Open Collections

A Novel Wireless Three-pad ECG System for Generating Conventional 12-lead Signals by Huasong Cao  B.Eng., Wuhan University, 2007  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF APPLIED SCIENCE in The Faculty of Graduate Studies (Electrical and Computer Engineering)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) April 2010 c Huasong Cao 2010  Abstract A wireless body area network (WBAN) is a radio-frequency (RF) based wireless networking technology that interconnects tiny nodes in, on or around a human body. Typically, the transmissions of these nodes cover a short range of about 2 meters. This thesis presents a complete survey on recent advances in WBAN, including the market needs, channel modelling, standardization of low-layer communication protocols, quality-of-service (QoS) provisions, developments of sensors/actuators, WBAN architectures and experimental platforms. A recent work employing the nonbeacon-enabled mode of the IEEE 802.15.4 standard for QoS provisions has motivated us to design a QoS framework based on the beacon-enabled mode of the same standard. The proposed QoS framework can better differentiate WBAN application traffic streams and serve periodic traffic more directly through the time-division-multiple-access (TDMA) based mechanism. A dominant feature of the proposed framework is the minimum adaptation to the existing standard, which makes it easy to adopt our platform and associated algorithms, as well as to implement them on off-the-shelf hardware platforms. Employing the proposed QoS framework, we propose a novel wireless three-pad electrocardiography (W3ECG) system. W3ECG furthers the pad design idea of single-pad wireless ECG systems. Inspired by the transformation possibility of signals obtained in vectorcardiographic (VCG) systems, we bring two more pads to the single-pad approach to gain spatial variety of the heart activity. Signals obtained from these three pads, plus the spatial information, make it possible to synthesize conventional 12-lead ECG signals. We have been able to manufacture the front-end ECG circuit, and combine it with an IEEE 802.15.4 hardware platform TelosB. Software for the server and pad has also been developed to make a fully running W3ECG possible. By explaining and evaluating our QoS platform designed for general WBAN applications, and our W3ECG system invented for particular healthcare area, we foresee a bright future for wide deployments of such kind of wireless networks on the human body.  ii  Table of Contents Abstract  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  ii  Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  iii  List of Tables  vi  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  ix  Dedication  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  x  1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  1  1.1  1.2  1.3  1.4  1.5  1.6  Wireless Networks Reaching Body Area . . . . . . . . . . . . . . . . . . . . . . . . .  1  1.1.1  WBANs versus Wider Area Wireless Networks . . . . . . . . . . . . . . . . .  1  1.1.2  WBANs versus WBASNs . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  2  1.1.3  WBANs versus WSNs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  3  Huge Needs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  3  1.2.1  Healthcare . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  4  1.2.2  Sports and Fitness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  4  1.2.3  Secure Authentication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  4  Technology Advances  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  5  1.3.1  Body Area Channel Models  . . . . . . . . . . . . . . . . . . . . . . . . . . .  5  1.3.2  Low-power Radio Technologies . . . . . . . . . . . . . . . . . . . . . . . . . .  5  1.3.3  QoS Provisions in WBAN  8  1.3.4  Sensors and Actuators  . . . . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11  Wired and Wireless ECG Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 1.4.1  Conventional 12-lead ECG Systems  1.4.2  Body Surface Mapping Systems  1.4.3  Vectorcardiographic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 14  1.4.4  Wireless Single-pad ECG Systems . . . . . . . . . . . . . . . . . . . . . . . . 15  Interconnections of WBANs  . . . . . . . . . . . . . . . . . . . . . . . 14  . . . . . . . . . . . . . . . . . . . . . . . . . 14  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15  1.5.1  One-tier Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15  1.5.2  Two-tier Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16  Existing WBAN Platforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 iii  Table of Contents  1.7  1.6.1  MITHril  1.6.2  CodeBlue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18  1.6.3  AID-N  1.6.4  WHMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18  1.6.5  MIMOSA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18  1.6.6  WiMoCA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19  Contributions  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19  1.7.1  A Survey and Outlook  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19  1.7.2  A QoS Provisioning Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . 19  1.7.3  A Novel Wireless ECG System . . . . . . . . . . . . . . . . . . . . . . . . . . 20  2 Employing IEEE 802.15.4 for QoS Provisioning in WBAN  . . . . . . . . . . . . 21  2.1  Topology Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21  2.2  Service Differentiation Techniques  2.3  2.4  . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22  2.2.1  QoS Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22  2.2.2  Traffic Prioritization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23  2.2.3  Traffic Accommodation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24  Admissions Control  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25  2.3.1  Association and Admission . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25  2.3.2  The Admission Controller  2.3.3  Two-step Decision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26  Traffic Scheduling  . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28  2.4.1  The Superframe Scheduler  . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28  2.4.2  The CFP Scheduler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30  3 Proposed Wireless Three-pad ECG System  . . . . . . . . . . . . . . . . . . . . . . 33  3.1  Electrode, Lead and Pad  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33  3.2  Heart Dipole Hypothesis  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34  3.3  3.4  3.2.1  Heart-vector Projection Theory  . . . . . . . . . . . . . . . . . . . . . . . . . 34  3.2.2  Analyzing Input Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36  3.2.3  Derivation of Transformation Equations . . . . . . . . . . . . . . . . . . . . . 36  Wireless Three-pad ECG System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 3.3.1  Architecture Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39  3.3.2  Pad Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39  3.3.3  Pad Placement Positions  Software Implementation  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43  3.4.1  Sampling and Radio Communications on The Pad . . . . . . . . . . . . . . . 43  3.4.2  Beacon-frame Generation and Data Forwarding on The Sink . . . . . . . . . 44  3.4.3  Graphical User Interface and MySQL Database on The Server . . . . . . . . 44  iv  Table of Contents 4 Performance Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 4.1  4.2  4.3  Performance Evaluations of ZigBee for WBASN  . . . . . . . . . . . . . . . . . . . . 51  4.1.1  Simulation Setup  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51  4.1.2  Average Reception Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51  4.1.3  Throughput  4.1.4  Latency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54  4.1.5  Fairness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56  4.1.6  Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52  Performance Evaluations of Proposed QoS Provisioning Platform  . . . . . . . . . . 59  4.2.1  Simulation Setup  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59  4.2.2  Number of Admitted Traffic Flows . . . . . . . . . . . . . . . . . . . . . . . . 60  4.2.3  Throughput  4.2.4  Energy Consumption  4.2.5  Constraint Compliance Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . 62  4.2.6  Latency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61  Performance Evaluations of The W3ECG System  . . . . . . . . . . . . . . . . . . . 63  4.3.1  Experimental Study Using Commercial Patient Monitor . . . . . . . . . . . . 64  4.3.2  Computer Simulations Based on Practical EASI Data Set . . . . . . . . . . . 66  4.3.3  Deployments of W3ECG and Experimental Studies  4.3.4  Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81  5 Conclusion  . . . . . . . . . . . . . . 70  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82  Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84  Appendices A Statement of Co-Authorship . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88  v  List of Tables 1.1  WBAN and WPAN technologies and comparisons. . . . . . . . . . . . . . . . . . . .  8  2.1  Traffic priorities and 2-bit representations. . . . . . . . . . . . . . . . . . . . . . . . . 23  2.2  Capability Information field of Association Request command. . . . . . . . . . . . . 24  2.3  IEEE 802.15.4 MAC command frames. . . . . . . . . . . . . . . . . . . . . . . . . . . 27  2.4  QoS notification command format. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27  2.5  Valid values of the Association Status field. . . . . . . . . . . . . . . . . . . . . . . . 28  3.1  Suggested W3ECG pad placement locations. . . . . . . . . . . . . . . . . . . . . . . 43  4.1  ZigBee simulation parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52  4.2  ZigBee application traffic parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . 52  4.3  QoS application traffic parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59  4.4  Cross-correlation coefficients between each of the three signals obtained from W3ECG. 72  4.5  Placement of electrodes for standard 12 leads. . . . . . . . . . . . . . . . . . . . . . . 73  vi  List of Figures 1.1  WBAN in the family of wireless networks. . . . . . . . . . . . . . . . . . . . . . . . .  2  1.2  Superframe structure of IEEE 802.15.4 beacon-enabled mode. . . . . . . . . . . . . .  9  1.3  Dalhousie’s two dimensional torso model with 352 nodes [18]. . . . . . . . . . . . . . 13  1.4  One-tier system architecture for interconnecting WBANs. . . . . . . . . . . . . . . . 16  1.5  Two-tier system architecture for interconnecting WBANs. . . . . . . . . . . . . . . . 17  2.1  Example of using star topology to directly link WBAN sensors in an IEEE 802.15.4 network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22  2.2  A successful admission and scheduling process in the proposed QoS platform. . . . . 32  3.1  Illustration of the heart vector and its projection on a lead vector. . . . . . . . . . . 35  3.2  Illustration of the heart vector and three lead vectors. . . . . . . . . . . . . . . . . . 37  3.3  Schematic of ECG front-end design. . . . . . . . . . . . . . . . . . . . . . . . . . . . 46  3.4  PCB layout of ECG front-end circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . 47  3.5  Snapshot of the manufactured ECG front-end circuits. . . . . . . . . . . . . . . . . . 47  3.6  ECG front-end circuit interfaced with TelosB platform. . . . . . . . . . . . . . . . . . 48  3.7  Placement positions of electrodes for RA, LA and LL, and three sets of electrodes for W3ECG system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49  3.8  A snapshot of the W3ECG server GUI demonstrating that three pads are synchronized and transmitting ECG signals. . . . . . . . . . . . . . . . . . . . . . . . . . . . 50  4.1  Average reception ratio for WBASNs with ExG sensors. . . . . . . . . . . . . . . . . 53  4.3  Throughput for WBASNs with ExG sensors. . . . . . . . . . . . . . . . . . . . . . . 53  4.2  Average reception ratio for WBASNs without ExG sensorss. . . . . . . . . . . . . . . 54  4.4  Throughput for WBASNs without ExG sensors. . . . . . . . . . . . . . . . . . . . . . 54  4.5  Delay for WBASNs with ExG sensors. . . . . . . . . . . . . . . . . . . . . . . . . . . 55  4.6  Delay for WBASNs without ExG sensors. . . . . . . . . . . . . . . . . . . . . . . . . 55  4.7  Average reception ratio for sensors of WBASNs positioned two hops away. . . . . . . 56  4.8  Average reception ratio for sensors of WBASNs positioned different numbers of hops away. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57  4.9  Throughput for sensors of WBASNs positioned two hops away. . . . . . . . . . . . . 57  4.10 Throughput for sensors of WBASNs positioned different numbers of hops away. . . . 58 4.11 Number of admitted traffic flows in QoS framework simulation. . . . . . . . . . . . . 60 4.12 Throughput in QoS framework simulation. . . . . . . . . . . . . . . . . . . . . . . . . 61  vii  List of Figures 4.13 Energy consumption in QoS framework simulation. . . . . . . . . . . . . . . . . . . . 62 4.14 Constraint compliance ratio in QoS framework simulation. . . . . . . . . . . . . . . . 63 4.15 Latency in QoS framework simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . 64 4.16 ECG signals of Lead I, II and III from patient monitor. . . . . . . . . . . . . . . . . 65 4.17 ECG signals of Pad 1, Pad 2 and Pad 3 locations from patient monitor. . . . . . . . 65 4.18 Comparison of cross-correlation coefficients between W3ECG system and EASI system for Patient 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 4.19 Comparison of cross-correlation coefficients between W3ECG system and EASI system for Patient 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 4.20 Comparison of cross-correlation coefficients between W3ECG system and EASI system for Patient 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 4.21 Comparison of cross-correlation coefficients between W3ECG system and EASI system for Patient 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 4.22 Comparison between synthesized 12-lead ECG signals and directly obtained versions. 70 4.23 Time domain of ECG graph corrupted by 50/60 Hz noise. . . . . . . . . . . . . . . . 71 4.24 Frequency domain of ECG graph corrupted by 50/60 Hz noise. . . . . . . . . . . . . 72 4.25 Coefficients for cross-correlation between observed and synthesized 12-lead signals. . 74 4.26 Comparison between observed and synthesized Lead I signals. . . . . . . . . . . . . . 75 4.27 Comparison between observed and synthesized Lead II signals. . . . . . . . . . . . . 75 4.28 Comparison between observed and synthesized Lead III signals. . . . . . . . . . . . . 76 4.29 Comparison between observed and synthesized Lead aVR signals. . . . . . . . . . . . 76 4.30 Comparison between observed and synthesized Lead aVL signals. . . . . . . . . . . . 77 4.31 Comparison between observed and synthesized Lead aVF signals. . . . . . . . . . . . 77 4.32 Comparison between observed and synthesized Lead V1 signals. . . . . . . . . . . . . 78 4.33 Comparison between observed and synthesized Lead V2 signals. . . . . . . . . . . . . 