perimeter of a coil is surrounded
by four turns and θ1 = θ2 = θ3 = 90◦ .
Through analysis as shown in reference [23], the distributive coil
model can be simplified as a lumped capacitor C self , in parallel with
and inductor, as

C self = C p,k (k − p)2 /n2t (A.8)

p<k
Consequently, the equivalent impedance and Q factor of the coil is

R AC + j ωL
Z e = ( j ωL + R AC )||(1/j ωC sel f ) = ,
(1 − ω2 L C sel f ) + j ω R AC C sel f
(A.9)
Q ( f ) ≈ 2π f L (1 − f 2 / fsel
2
f )/R D C (1 + f / f h ),
2 2
(A.10)
√ √ −1
where fh = 2 2/ πrs2 μ0 σ η β Nt Ns , fsel f = 2π L C sel f , and
R D C = Nt (D out + D i n )/2σ πrs2 Ns , as defined in previous sections.
References
1. Zhou, M., Yuce, M. R., and Liu, W. (2008) A Non-Coherent DPSK Data
Receiver With Interference Cancellation for Dual-Band Transcutaneous
Telemetry, IEEE Journal of Solid-State Circuits, 43(9), pp. 2003–
2012.
2. Buchegger, T., Oberger, G., Reisenzahn, A., Hochmair, E., Stelzer, A., and
Springer, A. (2005) Ultra-Wideband Transceivers for Cochlear Implants,
EURASIP Journal on Applied Signal Processing, 2005(18), pp. 3069–
3075.
3. Spelman, F. A. (May 1999) The Past, Present, and Future of Cochlear
Prostheses, IEEE Engineering in Medicine and Biology Magazine, 18(3),
pp. 27–33.
4. McDermott, H. (July 1989) An Advanced Multiple Channel Cochlear
Implant, IEEE Transactions on Biomedical Engineering, 36(7), pp. 789–
797.
5. Liu, W., Vichienchom, K., Clements, M., DeMarco, S. C., Hughes, C.,
McGucken, E., Humayun, M. S., de Juan, E., Weiland, J. D., and Greenberg,
R. (2000) A Neuro-Stimulus Chip With Telemetry Unit for Retinal
Prosthetic Device, IEEE Journal of Solid-State Circuits, 35(10), pp. 1487–
1497.
6. Weiland, J. D., Humayun, M. S. (July 2008) Visual Prosthesis, in
Proceedings of IEEE, 96(7), pp. 1076–1084.
7. Theogarajan, L., Wyatt, J., Rizzo, J., Drohan, B., Markova, M., Kelly, S.,
et al. (February 2006) Minimally Invasive Retinal Prosthesis, IEEE ISSCC
Digest of Technical Papers, 2(5), pp. 99–100.
8. Ortmanns, M., Rocke, A., Gehrke, M., Tiedtke, H. J. (December 2007)
A 232-Channel Epiretinal Stimulator ASIC, IEEE Journal of Solid-State
Circuits, 42(12), pp. 2946–2959.
9. Rothermel, A., Liu, L., Aryan, N. P., Fischer, M., Wuenschmann, J., Kibbel,
S., et al. (January 2009) A CMOS Chip With Active Pixel Array and
Specific Test Features for Subretinal Implantation, IEEE Journal of Solid-
State Circuits, 44(1), pp. 290–300.
10. Schwartz, G. D., and Barkin, J. S. (2007) Small-Bowel Tumors Detected
by Wireless Capsule Endoscopy, Digestive Diseases and Sciences, 52(4),
pp. 1026–1030.
11. Song, Y., Borton, D. A., Park, S., Patterson, W. R., Bull, C. W., Laiwalla,
F., Mislow, J., Simeral, J. D., Donoghue, J. P., and Nurmikko, A. V.
(2009) Active Microelectronic Neurosensor Arrays for Implantable
Brain Communication Interfaces, IEEE Transactions on Neural Systems
and Rehabilitation Engineering, 17(4), pp. 339–345.
12. Chestek, C. A., Gilja, V., Nuyujukian, P., Kier, R. J., Solzbacher, F., Ryu, S.
I., Harrison, R. R., and Shenoy, K. V., Hermes, C. (August 2009) Low-
Power Wireless Neural Recording System for Freely Moving Primates,
IEEE Transactions on Neural Systems and Rehabilitation Engineering, 17,
pp. 330–338.
13. Lin, C. T., Ko, L. W., Chiou, J. C., Duann, J. R., Huang, R. S., Liang, S. F., Chiu,
T. W., and Jung, T. P. (July 2008) Noninvasive Neural Prostheses Using
Mobile and Wireless EEG, Proceedings of the IEEE, 96(7).
14. Low, K. S., Lee, G. X., and Taher, T. (2009) A Wearable Wireless
Sensor Network for Human Limbs Monitoring, IEEE Instrumentation
and Measurement Technology Conference, pp. 1332–1336.
15. Harrison, R., Kier, R. J., Chestek, C. A., Gilja, V., Nuyujukian, P., Ryu, S.,
Greger, B., Solzbacher, F., and Shenoy, K. V. (August 2009) Wireless
References 463
Neural Recording With Signal Low-Power Integrated Circuit, IEEE

Transactions on Neural Systems and Rehabilitation Engineering, 17, pp.
322–329.
16. Akin, T., Najafi, K., and Bradley, R. M. (January 1998) A Wireless
Implantable Multichannel Digital Neural Recording System for a
Micromachined Sieve Electrode, IEEE Journal of Solid-State Circuits,
33(1), pp. 109–118.
17. Mohseni, P., and Najafi, K. (2003) A Wireless FM Microsystem for
Biomedical Neural Recording Applications, in Proceedings of IEEE
Southwest Symposium on Mixed-Signal Design, pp. 217–222.
18. Chae, M., Yang, Z., Yuce, M. R., Hoang, L., and Liu, W. (2009) A 128-
Channel 6 mW Wireless Neural Recording IC With Spike Feature
Extraction and UWB Transmitter, IEEE Transactions on Neural Systems
and Rehabilitation Engineering, 17, pp. 312–321.
19. Sodagar, A. M., Wise, K. D., and Najafi, K. A. (2009) Wireless Implantable
Microsystem for Multichannel Neural Recording, IEEE Transactions on
Microwave Theory and Techniques, 57, pp. 2565–2573.
20. Chow, E. Y., Ouyang, Y., Beier, B., Chappell, W. J., and Irazoqui, P. P.
(2009) Evaluation of Cardiovascular Stents as Antennas for Implantable
Wireless Applications, IEEE Transactions on Microwave Theory and
Techniques, 57, pp. 2523–2532.
21. Kelly, S. K., Shire, D. B., Doyle, P., Gingerich, M. D., Drohan, W. A., Rizzo, J. F.,
Theogarajan, L. S., Chen, J., Cogan, S. F., and Wyatt, J. L. (2009) The Boston
Retinal Prosthesis: A 15-Channel Hermetic Wireless Neural Stimulator,
2nd International Symposium on Applied Sciences in Biomedical and
Communication Technologies (ISABEL 2009), pp. 1–6.
22. Sivaprakasam, M., Liu, W., Wang, G., Zhou, M., Weiland, J. D., and
Humayun, M. S. (2006) Challenges in System and Circuit Design for
High Density Retinal Prosthesis, IEEE/NLM Life Science Systems and
Applications Workshop (LSSA), pp. 1–2.
23. Yang, Z., Liu, W., and Basham, E. (October 2007) Inductor Modeling
in Wireless Links for Implantable Electronics, IEEE Transactions on
Magnetics, 43, pp. 3851–3860.
24. Yu, H., and Najafi, K. (2003) Low-Power Interface Circuits for Bio-
Implantable Microsystems, IEEE ISSCC Digest of Technical Papers, 1, pp.
194–487.
25. Wise, K. D., Anderson, D. J., Hetke, J. F., Kipke, D. R., and Najafi, K.
(2004) Wireless Implantable Microsystems: High-Density Electronic
Interfaces to the Nervous System, Proceedings of the IEEE, 92(1),

pp. 76–97.
26. Tang, Z., Smith, B., Schild, J. H., and Peckham, P. H. (1995) Data
Transmission From an Implantable Biotelemeter by Load-Shift Keying
Using Circuit Configuration Modulator, IEEE Transactions on Biomedical
Engineering, 42(5), pp. 524–528.
27. Young, D. J. (2009) Wireless Powering and Data Telemetry for Biomed-
ical Implants, in 31st IEEE EMBS Annual International Conference, pp.
3221–3224.
28. Ghovanloo, M., and Najafi, K. (2004) A wideband Frequency-Shift
Keying Wireless Link for Inductively Powered Biomedical Implants,
IEEE Transactions on Circuits and Systems, 51(12), pp. 2374–
2383.
29. Jung, L. H., Byrnes-Preston, P., Hessler, R., Lehmann, T., Suaning, G.
J., and Lovell, N. H. (2007) A Dual Band Wireless Power and FSK
Data Telemetry for Biomedical Implants, in 29th IEEE EMBS Annual
International Conference, pp. 6596–6599.
30. Troyk, P. R., and DeMichele, G. A. (2003) Inductively Coupled Power
and Data Link for Neural Prostheses Using a Class-E Oscillator and FSK
Modulation, in 25th IEEE EMBS Annual International Conference, 4, pp.
3376–3379.
31. Xu, W., Luo, Z., and Sonkusale, S. (2009) Fully Digital BPSK Demodulator
and Multilevel LSK Back Telemetry for Biomedical Implant Trans-
ceivers, IEEE Transactions on Circuits and Systems, 56(9), pp. 714–
718.
32. Hu, Y., and Sawan, M. (2005) A Fully Integrated Low-Power BPSK
Demodulator for Implantable Medical Devices, IEEE Transactions on
Circuits and Systems, 52(12), pp. 2552–2562.
33. Chen, K., Yang, Z., Hoang, L., Weiland, J. D., and Humayun, M. S.
(September 2010) An Integrated 256-Channel Epiretinal Prosthesis,
IEEE Journal of Solid-State Circuits, 45(9), pp. 1946–1956.
34. Wang, G., Liu, W., Sivaprakasam, M., and Kendir, A. (October 2005)
Design and Analysis of an Adaptive Transcutaneous Power Telemetry
for Biomedical Implants, IEEE Transactions on Circuits and Systems,
52(10), pp. 2109–2117.
35. Proakis, J. G. (1995) Digital Communications, McGraw Hill, Singapore.
36. Vaughan, R. G., Scott, N. L., and White, D. R. (September 1991) The
Theory of Bandpass Sampling, IEEE Transactions on Signal Process,
39(9), pp. 1973–1984.
References 465
37. Buchegger, T., Ossberger, G., Reisenzahn, A., Hochmair, E., Stelzer, A., and
Springer, A. (2005) Ultra-Wideband Transceivers for Cochlear Implants,
EURASIP Journal on Applied Signal Processing, 18, pp. 3069–3075.
38. Porcino, D., and Hirt, W. (2003) Ultra-Wideband Radio Technology:
Potential and Challenges Ahead, IEEE Communications Magazine, 41, pp.
66–74.
39. Siwiak, K. (2001) Ultra-Wide Band Radio: Introducing a New Technol-
ogy, in IEEE VTS 53rd Vehicular Technology Conference, 2, pp. 1088–1093.
Chapter 16
Ultra Wideband for Wireless Body