78 4.34 Comparison between observed and synthesized Lead V3 signals. . . . . . . . . . . . . 79 4.35 Comparison between observed and synthesized Lead V4 signals. . . . . . . . . . . . . 79 4.36 Comparison between observed and synthesized Lead V5 signals. . . . . . . . . . . . . 80 4.37 Comparison between observed and synthesized Lead V6 signals. . . . . . . . . . . . . 80  viii  Acknowledgements My deepest gratitude goes to my supervisors, Dr. Victor C. M. Leung and Dr. Leo Stocco. Without their guidance and support, this work would not have been possible. My special thanks go to Dr. Milan Hor´ acek from the Dalhousie University and Dr. Ilangko Balasingham from Oslo University Hospital, Rikshospitalet. Dr. Hor´ acek kindly provided us clinical ECG data and Dr. Balasingham generously offered me access to the hospital facilities. I would also like to thank Haoming Li for insightful discussions and inspirations, Dr. Min Chen, Dr. Sergio Gonz´ alez-Valenzuela, Dr. Xuedong Liang, Dr. Henry Chan and Cupid Chow for kind advice and wonderful collaborations.  ix  Dedication To my parents, grandparents and family.  x  Chapter 1  Introduction 1.1  Wireless Networks Reaching Body Area  With growing needs in ubiquitous and human-centric communications and recent advances in verylow-power wireless technologies, there has been considerable interest in the development and application of wireless networks around humans. A wireless body area network (WBAN) is a radiofrequency (RF) based wireless networking technology that interconnects tiny nodes in, on or around a human body [25][4][51]. Typically, the transmissions of these nodes cover a short range of about 2 meters. WBANs target diverse applications including healthcare, sports and fitness, secure authentication, and safe-guarding of uniformed personnel. Before the advent of WBANs, there are wireless networks serving wider areas (wider than 2 meters). They are wireless metropolitan area networks (WMAN), wireless local area networks (WLAN) and wireless personal area networks (WPAN) [40]. As we can tell from the names, these wireless networks differ from each other mainly in communication ranges. They also have distinguishments in terms of data rate, latency and energy consumption. Following, we explain these design tradeoffs from the perspectives of different layers.  1.1.1  WBANs versus Wider Area Wireless Networks  The design of a WBAN, from the physical layer’s perspective, is a tradeoff between coverage, data rate and power consumption. Compared to the wider area wireless networks, decreasing the coverage to 2 meters and limiting data rate to 1 Mbps, the current draw of a typical low-power radio will be around 10 milliampere. The coverage of WBAN (2 meters) is determined by its definition, the data rate is determined by targeting applications, and the power consumption is a result of the above two plus surrounding environment and the wearer’s action. Moving to middle layers, the tradeoff transforms to be between reliability, latency and energy consumption. Compared to wider area wireless networks, WBANs have much less energy consumption, equivalently longer lifetime, by having very low duty cycle and simplified protocol stack. Obviously, the quality-of-service (QoS) requirements (reliability and latency) originate from applications, and the energy consumption reflects the duty cycle of transmissions and overall protocol complexity. When comparing the differences of WBANs and wider area networks at the top layer, WBAN applications (e.g., healthcare, sports and fitness, and secure authentication) focus more on interactions between electronic devices and the human. A WBAN system takes both physiological activities and human actions as inputs to process and communicate. This makes the signal pro-  1  1.1. Wireless Networks Reaching Body Area cessing more complicated than that of audio/video applications. However, it offers a unique way of serving people. Figure 1.1 gives an view of typical wireless networks and their applications.   ✁ ✄✂  ☎✝✆ ✞✟✞✝✠☛✡✟☞ ✌✝✍ ✍ ✎✟☞ ✌  ✂✚✙ ✛ ✔✖✜✘✢✟✣  ✏✒✑✄✓ ✔✖✕  ✦★✭✮✩✬✫   ✄✗✘✂   ✁✤✥✂  ✲ ✦★✳✖✴✵✩✶✫  ✦★✧✪✩✬✫ ✤ ✜✯✓ ✰ ✜✯✱  ✷✹✸  ✦★✺✻✩✶✫  ✷✹✸ ✼✾✽ ✷✹✿❁❀ ✽ ✿✹✷  Figure 1.1: WBAN in the family of wireless networks.  1.1.2  WBANs versus WBASNs  A wireless body area sensor network (WBASN) is a WBAN that emphasizes the sensor/actuator functionalities of nodes in the network. In a WBAN, nodes can be classified without exception into two groups: sensors and actuators, both of which are equipped with the low-power radio. Typical sensors in WBAN include those for monitoring physiological activities (e.g., blood pressure, respiration, electrocardiography (ECG), electroencephalography (EEG) and electromyography (EMG)) and those for registering human actions (e.g., accelerometers, gyroscopes, audio/video sensor and remote controller); typical actuators in WBAN are as various as those for outputting computer processed data (e.g., speaker and display), those for assisting organ functionalities (e.g., pace maker and drug delivery for patient with diabetes) and those for facilitating people with disabilities (e.g., robotic arm/leg/wheelchair). In this thesis, we use WBASN to refer to healthcare related scenarios, and WBAN for all kinds of applications.  2  1.2. Huge Needs  1.1.3  WBANs versus WSNs  There have been extensive researches in the area of wireless sensor networks (WSNs) in the past decade [1]. We deem that the research and development on WSNs are one of many factors that have stimulated great interest in WBAN. Actually, WBAN is a specific form of conventional WSN. Unlike WSNs however, WBANs have their own characteristics as discussed below, which distinguish themselves from WSNs and also create new technical challenges. Deployment and Density The number of sensor/actuator nodes deployed on the wearer depends on use cases. Typically, the nodes are placed strategically on the human body or hidden in clothing; they are not deployed with a high redundancy to tolerate node failures as in conventional WSNs, and thus do not require a high node density. WSNs however, are often deployed in the fields which are not easily accessible by personnel and as a result, require more nodes than necessary to be randomly placed and compensate node failures. Data Rate Most WSNs are applied for event-based monitoring, where events can happen irregularly. In comparison, WBANs are employed for registering human’s physiological activities and actions, which vary in a more periodic manner. As a consequence, the application data streams exhibit relatively stable rates. Latency As mentioned above, the latency requirement of a WBAN originates from the applications and may be traded for reliability and energy consumption. While energy saving is definitely beneficial, replacement of batteries in WBAN nodes is much easier than in WSNs, which nodes can be physically unreachable after deployment. Therefore, it may be necessary to maximize battery life-time in a WSN, compared to in a WBAN, at the expense of higher latency. Mobility Wearers of WBANs may move around. WBAN nodes affiliated with the same wearer move together and in the same direction. In other words, the nodes in one WBAN share the mobility pattern. In contrast, WSN nodes are usually considered to be stationary, and any node mobility does not occur in groups.  1.2  Huge Needs  Reaching the body area opens a new era for wireless networks. Huge needs for ubiquitous and human-centric communications drive the advances of WBAN, and appearing WBAN use cases are  3  1.2. Huge Needs creating a greater market that emphasizes these needs. The following gives a picture of what the huge needs are from the perspectives of healthcare, sports and fitness, and secure authentication.  1.2.1  Healthcare  According to the U.S. Census Bureau, worldwide population of elderly over age 65 is expected to more than double by 2020, and more than triple by 2050 [48]. According to World Health Organization, more than 1 billion people in the world are overweight, and at least 300 million of those are clinically obese; over 600 million people worldwide have chronic diseases. Statistics have also confirmed the aging trend of women giving first-time births. At the same time, electronic health (e-health) has evolved from telehealth to mobile health (m-health), enabling long-term ambulatory monitoring and point-of-care. Research projects have produced implantable or wearable devices for patients, the disabled, aging people, pregnant women and neonates.  1.2.2  Sports and Fitness  According to Bluetooth Alliance’s study on market potential for its low energy technology, the volume in sports and exercise will be 47 million in 2010, and over 100 million in 2012. The financial results briefing for the fiscal year for Nintendo shows that the most successful Wii game, Wii Sports, had been sold 50.54 million copies worldwide as of March 2009. The global trend of integrating unobtrusive devices for sports and fitness has been making them easy and merry. Professional equipments, smart phones and even watches are being connected to the wireless network to enhance the exercise or training experience. A good example is the Nike+Ipod Sports Kit, which connects the Nike shoes and Apple’s portable devices together, and even integrates with web services.  1.2.3  Secure Authentication  According to Federal Bureau of Investigation, 1 billion dollars are being spent to create a new biometric database, including DNA, fingerprints and other biometric data. Also in Germany, the market has confirmed an increasing revenues of biometrics, in hundreds of millions of euro dollars. The goal of secure authentication has involved both physiological and behavioral biometrics, among which facial patterns, finger prints and eye irises are employed extensively. The potential problems, e.g., proneness to forgery and duplicability however, have motivated the investigations into new physical/behavior characteristics of human body, e.g., EEG and gait, and multimodal biometric systems.  4  1.3. Technology Advances  1.3  Technology Advances  1.3.1  Body Area Channel Models  In the past few years, researchers have made considerable progress in characterizing the body area propagation environment through both measurement-based and simulation-based studies [19][44][53] in order to support: • Prediction of link level performance in alternative sensor deployment configurations; • Development of more effective antennas with, e.g., lower specific absorption and better coupling to the dominant propagation modes. These works have been conducted in both the Industrial, Scientific and Medical (ISM) bands between 400 MHz and 2.45 GHz and in the ultra-wideband (UWB) frequency allocation between 3.1 and 10.6 GHz. In each of the frequency bands, intra-body, on-body and off-body channels have been studied [9]. Significant progress has also been made toward: • Identification of the propagation mechanisms that affect signal transmissions between nodes; • Assessment of the effects of multipath reflections from the external environment to signal transmissions between nodes; • Characterization of the fading statistics on body links that occur With body motion and change of body position in both sparse and rich scattering environments; • Development of standard UWB channel impulse response models and evaluation of typical modulation schemes utilizing.  1.3.2  Low-power Radio Technologies  Following is a comparative study of emerging and existing low-power radio technologies, including Bluetooth Low Energy [3], ZigBee [2] and IEEE 802.15.4 [26], as well as UWB and IEEE 802.15.6 [25]. Bluetooth Low Energy Technology Bluetooth Low Energy technology, formerly known as Bluetooth Low End Extension (LEE), and later Wibree, provides ultra-low power consumption and cost while minimizing the difference between Bluetooth and itself. Introduced in 2004 by Nokia, Bluetooth LEE was designed to wirelessly connect small devices to mobile terminals. Those devices are often too tiny to bear the power consumption as well as cost associated with a standard Bluetooth radio, but are ideal choices for the health-monitoring applications discussed in Section 1.6. Bluetooth LEE was said to be a ”hardwareoptimized” radio, which means its major difference from Bluetooth resides in the radio transceiver, baseband digital signal processing and data packet format. After further development under the project MIMOSA, which targets use cases including both WBANs and WPANs, LEE was released 5  1.3. Technology Advances to public with the name Wibree in 2006. One year later, an agreement was reached to include it in future Bluetooth specifications as Bluetooth Low Energy technology. Bluetooth Low Energy technology is expected to provide a data rate of up to 1 Mbps. Using fewer channels for paring devices, synchronization can be done in a few milliseconds compared to Bluetooth’s seconds. This benefits latency-critical WBAN applications, e.g., alarm generation and emergency response, and enhances power saving. Bluetooth Low Energy products can be categorized into two groups: dual-mode chips and stand-alone chips. As the names indicating, stand-alone chips are intended to be equipped with sensors/actuators and talk to each other only, while dual-mode chips are to be equipped with a personal server, e.g., smart phone, and able to also connect to traditional Bluetooth devices. Similar to Bluetooth, Bluetooth Low Energy technology will likely operate using a simpler protocol stack and focus on short-range star-configured networks without complicated routing algorithms. This suits WBANs configured in star-topology, and provides better mobility support for them. Inter-WBAN communications can be realized through a second radio or using a dual-mode chip; however, the tradeoff is larger power consumption. ZigBee and IEEE 802.15.4 ZigBee/IEEE 802.15.4 targets low-data-rate and low-power-consumption applications. Specifically, ZigBee Alliance has been working on solutions for smart energy, home automation, building automation and industrial automation. The recently completed ZigBee Health Care public application profile provides a flexible framework to meet Continua Health Alliance requirements for remote health and fitness monitoring. These solutions better suit WBAN deployment scenarios in a limited area, e.g., a hospital or a house. ZigBee/IEEE 802.15.4 devices can operate in three ISM bands, with data rates from 20 Kbps to 250 Kbps. ZigBee supports three types of topologies - star, cluster tree and mesh. In the star topology, a coordinator initiates and controls the network (i.e., similar to a piconet in Bluetooth) but there is no need for synchronization. The major advantage of ZigBee is its capability of providing multi-hop routing in a cluster tree topology or a mesh topology. As a result, WBAN network coverage can be expanded to a WPAN area using the same radio. A ZigBee mesh network may include both full-function devices (FFD) and reduced-function devices (RFD), where a RFD is equivalent to a stand-alone chip in Bluetooth Low Energy and can only act as an end device while a FFD is equivalent to a dual mode chip and can also act as a coordinator or a router. There have been many academic research projects utilizing ZigBee for transporting healthrelated data. Most prototypes mentioned in Section 1.6, however, are based on IEEE 802.15.4 chips that do not employ the higher layer ZigBee protocol stack, either because networking capability is not a must, or researchers are interested in devising more appropriate protocols. In our view, ZigBee may have a better chance to be adopted in the area of home automation and industrial automation and control, while in the area of connecting low-power peripheral devices around the human body, e.g., watches, health-related monitors and sports sensors, Bluetooth Low Energy technology possesses a bigger potential to be widely employed, due to its association with Bluetooth  6  1.3. Technology Advances as well as lower cost and lower power consumption. UWB and IEEE 802.15.6 According to the Federal Communications Commission (FCC), UWB refers to any radio technology having a transmission bandwidth exceeding the lesser of 500 MHz or 20% of the arithmetic center frequency. FCC also regulates license-free use of UWB in 3.1 - 10.6 GHz band to have a relatively low power spectral density emission. This leads to the suitability of UWB applications in shortrange and indoor environments, and environments sensitive to RF emissions, e.g., in a hospital. Commercial products based on UWB provide extremely high data rates, e.g., ”Certified Wireless USB” devices work at up to 480 Mbps, enabling short-range wireless multimedia applications, such as wireless monitors, wireless digital audio and video players and other HCI use cases. These multimedia devices can be either wirelessly connected with WBANs or are themselves portable as part of a WBAN. UWB is also an ideal technology for precise localization, which complements global positioning system (GPS) in the indoor environment for WBAN tracking. At the same time, concerns with electronic and magnetic energy absorbed by human tissues from RF circuits placed in close proximity to humans mean that WBAN devices need to employ low transmission power and low transmission duty cycles. In this regard UWB outperforms conventional transmission methods and thus attracts much attention. An emerging WBAN standard, IEEE 802.15.6 - Body Area Networks (BANs), will likely employ UWB, according to recent proposals and meeting minutes. The standard intends to endow future generation electronics in close proximity to, or inside human body. However, when this standard and any electronics that utilize it will become available remain unknown. Other Technologies A summary of the above low-power radio standards is listed in Table 1.1 for comparison. Also summarized in the table are proprietary and open technologies like Insteon, Z-Wave, ANT, RuBee and radio frequency identification (RFID). Insteon and Z-Wave are both proprietary mesh-networking technologies for home automation. Z-Wave works in the 2.4 GHz ISM band while Insteon makes use of both power lines and the 900 MHz ISM band. ANT is another proprietary sensor networking technology, featuring a simpler protocol stack and lower power consumption. ANT has been embedded in Nike+Ipod Sports Kit mentioned above to collect workout data. RuBee and RFID are both used for asset management and tracking. These technologies are complimentary to each other in terms of frequency bands, battery life and use cases. They have all been implemented on silicon chips and are being sold in comparable volumes each year. With the advances of very large scale integration (VLSI), dual and multiple-standard radios can be intergraded into a single chip, greatly reducing the cost and power consumption, while fostering combining as well as merging of technologies.  7  1.3. Technology Advances Technology Bluetooth Low Energy ZigBee (IEEE 802.15.4) UWB (IEEE 802.15.6) Insteon  Z-Wave ANT RuBee (IEEE 1902.1) RFID (ISO / IEC 18000-6)  Frequency Band 2.4 GHz ISM  Data Rate (bps) 1M  Multiple Access Method FH + TDMA  Coverage Area (meter) 10  Network Topology Star  ISM  250 K  CSMA  30-100  Star/Mesh  3.1-10.6 GHz  480 M  CSMA/TDMA  <10  Star  131.65 KHz (Powerline) 902-924 MHz 900 MHz ISM 2.4 GHz ISM 131 KHz  13 K  Unknown  Home Area  Mesh  9.6 K 1M 9.6 K  Unknown TDMA Unknown  30 Local Area 30  860-960 MHz  10-100 K  Slotted-aloha / Binary tree  1-100  Mesh Star/Mesh Peer-topeer Peer-topeer  Table 1.1: WBAN and WPAN technologies and comparisons.  1.3.3  QoS Provisions in WBAN  IEEE 802.15 Task Group 6 - Body Area Networks - is actively discussing proposals for a new optimized standard for WBAN. Providing a reliable communication service to prioritized data traffic is highly desirable; however, a limited amount of work on this topic exists in the literature for such a network type. A pioneering research employing QoS provisioning for WBAN application traffic has been published in [54]. In this work, a radio-agnostic QoS framework, BodyQoS, which adaptively schedules bandwidth to prioritized WBAN applications, is proposed. BodyQoS has been implemented on a radio scheme that complies with IEEE 802.15.4 nonbeacon-enabled mode [26], which is based on a contention multiple access method. This motivates us to design a QoS framework directly based on the beacon-enabled mode of IEEE 802.15.4 standard utilizing both the contention access period (CAP) and contention-free period (CFP) mechanisms. The proposed QoS framework will better differentiate WBAN application traffic, serve periodic traffic more directly through contention-free multiple access method, and provide easier adaptation to the standard for implementat IEEE 802.15.4 Beacon-enabled Mode The IEEE 802.15.4 standard specifies physical and media access control (MAC) layer protocols that are designed for low data-rate, short-range WPAN. Although a WPAN by definition covers a comparatively larger area than a WBAN, this coverage is achieved in IEEE 802.15.4 by transmitting at a higher power and through multi-hop relay. Therefore IEEE 802.15.4 has been selected for use in several WBAN platforms [4][33], which limit the transmission range by setting an appropriate transmission power and configuring the network in a star topology. Two operation modes are supported in the IEEE 802.15.4 MAC layer, namely beacon-enabled 8  1.3. Technology Advances mode and non-beacon-enabled mode. Beacon-enabled mode utilizes a slotted-CSMA (carrier sense multiple access) approach in CAP and a time division multiple access (TDMA) approach in CFP to provide low-latency communications. On the contrary, non-beacon-enabled mode uses an unslottedCSMA approach, which enables a more distributed network and power savings. The larger power consumption of the beacon-enabled mode originates from the need for a device to periodically wake up and listen to beacon frames. However, in our QoS platform proposed in Chapter 2, satisfying the time constraints of critical WBAN applications outweighs power-saving concerns.   ✂✁☎✄☎✆✞✝✠✟   ✂✁❀✄✬✆✞✝✠✟ ✡☞☛✍✌  ✡✠✎✏✌ ✑✠✒✔✓  ✑✠✒✕✓  ❃ ❄✕❅✕❆ ❇ ❈ ❉✞❊  ✖ ✗✘✖ ✙✠✖ ✚✂✖ ✘ ✛ ✖ ✜✢✖ ✣✘✖ ✤✂✖ ✥✘✖ ✦✘✖ ✧✘✖ ✙ ✗✕✖ ✙ ✙☞✖ ✙ ✚✞✖ ✙ ✛✕✖ ✙ ✜✔✖ ✙ ✣✕✖ ★✏✩✫✪ ✄✬ ✂✄☎✭☞✁ ✏★ ✮✠✯ ✁☎✰ ✱✕✰ ✄✬✲✳✁ ✩✴✮ ✰ ✄ ✵✬✶ ✝✠✟✸✷☎✹ ✺ ✻ ✼ ☛ ✆ ✵✏✶✾ ✿✢✁❀✽ ✪ ✄✬ ✂✄✬✭☞✁ ★✏✮✠✯ ✁☎✰ ✱✔✰ ✄✬✲✳✁ ✩✴✮ ✰ ✄ ✵✏✶ ✝✠✟✸✷☎✹ ❋ ✻  ✂❁❂  Figure 1.2: Superframe structure of IEEE 802.15.4 beacon-enabled mode.  Figure 1.2 illustrates the superframe structure in the beacon-enabled mode. It consists of two parts: one active and one inactive. The active part is equally divided into aNumbSuperframeSlots slots, which are then grouped into beacon, CFP and CAP. The beacon is transmitted at the start of slot 0, with CAP commencing immediately after it, and followed by the CFP. The length of the superframe and its active part, given by the beacon interval (BI) and superframe duration (SD), are defined as follows, BI = aBaseSuperf rameDuration × 2BO  (1.1)  SD = aBaseSuperf rameDuration × 2SO  (1.2)  aBaseSuperf rameDuration = aBaseSlotDuration × aN umSuperf rameSlots  (1.3)  = 60 × 16 = 960symbols  (1.4)  where BO is the beacon order, and SO is the superframe order (0 ≤ SO ≤ BO ≤ 14). These two parameters are determined and communicated by the coordinator of the WBAN through beacons delimiting the superframes. The CAP is required to have a minimum length of aMinCAPLength (440 symbols) to ensure enough bandwidth for MAC layer frames. Communications in the CAP 9  1.3. Technology Advances are based on p-persistent slotted-CSMA, with each backoff period spanning aUnitBackoffPeriod time units (20 symbols). The length of the CFP can grow or shrink depending on the total length of all the combined GTSs. Each GTS can occupy contiguous slots. The length of one slot is aBaseSlotDuration × 2SO symbols. It has to be exclusively occupied by a single traffic flow. During an inactive period, devices turn off their radios to minimize power consumption. The longer the inactive portion of the superframe is, the more power-efficient the network becomes. A larger BI and a smaller SD lead to higher power savings, while prolonging the delay. If the BI has been pre-determined according to a latency requirement, then the SD trades off the bandwidth for power consumption; i.e., a larger SD results in longer unit backoff periods in the CAP, and a wider GTS slot in the CFP. As a consequence, when the context of a WBAN changes, e.g., a device joins the network, updating the BI and SD in real-time is desirable. However, the length of the active portion, and thus each slot, changes exponentially according to Equation 1.1 and 1.2, which complicates the adaptation. This is addressed by our scheduler in the next section. Currently, there are two limitations when scheduling GTSs according to the IEEE 802.15.4 specifications. Firstly, only up to 7 GTSs can be allocated in one superframe. It is far from enough for a certain WBAN setup, e.g., a motion-monitoring application involving tens of accelerometers or gyroscopes. Secondly, allocation of GTSs works in a first-come-first-served (FCFS) way, which can starve high-priority streams. Because of these limitations, several improvements on scheduling GTSs have been proposed. In [38], an optimal on-line scheduling of all traffic is provided, given the knowledge of a superframe structure and all current transactions, which allows sharing of GTS by multiple transactions. In [52], an off-line scheme to generate a superframe structure has been proposed and implemented. Authors of [31] study the GTS allocation from a network calculus perspective, and provide an approach for sharing GTSs based on round-robin scheduling. In [30], a new superframe structure is proposed to accommodate traffic with different data rates by feeding them to different sub-beacon frames. In [24], authors propose a way of adapting traffic priorities by examining their activities in previous superframes. An improvement proposed in [7] accommodates more GTSs in the CFP by interpreting a beacon message in a different way. Nevertheless, three problems have not been addressed in employing IEEE 802.15.4 for WBAN applications. First, analysis and simulation results have shown that the performance of GTS allocation is largely dependent on the traffic characteristics [41][5]. However, no GTS allocation scheme has taken into account specific QoS requirements of WBAN application traffic. Second, all of the above works assume that all applications wish to transmit within the CFP rather than utilizing both CAP and CFP, which leads to an underutilization of the total bandwidth. Third, adapting a superframe structure is desirable to accommodate the changing context of a WBAN bearer; however, it is always assumed fixed in existing approaches after the network starts. In order to address these problems, in Chapter 2 we propose a WBAN QoS framework that employs the IEEE 802.15.4 beacon-enabled mode.  10  1.3. Technology Advances  1.3.4  Sensors and Actuators  Sensors and actuators are the key components of a WBAN. They bridge the physical world and electronic systems. As sensors/actuators are going to be put on human bodies or even implanted, their size, form factor and physical compatibility to human tissues are crucial. This motivates the search and synthesis of novel materials. Micro-electromechanical Systems With the advances in micro-electromechanical systems (MEMS), sensors/actuatos are getting tinier and tinier in size, in the range of 1 to 100 micrometres. Accelerometers and gyroscopes are good examples of this advancement. They are widely used for motion capture. With accelerometers/gyroscopes mounted on certain part of a human body, the system can effectively register the subject’s movement. Also, it is reported that MEMS devices have been manufactured for drug delivery. By fabricating little spikes on silicon or polymers, the liquid drug can be injected through epidermis under pre-defined instructions or remote control. Textile Electrodes As mentioned above, ECG monitoring is a typical application in WBAN that can help identify the wearer’s health status. It measures potential differences across electrodes attached to corresponding parts of the torso. For bed-side monitoring, disposable electrodes, traditionally made of silver chloride (Ag/AgCl), are widely used. However, long-term usage of these types of electrodes may cause artifacts as well as skin problems. A recently developed solution is to use textile-structured electrodes, which are embedded inside the garment, such as fiber, yarn and fabric structure [50]. These textile-structure electrodes, possibly woven into clothes, are free from skin problems and thus are comfortable and suitable for long-term monitoring. Compared to Ag/AgCl electrodes, they are much more flexible since the shape can be adapted with the human movement. The same kind of electrodes can be also employed for ambulatory EEG and EMG systems. Translating Images Thanks to the maturing of both charge-coupled device (CCD) and complementary metal-oxidesemiconductor (CMOS) active-pixel sensors, cameras can be made tiny to be embedded into a pair of glasses. The images captured can be processed in real-time and converted into voice format, to assist people who have eyesight problems. The images can even be translated to other kinds of formats, e.g., gentle electrical impulses, to be imposed on the tongue. Together with a lollipop-sized electrode array in their mouths, blind peopled can be trained to regain ”vision”. Energy Harvesting As all WBAN nodes require an energy source for data collection, processing and transmission, development of suitable power supplies becomes paramount. One solution to this problem is energy harvesting, e.g. based on body movements or temperature difference; another solution reported 11  1.4. Wired and Wireless ECG Systems recently is to utilize wireless energy transmission over the short range (several meters), using evanescent waves [32]. Both approaches require appropriate energy conversion and storage devices.  