Area Networks
Mehmet Rasit Yucea,b and Ho Chee Keongb

a Department of Electrical and Computer Systems Engineering,
Monash University,
Clayton, VIC 3800, Australia
b School of Electrical Engineering and Computer Science,
The University of Newcastle, Callaghan,

NSW, 2308 Australia
[email protected]
Wireless technologies based on narrow bands, especially 2.4 GHz

ISM, have been used in medical monitoring widely. This chapter
discusses the use of ultra-wide band technology (UWB) for medical
monitoring systems. UWB presents some unique benefits when
compared with the narrow-band systems: (i) low-power transmitter
design, (ii) low radio frequency (RF) and electromagnetic interfer-
ence (EMI) effects in medical environment, (iii) small-size antenna,
and (IV) high data rate. Especially the UWB has efficiently been used
for simultaneous monitoring of many continuous physiological sig-
nals such as ECC (electrocardiogram) /EEG (electroencephalogram)
and neural signals for brain-computer interfaces because of its high
data rate capability.

www.panstanford.com
468 Ultra Wideband for Wireless Body Area Networks
16.1 Introduction
Currently, there is no wireless standard for a wireless body area

network (WBAN). Devices designed by different manufacturers
utilize different frequency bands and protocols. These resulted in
a lack of interoperability among the different WBAN solutions.
Therefore, standardization is required to ensure interoperability.
In November 2007, IEEE 802.15 TG6 was formed to develop a
communication standard optimized for body area network [1]. The
goal of this task group is to standardize both the physical (PHY)
and media access control (MAC) layers for a short-range, low-power,
and highly reliable wireless communication scheme to operate in or
near human body. Several wireless bands such as MedRadio, Medical
Implant Communication Service (MICS), industrial, scientific, and
medical (ISM), Wireless Medical Telemetry System (WMTS), and
UWB are considered for WBAN standardization. Low data rate
UWB technology can find applications in biomedical monitoring
especially for multichannel continuous signal monitoring such as
EEG, ECG, EMG (Electromyography), and neural signals. Herein we
will take a closer look at how UWB can be applied to some of WBAN
applications.
UWB is a narrow pulse-based transmission system whose
spectrum is spread across a wide range of frequencies. Low data
rate UWB is attractive for vital signs monitoring systems because
its transmission power is lower than that of WLAN, Bluetooth, and
ZigBee systems, which is less likely to affect human tissue and cause
interference to other medical equipment. Furthermore, it is able to
transmit higher data rate, which is suitable for real time continuous
monitoring of many physiological signals. In this part of the book, we
will discuss the design of sensor node electronics using the low data
rate UWB transmission and its characteristics when operating in or
around human body.
16.1.1 Background of UWB

In 2002, the Federal Communications Commission (FCC) approved
the use of UWB wireless scheme. UWB is a narrow pulse baseband
system, whose spectrum is spread across a wide range of frequency.
Introduction 469
Figure 16.1. UWB and narrowband spectrum. See also Color Insert.
A generic definition of UWB is that the emitting signals have a

fractional bandwidth larger than 0.2 or at least 500 MHz. The
fractional bandwidth is defined as the bandwidth between the −10
dB lower and upper corner frequency. UWB is allowed to operate
in 0–960 MHz and 3.1–10.6 GHz bands, but the effective isotopic
radiated power has to be kept below −41 dBm/MHz [2]. The major
difference between an UWB signal and a narrowband signal is that
an UWB signal occupies a large bandwidth with a transmission
power below the noise floor of narrowband system. Figure 16.1
shows the spectrum mask of an UWB signal that highlights the major
differences between a wideband and a narrowband system.
Basically, there are two types of UWB system: the multiband
carrier-based UWB system and the impulse radio UWB system (IR-
UWB). A multiband system is supported by WiMedia, and it uses
the OFDM (orthogonal frequency-division multiplexing) modulation
scheme with the spectrum divided into bands of 528 MHz. The
power consumption of multiband UWB system is high due to
the complex real-time signal processing required for the OFDM
modulation and demodulation. The power consumption of WiMedia
compatible chip sets is typically more than 500 mW; therefore, these
are not suitable for battery-powered devices. Impulse radio systems
send very short pulses and use simpler modulation schemes such
as pulse-position modulation (PPM) and on-off keying (OOK), which
result in significant power saving and, therefore, more suitable for
battery-operated WBAN applications.
In March 2007, a low data rate UWB standard (IEEE 802.15.4a)

based on impulse UWB was approved [3]. This standard is
developed mainly for high precision ranging application. Therefore,
it is not suitable to be directly applied to a WBAN application.
Modifications are required in the PHY and MAC layers in order to
match the WBAN requirements.
A WBAN system requires a cross layer solution. In order to design
an efficient WBAN system, it is important to consider the hardware
design, PHY layer, and the MAC protocol. In this chapter, the
advantages and limitations of UWB, UWB sensor node design, UWB
PHY, and UWB MAC layers developments for WBAN applications will
be covered.
16.2 Advantages and Limitations of UWB for WBAN
In a WBAN system, real-time data acquisition and analysis is needed;

therefore, the ability to sustain higher data rate and provide an
efficient communication protocol is essential. A WBAN system is
not only limited to medical usage, it can also support personal
entertainment applications. Therefore, a scalable data rate from
10 bps to 10 Mbps is required. A WBAN sensor node for medical
usage can be wearable or implantable. Therefore, ultra-low power
sensor node is extremely important, especially for the case of an
implantable node. As WBAN system is designed to operate near or
in human body, special absorb ratio (SAR) is an important factor
to be considered. This will translate to a need for using lower
transmission power. The basic requirements for a WBAN are as
follows [4]:
• Supports different types of sensors, both inside and outside

body
• Supports scalable data rates from 10 bps to 10 Mbps
• Data from the sensor nodes need to be secure and reliable
• Ultra-low power nodes
• Coexistence with existing networks
• Low transmission power
• Small form factor
Advantages and Limitations of UWB for WBAN 471
16.2.1 Favorable Factors for Use of UWB in

WBAN Applications
Table 16.1 shows the power consumption of various wireless sensor
platforms used for a WBAN application. Energy per bit for UWB
platform is much lower as compared to the other narrowband
wireless platforms. Therefore, it is an advantage to use the UWB
wireless scheme for WBAN applications considering the transmitter
power consumption.
Table 16.2 shows the parameters of the commonly used wireless
schemes for WBAN applications. A WBAN system needs to support
scalable data rates up to 10 Mbps; only the UWB and WLAN wireless
schemes can meet this requirement. UWB is preferred to WLAN,
as a WLAN system is much more power hungry as compared to
UWB.
In terms of coexistence, Bluetooth, ZigBee, and WLAN operate
in the 2.4 GHz ISM band, which is overcrowded and, therefore,
has a higher probability of interfering with the existing wireless
Table 16.1. Power consumption for various wireless sensor platforms [4]
Current
Model Company Frequency Data rate Tx Rx
UWB University of Newcastle 3.1–10.6 GHz 10 Mbps 15 mA −
Mica2 Crossbow 868/916 MHz 38.4 Kbps 27 mA 10 mA
(MPR400)
MicAz Crossbow 2.4 GHz 250 Kbps 17.4 mA 19.7 mA
Mica2DOT Crossbow 433 MHz 38.4 Kbps 25 mA 8 mA
CC1010 Texas Instrument 300–1000 MHz 65 Kbps 26.6 mA 11.9 mA
CC2400 Texas Instrument 2.4 GHz 1 Mbps 19 mA 23 mA
Table 16.2. Wireless scheme comparison table
Wireless scheme Max data rate Frequency Max EIRP

Ultra Wideband > 10 Mbps 3.1–10.6 GHz −41dBm
Bluetooth 3 Mbps 2.4 GHz 0 dBm
ZigBee 250 Kbps 2.4 GHz 0 dbm
WLAN (802.11b) 11 Mbps 2.4 GHz 30 dBm
MICS 300 kbps 402–405 MHz −16 dBm
infrastructure. UWB and MICS wireless scheme are less likely to

cause interference to the existing wireless infrastructure as UWB
operates below the noise floor of existing wireless system, while
MICS is a dedicated frequency band for medical usage.
From Table 16.2, UWB’s max effective isotropically radiated
power (EIRP) is the lowest as compared to other wireless schemes.
The low transmission power reduces the chances of heating effects
on human tissue and increases the difficulty level for unauthorized
signal tapping.
Another added advantage of an UWB system is hardware
simplicity for the design of UWB transmitter, which in turn will lead
to a more compact sensor node design. As described so far, it is
obvious that there are important favorable factors for the use of
UWB wireless scheme in a WBAN application.
16.2.2 Limitations of UWB

A major challenge faced in the design of UWB is its receiver
complexity required to demodulate the received low-power level
signal and the narrow pulses. In wireless communication systems,
there are basically two types of receivers: coherent receiver and
noncoherent receiver. Coherent receivers synchronize the receiver
clock (i.e., local oscillator) with the transmitter clock and the re-
ceiver generates template waveforms locally. A noncoherent receiver
does not require the synchronization between the transmitter and
the receiver, and does not generate a local template waveform. For
noncoherent receiver, detection is commonly achieved by means
of energy detection or using the reference waveform generated
from the transmitted signal. In general, noncoherent receivers
have a simpler implementation, but are penalized on performance;
however, this may not be the case for UWB. An UWB coherent
receiver requires a very high speed ADC with a large analog input
bandwidth and needs to achieve precise synchronization. These two
factors are critical for a reliable operation of a coherent receiver,
which is very difficult to achieve due to the extremely narrow pulse
nature (typically less than 2 ns) of an UWB system. As a result of
an imperfect synchronization in case of a coherent UWB receiver,
significant performance degradation will occur.
UWB Hardware Development 473
The power consumption of an UWB receiver is much higher than

that of a UWB transmitter whether it is coherent or non-coherent.
High power consumption is not acceptable for a power limited
sensor nodes.
16.3 UWB Hardware Development
The basic hardware components of any wireless scheme are the an-
tenna, transmitter, and receiver. The large bandwidth characteristic
of an UWB signal has posed a new set of problems for hardware
designers, which is different from what they faced in a narrowband
system. To name a few differences, the antenna requires a flat pass
band of a few gigahertz, the transmitter employs direct modulation
techniques, and the receiver now has to sample and synchronize
very narrow pulses. In this section, we will take a look at some of
the UWB hardware components developed for WBAN applications.
16.3.1 UWB Antennas for WBAN Applications