1.4  Wired and Wireless ECG Systems  ECG is a widely accepted approach for monitoring of cardiac activity and clinical diagnosis of certain heart diseases, particularly those caused by damage to the conductive tissue or levels of dissolved salts, and observed in the form of abnormal rhythms [49]. By putting electrodes on corresponding locations of the torso, an ECG system gives body-surface potential differences across these electrodes. There is no consensus on the optimal quantity and positions of electrodes that shall be used in an ECG system; they largely depend on the particular application [21]. While the two parameters are yet to be identified, various ECG systems have been developed. They can be classified into four groups, namely conventional 12-lead ECG systems, electrocardiographic body surface mapping (BSM) systems, vectorcardiographic (VCG) systems, and wireless single-pad ECG systems. Each ECG system can be seen as a particular configuration of the above two parameters. The portable and wireless versions of ECG system are typical WBAN applications for ambulatory monitoring of patients, workers and soldiers. Figure 1.3 is a two-dimensional representation of the Dalhousie torso model of electrode placement. We use this model to describe the configurations of different ECG systems below. The left and right half of the display represent the anterior and posterior torso respectively. From neck to waist, the numbered transverse levels are 1 inch apart, and the positions around transverse sections are supposed to be equiangularly divided.  12  I  323  288  248  207  165  121  99  79  38  H’  338  308  268  227  186  145  H  101  60  20  122  100  59  19  G’  324  289  249  208  166  123  80  39  21  83  63  43  25  84  64  44  26  85  65  45  27  5  86  66  46  28  6 7  87  67  47  29  15  88  68  48  30  89  69  49  31  103 104 105 106 107 108 109 110  82  62  42  24  14  4  90  50  32  91  51 71  33  111 112 113  70  33  92  52  32  147 148 149 150 151 152 153 154 155 156  157  124 125 126 127 128 129 130 131 132 133 134 135 136 137  81  41  23  3  158  114  72  31  138  93  53  30  159  115  73  34  15  139  94  54  16  7 8  160  116  74  35  9  140  95  55  17  10  161  117  75  36  11  141  96  56  18  12 1  162  118  76  37  13  142  97  57  22  163  119  77  21  120  78  19  99  79  38  164  143 144 121  98  58  20  5  4’  4  3’  3  2’  2  1’  G  F’  F  E’  E  D’  D  C’  C  B’  B  A’  A  P’  P  O’  O  N’  N  M’  M  L’  L  K’  K  J’  J  I’  I  5’ 167 168 169 170 171 172 174 175 176 177 178 179 180 181 182 183 184 185 165 6 173 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 6’ 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 207 7 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 7’ 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 248 8 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 8’ 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 288 9 309 310 311 312 313 314 315 316 317 318 319 320 321 322 9’ 325 326 327 328 329 330 331 332 333 334 335 336 337 323 10 339 340 341 342 343 344 345 346 347 348 349 350 351 352 10’  146  102  61  40  22  13  1 2  Dalhousie 352 nodes with 120 torso leads  1.4. Wired and Wireless ECG Systems  Figure 1.3: Dalhousie’s two dimensional torso model with 352 nodes [18].  13  1.4. Wired and Wireless ECG Systems  1.4.1  Conventional 12-lead ECG Systems  Conventional 12-lead ECG systems require 10 electrodes on the patient’s torso: 6 unipolar electrodes (green nodes in Figure 1.3), 3 bipolar electrodes (yellow nodes in Figure 1.3), and 1 reference electrode (normally placed around node 338 in Figure 1.3). These electrodes then form 12 leads (leads I, II, III, aVR, aVL, aVF, V1, V2, V3, V4, V5 and V6) to facilitate fast interpretation by cardiologists. Since they have been well-trained to accept 12-lead ECG information, a huge number of ECG systems are using such electrodes and placement configuration [14].  1.4.2  Body Surface Mapping Systems  BSM systems originate in the 1960s [21]. In such a system, a large number of electrodes (32 to 219) are placed on strips which are arranged around the circumference of the human torso. The potentials are simultaneously recorded and displayed on a map of the body surface model, as in Figure 1.3, and caregivers/researchers can have a full knowledge of the bio-potential distributions on torso [15][29][23]. Better diagnostic accuracy which cannot be achieved in ECG systems with less leads, will be achieved. However, as each electrode is wired to a data acquisition board, the complexity of BSM systems hinders their wide employment in clinical environment. The BSM system from Dalhousie University is an example in this category. There are 120 measurement sites (circled nodes in Figure 1.3) connected with 1-mm-diameter Ag/AgCl electrodes, at which potentials are recorded. There are another 232 nodes (square nodes in Figure 1.3), at which potentials can be interpolated from the set of the 117 recorded potentials. The researchers at Dalhousie University have also developed a boundary-element model of a realistic three-dimensional human torso, employed to simulate potentials at the above 352 nodes.  1.4.3  Vectorcardiographic Systems  A VCG system [35] uses a minimum of 4 electrodes (3 leads) and registers the electrical heart activity in three orthogonal leads. Based on the hypothesis that electrical heart activity can be represented by a stationary dipole, the potentials recorded at three leads are supposedly proportional to one rectangular component of the assumed heart dipole vector [22]. Under this assumption, data recorded by a VCG system can be transformed to that of a 12-lead system by the means of linear equations. The advantage of this approach is the reduced number of leads, which benefits caregivers and patients in setup and long-term monitoring. However, wires are still required to connect electrodes, thus causing artifacts/noise and discomfort. The EASI system [11] (4 electrodes are shown in Figure 1.3 by red nodes, 1 reference electrode is the same as that in the 12-lead system) is an example of the modified VCG system. It has been approved by the US Food and Drug Administration for assessing normal, abnormal, and paced cardiac rhythms and for detecting myocardial ischemia or silent ischemia.  14  1.5. Interconnections of WBANs  1.4.4  Wireless Single-pad ECG Systems  With the desire to make ECG system portable, easy to setup, comfortable to patient and tolerant of artifact, wireless single-pad ECG systems have been developed [37][27][12]. The so called singlepad is a tiny printed circuit board (PCB) with two or three electrodes attached or embedded. It performs front-end ECG data acquisition, analog-to-digital conversion and wireless transmission. Digital ECG data is transmitted wirelessly to a personal server for storage or diagnosis. Such a pad is usually placed on the left chest (around node 130 in Figure 1.3) to obtain a clear QRS complex (Q-, R-, S- waves in the ECG signal). As there is no wire involved, the sticky-pad based ECG system benefits patients in both setup and monitoring stages. However, the problem comes to cardiologists’ diagnosis of the signals collected from such an ECG system, as no standard specifies where to put the pad or how close to each other the electrodes should be in order to obtain meaningful measurements. Further, there can be a significant information loss compared with conventional 12-lead ECG systems, which may limit wireless ECG system’s efficacy. In order to address these problems, in Chapter 3 we propose an upgraded version of the wireless single-pad ECG system.  1.5  Interconnections of WBANs  Unlike WSNs as autonomous systems, a WBAN seldom works alone. Multiple WBANs can be interconnected to facilitate data exchange, storage, sharing and management. Based on the scale of the system and the number of underlying radios employed, we discuss in this section two architecture approaches of WBAN platforms, namely one-tier and two-tier architectures.  1.5.1  One-tier Architecture  A one-tier system architecture considers a limited environment, e.g. home, office and playground, where one low-power radio is enough to cover the whole span through multiple hops. Two categories of nodes exists in this architecture setup, i.e., sensor/actuator nodes in or on a human body, and router nodes around WBAN wearers. These router nodes equipped with the same radio as that with the sensor/actuator node are deployed across the infrastructure to facilitate multi-hop routing. This architecture setup is similar to that of a traditional WSN and both of them often employ a gateway to interface with the outside world. In WSNs, however, every node functions as a sensor node as well as a router node. Figure 1.4 illustrates an example of the one-tier system deployment in a building with three floors. Router nodes are distributed across each floor with the intention to have better coverage. Data exchanges between WBANs are through router nodes; data storage, sharing and management through the web for example are through the gateway. Since there is only one radio, all communications share the same bandwidth and thus collision can easily occur while the number of nodes, including both routers and sensor/actuator nodes, increase in the area. Normally asynchronous MAC mechanism, e.g., the one in ZigBee/IEEE 802.15.4, is used to deal with collisions.  15  1.5. Interconnections of WBANs  ✛✢✕✖✌☛✛✗✠☛✁ ✣✥✤✥✦✧✎✑★✡✤✗✎✓✠✡✁✏✌✡✠✡✄☛✕✖✛   ✂✁☎✄✝✆✟✞ ✠✡✠☛✁  ✞ ☞ ✌✡✍✏✎✑✠✒✠✓✎✑✔☛✕✖✁ ✌☛✕✗✎✑✘✙✠✡✁✚✍✜✛  ✫✂✌✡✄✝✆✟✞ ✠✡✠☛✁ ✎✑✔✡✕✭✬✮✤✯✎✓✕✖✘✰✤✲✱  ✩ ✛ ✎✪✆✟✞ ✠✡✠☛✁ ✁✚✠✡★✓✎✓✕✖✁✏✌☛✠✡✄✡✕✖✛  Figure 1.4: One-tier system architecture for interconnecting WBANs.  While the coverage of a WBAN is limited to about two meters, this way of interconnection extends the system to hundreds of cubic meters or so. It suits both short-term setup, e.g., on a conference site, and long-term setup, e.g., at home.  1.5.2  Two-tier Architecture  Compared to the one-tier approach, two-tier architecture is applied for use cases in a larger area, e.g., metropolis area. To cover such range using certain low-power radio, like Bluetooth, is simply impossible. Therefore, an larger area wireless network, e.g. cellular system, is needed in addition to the low-power radio serving the 2-meter range. There is no router nodes deployed around the wearer, since the wearer is supposed to travel at will. In order to bridge these two radio networks, a gateway device, increasingly realized by a smart phone, sits in between and helps interpret data from two ends. Figure 1.5 illustrates the idea of a two-tier system. Data exchanges, storage, sharing and management are all through the second radio, and a server is often involved. As two underlying radios are normally utilizing different frequency bands, collisions are limited to communications within WBANs. Consequently, both synchronous and asynchronous MAC mechanisms can be used for scheduling and collision avoidance. While this architecture can effectively extend the system coverage to a larger area, the increased power con16  1.6. Existing WBAN Platforms  ✢ ✣ ✤✦✥★✧ ✩✪✩✫✧ ✬✫✭✎✮✯✤✦✭✎✧✱✰✲✩✦✮✱✥✴✳  ✠☛✡✌☞✎✍✑✏✓✒✓✏✓✔ ✕ ☞✗✖✙✘✟✄✚✍✛✡✜✖   ✂✁☎✄✝✆✟✞   ✂✁✎✄✚✆✵✞  ✶✁✎✄✝✆✟✞  ✶✁✎✄✚✆✵✞   ✂✁☎✄✝✆✟✞  Figure 1.5: Two-tier system architecture for interconnecting WBANs.  sumption and associated cost would hinder its applications in long-term cases before further optimizations. The dual-radio approach can also be used in the environments mentioned for the one-tier approach when power and cost are less of a concern. For example, on a disaster site, another radio is desirable to increase overall system performance. Further, there can be a hybrid approach where both router nodes and a second radio are used.  1.6  Existing WBAN Platforms  Based on the discussions of the two architecture approaches of WBAN platforms, we enumerate in this section several pioneer WBAN research projects and platforms.  1.6.1  MITHril  Early research efforts at MIT Media Lab has produced MITHril [42], a wearable computing platform that employs ECG, skin temperature and galvanic skin response sensors for wearable sensing and context-aware interaction. MITHril is not a real WBAN in that multiple sensors are wired to a single processor; there is no wireless network connecting sensors. A later version of this platform, 17  1.6. Existing WBAN Platforms MITHril 2003, extends MITHril to a distributed multi-user wireless platform by utilizing Wi-Fi function available on personal digital assistant (PDAs), i.e., a PDA acts as a personal server and relays data of each person to a central station. This makes MITHril 2003 a two-tier architecture.  1.6.2  CodeBlue  The CodeBlue project at Harvard University [43] considers a one-tier architecture in the hospital environment, where multiple router nodes can be deployed on the wall. All nodes use the same ZigBee radio. Patients/caregivers can publish/subscribe to the mesh network by multicasting; there is no centralized or distributed server or database for control and storage. Localization functionality is provided by MoteTrack with an accuracy of 1 meter, based on the same radio. As a result of mobility and multi-hop transmissions, the system experiences considerable packet losses and is limited to 40 Kbps aggregate bandwidth per receiver.  1.6.3  AID-N  Based on the CodeBlue architecture, the Advanced Health and Disaster Aid Network (AID-N) is developed at Johns Hopkins University [17] for mass casualty incidents where electronic triage tags can be deployed on victims. Additional wireless capabilities, e.g., Wi-Fi and cellular networks, are introduced to facilitate the communications between personal servers and the central server where data is stored. This makes AID-N a hybrid two-tier architecture. Furthermore, a web portal is provided to multiple types of users, including emergency department personnel, incident commanders and medical specialists. A GPS module is employed for outdoor localization while THE MoteTrack system is designed for tracking indoors. However, patients have mobility constraints due to lack of routers in the network, and a very limited number of sensor nodes can be put on each patient because of the limited bandwidth.  1.6.4  WHMS  The Wearable Health Monitoring Systems (WHMS) is developed at the University of Alabama [36] and targets at a larger-scale telemedicine system for ambulatory health status monitoring. It is an example of the two-tier architecture. WHMS has a star-topology network for each patient, which is connected via Wi-Fi or cellular network to a healthcare provider. The personal server, implemented on a PDA, cell phone, or personal computer (PC), coordinates the data collection from sensor nodes using a time-division multiple access (TDMA) mechanism, provides interface to the users and transfers data to a remote central server. Physicians can access data via the Internet and alerts can be created by an agent running on the server. However, the power consumption and cost associated with long-term data uploading can hamper system realization.  1.6.5  MIMOSA  Microsystems platform for MObile Services and Applications (MIMOSA) [28] is a research project involving 15 partners from 8 different European countries to create Ambient Intelligence. MI18  1.7. Contributions MOSA’s approach is similar to WHMS while it exclusively employs a mobile phone as the usercarried interface device. Wibree, later renamed as Bluetooth Low Energy technology, and RFID tags are used for connecting local sensor nodes. NanoIP and Simple Sensor Interface (SSI) protocols are integrated into MIMOSA to provide an application programming interface (API) for local connectivity and facilitate sensor readings.  