UWB operates in the gigahertz frequency range, which makes it
easier to achieve an antenna with smaller form factor. This is
due to the fact that antenna size is relatively proportional to the
wavelength. As shown in Eq. 16.1, wavelength reduces as frequency
increases.
c
Wavelength, λ = (16.1)
f
where c is the speed of light (3 × 108 m/s) and f is the frequency.
In recent years, several compact antennas that are suitable
for use in WBAN applications are proposed. Textile antenna [5]
and button antenna [6] are such examples that are designed
to seamlessly integrate into the clothing for on-body WBAN
applications. The textile antenna is only 0.5 mm thick; it uses a highly
flexible and conductive metalized nylon fabric as the conductor and
an acrylic fabric as dielectric substrate. The button structure used
in the button antenna is constructed using polytetrafluoroethylene
(PTFE) material. The PTFE button structure is fixed to a metal disc.
The metal disc is connected to a 50 micro-strip line with a ground
plane at the bottom.
Placement of the antenna on the human body plays an important

role in determining the performance of a WBAN system. Typically,
an antenna is placed at least 10 mm away from the body. If the
antenna is placed very close to the body, the signal will be severely
attenuated. Distance between the body and the antenna only affects
the signal strength; there is no significant impact on the pulse shape
and does not cause pulse spreading [5].
Another factor that should be taken into account when selecting
the appropriate antenna for WBAN application is the beam width. A
directional antenna is able to achieve higher gain at the expense of
a smaller angle of coverage, while an Omni antenna covers a wider
area but with a lower antenna gain. A directional antenna with 60◦
beam width is sufficient, if both transmit and receive antennas are
located on the torso of the body. If the transmit or receive antenna
is positioned on a moving part of the body such as arm or head, an
Omni antenna would be a better choice.
Antenna designed for use in an implantable device should take
dielectric permittivity and conductivity of the surrounding tissues
into consideration. The implantable environment will cause the
antenna operating frequencies to shift and this would affect the
impedance matching for the frequency band of interest.
Apart from the normal design consideration for designing a
typical antenna used in free space condition, four additional factors
should be considered when designing an antenna for a WBAN
application.
(1) How can it be seamlessly integrated into clothing?

(2) Effect of body on the radiation pattern.
(3) What is the minimum distance between the antenna and the
body?
(4) Where will the antennas be located? (i.e., whether it is mostly
static or there is much movement)
16.3.2 UWB Transmitters for WBAN Applications

Basically there are two methods of generating an UWB pulse: trans-
mission line method and CMOS-based methods. UWB monocycle
pulse can be generated by passing a square wave through a step
Figure 16.2. Block diagram of a carrier-less UWB transmitter for WBAN

application.
recovery diode (SRD) and a pair of shorted circuit transmission line

[7, 8]. But this method requires constant biasing of the diode and,
therefore, is not suitable for low-power WBAN applications. Further-
more, the size of the transmitter designed using transmission line
method is much larger than that of a CMOS UWB transmitter. An
alternative method of implementing UWB transmitter is using CMOS
integrated circuit (IC) technology. CMOS-based UWB transmitters
are more energy efficient and have much smaller form factors, which
is more suited for WBAN applications.
CMOS-based transmitter solutions suitable for use in WBAN
have been proposed in reference [9–11]. The UWB transmitter
proposed in reference [9] uses a narrow pulse generator followed
by a buffer amplifier as shown in Fig. 16.2. The pulse generator
is able to generate pulses with pulse width ranging from 50 ps
to 500 ps. Spectral shaping is achieved by means of an external
filter and an antenna. A CMOS UWB transmitter for neural recording
systems has been proposed in reference [10]; the working principle
of the UWB transmitter is similar to that of reference [9]. In
reference [11], a carrier-based CMOS UWB transmitter for WBAN
applications is proposed. The block diagram for the transmitter
proposed in reference [11] is shown in Fig. 16.3. The triangular
pulse determines the bandwidth, while the oscillator determines
the center frequency (it is similar to a carrier frequency used in
narrowband radio systems). Here the triangle-pulse generator can
also be replaced with a narrow rectangular-pulse train generator.
An UWB transmitter for ECG monitoring assembled using off-the-
shelves CMOS components is documented in reference [12]. The
working principle of the transmitter is similar to that of references
[9] and [10], which uses direct modulation technique (carrier-less
based).
Figure 16.3. Block diagram of carrier-based UWB transmitter for WBAN

application.
XOR
Delay
AND
Figure 16.4. Narrow UWB pulse generation. See also Color Insert.
The basic building block for most CMOS-based UWB transmitters

suitable for WBAN applications is formed using either a narrow
rectangular pulse train generator or a triangle pulse generator.
Figure 16.4 shows an example of how a narrow rectangular pulse
can be generated in a CMOS circuit using a delay unit [12]. The AND
is used to allow the narrow pulse at rising edges only. The triangular
pulse can be achieved by using inverter together with NAND or NOR
gates as shown in Fig. 16.5 [13]. Here the inverter causes a very short
delay, which results in the formation of triangular pulse; the pulse
width depends largely on the rise and fall time of the input signals.
An UWB transmitter using direct modulation does not require
an oscillator, which reduces the size of the transmitter and allows
lower power consumption. Therefore, it is more suitable for WBAN
applications as compared to carrier-based UWB transmitter. For a
carrier-less UWB transmitter, selection of the pulse width, rise and
fall time will affect the performance of the system, as they determine
the bandwidth and the peak and null of the UWB spectrum. Through
proper selection of the pulse width, rise and fall time, it is possible
(a) Nand gate implementation
(b) Nor gate implementation
Figure 16.5. Triangle pulse generation. See also Color Insert.
to enhance the bit error rate and reduce the power consumption of
an UWB WBAN system. The effects of the narrow rectangular pulse
parameters on the UWB spectrum are illustrated in the next section.
16.3.2.1 Effects of pulse width on UWB spectrum

Equation 16.2 shows the Fourier series expansion of a rectangular
pulse train [14]. In order to satisfy the FCC’s spectrum requirement,
the signal must be band limited, which can be described by Eq. 16.3
[12]. Equation 16.3 is based on an ideal pulse with zero rise and
fall time. A realistic pulse with finite rise and fall time can be
represented by Eq. 16.4. Finite rise time adds an additional null to
the circuit as shown in Fig. 16.6. The pulse width for both spectrums
in Fig. 16.6 is selected such that it is optimized to operate at center
frequency of 4 GHz, with the spectrum peak at 4 GHz and nulls
at 3 GHz and 5 GHz. The red spectrum has a rise time of 100 ps,
while the blue spectrum has a rise time of 250 ps. It can be seen
from Fig. 16.6 that the null of blue spectrum due to rise time
appears at 4 GHz, which severely degrades the performance of the
UWB transmitter. Null due to rise/fall time occurs every 1/tr , and
null arising from the pulse width appears every (τ + tr ) interval.
Figure 16.6. Spectrum of an UWB rectangular pulse in frequency domain

and nulls in the UWB spectrum due to rise time. See also Color Insert.
Therefore, the format of the transmitted pulse should be arranged

to ensure that the null does not appear at the frequency band
of interest. It is desirable to keep the spectrum peak of the sinc
envelope at the center frequency. The effect of null in the band of
interest for a bandlimited signal is shown in Fig. 16.7. In the worst
case scenario, a null at the center of the band of interest will result
Figure 16.7. Effect of null in a band-limited signal. See also Color Insert.
in excessive ringing, which will limit the data rate and increases the
noise floor.
∞
Aτ 2Aτ sin (π kτ / Tb )
x (t) = + cos (kω t) (16.2)
Tb Tb k=1 π kτ / Tb
2Aτ sin (π kτ / Tb )
n2
x (t) = cos (kω t) (16.3)
Tb k=n1 π kτ / Tb
2A (τ + tr ) sin (π ktr / Tb )
n2
x (t) =
Tb k=n1
π ktr / Tb
sin (π k (τ + tr ) / Tb )
cos (kω t) (16.4)
π k (τ + tr ) / Tb
where A is the pulse amplitude, Tb is the period of the clock, τ is
the pulse width, tr is the rise time, fc is the centered frequency, ω =
2π/Tb , n1 = ω1 /ωn2 = ω2 /ω, ω1 and ω2 are the lower and upper
cutoff frequencies of the band-pass filter.
It is important to select the appropriate rise time and pulse width
to ensure optimized performance for an UWB transmitter. Rise time
can be optimized using Eq. 16.5, where 0.5 corresponds to a 4 dB
loss. The peak of the spectrum (i.e. the sinc envelope) can be put at
the center frequency by using Eq. 16.6, where α represents the side
lobes, “1” is the main lobe, and “3” is the first side lobe. Optimizing
the nulls and peak of the spectrum enhances the performance of the
transmitter, which is achieved by adjusting the pulse width.
fc tr < 0.5 (16.5)
fc (τ + tr ) = 0.5 α, where α = 1, 3, 5, 7, 9, . . . (16.6)
where fc is the center frequency, τ is the pulse width and tr is the

rise time.
Impulse UWB transmitter without carrier is the simplest and
most popular form of UWB transmitter used in WBAN applications.
When using this type of UWB transmitter, proper selection of
pulse width and rise/fall time is required to ensure an optimized
performance.
16.3.3 UWB Receiver

As seen from the previous section, the UWB transmitter design is
simple, provides low power consumption, and is able to support
a very high data rate. Typically, an UWB receiver is much more
complex as compared to an UWB transmitter, and the receiver power
consumption is much higher than that of the UWB transmitter. As
discussed in Section 16.2, most of the favorable factors for using
UWB in WBAN applications lie in the transmitter portion, while
most of the limitations are found in the UWB receiver portion. In
this section, we will look at the UWB receiver for use in WBAN
applications.
Figure 16.8 shows the block diagram of the UWB receiver front-
end designed for a WBAN application [9] using energy detection
principle. The input signal is first amplified by low-noise amplifier
(LNA), followed by an optional fixed-gain amplifier before the self-
mixing stage. The self-mixer is a square-law device that translates
the high frequencies received signal to baseband signals. The
baseband signal is further amplified by a variable-gain amplifier
before been fed to the integrator bank. The integrator bank is made
up of eight integrators, each having a minimum integration time
of 8 ns. Therefore, the maximum data rate allowed for this system
is 15.625 Mb/s. Demodulation is performed at the output of the
integrator bank using a comparator with a threshold set by a 5-bit
digital-to-analog converter (DAC). Synchronization and bit detection
are performed using the field-programmable gate array (FPGA).
Figure 16.8. Block diagram of an UWB receiver [9].

PHY Layer for UWB WBAN 481
The key objectives to be considered when designing an UWB

receiver for WBAN applications are to keep the circuitry simple and
the power consumption low. The most power-hungry component in
an UWB receiver front-end is the analog-to-digital converter (ADC),
especially for the case of a fully digital UWB receiver. For a fully
digital UWB receiver, the ADC is placed after the LNA before the
analog down-conversion. An ADC with sampling rate and input
bandwidth of a few gigahertz is required for a fully digital UWB
receiver, and these would result in significant power consumption,
which is not suitable for battery-powered applications. Therefore,
the receiver proposed in reference [9] chooses to place the
comparator after the analog integrator bank, where the comparator
is only required to sample at baseband data rate (tens of MHz). The
second highest power consumption component in an UWB receiver
front-end is template generation circuitry. A matching template is
difficult to generate, and synchronizing the received signal with
the generated template is very challenging, thus increasing the
power consumption and complexity of the receiver. Therefore, self-
mixing energy detection method is selected in reference [9] for
implementing a suitable UWB receiver for WBAN applications.
Although a noncoherent energy detection receiver may not
perform as well as a coherent UWB receiver or may not have the
flexibility of a fully digital receiver, it is, however, the best UWB
receiver choice for a WBAN application. The choices of antenna,
UWB transmitter, and receiver have been highlighted in this section.
The goal for hardware design of a WBAN application is to keep the
circuit small, simple, and the power consumption low.
16.4 PHY Layer for UWB WBAN
A WBAN system can be broadly classified into two groups: wearable

BAN and implantable BAN. Wearable BAN can be for both medical
usage and entertainment purpose. Implantable BAN is mainly
for medical applications. As the environment that wearable and
implantable BAN are operating in is very different, it has been
suggested to use two separate PHY layers for implantable and
wearable BAN [15].
Table 16.3. UWB PHYs proposed for IEEE 802.15.6
Organization Signal source Modulation Data rate Frequency band (MHz)

ETRI-Samsung IR GPPM 100 k∼10 Mbps 3.1∼4.6 GHz
7.25∼8.5 GHz
FT-CEA-Thales IR PPM, DBPSK 1.95∼27.24 Mbps 3.1∼10.6 GHz
IMEC IR BPM, OOK, DBPSK 100 k∼27.2 Mbps 3∼10 GHz
TI IR PPM 500 Kbps 6.4∼8.9 GHz
NICT-YNU-Meiji IR/ Chirp BPM/ DMPSK 200 k∼10 Mbps 3.1∼10.6 GHz
CSEM FM FSK-FM 30 ∼ 250 Kbps 6∼9 GHz
Several UWB WBAN PHY layer solutions have been proposed

to IEEE 802.15.6 task group. The proposed UWB PHYs are largely
variations of IEEE 802.15.4a. Impulse radio-based UWB is the most
popular choice of signal generation method [16]. Table 16.3 shows
the proposed PHY specifications by the different organizations
compiled by reference [16].
16.5 UWB WBAN Channel
Based on the UWB channel model provided by IEEE 802.15.4a, it

can be seen that no energy is penetrating the body, instead energy
is diffracting around the body. Therefore, the distance between the
transmitter and the receiver is the distance along the body perimeter
instead of direct distance between the transmitter and the receiver.
Based on measurement data obtained from IEEE 802.15.4a, there
are always two clusters of multipath components for the WBAN
channel. The first cluster is the signal diffracting around the body,
and the second cluster is due to the reflection of the ground.
Table 16.4 shows the channel model for four scenarios defined by
IEEE 802.15.6 task group for WBAN applications. Only CM3 and
CM4 channel models are defined for UWB. Path loss and power
delay profile (PDP) models for UWB are provided for both CM3
and CM4 scenarios. Log-normal distribution can be used to describe
the path amplitude for the static case. As for the nonstatic case,
where the object has large movement, the path amplitudes fit well
to Weibull distribution [17]. The path arrival time is modeled by
Poisson distribution.
MAC scheme for UWB WBAN 483
Table 16.4. Channel models for WBAN
Scenarios Description
CM1 Implantable to Implantable
CM2 Implantable to Body Surface/External
CM3 Body Surface to Body Surface
CM4 Body Surface to External
For the implantable case, reference [18] has developed a model

based on the time domain electromagnetic simulation using the
anatomy model of a human body. According to reference [18], the
path loss for inside the body can be modeled as a power function as
shown in Eq. 16.7. It also concludes that it is possible to improve
the communication link from inside the body to body surface by
ensuring the transmitted signal is polarized along the width of the
body instead of polarizing along the height of the body.
P (d) = a d b + a1 (16.7)
where d is the distance, a = −2.7 and b = 0.7 for the horizontally
polarized antenna, and a = −3 and b = 0.6 for the vertically
polarized antenna.
16.6 MAC scheme for UWB WBAN
Designing a MAC protocol for WBAN is challenging, as a single

protocol needs to satisfy two different applications, each with a very
unique set of requirements. Medical and personal entertainment
applications have different requirements on the MAC layer. Medical
applications typically have lower data rate, and the traffic is periodic.
Data reliability and low latency are criteria for life criteria medical
applications. Normally, the data rate for personal entertainment
applications is higher and busty in nature. Data reliability and
latency requirements are not as stringent as that of medical
applications. In this section, we will look at a few MAC protocols
designed for WBAN applications and discuss whether it is suitable
for UWB PHY layer.
Most of the MAC schemes designed for WBAN are a modification
from the IEEE 802.15.4 MAC scheme. An IEEE 802.15.4 network
Figure 16.9. IEEE 802.15.4 super-frame.
can operate in beacon and nonbeacon modes. In the beacon mode,

a periodic beacon is broadcasted by the personnel area network
(PAN) coordinator. The period between two consecutive beacons
is the super-frame structure. A super-frame structure is made up
of an active and an inactive period as shown in Fig. 16.9. The
active period consists of a contention access period (CAP) and a
contention free period (CFP). CSMA-CA mechanism is used to access
the channel during the CAP period, and reservation of guarantee
time slot (GTS) is used for channel access for the CFP portion. In
order to reserve a GTS slot, the node has to send a GTS request
during the CAP period using CSMA-CA, and the network coordinator
decides the GTS allocation. This MAC scheme is not entirely suitable
for WBAN applications as it permits only up to seven GTS, which
does not provide sufficient channel access for time critical WBAN
data. Furthermore, quality of service (QoS) could not be ensured
because in order for the node to obtain a GTS, it needs to be able
to send a GTS request during the CAP period. Nonbeacon mode is
not considered for a WBAN application because in this mode the
network is always active and, therefore, is not energy saving [19].
A priority-guaranteed MAC protocol for WBAN application is
proposed in reference [20]. In this scheme, there are two contention-
free periods; CFP1 is for timeslot reserved for periodic traffic (TSRP)
and CFP2 is for timeslot reserved for burst traffic (TSRB). TSRP is
mainly used for periodic medical signal, while TSRB is catered for
burst data package transmitted by personal entertainment devices.
Two separate control channels AC1 and AC2 are used for medical
and entertainment devices, respectively. Randomized slotted ALOHA
is the mechanism used for accessing the control channel in the
CAP period. The advantages of this scheme are that there are
two separate control channels for medical and entertainment
Figure 16.10. Priority-guaranteed MAC [20].
applications, therefore ensuring that the QoS of medical application

is not affected during the period where there is high traffic for
entertainment devices. Furthermore, all the data are transmitted
in the dedicated contention-free slot and, therefore, is able to
ensure low latency. In a multiuser scenario, nodes on different
users are not synchronized; collision is, therefore, unavoidable, and
this would significantly degrade the performance of the scheme.
From an UWB hardware perspective, constantly listening for the
beacons for synchronizing purposes would require extensive use of
an UWB receiver, which is not preferred as this leads to higher power
consumption.
A TDMA (time division multiple access)-based MAC scheme, the
BodyMAC, is proposed in reference [21]. This scheme utilizes sleep
mode to enhance the energy efficiency of the network. The frame
structure of BodyMAC is shown in Fig. 16.11. As most of the traffic is
from the nodes to the coordinator, a shorter duration is allocated for
downlink. The uplink portion occupies majority part of the frame.
CSMA-CA mechanism is the mechanism used for channel access in
the CAP zone. Typically, nodes in a WBAN system have low duty
cycle; therefore, a good sleep-mode mechanism can greatly enhance
the energy efficiency of the system but at the expense of flexibility.
In this scheme, in order for the nodes to enter into sleep mode,
the nodes should send a request to the gateway using CSMA-CA
during the CAP period. The nodes will enter sleep mode when
an ACK (acknowledgment or acknowledged) is received from the
gateway. In the sleep mode, the nodes will not be able to listen
to beacons, broadcast information, and receive data. But these can
transmit data to the gateway if a GTS has been allocated. GTS is
Figure 16.11. BodyMAC frame structure.
used as the synchronization mechanism when the node is in sleep

mode. After the sleeping node successfully transmitted a data packet
in the allocated GTS, the gateway will send an ACK back to the
node containing timing information. The node will then adjust its
time according to the received timing parameter. This is to ensure
that the node remains synchronized when the node wakes up after
a long period of sleep. Sleep mode will be terminated and re-
synchronization process will be activated if the ACK is not received
correctly. BodyMAC solves the problem of excessive listening, which
is not preferred due to UWB hardware limitation. But the use of
CSMA-CA mechanism for UWB PHY will significantly degrade the
reliability of the system due to the low signal-to-noise ratio (SNR)
nature of an UWB signal.
A transmit-only (Tx-Only) MAC scheme for UWB WBAN is used
in reference [22]. The main goal of this scheme is to overcome
the hardware limitation of UWB technology to allow higher energy
efficiency and to tackle the problem faced in the multiuser scenario.
In this scheme, there is no synchronization between the sensor
nodes and the gateway, therefore eliminating the need for a
power-hungry UWB receiver in the sensor node. The data is
transmitted at a much higher data rate to achieve extremely low
duty cycle, which aids in reducing the energy consumption and
reduces the chances of data packet collisions. In this scheme,
each sensor is allocated a fixed transmission interval as shown in
Fig. 16.12. Through careful selection of the transmission interval
(Tint ) and transmission timeslot (Tslot ), it is possible to minimize
the collision. The probability of collision is deterministic although
the transmission start time is a random variable. The main factors
affecting the probability of collision (PC ) are Tslot , Tint , and the
number of sensors on a user. The average collision probability for
Figure 16.12. Tx-Only transmission interval.
sensor M , as shown in Eq. 16.8, is given in reference [23].