1.6.6  WiMoCA  The Wireless sensor node for a Motion Capture system with Accelerometers (WiMoCA) [10] project at several Italian universities is concerned with the design and implementation of a distributed gesture recognition system. It is a two-tier architecture targeting home/office use cases. The system has a star topology with all sensing nodes sending data to a non-sensing coordinator node using a TDMA-like approach, and the coordinator in turn relays the data to an external processing unit using Bluetooth. The sensing modules, each made up of a tri-axial accelerometer, can be put on multiple parts of the body for motion detection. The radio modules of all nodes work in the 868 MHz European license-excempt band, with up to 100 Kbps data rate. A Java-based graphical user interface (GUI) at the processing unit side interprets data stream for posture recognition.  1.7  Contributions  The technical contributions of this thesis are summarized in the following three subsections.  1.7.1  A Survey and Outlook  This thesis presents a complete survey on recent advances in WBAN, including the market needs, channel modelling, standardization of lower-layer communication protocols, QoS provisions, developments of sensors/actuators, WBAN architectures and experimental platforms. Additionally, existing wired and wireless ECG systems are compared, as a basis to explain our proposed novel wireless ECG system. By explaining our QoS platform designed for general WBAN applications, and our wireless three-pad ECG system (W3ECG) invented for particular healthcare areas, we foresee a bright future for wide deployments of such kinds of wireless networks on the human body, while employing our platform and system.  1.7.2  A QoS Provisioning Platform  As WBAN will be serving human-centric communication applications, providing a reliable communication service to prioritized data traffic is highly desirable, while a limited amount of work on this topic exists in the literature. A recent work employing nonbeacon-enabled mode of IEEE 802.15.4 standard has motivated us to design a QoS framework based on the beacon-enabled mode of the same standard. Utilizing both the CAP and CFP mechanisms, the proposed QoS framework can better differentiate WBAN application streams and serve periodic traffic more directly through contention-free multiple-access methods. 19  1.7. Contributions A dominant feature of the proposed framework is minimum adaptation to the existing standard. The Admissions Controller, Superframe Scheduler and CFP Scheduler are all based on existing IEEE 802.15.4 frame structures and timing requirements. This made it easy to adopt our platform and associated algorithms, as well as to implement them on off-the-shelf hardware.  1.7.3  A Novel Wireless ECG System  To tackle the problems raised by wireless single-pad ECG system, we propose an upgraded version of it, the W3ECG. W3ECG furthers the pad design idea of the single-pad approach. Inspired by the transformation possibility of signals obtained in VCG systems, we bring two more pads in W3ECG to gain spatial variety of heart activity. The totalled three pads are uniformly designed so that each can be seen as a rectangular component of the assumed heart dipole vector or its rotated/shifted version. Signals obtained from these three pads, plus the spatial information, make it possible to generate conventional 12-lead ECG signals. We have been able to manufacture the front-end ECG circuit, and combine it with IEEE 802.15.4 radio platform. Software packages for the server and pad have also been developed to make a fully running W3ECG possible.  20  Chapter 2  Employing IEEE 802.15.4 for QoS Provisioning in WBAN As WBAN will be serving human-centric communication applications, providing a reliable communication service to prioritized data traffic is highly desirable, while a limited amount of work on this topic exists in the literature. A recent work employing nonbeacon-enabled mode of IEEE 802.15.4 standard has motivated us to design a QoS framework based on the beacon-enabled mode of the same standard. Utilizing both the CAP and CFP mechanisms, the proposed QoS framework can better differentiate WBAN application streams and serve periodic traffic more directly through contention-free multiple-access methods. A dominant feature of the proposed framework is minimum adaptation to the existing standard. The Admissions Controller, Superframe Scheduler and CFP Scheduler are all based on existing IEEE 802.15.4 frame structures and timing requirements. This made it easy to adopt our platform and associated algorithms, as well as to implement them on off-the-shelf hardware. Details of the traffic differentiation mechanism, and Admissions Controller, Superframe Scheduler and CFP Scheduler are discussed in this chapter. Evaluations are provided in Chapter 4.  2.1  Topology Selection  We reviewed the superframe structure of IEEE 802.15.4 beacon-enabled mode in Chapter 1. Beaconenabled mode utilizes a slotted-CSMA approach in CAP to provide low-latency channel access, and a TDMA approach in CFP to allow bandwidth reservation. The benefits of employing the beaconenabled mode for QoS provisioning is obvious and becomes clearer later in this chapter. However, there is a tradeoff: the WBAN network has to be organized in a star topology, as required by the standard. This is because of the need to synchronize all nodes using beacons in a timely manner. Figure 2.1 gives an example of using this topology to directly link WBAN sensors/actuators. In our platform, the personal server (the smart phone in Figure 2.1) is acting as the IEEE 802.15.4 PAN coordinator; sensor/acuator nodes are end devices in the IEEE 802.15.4 network. Although the scale of the network is limited by this topology, it is large enough to serve the needs of most WBAN use cases. The No. 1 goal of designing the QoS framework is to satisfy time constraints of critical WBAN applications, and increase utilization of the channel.  21  2.2. Service Differentiation Techniques  ✟✂✟✂✠  ✂✁ ✄✂✄☛✡✌☞☛✍ ✎✑✏✒✏✒✓☛✍ ✎  ✟✂✔✕✠  ☎✝✆✂✞ ✖✘✗  ✠✌✙ ✚ ✖✘✛✕✜ ✱  ✟✣✢✤✠ ✢ ✛✥✞ ✦ ✛✥✧ ★✪✩✬✫ ✘✖ ✛✕✛✥✭ ✮✕✦ ✧✕✆✂✞ ✛✥✭ ✯ ✰✥✱ ✭ ✜✲✥ ✛ ✧✕✆ ✙ ✜ ✱ ✭✴✳ ✱ ✭✒✵ ✱ ✕ ✧ ✮✶✮ ✱ ✳✪✦ ✖ ✱ ✜ ✯ ✜ ✱ ✧✕✜✷✛✕✭✒✸✹✆✂✖✘✞ ✚ ✆✂✞ ✛✥✭✺✮ ✱ ✳☛✦ ✖ ✱ ✵  Figure 2.1: Example of using star topology to directly link WBAN sensors in an IEEE 802.15.4 network.  2.2  Service Differentiation Techniques  2.2.1  QoS Vector  For each WBAN application traffic stream, four QoS parameters are considered. They can be specified in a vector format as (for traffic stream i ): Ai = (pi , di , ri , bi )  (2.1)  where Ai denotes the i -th application, pi the priority, di the time constraint, ri the arrival rate, and bi the burst size. We call this a QoS vector. Parameters pi and di are employed for differentiating traffic: priority roughly addresses how urgent the traffic needs to be sent, while time constraint specifies the exact tolerable delay during runtime. The reasons to have these two parameters at the same time are: • The priority helps identify the application’s degree of emergency, while time constraint is a combination of technical and user experiences. • For implementations, priority can be represented by two bits, while time constraint requires higher resolution. As a result, the priority bits can be carried in existing IEEE 802.15.4 association request to facilitate the coordinator’s preliminary decision on admissions. This is 22  2.2. Service Differentiation Techniques further explained in the next subsections. The definitions of the latter two parameters are dependent on the periodicity of the data traffic. For WBAN monitoring applications that generate periodic data traffic, e.g., ECG/EEG monitoring, ri and bi are the corresponding sampling frequency and resolution of the embedded analog-to-digital converter (ADC). In comparison, for event-based WBAN applications, e.g., remote control of an actuator, sensor status inquiry and triggering of an emergency alarm, which generate aperiodic traffic, ri and bi are the mean values of the traffic’s respective distributions. After the WBAN coordinator’s initiation, it maintains a group of lists admitted traf f ic list pi to register admitted traffic with the same priorities. Each of the list entries is a QoS vector, and is sorted with decreasing time constraint. These lists are used in our Superframe scheduler and CFP scheduler discussed in the following sections.  2.2.2  Traffic Prioritization  WBAN application traffic is differentiated into three priority categories with respect to their importance. They are, in decreasing priority levels: alarm/control (AC) traffic, command/data (CD) traffic, and routine (RT) traffic. The priorities are represented by 2-bit binary constants during implementation as listed in Table 2.1. Priority (binary) 0b00 0b01 0b10 0b11  Traffic Category Reserved Alarm/control traffic Command/data traffic Routine traffic  Abbreviation Reserved AC CD RT  Superframe Period Reserved CAP CAP CFP  Table 2.1: Traffic priorities and 2-bit representations. AC traffic refers to alarms triggered by sensory devices or crucial controls of actuators. These signals can indicate life-threatening conditions at the patient and thus need to be dealt with immediately. CD traffic includes higher layer commands and low-rate aperiodic data streams, e.g., commands for QoS provisioning and WBAN device configurations. This traffic is treated with a lower priority since responsiveness, although desirable, is not a critical requirement. A higher reliability of CD traffic can be obtained by increasing the number of retransmissions whenever necessary. The third category, RT traffic, is the major bandwidth consumer in a WBAN. It consists of all periodic traffic and possibly high-rate aperiodic traffic. As they possess the lowest priority in the framework, the maximum number of traffic streams in this category that can be accommodated in the network is determined by the traffic in the first two categories. This dependence will be further discussed in the subsections presenting our admission control and scheduling algorithms. Besides defining the above three traffic categories to differentiate services, admitted traffic streams of each category are maintained in a list at the WBAN coordinator and sorted according to the time constraints as aforementioned. In other words, admitted traffic is further prioritized with decreasing time constraint. The grounds for this differentiation come from the understanding that the more urgent the traffic is, the more important it is likely to be. When a certain traffic 23  2.2. Service Differentiation Techniques stream has to be dropped because of the need to admit/schedule more urgent traffic, the process starts from the traffic with the lowest priority, i.e., RT traffic with the longest delay tolerance. Implementations The functionalities of a sensor/actuator determine what traffic type a device generates. Since the device is equipped with a sensor/actuator before deployed on the human body, its priority is configured prior to joining the network. When a device requests admission to the network, its priority level is conveyed to the coordinator through an IEEE 802.15.4 association request. Table 2.2 shows the current Capability Information field of an Association Request command. We use the 2-bit Reserved field to carry the 2-bit priority information. bits:0 Alternate PAN Coordinator  1 Device Type  2 Power Source  3 Receiver On When Idle  4-5 Reserved  6 Security Capability  7 Allocate Address  Table 2.2: Capability Information field of Association Request command. The admission response is then carried by association response, where the status of association is expanded to reflect the admission decision.  2.2.3  Traffic Accommodation  According to our survey in Chapter 1, previous work either evaluated the performance of CAP, or assumed that all applications wish to transmit within the CFP. This limits both the capabilities of the beacon-enable mode and the utilization of the bandwidth. To accommodate all three categories of traffic in the context of an IEEE 802.15.4 superframe, our framework utilizes both CAPs and CFPs. CAPs and CFPs have distinct prosperities and are choices for different categories of traffic. The accommodations have been summarized in Table 2.1, and below we explain the rationales in detail. Recall communications in CAPs are based on the slotted-CSMA mechanism, which provides radios with quicker channel access when the network load is low. This advantage and constraint coincide with the objective and characteristics of AC and CD traffic, and thus they are arranged to be sent in CAPs. To differentiate these two traffic types in a CAP, the parameter backoff exponent (BE) [26], is configured with distinct values for each. BE determines the initial range of backoff periods a device shall wait before attempting to assess the channel state, either during the first access period, or when a non-idle channel is detected. A smaller BE indicates a larger possibility of obtaining a higher priority when devices compete for the channel. Therefore, AC traffic is assigned a smaller BE (BE = 0 in the simulations) to access the channel quicker, while CD traffic is configured with a larger BE (BE = 2 in the simulations). The acknowledgement (ACK) mechanism is enabled for both AC and CD packets to increase reliability. In comparison to AC and CD traffic, RT traffic arrives in a larger volume and is mostly periodic, which make it naturally suited for transmissions in CFPs. Communications in CFPs are based on 24  2.3. Admissions Control the TDMA mechanism, which guarantees exclusive channel access to traffic streams. Plus, no ACK mechanism is necessary for either flow or error control, assuming there is no other radio interfering in close proximity.  2.3  Admissions Control  Figure 2.2 summarizes a successful admission and scheduling process, which will be elaborated in this and the following sections.  2.3.1  Association and Admission  According to the IEEE 802.15.4 specifications, each device that wishes to communicate in the network is required to associate with the coordinator first. However, it is out of the scope of the standard to determine whether the WBAN coordinator and the network have sufficient resources available to allow another device to associate. It is reasonable to assume in a WBAN that the coordinator always has enough memory and network addresses to assign to a new traffic flow, since the total number of traffic flows in a typical WBAN is well within 255, which is the limit of the current standard. The question left is whether there is adequate bandwidth to fulfill the QoS requirements of the requesting traffic flow. This is where our Admission Controller comes into play. The Admission Controller sits upon IEEE 802.15.4 MAC layer, and handles associations. An admission request is processed only with regard to the corresponding traffic’s priority. Admission decision of AC or CD traffic is determined immediately, while that of RT traffic occurs in two steps. The exchange of admission request/response information is based on the existing association request/response command packets in the standard. The decision for AC or CD traffic is made by the Admission Controller itself upon receiving the traffic’s priority information (Table 2.2) carried by the association packet. The decision for this request is passed on to the requesting device immediately through an association response (in-direct transmission). In comparison, the Admission Controller evaluates RT traffic’s priority and provides a preliminary decision based on the available resources, leaving the final decision to the Superframe Scheduler and CFP Scheduler. In return the preliminary result is then sent via an association response to the requesting device. If the RT traffic can be scheduled in the CFP, then the relevant allocation information is set in the GTS descriptor embedded in the next beacon frame, as defined by the standard. Otherwise, the requesting device learns of the declined request after a timeout, similar to that of the original GTS request. Figure 2.2 illustrates this process.  