( j)
Tslot
Pc (M ) = 1 − 1 − ( j) (16.8)
j ∈ m\{M } Tint
The Tx-Only MAC scheme also deals with a multiuser scenario by
using different pulse rate for different user as shown in Fig. 16.13.
As shown in Fig. 16.13a and Fig. 16.13b, sensors on user 1 uses the
same pulse rate, while user 2 uses a different pulse rate (Fig. 16.13c).
Figure 16.13d shows the received signal seen at user 1 receiver,
when both sensor 1 from user 1 and user 2 transmit at the same
time. As illustrated in Fig. 16.13e, the receiver can easily differentiate
signals from another user, as the sampling interval between each
Figure 16.13. Different pulse rate for Tx-Only MAC. See also Color Insert.
pulse is known to the receiver. The receiver is usually the control

unit (gateway device) carried by each user in a WBAN application.
Another characteristic of this system is that different number
of pulses is allocated to the different sensors on the same user.
The assignment of the number of pulses is dependent on the
distance between the sensors and the receiver. Transmission power
management is the main reason for assigning different number
of pulses to the sensors; it is to ensure sufficient link budget is
achieved. The different number of pulses per data bit can also be
used as an identifier for the sensors.
The advantages of Tx-Only MAC are energy efficiency and circuit
simplicity and have accounted for multiuser scenario. The major
disadvantage of this MAC is that there is not feedback channel, where
signal reliability could be compromised.
16.7 UWB WBAN Applications
In this section, four different medical applications for both on-body

and implantable scenarios using the UWB wireless scheme will be
discussed.
16.7.1 Eight-Channel ECG (On-Body)

An eight-channel UWB recording system for monitoring of multiple
continuous ECG and EEG signals is designed in reference [12]
using commercial-off-the-shelves (COTS) components. The system
is depicted in Fig. 16.14. Signals from the electrodes are transferred
to the transmitter through the front-end circuit and is processed by
a microcontroller. The processed data is transmitted wirelessly UWB
signals to the receiver (i.e., base station). A noncoherent receiver is
used to demodulate the data. Figure 16.15 shows the photograph of
the complete system prototype.
Typically ECG and EEG signals have amplitude of less than
500 μV with frequency less than 100 Hz. The front-end of the UWB
transmitter uses an instrumental amplifier (INA321) [12]. The input
ECG signal is amplified by 60 dB together with a low-pass filter with
a cutoff frequency at 100 Hz.
UWB WBAN Applications 489
Figure 16.14. Multichannel UWB monitoring system block diagram. See

also Color Insert.
Figure 16.15. Photograph of complete system prototype. See also Color

Insert.
The microcontroller operates with an external 20 MHz crystal.
Each analog front-end output is connected to one of the dedicated
pins for the ADC module. The 10-bit ADC is programmed to sample
at 4.8 KHz with each channel sampling at 600 Hz. The 10-bit data
from each sample is separated into two bytes as shown in Fig. 16.16.
Low byte contains six least significant data bits (LSB) and two
Figure 16.16. Transmit data frame. See also Color Insert.
header bits. High byte contains the four most significant data bits
(MSB), three channel bits and one header bit. The MSB for the low
byte is always “1” and “0” for high byte. The three channel bits range
from “000” to “111”, each representing one of the eight channels.
High and low bytes are sent to the pulse generator through the
universal asynchronous receiver/transmitter (UART) transmit pin,
where a start bit, parity bit and stop bit is added. Output data rate
can range from 1.2 Kbps to 1.25 Mbps. The maximum data rate is
limited to 1.25 Mbps because this is the highest UART baud rate that
the microcontroller can support with a 20 MHz crystal in our current
prototype.
16.7.1.1 UWB pulse generators

Two UWB pulse generators for this monitoring system have been
incorporated onto the PCB board. The block diagram of the two
pulse generators and the generated pulses are depicted in Fig. 16.17.
A programmable oscillator (LTC6905) that produces an output
frequency from 17 MHz to 170 MHz with rise and fall time of 0.5 ns
is a common component for both transmitters.
Transmitter 1 has a rise/fall time of 0.5 ns, which is determined
by the clock generated via LTC6905. A buffer (SN74AHC1G125) with
propagation delay of 8 ns is used in the delay line. LTC6905 is
configured to generate a clock signal of 60 MHz, which corresponds
to a half duty cycle period of 8.33 ns. The clock is adjusted slightly
to account for the variation in the delay line. In the ideal case, an
UWB pulse less than 500 ps will be generated, but due to finite
rise/fall time, an UWB pulse between 1 to 2 ns is achieved with
the current setup. As shown in Fig. 16.17c, a narrow UWB pulse
Figure 16.17. UWB transmitter block diagrams: (a) gate-based narrow

pulse generation, (b) flip-flop-based narrow pulse generation, and (c)
generated UWB pulse. See also Color Insert.
is generated by performing an AND (NC7SZ08) operation of the

generated pulse train with its delayed version. Additional variation
in the actual delay (or pulse width) may come from the delay path
because of parasitic capacitances of the PCB trace or variation from
the buffer propagation delay specifications. In our proposed design,
the variation can be controlled by adjusting the clock rate. Another
advantage of using a clock signal is that it helps to make the pulse
width adjustable, which is used to optimize the spectrum shapes.
The second UWB transmitter (transmitter 2) is designed using
positive emittercoupled logic (PECL), which has a rise time of 130 ps.
Fast rise time is required to ensure more transmission power at
high frequency as derived from Eq. 16.4. The transistor–transistor

logic (TTL) signals from the microcontroller and oscillator clock
are converted to PECL logic using SY10ELT22LZG and the output is
fed a D-flipflop (MC10EP31) (Fig. 16.17b). The narrow UWB pulses
are generated by connecting the output of the D-flipflop to the
asynchronous reset pin. The propagation delay from reset to output
is 400 ps, which defines the pulse width. One pulse is generated at
every rising edge of the clock signal when the input data is high. The
pulse repetitive period (PRF) is determined by the clock. As the PRF
increases, the number of spectrum lines drops but the amplitude
of the spectrum lines increases. Changing the PRF can be used to
control the output power level to suit the different environment and
to support a wide range of output data rate.
Both transistors use a 4 GHz bandpass filter to shape the
spectrum to meet the FCC’s spectrum mask. A wideband antenna
designed for 3 to 5 GHz with an antenna gain of 2 dBi is used.
Effective isotopic radiated power transmitted is −45 dBm.
16.7.1.2 UWB receiver front-end

The received signal is passed through a bandpass filter centered at
4 GHz with 1 GHz bandwidth to remove the narrowband interfering
signals. Three LNAs, each with a gain of 14 dB is used to amplify
the signal, which gives a total gain of 42 dBm. The amplified signal
goes through a diode detector, which translates the high frequency
components to their low frequency components. The recovered low
frequency components are passed through a lowpass filter to form
an envelope of the data signal. The signal is further amplified by an
op-amp circuit by 60 dB before sending to the ADC. Figure 16.18
shows the recovered signal pattern from the transmitted UWB
modulated signal (Note that there is a time delay between the
transmitted and recovered signals due to the wireless transmission).
16.7.1.3 Data recovery

The recovered data is converted to digital signal using a 12-bit ADC
embedded in the Stratix II development board. The demodulation is
carried out with a FPGA chip running at 100 MHz. The receiver is
Figure 16.18. Recovered data from an OOK UWB modulated signal. See
also Color Insert.
capable of demodulating data of 50 Mbps. The demodulation

process is performed in two stages. In the first stage, the demodu-
lator waits for the start signals to arrive, which is a low bit. Once the
low bit is received, the demodulator samples the start bit until half
of the period before sending a signal to start the demodulation clock.
In the second stage, the demodulation clock goes high once the start
signal is received. The first bit of data is sent to the serial port, which
is the start bit. The demodulation clock runs at the same rate as the
input data and samples the data at the middle of the bit. After the
demodulation clock has generated 11 clock cycles, it goes low and
waits for the start signal from stage 1.
ECG signals are transmitted and received wirelessly using the
UWB pulses. The result is displayed using MATLAB in Fig. 16.19
on a remote computer. The signal is corrupted by the 50 Hz noise
as can been seen in the waveform obtained from the oscilloscope
before transmitting (Fig. 16.19a), after receiving and monitoring
in MATLAB in time (Fig. 16.19b) and the frequency domain
(Fig. 16.19c). The signal is passed through a 50 Hz digital notch filter
designed using a MTLAB program. The 50 Hz noise is successfully
removed and the ECG signal recovered. Removing the 50 Hz noise
at the PC instead of the receiver helps to reduce the complexity
and the programming power required at the receiver. The whole
measurement has been carried out in a lab where there were other
wireless standards (e.g., Wi-Fi) and equipment operating. The ECG
signal has successfully been monitoring without any error.
The demodulated signal is transmitted through the serial port
to the PC. A program written using Visual Basic is also developed
Figure 16.19. Monitored ECG waveforms wirelessly using UWB

transmission. See also Color Insert.
to display the received signals on the screen. Parity bit check is

performed on the received data to ensure that all data are received
correctly. Once the received data is decoded, it is formatted back into
a 10-bit word and separated based on the information embedded
in the channel bits (Fig. 16.16). Digital filtering is performed on the
received signal to remove the 50 Hz noise, which comes from the
power supply. The ECG signal in Fig. 16.20 is successfully monitored.
The graphical user interface program can display any eight channels
by changing the button “channel selection” shown in the window.
16.7.2 Implantable UWB WBAN

Medical telemetry can be categorized into two groups: high data rate
and low data rate systems. For example, the wireless endoscope and
Figure 16.20. Multichannel ECG signal.
the retina prosthesis are the recent implanted systems requiring

a large amount of data that should be delivered to outside or
into the body. Multi implantable neural recording systems and
multichannel monitoring of continuous signals such as ECG and EEG
also necessitate a high data rate communication. As an example,
scientists aim to achieve the recording of more than 100 channels
in order to simultaneously record brain functions; a data rate more
than 20 Mbps is required [14, 26]. A similar figure is also useful
for wireless endoscope implants for higher resolution pictures and
images. The MICS has channels with 300 KHz width and thus cannot
provide such data rates.
An endoscope capsule requires high data rate to increase the
resolution and image quality. For high data rate and short-range
applications of this kind, UWB communication is an ideal physical
layer solution by achieving a data rate equal or higher than
100 Mbps.
16.7.2.1 Multichannel neural recording systems

UWB is suitable for this application as the number of neuron
channels been monitored increases, the data rate will reach a speed
of 100 Mbps and higher [26]. Figure 16.21 shows the block diagram
of a multichannel neural recording system. A general multichannel
Figure 16.21. A multichannel neural recording system with UWB

transmitter [14]. See also Color Insert.
wireless neural recording system consists of preamplifiers, filters,

an analog multiplexer, a 2nd amplifier, an ADC, and a wireless
telemetry.
16.7.2.2 Electronic pills (wireless endoscope)

A small miniaturized electronic pill can reach areas such as small
intestines and deliver real-time video images wirelessly to an
external console. Figure 16.22 shows a wireless endoscope (i.e.,
electronic pill) for a medical monitoring system. The device travels
through the digestive system to collect image data and transfers
them to a nearby computer for display with a distance of 1 m or
more. A high resolution video-based capsule endoscope produces a
large amount of data, which should be delivered over a high capacity
wireless link.
Table 16.5 summarizes the commercially available electronic pill
technologies that are already been used in clinical environments.
One of the current state-of-the-art technologies for electronic pills is
the commercially available PillCam by the company “Given Imaging”
[27]. The pill uses Zarlink’s RF chip [28] for wireless transmission
based on the MICS band. The allowable channel bandwidth for this
band is only 300 kHz. It is difficult to assign enough data rate for
the high quality video data at the moment for real-time monitoring.
It is quite obvious that there is a need for higher bandwidth
data transmission for electronic pills that could facilitate a better
diagnosis.
Figure 16.22. A wireless endoscope monitoring system. See also Color

Insert.
Table 16.5. Comparison of hardware designs from vendors
Camera Image rate and

Model Company (sensor) Freq. (MHz) Data Rate resolution
PillCam (SB) Given Imaging Micron, CMOS 402–405 and 800 Kbps 14 images per,
433 (Zarlink) second or 2,600
color images
EndoCapsule Olympus Optical CCD camera, — — 2 images per
second
1920 ×1080
MiroCam
R
IntoMedic 320 × 320 pixel — — 3 images/second
SmartPill Smartpill Corp. acidity (pH), — — Only sensor
press., temp. discrete data
Wideband technology (UWB communication) is an ideal physical

layer solution that achieves a data rate equal or higher than
100 Mbps. Its current applications are mostly for in-door entertain-
ment, radar, and imaging. Due to high losses in body tissue at high
frequencies many are skeptical about using UWB for implanted and,
moreover, ingested devices. This challenge can only be addressed
by combined design of antennas, transmitters, and receivers as
described in reference [24]. Figure 16.23 shows an endoscope
application using UWB proposed in reference [24].
Figure 16.23. A wireless electronic pill system with main blocks

identified [24]. See also Color Insert.
A wideband electronic pill can transmit raw video data without

any compressing, resulting low-power, less delay in real-time and
increased picture resolution. With a high definition camera such as
2 megapixels, UWB telemetry can send up to 10 frames per second
(fps). When wideband technology is used with electronic pills, it
will provide better diagnostic systems for medical professionals to
analyze real-time video and image data wirelessly as a less invasive
method. Although it is known that tissue imposes strong attenuation
at higher frequencies, it may, however, be possible to send higher
power levels than the regulated UWB signal level [10, 24].
Small batteries are used or being considered to supply energy
to electronic pills applications, which take a considerable amount of
space. A wireless power can substantially reduce the overall implant
size and weight. For implantable systems like the cochlear implant
and the bionic eye, the inductive link has been used to transfer
power from the external unit to inside the skin to supply power
for electronics. In order to transfer enough energy, inductive links
Figure 16.24. An electronic pill technology with wireless power embed-

ded and wireless gateway for long-range data transmission. The available
electronic pill devices have a dimension of 26 mm × 11 mm. See also Color
Insert.
require the external unit to be very close to the implanted unit.
Ideally, the external unit should be placed on the body for an efficient
wireless power. A wireless power mechanism can also be used to
recharge the battery of the electronic pill if a rechargeable battery is
used.
Figure 16.24 shows the basic building blocks of an electronic
pill with wireless power mechanism. The power signal is received
and then regulated to provide power supply for the wireless data
link, the sensors, and the signal acquisition unit, which processes
and amplifies the sensors’ data. The external unit contains a
second wireless link that controls the communication between the
monitoring station and the implant devices. Overall length and
width of the commercially available electronic pill is usually 26 mm
× 11 mm.
In this system, the patient will wear the external unit, which
will be wearable on the body close to skin and, therefore, to the
swallowed electronic pill. A remote wireless receiver will be used
to receive data from the external unit (i.e., gateway) for monitoring,
Figure 16.25. An UWB telemetry link: (a) UWB pulse generation from the
power signal, (b) the UWB spectrum. See also Color Insert.
analyzing, and displaying the data. As the required specifications

are not restricted for this link, commercial wireless devices could
be used to undertake this duty. This additional wireless link will
provide patient freedom in a room in the hospital environment. With
this scenario, it is also possible for a health professional to see the
data online through the Internet when the patient is home.
An interesting UWB signal generation technique for implantable
electronics having a wireless power link is described in Fig. 16.25.
The UWB telemetry link is based on generating the narrow UWB
pulses by manipulating the received power signal. The analog power
signal can be converted to a square wave using an inverter. Then the
UWB narrow pulse can be generated in a similar fashion explained
in Section 16.3.2. In this example, the digitized power signal and
the delayed replica are passed through the XNOR gate or an AND
gate to obtain a narrow square pulse. The narrow square wave is
then passed through a BPF to fit the spectrum into the UWB mask
(Fig. 16.25b). The selected bandwidth of the BPF is 3.5–4.5 GHz,
which avoids the narrowband systems operating in the ISM bands.
This UWB telemetry link eliminates the need of a reference clock
at the implant site, which results in extremely low for such a high-
capacity telemetry link.
16.8 Design and Implementation of an

UWB-WBAN System
Figure 16.26 shows the photo of the UWB sensor node designed
for a transmitter-only UWB WBAN system described earlier. The
Design and Implementation of an UWB-WBAN System 501
Figure 16.26. UWB-WBAN sensor node: (a) block diagram of an UWB-

based WBAN sensor node, (b) photo of UWB sensor node (2.5 cm by 2.7 cm).
See also Color Insert.
sensor nodes are assembled on a four layers printed circuit board

with dimensions of 2.5 cm by 2.7 cm (Figure 16.26b), which is
sufficiently compact for use in a WBAN application. The block
diagram of the sensor node is shown in Fig. 16.26a. The narrow
pulse generator is formed using a variable voltage controlled
oscillator (VCO), delay circuitry, and a XOR gate. The details of
UWB pulse generation were explained with Figs. 16.4 and 16.5
earlier. The sensor board is able to produce narrow pulses ranging
from 0.5 ns to 2 ns, with a variable pulse repetitive frequency
between 17 MHz to 170 MHz. The generated narrow pulse passes
through a pulse-shaping filter (band-pass filter), which shapes
the pulse spectrum to conform to the UWB requirement. After
passing through the filter, the UWB pulse is amplified using a
wideband LNA to meet the −41.3 dBm transmission power level.
The microcontroller performs the analog-to-digital conversion,
determines the transmission format, modulation scheme, and sets
the data rate. The UWB pulse rate is independent of the data rate,
therefore enabling the system to vary the number of UWB pulses per
data bit. The logic gates combine the data with the generated UWB
pulse before transmitting the information using a wideband UWB
antenna. This sensor node is designed to operate in the 3.1 to 5 GHz
frequency band, using direct modulation technique (i.e., no carrier is
used).
16.8.1 UWB Receiver Circuitry

The UWB sensor node communicates with an UWB receiver. The
receiver circuitry is assembled using off-the-shelves modules. The
receiver antenna is placed on body, and the receiver circuitry is
placed off body. The block diagram of the receiver circuit is shown
in Fig. 16.27. The signal entering the receiver passes through a 3
to 5 GHz band-pass filter to eliminate the unwanted out-of-band
signals. The filtered signal is amplified by 48 dBm using three
wideband LNAs before down converting to baseband signal using
a mixer and a 4 GHz VCO. The baseband signal passes through a
low-pass filter with 100 MHz bandwidth before going through the
baseband amplification stage.
Figure 16.27. UWB receiver block diagram. See also Color Insert.
The recovered UWB pulse is digitalized using a high-speed ADC

and processed and demodulated by the FPGA before transferring the
data to the laptop using a serial cable. The role of the FPGA is to
process the received multiple UWB pulses and determine whether
it is bit “1” or bit “0” before sending to the laptop and set the
appropriate baud rate for data transfer to the laptop. Preprocessing
is necessary as the data transfer rate via the serial cable is much
lower than the data rate the sensor nodes is sending to the UWB
receiver using the gating method. After FPGA processing, the data
is transferred to the laptop based on the baseband data rate. The
sensor node and the receiver in Fig. 16.27 are used to maintain the
best position of sensors on the body when UWB communication is
used. The BER performance of the WBAN system is computed in the
laptop using MATLAB program based on a known pseudorandom
data sequence that the transmitter is sending [29].
16.8.2 Experimental Setup and Measurement Result

Table 16.6 shows the parameters used for the experimental setup.
Four locations commonly used for physiological signal monitoring
are chosen to evaluate the performance of UWB WBAN at different
receiver positions. Figure 16.28 shows the sensors and receiver
position evaluated in this measurement. Sensor 1 is at the ECG
position, sensor 2 at the EEG position, sensor 3 on the wrist, and
sensor 4 on the waist. All the readings are taken in a controlled
environment where the subject under test remains static throughout
each measurement. For each sensor and receiver pairs, five sets of
readings are taken. Each set of readings consists of 10,000 data bits.
Table 16.6. Experimental parameter
Parameter Values
UWB pulse rate 25 MHz
UWB pulse width 2 ns
UWB frequency band 3.9–4.5 GHz
Data rate 1 Mbps
Modulation OOK
Maximum transmissin Peak power −24.4dBm
Figure 16.28. Sensor and receiver locations. See also Color Insert.
All the readings are taken using antenna orientation that gives the
strongest signal strength.
Figures 29–32 show the BER plots obtained from four sensor
positions across the body. Most of the commonly monitored
physiological signals are located on the front upper half of the body.
Therefore, we have selected four commonly used sensor locations
that are well spread throughout the upper body to perform this
Figure 16.29. BER of ECG sensor for different receiver positions [29]. See
also Color Insert.
Figure 16.30. BER of EEG sensor for different receiver positions [29]. See
also Color Insert.
Figure 16.31. BER of sensor on wrist for different receiver positions [29].
evaluation. All the sensor nodes and receiver are placed 10 mm
away from the body in order to reduce the effect of the body on the
antenna.
These experiment results show that by carefully selecting the
receiver position, it is possible to reduce the amount of transmission
power to up to 20 dB. There are numerous configurations for
the sensor nodes in a WBAN system; therefore, there is not one
receiver position that is optimized for all configurations. However,
the receiver position at the center of the chest is the preferred choice
for most of the configuration especially when ECG monitoring is
used.
Figure 16.32. BER of sensor on waist for different receiver positions [29].
The optimum receiver position can be selected based on the