2.3.2  The Admission Controller  The admission control criteria are outlined in Algorithm 1 in pseudocode. The idea is to always admit AC traffic, while the decisions for CD and RT traffic are based on the available network resources. Specifically, CD traffic is admitted whenever there is a sleep period (SO < BO), when there is traffic with a lower priority (RT), or when the utilization of CAP is below a threshold as defined by the parameter α. RT traffic is preliminarily admitted whenever there is a sleep period, 25  2.3. Admissions Control CAP is underutilized (utilization of CAP < α), or when the utilization of CFP is below a threshold, as defined by another parameter, β. Algorithm 1 Admission Controller switch pi of Ai case AC: // always admit admit(Ai ); // add to list admitted traffic list AC case CD: if ((SO < BO)||(sizeof (admitted traf f ic list RT ) > 0)||(UCAP < α)) then admit(Ai ); // add to list admitted traffic list CD else reject(Ai ); end if case RT: if ((SO < BO)||(UCAP < α)||(UCF P < β)) then admit(Ai ); // add to list admitted traffic list RT else reject(Ai ); end if  Utilization Parameters The utilization calculation of CAP and CFP is performed by the scheduler, and is updated whenever there is newly admitted traffic. The values for the parameters α and β range between 0 and 1, and determine the estimated maximum allowable utilization of CAP and CFP respectively. Extremely high utilization of CAP can only be achieved by saturating the network in that traffic is not exclusively assigned bandwidth; it is also not always desirable because in our case certain bandwidth should be left for newly joining traffic. For CFP, bandwidth shall also be made available for the overhead and inter-frame spacing (IFS) resulting from variable GTS allocation that cannot be predetermined. According to our scheduling scheme, RT traffic can be assigned GTSs in different beacon intervals, and the allocations, i.e., number of slots and the start slot, change whenever there is newly admitted traffic. Therefore packet payload is truncated in a feasible length with regards to the length of the assigned GTS, which is unknown during scheduling. This results in variable packet overhead and waste of bandwidth due to the requirement of ensuring IFS in the standard.  2.3.3  Two-step Decision  If the admission request is successful, the requesting traffic shall further send its QoS vector to the coordinator for scheduling. AC traffic can start communications immediately; CD traffic is designed to start from the next beacon interval in case it is stopped due to unavailability of bandwidth; RT traffic has to wait for its own GTS descriptor in the beacon frame. If the corresponding association fails, the higher layer application will be informed, and shall subsequently decide whether or not to request admission again. The scheduler may or may not schedule enough bandwidth to the device 26  2.3. Admissions Control requesting RT traffic even though it has been associated successfully. Further, RT or even CD traffic may be terminated whenever there is not enough bandwidth to admit a new traffic flow with a higher priority; i.e., service pre-emption is employed. This termination triggers the disassociation of the corresponding device. The rationale of this design is to meet the time constraints of all three traffic categories, while maintaining compatibility with the standard. In other words: • There is only one reserved field available in the existing association request packet for extension. Only 2-bit priority information of a QoS vector can be put in, as shown in Table 2.2. • The association decision shall be made as soon as possible, both to fit the macResponseWaitTime of the standard, and to better serve the traffic in terms of less latency. Therefore, combining association and admission together saves time. • GTS allocations take effect in the next beacon interval, before which a higher priority traffic flow can be admitted into the WBAN, and an admitted RT traffic flow can be preempted. Implementations In order to realize the two-step decision, an additional command frame needs to be added to the IEEE 802.15.4 MAC layer. Table 2.3 lists the exiting command frames in the specifications with our added QoS notification (0x0A) command frame. And the fields of the QoS notification command is specified in Table 2.4. Note that the time constraint, arrival rate, and burst size are all quantified to 1 byte. As QoS notification is a separate command, the length of the fields for the three parameters can be modified easily if necessary, and additional fields can be added. Command Frame Identifier 0x01 0x02 0x03 0x04 0x05 0x06 0x07 0x08 0x09 0x0A  Command Name Association request Association response Disassociation notification Data request PAN ID conflict notification Orphan notification Beacon request Coordinator realignment GTS request QoS notification  Tx/Rx Tx Rx Tx/Rx Tx Tx Tx / Rx / Tx  Table 2.3: IEEE 802.15.4 MAC command frames. octets MHR fields  1 Time Constraint  1 Arrival Rate  1 Burst Size  Table 2.4: QoS notification command format.  27  2.4. Traffic Scheduling Also, one value is added to the Association Status field of an Association Response command, to reflect the temporary decision for RT traffic (0x03 in Table 2.5). Association Status 0x00 0x01 0x02 0x03 0x04 - 0x7f 0x80 - 0xff  Description Association successful. PAN at capacity. PAN access denied. On wait list. Reserved. Reserved for MAC primitive enumeration values.  Table 2.5: Valid values of the Association Status field.  2.4  Traffic Scheduling  The proposed scheduling algorithm consists of two stages, Superframe Scheduler and CFP Scheduler. The scheduler’s name tells its functionality. The first stage of the scheduling is the Superframe Scheduler, as shown in Algorithm 2. Superframe Scheduler is responsible for adapting the superframe structure by tuning BO, SO and the length of CAP/CFP, with the intent to serve as many admitted traffic streams as possible. This objective is given a higher priority than the objective of power saving. The algorithm is executed before sending the next beacon frame, assuming there is a newly admitted device in the past beacon interval. The second stage of scheduling employs the CFP Scheduler, which schedules GTSs in the following beacon interval based on the most recent superframe structure as determined by the Superframe Scheduler. The CFP Scheduler is invoked before each beacon frame is sent as GTS allocations may vary in each superframe even if the superframe structure is unchanged. Following we describe how these two schedulers work.  2.4.1  The Superframe Scheduler  In Algorithm 2, Superframe Scheduler calculates parameters BO and SO in steps 2 through 4. In step 5, the length of CAP and CFP are estimated. Based on the maximum allowable CFP length estimated in step 6, it is decided if RT traffic can be accommodated in such a superframe structure, and actions are taken accordingly. After the algorithm converges, step 7 updates the utilization indicator of CAP and CFP for the Admission Controller discussed above. With lowering power consumption in mind, Superframe Scheduler works to fulfill QoS requirements of all admitted traffic. Examining the parameters, BO is the most important one for determining delay, while SO determines the bandwidth and power consumption if BI is fixed. When attempting to meet the QoS requirements, having a longer BI and a smaller SD is advantageous in terms of power consumption because of the increased sleep period, which is the difference between BI and SD. The delay of RT traffic is at least one BI, assuming scheduling a GTS in every beacon interval with the same start slot and the minimum number of slots as required in the standard. Therefore, to fulfill all RT traffic’s time constraints, the smallest RT traffic delay allowed should 28  2.4. Traffic Scheduling  Algorithm 2 Superframe Scheduler //1.Init Initiate aM inCAP Length, BO, SO //2.Determine maximum BO min CF P delay = di of first Ai in admitted traf f ic list RT ; //sorted list BO = argmax(0≤BO≤15) {aBaseSuperf rameDuration × 2BO ≤ min delay CF P }; //3.Determine minimum SO min delay CAP = min(di of first Ai in admitted traf f ic list AC, di of first Ai in admitted traf f ic list CD); SO = argmin(0≤SO≤BO) {aBaseSuperf rameDuration × (2BO − 2SO) ≤ min delay CAP }; //4.Determine superframe structure update BI, SD; sleep period = BI - SD; slot druation = aBaseSlotDuration × 2S O; //5.Determine CAP traffic and maximum allowable CFP slots for each in admitted traf f ic list AC and admitted traf f ic list CD traf f icC AP + = ri × get packet length(bi ) × BI; CAP = max(aM inCAP Length, traf f ic CAP/α); max CF P = min(min CAP delay − beacon period − sleep period, SD − beacon period − CAP ); //6.GTS scheduling for each in admitted traf f ic list RT do delay in BI = f loor(di /BI); num GT S+ = 1/delay in BI; GT S length = ceil(traf f ic CF Pi /β/slot duration); // in slots num slot+ = GT S length/delay in BI; end for if ((num GT S <= 7) AND (num slot <= max CF P )) then CF P scheduler(); else if SO < BO then SO + +; goto step 4; else if BO > 0 then BO − −; SO = 0; goto step 3; else stop RT with lowest priority; goto step 2; end if //7.Update UCAP , UCF P UCF P = sum(traf f ic CF Pi )/(num slotxslot duration); UCAP = traf f ic CAP/(SD − beacon period − num slotxslot duration);  29  2.4. Traffic Scheduling be longer than the time of one beacon interval (see step 2). Assuming traffic in CAP is much lower than the available bandwidth, the largest delay seen by CAP traffic will be the sleep period. As a result, the traffic with the smallest allowable delay in CAP shall be longer than the sleep period. Otherwise, the traffic should be terminated (see step 3). In step 4, the superframe structure is then determined based on the values computed for BI and SD. To estimate the length of CAP, we need to evaluate the average required bandwidth (in symbols) of AC and CD traffic flows in one beacon frame. In step 5, the get packet length() function takes in the average burst size of each traffic flow as the required parameter, and calculates the equivalent length this traffic flow requires in terms of symbols. The function considers lower layer headers, IFS and ACK packets. Summing up the numbers of symbols required gives an estimation of the effective physical layer throughput in CAP. Further, this sum is divided by α to get an estimation of the CAP length required, while taking into account possible backoffs and collisions. After that, the maximum allowable CFP length is simply the length left by CAP and the beacon frame, which also has to fulfill time constraint obtained in step 3. In step 6, where the scheduler checks the ability of scheduling all RT traffic, a similar method is employed to estimate the required bandwidth in symbols. However, in comparison, the bandwidth required is done by summing up the quantized bandwidth requirements of each RT traffic flow, since GTS length has to be a multiple of consecutive time slots. In step 6, get gts packet length() has a similar functionality as that of get packet length() in step 5, excluding the need for the ACK packet. Besides ensuring the estimated bandwidth requirement to be smaller than the maximum allowable CFP length (num slot ≤ max CF P ), the estimated number of all GTS slots assigned in one beacon interval has to be less than 7 (num GT S ≤ 7) according to the standard. The current IEEE 802.15.4 standard applies a FCFS method to schedule GTSs in the CFP, with a maximum of 7 GTSs in each beacon interval. This impedes scheduling new traffic as soon as the number of associated RT traffic flows reaches 7. Improving this using a round-robin approach can definitely enlarge the number of traffic flows accommodated; however, it may fail to meet time constraints [38][31]. Therefore, in our scheduling algorithm, the number of slots obtained by a device requesting an RT service depends on its sampling rate and resolution, as well as on its time constraint.  2.4.2  The CFP Scheduler  The second stage is to schedule GTSs for the next beacon interval after Superframe Scheduler has completed its task. The CFP schedules GTSs according to the computation results of Superframe Scheduler using a greedy policy, i.e., as long as there is slot left for CFP in the next superframe (CF P ≤ max CF P ), a GTS is scheduled for a traffic flow which is in admitted traf f ic list RT , if the traffic flow’s estimated length of GTS has not been met before the next timeout (GT S length > 0), and the total number of GTSs allocated in the next beacon interval is less than 7 (num GT S ≤ 7) (See Algorithm 3). Whenever a new GTS is scheduled, the corresponding GTS descriptor is added to the outgoing beacon frame. Taking the greedy approach rather than spreading the data widely in order to get a smooth throughput is our preferred method because of the need to accommodate potential new traffic, which will in turn impact the current scheduling result.  30  2.4. Traffic Scheduling  Algorithm 3 CFP Scheduler for each in admitted traf f ic list RT do if (− − delay in BI == 0) then // timeout reset delay in BI; reset GT S length; // plus the value left in GT S length before reset end if if ((GT S length > 0)AN D(CF P < max CF P )AN D(num GT S ≤ 7) then slot assigned = the maximum no. of slots can be assigned in CFP. //update CFP, num GTS, GTS length CF P = CF P + slot assigned; num GT S + +; GT S length = GT S length − num assigned; // update GTS descriptor in beacon frame ...... end if end for  31  2.4. Traffic Scheduling  Figure 2.2: A successful admission and scheduling process in the proposed QoS platform. 32  Chapter 3  Proposed Wireless Three-pad ECG System To tackle the problems raised by wireless single-pad ECG system, we propose an upgraded version of it, the wireless three-pad ECG system (W3ECG). Each of the three pads has three electrodes on board. Inspired by the transformation possibility of signals obtained in VCG systems, we bring two more pads to the single-pad ECG system to gain spatial variety of heart activity. The total three pads are uniformly designed so that each can be seen as a rectangular component of the assumed heart dipole vector or its rotated/shifted version. Signals obtained from these three pads, plus the placement information, make it possible to generate conventional 12-lead ECG signals. Details of the rational, data analysis, pad design and software implementation are discussed in this chapter. Evaluations are provided in Chapter 4.  3.1  Electrode, Lead and Pad  Three terms are used in the context of this thesis, electrode, lead and pad. They have distinct meanings as defined below. Electrode An electrode is a physical object, which is attached to the torso and connected with wires to the electronic system. It often comes with gel to enhance the connection between dermis and the wire. However, new technologies have made textile-structure electrodes possible, which can be woven into clothes and working without gel. We refer to only the traditional type of electrode in this thesis. As any ECG signal is the differential potentials recorded on two electrodes, one electrode does not give any meaningful output. Lead A lead is a combination of two electrodes. As discussed in the following section, any combination of two electrodes corresponds to a vector representation in the space of heart activity. In addition to these two electrodes, a typical ECG circuit often has a third electrode placed on the torso to establish a reference voltage.  33  3.2. Heart Dipole Hypothesis Pad A pad is a self-contained board, which has the ECG front-end circuit and two/three electrodes attached. It is also equipped with a wireless radio to enable transmissions of recorded ECG signal to a server in short range.  3.2  Heart Dipole Hypothesis  The transformation from signals obtained in VCG systems to conventional 12-lead ECG signals are based on the heart dipole hypothesis, and so is W3ECG’s transformation. Below we first review the heart-vector projection theory [16]. Then by analyzing the data set available, we derive the practical equations for transformation from W3ECG signals to 12-lead ECG signals.  3.2.1  Heart-vector Projection Theory  It is widely accepted that the electrical heart activity can be modelled as a stationary dipole [34][35]. Under this assumption, the heart dipole moment is a function of time and represented by a vector in the three-dimensional space as p = px x + py y + pz z  (3.1)  where x, y and z are standard unit vectors of a rectangular coordinate system, e.g., x and y axes span the frontal plane, x and z axes span the transverse plane, and y and z axes span the sagittal plane. In Equation 3.1, px , py and pz are projections of the heart dipole moment p on three axes, representing scalar components of p. This is illustrated in Figure 3.1. Elements of p are in units of mA × cm. The potential Vi (subscript i is for distinguishing torso locations) p produces on the torso is the multiplication of p and a resistive component ci . ci is a function of shape, size, and characteristics of the medium, the position of the dipole as well as where the potential is measured (position of electrode on torso). Assuming homogeneity and thus linearity of the medium, the electric potential appearing at any torso location is represented as the projection of p on ci as Vi = ci · p  (3.2)  where ci is called the lead vector with units of Ω/cm. For example, as shown in Figure 3.1, the lead V1 in 12-lead system can be expressed as VV 1 = cV 1 · p  (3.3)  Unipolar and Bipolar Leads The above expressed potential Vi is supposed to be the potential difference between where the electrode is placed on the torso and the midpotential of the heart dipole. The corresponding lead ci  34  3.2. Heart Dipole Hypothesis     ✂  ✁  Figure 3.1: Illustration of the heart vector and its projection on a lead vector.  is called a unipolar lead. The lead whose potential represents the differences between two unipolar leads is called a bipolar lead. Although Equation 3.2 is for unipolar leads, the same expression applies to bipolar leads, if cV 1 is replaced by the corresponding lead vector of a bipolar lead. Coordinate System Above, we represent the heart vector and lead vector in Cartesian space. However, they can be represented in any three-dimensional rectangular coordinate system, and the measured potential difference is independent of it, as indicated in Equation 3.2. Since a unit vector is defined by two points, a minimum of four points are necessary for defining the space. In other words, the minimum number of electrodes put on torso is four, in order to fully register the cardiac activity in three-dimensional space. This is the essence of the VCG approach. Further, the orthogonality of three unit leads is a sufficient condition, while linear independence is the necessary condition. Take the EASI lead system for example. It specifies 5 electrode locations on the torso (node E, A, S, I and the reference), which makes 3 bipolar leads (ES, AS, AI). Distances between these 5 electrodes are generally a couple of centimeters, and electrodes are connected by wires.  35  3.2. Heart Dipole Hypothesis  3.2.2  Analyzing Input Data  The Objective Since most cardiologists are well-trained to accept ECG information in the conventional 12-lead format, it is desirable for an ECG system to render such standard information for interpretation. However, this capability is not present in current wireless single-pad ECG systems. Since they only register cardiac activity in one direction, it is not possible to display signals in the 12-lead format. If we use the VCG’s approach, we can render 12-lead ECG information, but we lose the benefits of wireless pads. Even placing 4 (5 for an EASI lead system) electrodes on the torso with wires in between is cumbersome, especially for long-term monitoring. Plus the wires can cause artifacts and involve 50/60 Hz noise [51]. Our goal is to design a wireless ECG system which provides standard 12-lead ECG information without wires between electrodes. To eliminate the wires, one way, is to attach 4 electrodes to one pad, and make the pad as small as a typical electrode. When all electrodes are placed close to each other, however, spatial variety of leads can be lost and it makes the leads highly dependent on each other. Another way, if we separate these electrodes, we can trade complexity for spatial variety by separating electrodes and embedding them into different pads. Particularly, one lead is one pad with two/three electrodes. This requires introduction of at least two more pads to achieve spatial variety and dealing with consequent complexity. We need to ensure linear independence between the three corresponding leads. Input Data Two sets of data are obtained from Dalhousie University [22][11]. The first set includes 352 lead vectors for the 352 nodes on the display of Figure 1.3 derived from computer simulations. The second set includes coefficients for predicting the 352 nodes from EASI leads obtained through practical experiments and interpolations. Essentially, these two sets of data are representations of same leads in two coordinate systems: one is the heart dipole rectangular coordinate system as shown in Figure 3.1; the other is the non-rectangular coordinate system spanned by leads ES, AS and AI. This data is used to derive the equation for transformation.  3.2.3  Derivation of Transformation Equations  Based on Lead Vectors If the lead vectors for three pads are cA , cB and cC , according to Equation 3.2, the potentials measured at the pads can be represented as in Equation 3.4 - Equation 3.6. VA = cA · p  (3.4)  VB = cB · p  (3.5)  VC = cC · p  (3.6)  36  3.2. Heart Dipole Hypothesis Combining the three measured potentials into a vector overlineV3 , and three lead vectors into a 3 × 3 lead matrix c3 , we have Equation 3.7. This is illustrated in Figure 3.2. V3 = c3 p  (3.7)     ✂  ✁  Figure 3.2: Illustration of the heart vector and three lead vectors.  When the three lead vectors are linearly independent, the heart vector can be computed as p = c−1 3 V3  (3.8)  The heart vector moment p can be computed from the three lead vectors if and only if c−1 3 exists and is obtainable. It is not possible to fully represent cardiac activity by any single lead. At least three are necessary. Once the heart vector moment is known, the 12-lead standard potentials can be computed by projecting p onto the corresponding 12-lead vectors. Take lead V 1 for example. Combining Equation 3.3 and Equation 3.8, lead V 1 can be computed as in Equation 3.9. VV 1 = cV 1 c−1 3 V3  (3.9)  The same approach can be used to compute the other eleven leads in the conventional 12-lead system, which can be expressed in vector format as in Equation 3.10, where each row in c12 is  37  3.2. Heart Dipole Hypothesis the lead vector for one lead in a 12-lead system, and the corresponding row in V12 is the electric potential of that lead measured. V12 = c12 c−1 3 V3  (3.10)  V12 = T V3  (3.11)  We rewrite Equation 3.10 as  We define T in equation 3.11 as the transformation matrix, representing the transformation coefficients between two vectors, V3 and V12 . The lead vectors needed for the matrix c12 in Equation 3.10 are directly available in the computer-simulation-based data set as mentioned above. The lead vectors for the three pads to compute c−1 3 in Equation 3.10 are calculated by substracting two corresponding lead vectors, each of which is one node in the same data set. Equation 3.10 provides a method of generating standard 12-lead ECG signals from 3-lead ECG signals. Specifically, if three electric potential traces are recorded at the three leads, traces for all 12-lead ECG can be synthesized. Based on Coefficients for Predicting Nodes from EASI Leads Similar to equation 3.11, for an EASI lead system, we have Equation 3.12, where U12 is equivalent to T in Equation 3.11, representing transformation coefficients from EASI leads to the 12 leads. V12 = U12 VEASI  (3.12)  Equation 3.12 can be extended to express the transformation from EASI leads to our three-pad leads as in Equation 3.13. V3 = U3 VEASI  (3.13)  Combining equation 3.12 and equation 3.13, results in Equation 3.14. V12 = U12 U3−1 V3  (3.14)  Equation 3.14 is in the same expression as Equation 3.10, but with different notations. The transformation matrices U12 and U3−1 in Equation 3.14 are counterparts of c12 and c−1 3 in Equation 3.10 respectively. In essence, U12 and U3−1 represent the corresponding transformation coefficients in the coordinate system spanned by EASI leads.  38  3.3. Wireless Three-pad ECG System  3.3  Wireless Three-pad ECG System  3.3.1  Architecture Selection  Generally, two categories of system architecture exist in WBAN systems, as summarized in Chapter 1. The two-tier approach is a better choice for W3ECG when multiple systems are interconnected, because of the following reasons: • Benefits of a wireless ECG system stand out during ambulatory monitoring. To cover an unlimited area, the two-tier approach is a must. • The data rate of an ECG channel is relatively large in WBASN applications. Introducing two more channels than a wireless single-pad ECG system has can saturate the already limited bandwidth of a one-tier approach. The performance evaluations of ZigBee’s support for multiple WBASNs in a hospital environment (see Chapter 4) supports this. • Synchronization, which is required during W3ECG implementation, can be more easily realized at the local tier of the two-tier approach. • Employing a two-tier approach, signal processing and transformations can be done at each user’s local personal server. This reduces computing intensity at the central server and possibly the aggregate data rate of each user. Topology The local tier of each W3ECG system is based on our proposed WBAN QoS provisioning framework and therefore is organized in a star topology. This topology configuration makes it easy to utilize the framework for both QoS provisioning and synchronization purposes. As also specified in Chapter 2, a personal server coordinates communications among WBAN nodes. Data can be stored locally at the personal server, forwarded to caregivers’ central server, or interpreted for real-time analysis. 12-lead ECG signals are synthesized whenever needed.  3.3.2  Pad Design  There are various implementations of wireless single-pad ECG system as mentioned in Chapter 1. It is impossible to have an identical interpretation of signals collected from all these implementations. However, as in a W3ECG system, cross-processing of signals obtained from each pad is necessary, we need to standardize the pad design in the following aspects: • ECG front-end circuitry (common-mode noise reduction and degree of amplification). • Voltage reference. • Distance between electrodes on each pad. • Synchronization of sampling of ECG signals on three pads. 39  3.3. Wireless Three-pad ECG System Following we explain our implementation of the front-end circuit and its connection to a radio platform TelosB. ECG Front-end Circuit The prerequisite of any ECG circuitry is to have a high common-mode rejection ratio (CMRR) and a low offset voltage to reduce the effect of 50/60 Hz noise. This is not difficult because most off-the-shelf instrumentation amplifiers can easily achieve a CMRR of 80 dB and relatively low offset voltage. In addition to that, portable ECG circuitry imposes another requirement, lowpower consumption. In other words, it is desirable that the amplifiers work with a low voltage supply and drain small amount of currents. Based on these criteria, Texas Instruments’ instrumentation amplifier INA333 and operational amplifier OPA333 are chosen for our ECG circuitry [46]. The INA333 is reported to be the industry’s lowest power zero-drift instrumentation amplifier. It operates with a power supply as low as 1.8V and quiescent current of 75 uA, and features a minimum CMRR of 100 dB with a 25 uV offset voltage. The OPA333 features a 1.8 V power supply and very low offset voltage (10 uV). Figure 3.3 presents the schematic of the ECG front-end circuitry. It consists of five blocks, namely a separation block, a common-noise reduction block, an amplification block, a feedback block and, a voltage supply block. The circuit is based on the reference design provided in INA333’s data sheet. However, instead of using three electrodes as inputs and one as feedback, we have two inputs and one feedback. The separation block has two voltage followers, realized by an OPA2333 chip. The common noise reduction block is mainly the INA333 instrumentation amplifier, which reduces the common-mode 50/60 Hz noise and amplifies the differential signal with an amplification of 5. The amplification block is again realized by an OPA2333 chip, and amplifies 200 times the output of the INA333. It also provides feedback to INA333 to adaptively tune the voltage reference level, and combines a first-order filter with a cut-off frequency of 150 Hz. The feedback block sets the body at the appropriate voltage level, and attenuates the 50/60 Hz noise on the body. As there is no actual ground for this portable pad, the voltage supply block divides the voltage supply (coin cell) by two and provides the middle point as the virtual ground to reference the whole front-end circuit. Figure 3.4 shows the PCB layout of our front-end circuit, and Figure 3.5 is a snapshot of the manufactured prototypes. The four items in Figure 3.5, from left to right, are the circuit’s front side, back side, back side with mount buttons, and back side mounted with electrodes respectively. The front-end is interfaced with a TelosB platform to enable radio communications, as shown in 3.6. The combined prototype is what we defined as a ”pad”. Figure 3.6, shows a front and side view of the pad. Calibration of