minimum total transmit power required to attain a reasonable BER
performance for all the sensor nodes in a particular configuration.
As an example for EEG sensor, the best receiver position is on the
chest, and the worst receiver position is on the waist. For a BER of
10−3 , there are only 8 dB differences between the best receiver and
the worst receiver positions [29].
16.8.3 Summary
In this section, we have looked at four different medical applications
using UWB wireless schemes. The feasibility of UWB in medical
WBAN applications can be clearly seen in this section. It is
possible to design an UWB transmitter using COTS parts or CMOS
implementation. Actual hardware measurement result shows that
UWB is a promising option for both on-body and implantable
applications.
One big advantage of UWB wireless technology is that its high
data rate ranges from 850 Kbps to 20 Mbps, which can be used for
simultaneous monitoring of many continuous physiological signals
such as ECC, EEG, and EMG. In addition, UWB wireless technology
also does not present an EMI (electromagnetic interference) risk to
other narrowband systems and medical equipment in health care
as its transmitter power is quite low and the frequencies used are
References 507
at very high frequencies (>3.5 GHz). The biggest drawback for the
use of the UWB technology is that the UWB chips are not available
commercially to apply in a telemedicine system at the moment.
16.9 Conclusion
The UWB WBAN system is indeed a multilayer problem. In order to

design an efficient UWB WBAN system, the designer has to consider
the hardware level, PHY layer, and MAC protocol. UWB is a promising
wireless technology that could give WBAN a breakthrough, but more
research and testing on the hardware and system level is required
before it can be successfully implemented. An interesting application
will be to integrate UWB sensor nodes together with narrowband
wireless technologies form a WBAN application for health care
systems.
UWB-based WBAN sensor nodes can be applied to both non-
invasive and invasive monitoring systems in medical applications
such as multichannel EEG, tracking patients, and neural recording,
also in gaming and entertainment industry where a high data rate
transmission is required without affecting the operation of other
wireless systems in the environment.
References
1. http://www.ieee802.org/15/pub/TG6.html, 2010.
2. FCC 02–48 (First Report and Order), 2002. http://transition.
fcc.gov/Bureaus/Engineering Technology/Orders/2002/fcc02048.pdf.
3. http://standards.ieee.org/getieee802/download/802.15.4a-2007.pdf.
4. L. Huan-Bang and R. Kohno (2007) Introduction of SG-BAN in IEEE
802.15 With Related Discussion, in IEEE International Conference on
Ultra-Wide Band 2007, pp. 134–139.
5. M. Klemm and G. Troester (November 2006) Textile UWB Antennas
for Wireless Body Area Networks, IEEE Transactions on Antennas and
Propagation, 54(11), pp. 3192–3197.
6. B. Sanz-Izquierdo, J. C. Batchelor, and M. I. Sobhy (2007) Compact
UWB Wearable Antenna, in Loughborough Antennas and Propagation
Conference 2007 (LAPC 2007), pp. 121-124.
7. L. Jeong Soo, C. Nguyen, and T. Scullion (2001) New Uniplanar

Subnanosecond Monocycle Pulse Generator and Transformer for Time-
Domain Microwave Applications, IEEE Transactions on Microwave
Theory and Techniques, 49, pp. 1126–1129.
8. H. Jeongwoo and N. Cam (2002) A New Ultra-Wideband, Ultra-Short
Monocycle Pulse Generator With Reduced Ringing, IEEE Microwave and
Wireless Components Letters, 12, pp. 206–208.
9. D. Simic, A. Jordan, R. Tao, N. Gungl, J. Simic, M. Lang, L. V. Ngo,
V. Brankovic (1–5 July 2007) Impulse UWB Radio System Architecture
for Body Area Networks, in 16th IST Mobile and Wireless Communica-
tions Summit 2007, pp. 1–5.
10. M. R. Yuce, H. C. Keong, and M. S. Chae (October 2009) Wideband
Communication for Implantable and Wearable Systems, IEEE Trans-
actions on Microwave Theory and Techniques, 57(10), pp. 2597–
2604.
11. J. Ryckaert, C. Desset, A. Fort, M. Badaroglu, V. de Heyn, P. Wambacq,
G. van der Plas, S. Donnay, B. van Poucke, and B. Gyselinckx (December
2005) Ultra-Wide Band Transmitter for Low-Power Wireless Body Area
Networks: Design and Evaluation, IEEE Transactions on Circuits and
Systems I, 52(12), pp. 2515–2525 .
12. H. C. Keong and M. R. Yuce (2008) Low Data Rate Ultra Wideband ECG
Monitoring System, in 30th IEEE-EMBS Annual International Conference
2008.
13. K. Hyunseok, J. Youngjoong, and J. Sungyong (2005) Digitally Con-
trollable Bi-Phase CMOS UWB Pulse Generator, in IEEE International
Conference on Ultra-Wideband, pp. 442–445.
14. M. R. Yuce, W. Liu, M. S. Chae, and J. S. Kim (24–26 September
2007) A Wideband Telemetry Unit for Multi-Channel Neural Recording
Systems, in IEEE International Conference on Ultra-Wideband, pp. 612–
617.
15. R. Kohno, K. Hamaguchi, H. -B. Li, and K. Takizawa (10–12 September
2008) R&D and Standardization of Body Area Network (BAN) for
Medical Healthcare, in IEEE International Conference on Ultra-Wideband
2008, 3, pp. 5–8.
16. C. Lee, J. Kim, H. S. Lee, and J. Kim (28–30 September 2009)
Physical Layer Designs for WBAN Systems in IEEE 802.15.6 Proposals,
in 9th International Symposium on Communications and Information
Technology 2009, pp. 841–844.
Conclusion 509
17. H. Viittala, M. Hamalainen, J. Iinatti, and A. Taparugssanagorn (24–27

November 2009) Different Experimental WBAN Channel Models and
IEEE802.15.6 Models: Comparison and Effects, in 2nd International
Symposium on Applied Sciences in Biomedical and Communication
Technologies 2009, pp. 1–5.
18. A. Khaleghi and I. Balasingham (26–29 April 2009) On the Ultra
Wideband Propagation Channel Characterizations of the Biomedical
Implants, in IEEE Vehicular Technology Conference 2009, pp. 1–4.
19. C. Li, L. Wang, J. Li, B. Zhen, H. –B. Li, and R. Kohno (13–16 September
2009) Scalable and Robust Medium Access Control Protocol in Wireless
Body Area Networks, in 20th IEEE International Symposium on Personal,
Indoor and Mobile Radio Communications, pp. 2127–2131.
20. Y. Zhang and G. Dolmans (23–29 August 2009) A New Priority-
Guaranteed MAC Protocol for Emerging Body Area Networks, in 5th
International Conference on Wireless and Mobile Communications 2009,
pp. 140–145.
21. G. Fang and E. Dutkiewicz (28–30 September 2009) BodyMAC: Energy
Efficient TDMA-based MAC Protocol for Wireless Body Area Networks,
in9th International Symposium on Communications and Information
Technology 2009, pp. 1455–1459.
22. K. H. Chee and M. R. Yuce (2009) Analysis of a Multi-Access Scheme and
Asynchronous Transmit-Only UWB for Wireless Body Area Networks,
in IEEE Engineering in Medicine and Biology Conference 2009, pp. 6906–
6909.
23. M. Weisenhorn and W. Hirt (2005) Uncoordinated Rate-Division
Multiple-Access Scheme for Pulsed UWB Signals, IEEE Transactions on
Vehicular Technology, 54, pp. 1646–1662.
24. M. R. Yuce, T. Dissanayake, and H. C. Keong, (October 2009) Wireless
Telemetry for Electronic Pill Technology, in IEEE Conference SENSORS.
25. M. R. Yuce, T. N. Dissanayake, and H. C. Keong (2009) Wideband
Technology for Medical Detection and Monitoring, in Recent Advances in
Biomedical Engineering, (ed. G. R. Naik), ISBN: 978-953-307-004-9, IN-
TECH, available from http://sciyo.com/articles/show/title/wideband-
technology-for-medical-detection-and-monitoring.
26. M. Chae, Z. Yang, M. R. Yuce, L. Hoang, and W. Liu (August 2009) A
128-Channel 6 mW Wireless Neural Recording IC With Spike Feature
Extraction and UWB Transmitter, IEEE Transactions on Neural Systems
and Rehabilitation Engineering, 17, pp. 312–321.
27. http://www.givenimaging.com/2010.
28. P. Bradley (2006) An Ultra Low Power, High Performance Medical
Implant Communication System (MICS) Transceiver for Implantable
Devices, in IEEE Biomedical Circuits and Systems Conference, 16,
pp. 158.
29. H. C Keong, M. R. Yuce, T See, and T. M. Chiam (2010) On-Body Evaluation
of UWB Receiver Position for Wireless Body Area Networks, in IEEE
International Conference on Ultra-Wideband 2010.
October 19, 2011 18:0 PSP Book - 9in x 6in Mehmet-color-Insert
Color Insert
Figure 1.2. Wireless standards.
Figure 1.3. Wireless body area network implementation scenarios:

(a) single WBAN node, (b) a WBAN node with multiple bodies, and
(c) multiple WBAN nodes.
C2 Color Insert
Figure 2.1. Patient PDA with ECG leads and SpO2 sensor.
Figure 2.4. SMART central main GUI.

Color Insert C3
Figure 2.7. Diagnosis estimation and ECG-computed HR of an average

patient.
Figure 2.8. ECG – SpO2 HR comparison and ECG pacemaker activity when
patient HR dropped below 60 BPM.
C4 Color Insert
Figure 3.1. Architecture of PCC system.
Figure 3.2. Wireless ECG sensor.
Figure 3.3. Remote surveillance server.

Color Insert C5
Figure 3.8. QRS complex waveform after different impact factor in the
adaptive filters.
Figure 3.11. Positive states of a QRS complex in D (t).
Figure 3.12. Illumination of geometric analysis method.

C6 Color Insert
Figure 3.13. Application example of PCC system.
Figure 4.1. (a) A sensor node composed of processing unit and custom-
designed sensor board. The motion sensor board has a triaxial accelerome-
ter and a biaxial gyroscope Sensor node and subject (b) A subject wearing
nine sensor nodes.
Figure 4.4. HMEM model and structure.

Color Insert C7
S = 10, and P = (a) 25%, (b) 15%, (c) 5%.
C8 Color Insert
S = 20, and P = (a) 25%, (b) 15%, (c) 5%.
Color Insert C9
Figure 6.2. An example of sensor node hardware: (a) single wireless

sensor node, (b) multi-channel wireless sensor node.
Figure 6.5. ECG with 60 Hz (or 50 Hz) noise (a) and without noise (b).
C10 Color Insert
Figure 6.6. A Mica2DOT board (taken from www.xbow.com).
Figure 6.7. A T-node sensor node [16].
Figure 6.8. An implanted sensor node with the external unit and wireless
data links.
Color Insert C11
Figure 6.9. Figure PiiX sensor node from Corventis [38] and
micropaq
R
Wearable Monitor from WelchAllyn [37].
Figure 6.11. A 4-channel sensor node.
Figure 6.12. Block diagram of BCUs: (a) wired BCU, (b) portable BCU.
C12 Color Insert
Figure 6.13. A wired BCU (BCU-1).
Figure 6.14. Intermediate body control unit (BCU-2). This device is

shared by more than one patient and is portable. It contains dual wireless
transceiver to support two directional wireless links.
Figure 6.17. Software programs for the proposed WBAN application.

Color Insert C13
Figure 6.20. An example of ECG Monitoring.
Figure 7.1. The Berkeley Tricorder

C14 Color Insert
Figure 7.2. Power spectrum of ECG signal. The peak at 120 Hz is a

harmonic of the 60 Hz power-line interference.
Figure 7.6. ECG waveform.

Color Insert C15
Figure 7.7. Einthoven ECG lead locations (from wikiCommons).
Figure 7.15. Sample data from the Berkeley Tricorder.

C16 Color Insert
Figure 7.16. ECG signal (lower plot) captured while subject in motion
(upper plot).
Figure 8.1. Change of total expenditure on health (percentage of gross

domestic product) over years.
Color Insert C17
Figure 8.3. Frequency and amplitude characteristics of biopotential

signals — most commonly monitored signals in medical practice [3].
Figure 8.4. Different measurement and biasing methods for biopotential

signal acquisition.
C18 Color Insert
Figure 8.5. Measurement of biopotential signals (Type I measurement)

introducing the nonidealities from different sources.
Figure 8.11. The complete architecture of a compensated instrumenta-

tion amplifier using a DC servo loop for implementing high-pass filtering
characteristics [10 ,11].
Color Insert C19
Figure 8.13. Comparison of different instrumentation amplifiers in terms

of their power efficiency. Y-axis indicates the current consumption of the
instrumentation amplifier and x-axis indicates the noise density of the
instrumentation amplifier. Dashed lines indicate the constant contours of
noise efficiency factor (NEF) lines [8]. The diameter of a circle indicates the
total integrated noise of the instrumentation amplifier.
Figure 8.16. The implementation of electrode-tissue impedance mea-

surement circuit operating simultaneously with the biopotential measure-
ment circuit [28]. Instead of sinusoidal current sources and analog modula-
tors/demodulators, chopper-modulation technique has been employed.
C20 Color Insert
Figure 9.4. A wireless polling network showing the polling sequence.
Figure 9.5. IEEE 802.15.4 MAC super frame structure.
Figure 9.9. WBAN packet loss and packet generation rate for different
data aggregation rate/payload size.
Color Insert C21
Figure 10.1. Toumaz Sensium TM digital Plaster.
Figure 10.3. RSSI vs. time for various patient scenarios.

C22 Color Insert
Figure 10.4. Optimal transmit power and associated RSSI for a normal
walk, slow walk, and resting position.
Color Insert C23
Figure 10.8. RSSI vs. time using the Sensium TM platform.
Figure 11.1. Overview of on-body, off-body, and body-to-body

communications.
C24 Color Insert

estimated theoretical plots while the user was mobile in the anechoic
chamber.

estimated theoretical plots while the user was mobile in the reverberation
chamber.
Color Insert C25
Figure 12.6. The loaded antenna inside a tissue simulating cylinder in a

CST simulation.
Figure 13.2. Frequency characteristics of commercially available micro

ovens for home use [6].
C26 Color Insert
Figure 14.1. Wavelength in air and typical muscle vs frequency. Note: Log
scales.
Figure 14.2. Silver on alumina patch antenna mounted on test implant

case.
Figure 14.5. Matching schematic from Fig. 14.4.

Color Insert C27
Figure 14.6. SAW filter input and (differential) output; 400 to 420 MHz;
Ansoft SV software.
Figure 14.12. Impedance and series resistance for 10 pF 0402 capacitor

(courtesy AVX).
C28 Color Insert
Figure 15.8. Measured results. (a) and (b) Data telemetry. Trace 1:
the measured voltage across the transmitter coil; Trace 2: the measured
voltage across the receiver coil; Trace 3: the measured data receiver output.
(c) Power telemetry. Trace 1: the measured voltages across the power
transmitter coil; Trace 2: the measured voltage across the power receiver
coil (input to rectifier); Trace 3: rectifier output for powering secondary
electronics; (d) output of two stimulator drivers that are powered and
controlled by the power and data telemetry [33].
Color Insert C29
Figure 16.1. UWB and narrowband spectrum.
XOR
Delay
AND
Figure 16.4. Narrow UWB pulse generation.
(a) Nand gate implementation
(b) Nor gate implementation
Figure 16.5. Triangle pulse generation.

C30 Color Insert
Figure 16.6. Spectrum of an UWB rectangular pulse in frequency domain

and nulls in the UWB spectrum due to rise time.
Figure 16.7. Effect of null in a band-limited signal.

Color Insert C31
Figure 16.13. Different pulse rate for Tx-Only MAC.
Figure 16.14. Multichannel UWB monitoring system block diagram.

C32 Color Insert
Figure 16.15. Photograph of complete system prototype.
Figure 16.16. Transmit data frame.

Color Insert C33
Figure 16.17. UWB transmitter block diagrams: (a) gate-based narrow

pulse generation, (b) flip-flop-based narrow pulse generation, and (c)
generated UWB pulse.
Figure 16.18. Recovered data from an OOK UWB modulated signal.

C34 Color Insert
Figure 16.19. Monitored ECG waveforms wirelessly using UWB

transmission.
Figure 16.21. A multichannel neural recording system with UWB

transmitter [14].
Color Insert C35
Figure 16.22. A wireless endoscope monitoring system.
Figure 16.23. A wireless electronic pill system with main blocks

identified [24].
C36 Color Insert
Figure 16.24. An electronic pill technology with wireless power embed-

ded and wireless gateway for long-range data transmission. The available
electronic pill devices have a dimension of 26 mm × 11 mm.
Figure 16.25. An UWB telemetry link: (a) UWB pulse generation from the
power signal, (b) the UWB spectrum.
Color Insert C37
Figure 16.26. UWB-WBAN sensor node: (a) block diagram of an UWB-

based WBAN sensor node, (b) photo of UWB sensor node (2.5 cm by 2.7 cm).
Figure 16.27. UWB receiver block diagram.

C38 Color Insert
Figure 16.28. Sensor and receiver locations.
Figure 16.29. BER of ECG sensor for different receiver positions [29].
Color Insert C39
Figure 16.30. BER of EEG sensor for different receiver positions [29].
Figure 16.31. BER of sensor on wrist for different receiver positions [29].
Figure 16.32. BER of sensor on waist for different receiver positions [29].

Wireless Body Area Networks Technology, Implementation, and Applications PDF

Uploaded by

Copyright:

Available Formats

Wireless Body Area Networks Technology, Implementation, and Applications PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Wireless Body Area Networks Technology, Implementation, and Applications PDF

Uploaded by

Copyright:

Available Formats

What are some of the applications discussed for wireless body area networks?

What are some of the applications discussed for wireless body area networks?

What are some of the design requirements discussed for wireless body area networks?

What are some of the design requirements discussed for wireless body area networks?

Wireless.

indd 1 10/19/11 12:53:03 AM

No claim to original U.S. Government works

International Standard Book Number-13: 978-9-81424-157-1 (eBook - PDF)

1 Introduction to Wireless Body Area Network 1

2 Wireless Patient Monitoring in a Clinical Setting 19

3 Real-Time Cardiac Arrhythmias Monitoring for

3.2 History of PHC Research 44

4 Human Bio-Kinematic Monitoring with Body

4.5.2.1 Preprocessing and feature extraction 90

5 Signal Processing In-Node Frameworks for Wireless

6 Hardware Development and Systems for Wireless

6.2 Wireless Body Sensors 138

7.4 Wireless Interface 211

9.4 Scheduled Protocols 268

10 Power Management in Body Area Networks for

11 Channel Modeling of Narrowband Body-Centric

11.2 Channel Modeling for Wireless Body Area

12 Antenna Design and Propagation for WBAN

12.4 Volume Conduction Antennas 363

13 Coexistence Issues with Wireless Body

13.5.3 Frequency Spreading and Code Division

14 Implanted Wireless Communication Making a

14.11 Base Station 431

15 Wireless Power and Data Telemetry for Wearable

16 Ultra Wideband for Wireless Body Area Networks 467

16.7 UWB WBAN Applications 488

Sensor-based measurements and monitoring techniques have been

introduced, which can be used to design basic low-cost wireless

Mehmet Rasit Yucea and Jamil Y. Khanb

Monash University, Clayton, VIC 3800, Australia

The University of Newcastle, Callaghan, NSW 2308, Australia

Recent advances in wireless technologies and ICT (information and

Wireless Body Area Networks: Technology, Implementation, and Applications

2 Introduction to Wireless Body Area Network

integrated with sensors, it is possible to acquire and monitor

Figure 1.1. A wireless body area network scheme.

introduce numerous possibilities to improve health care and sports

4 Introduction to Wireless Body Area Network

With the widespread use of portable devices such as mobile

it is expected that the remote patient monitoring technology will be

Various medical equipment vendors have started to produce

6 Introduction to Wireless Body Area Network

Aged care is one of the prime areas of WBAN applications. Aging

1.3 Wireless Personal Area Network (WPAN)/Wireless

Wireless Personal Area Network (WPAN)/Wireless Local Area Network (WLAN) 7

Table 1.1. Wireless technologies for short-range data communications

Frequency DataRate Range Standard Transmission

Wi-Fi 2.4 GHz , 5 GHz Up to 400 Mbps 300 m IEEE802.11(b/g/n) 250–1000 mW

to exchange data. Using these technologies, it is possible to

8 Introduction to Wireless Body Area Network

can select the most appropriate link based on availability, quality, or

1.4 Wireless Body Area Network

A body area network is a collection of wireless sensors (i.e., sensor

Figure 1.2. Wireless standards. See also Color Insert.