Voltage Reference W3ECG requires that each pad have the same amplification, so that signals can be combined with different weights at the personal server. This assumes that each pad is supplied with exactly the same voltage level over time. However, pads must be battery powered. They do not share a 40  3.3. Wireless Three-pad ECG System common power source, which makes the cross-pad data processing difficult. As batteries discharge over time, the voltage supply of each pad decreases. It is necessary to regulate the reference voltage supply to the ADC on each pad. This can be achieved by using a voltage regulator on every pad. Many microcontrollers are manufactured with a built-in voltage regulator. The output voltage of this regulator is lower than that of which powers the chip, e.g., MSP430F1611 on TelosB provides two reference voltages: 1.5 V and 2.5 V, which requires a minimum power supply voltage of 2.2 V and 3 V respectively. As a result, we choose 1.5 V as the reference to the internal ADC of the MSP430F1611. Standardization of Distance between Electrodes According to Figure 1.3, from neck to waist, the numbered transverse levels are 1 inch apart, and the positions around transverse sections are supposed to be equiangularly divided. Since the average waistline of an adult is 37 - 39.7 inches [6], the estimate of the distance between two adjacent nodes in the transverse plane is 1.15 - 1.24 inches, as calculated in Equation 3.15, where 32 is the number of labelled vertical lines in Figure 1.3. distance = waistline ÷ number of vertical lines  (3.15)  In order to standardize the distance between two input electrodes, irrespective of the placement direction, we round the distance in the model to 1 inch for both directions. As a result, the distance between two input electrodes shall be a multiple of this minimum distance. The distance has been determined to be 2 inches for our prototype, considering both the size of a TelosB board (the TelosB board is 2.5 inch in length), and the desire to gain spatial variety. On one hand, the smaller the pad is, the more comfortable it is; on the other hand, the farther away the two input electrodes are, the larger magnitude it records (larger distance corresponds to a better defined lead vector in the heart vector domain, as in Figure 3.2, which is more tolerant to placement noise). The 2-inch decision is a tradeoff of the above two. Figure 3.5 shows the size of the pad and the distance between two input electrodes. The feedback electrode is arranged at a location to allow enough separation between the two other electrodes. It is not necessary to put the feedback electrode exactly in the middle of the layout, as the circuit will be measuring a differential signal from the two input electrodes. Synchronization of Sampling Aside from the above standardizations, the need to process data from three pads also imposes the requirement to synchronize the sampling process at each pad, so that the values from three pads can be combined to synthesize 12-lead signal values, presumably recorded at the same sampling moment. Equation 3.11 and Equation 3.14 also explains that the transformation of leads are based on the potentials recorded at the same moments on each pad. The synchronization can be achieved by starting the sampling at each pad at the same moment. Due to the drift of individual crystal on each pad, re-synchronization is needed periodically. Having IEEE 802.15.4 beacon-enabled mode as the communications protocol, the time to start sampling 41  3.3. Wireless Three-pad ECG System and re-synchronization can be indicated by adding a payload in the beacon frame. The IEEE 802.15.4 superframe is an ideal structure for synchronizing both the communications and sampling processes. The reason to have the indications in the beacon-frame payload is that each node is required to be active for receiving a beacon frame; if it’s broadcasted in the CAP, the nodes may be asleep during that period.  3.3.3  Pad Placement Positions  Since the pad design has been standardized, they share the same amplification factor, and the signals registered at three pads can be represented as in Equation 3.16, where amplif ication = 1000, and ref erence voltage = 1.5(V ) (the ADC in MSP430F1611 is 12-bit). V3 = V3 · amplif ication ·  212 − 1 ref erence voltage  (3.16)  While the pads are aligned to the node positions in Figure 1.3, Equation 3.10 and Equation 3.14 can be used to synthesize the 12-lead ECG signal from observations on the three pads. The only requirement is that the corresponding vectors are linearly independent of each other in the corresponding domain, i.e., Equation 3.10 corresponds to the domain spanned by heart vectors, while Equation 3.14 corresponds to the domain spanned by EASI leads. Two Approaches The pad placement is crucial in W3ECG as it defines the transformation matrix. Two approaches are suggested when instructing where to place the pads: • Suggest one set of placement locations, and instruct the caregiver/patient to place the pads at exactly these locations. The popularity of EASI leads suggest the need to provide locations that can be easily identified in anatomy. In this case, the transformation matrix is unique and pre-calculated, and linear independence shall be ensured. • Suggest sets of areas for placement, and instruct the caregiver/patient to place the pads at convenient locations. This approach caters to the need of an operation, female users, or comfort of different individuals. A localization mechanism is needed to determine the pad placement locations during deployment, and tune the coefficients in the transformation matrix in real-time. The user shall be notified if linear independence is lost. Following, we study the first approach only. The localization mechanism required for the second approach is out of the scope of this thesis. Suggested Placement Locations In an EASI lead system, the electrodes are placed at position 190(E), 198(A), 84(S) and between 165&207(I) in Figure 1.3. Three leads are ES, AS and AI. Referring to the lead directions of the EASI lead system, and considering the importance of V1 to V6 electrodes in 12-lead system, we 42  3.4. Software Implementation design the combination of pad placements as specified in Table 3.1. Figure 3.7 illustrates the actual placement locations on the body. Pad Pad Pad Pad  Number 1 2 3  Positive Input 109 170 178  Negative Input 107 127 177  Table 3.1: Suggested W3ECG pad placement locations. Ideally, each of three pads is aligned with one axis of the heart vector. However, human organs and skin are not linear resistive components, which makes the alignment to the three axes difficult. The attempt to achieve orthogonality of three vectors leads to the above three placement locations, i.e., pad 1, pad 2 and pad 3 are estimated directions of x, y and z in Figure 3.2. As a result of the above placements, we obtain U3 as −0.0974 0.2463 U3 =  0.4385  0.1014  0.0837 −0.0872  −0.2224 0.3478 −0.5245 U3 has a condition number of 2.6069 (EASI lead system has a condition number of 1.0162), which indicates a relatively stable linear transformation. In other words, it is tolerant to placement errors.  3.4  Software Implementation  The W3ECG software consists of three parts, which are all based on TinyOS platform. Two of the three are nesC codes for TelosB motes (one for TelosB on the pad side, the other one for TelosB on the server side), and one is Java code for the server.  3.4.1  Sampling and Radio Communications on The Pad  The nesC codes on each pad and on the sink are based on the existing implementation of the beacon-enabled mode of IEEE 802.15.4 in TinyOS [20]. On each pad, sampling and data storage functionalities have been added. Following we describe the configuration of the ADC and the design of the data buffer in detail. ADC Configurations The ADC equipped with MSP430F1611 chip on TeloB mote provides 12-bit resolution, and can be configured in four different modes. The single-channel-repeat mode is chosen because on each pad, one channel is used, continuously. The ALCK clock has is soured to provide a more accurate sampling frequency, which is set to 250 Hz. The event SingleChannel.multipleDataReady() is triggered whenever there are 16 samples available, which is the current maximum allowable value. All 43  3.4. Software Implementation the configurations are done through TinyOS interface Msp430Adc12SingleChannel of component Msp430Adc12ClientAutoRVGC(). Design of Data Buffer The data buffer is designed to temporarily store ECG samples before getting transmitted. It is separated from the payload buffer of the TinyOS packet. In other words, ECG samples are moved from the data buffer to the payload buffer, and then forwarded to the radio. Cyclic array has been chosen as the data structure for the data buffer, and two pointers (write and read) are assigned to provide synchronized access to the buffer. The synchronization is achieved by using the atomic keyword for operations involving the two pointers. Each time the event for new ADC samples is triggered, the samples are stored to the buffer with the write pointer as the offset. As the event is triggered for every 16 samples, the offset specified by either the write or the read pointer is a multiple of 16.  3.4.2  Beacon-frame Generation and Data Forwarding on The Sink  The sink node, known as the PAN coordinator in IEEE 802.15.4, generates a beacon-frame at the beginning of every superframe, and accepts association from all three pads, which act as End Devices in IEEE 802.15.4. This has been realized in exiting the code base. We added the implementation to enable serial communications of commands between the sink and server. Whenever the sink acknowledges association, a message is created to notify the server of this decision and the corresponding pad ID. As discussed above, we also implemented the functionality to instruct all pads to synchronize sampling processes, by adding specific beacon-frame payloads. This is triggered by a command from the server end through a serial connection. Everytime the sink node sends a beacon-frame, it checks weather there is a need to indicate in the payload to start or synchronize the sampling processes on the pads. If so, it sets the corresponding fields and transmits the beacon-frame. As soon as it is done, these fields are reset. ECG data packets’ payload from each pad will be wrapped up in a TinyOS frame at the sink node, and forwarded to the server over the serial link. No processing on the data is done at the sink. When wrapping up the payload data, the length, AM type fields in the serial TinyOS frame are set through the AMSend interface of TinyOS component SerialActiveMessageC. The AM Type is used to differentiate this packet from the above one for the notification of association.  3.4.3  Graphical User Interface and MySQL Database on The Server  The server runs a graphical user interface (GUI) implemented in Java. The implementation is based on the the TinyOS Oscilloscope application. We polished the GUI so it can display three ECG graphs at the same time, and added tabs to switch the monitoring screen between different channels. When the server has been notified of associations from all the three pads, it pops up a window and asks if the user wishes to start the system. The system can also be started by clicking the 44  3.4. Software Implementation ”Start” button on the left bottom of the GUI, if the user decides to deploy less than three pads for simpler monitoring. See Figure 3.8. A running server displays a limited duration of ECG graphs on the screen, but stores all samples in the MySQL database through Java Database Connectivity interface. The database has three tables, each of which registers data from one pad. Each table has a column for the sampling time, and a column for the potential value.  45  3.4. Software Implementation  Figure 3.3: Schematic of ECG front-end design.  46  3.4. Software Implementation  Figure 3.4: PCB layout of ECG front-end circuit.  Figure 3.5: Snapshot of the manufactured ECG front-end circuits.  47  3.4. Software Implementation  Figure 3.6: ECG front-end circuit interfaced with TelosB platform.  48  3.4. Software Implementation  Figure 3.7: Placement positions of electrodes for RA, LA and LL, and three sets of electrodes for W3ECG system.  49  3.4. Software Implementation  Figure 3.8: A snapshot of the W3ECG server GUI demonstrating that three pads are synchronized and transmitting ECG signals.  50  Chapter 4  Performance Evaluations 4.1  Performance Evaluations of ZigBee for WBASN  In this section, we present our simulation studies of ZigBee’s support for multiple WBASNs sharing the same channel. The objective of this work is to evaluate the reliability of ZigBee/IEEE802.15.4 protocols when accommodating relatively high-data-rate applications, e.g., W3ECG, for multiple users. This lays the foundation for selecting the architecture/topology of our W3ECG system, as discussed in Chapter 3. The OPNET Modeler 14.0 with ZigBee model is used [39]. In the simulations, we evaluate the system’s average reception ratio, throughput, latency and fairness of WBASN sensor nodes by varying the number of WBASNs and the distance from the WBASNs to the gateway.  4.1.1  Simulation Setup  We used a 100 m x 100 m area to simulate one floor of a hospital, as shown in Fig. 4. At the centre of this floor sits the ZigBee coordinator. It also acts as gateway server, collecting data from different WBASNs. Routers are equally distributed (20 meters apart) across this floor. Transmit power of devices is configured based on the datasheet of CC2420 radio chip [47], to reflect different roles in the network, i.e., end devices transmit at a lower power level and thus last for a longer time. In order to mimic practical deployment of routers along hallways and simulate the signal propagation along them, path loss model and radio sensitivity are specifically configured. Although it is an square area, the signal is more likely to reach the most nearby neighbor routers, and thus travels perpendicularly or horizontally as in the figure shown in [5]. In order to simulate a WBASN, each five end devices are grouped together at one location. They represent sensor nodes, supposedly worn by one person. Details of simulation parameters are listed in Table 4.1. Table 4.2 lists simulated applications for sensor nodes and corresponding data rates. As ExG sensors generate relatively larger data rates, in the following simulations we evaluate the performance by including and excluding these sensor nodes respectively, to see their effects.  4.1.2  Average Reception Ratio  First, we evaluate the average reception ratio of the whole network. For all scenarios, we position each WBASN along one route to the gateway. For each scenario, all WBASNs are positioned with an equal distance to the gateway, with the intent to set the same number of hops for them. Devices are not mobile in these cases.  51  4.1. Performance Evaluations of ZigBee for WBASN Parameter Simulation Area Number of Coordinator Number of Router Tx. Power (coordinator/router) Tx. Power (end device) Maximum Routers per Device Maximum Children per Device Maximum Network Depth Simulation Time  Value 100 m x 100 m 1 20 1 mW (< 25m) 0.05 mW (< 20m) 8 30 5 10 minutes  Table 4.1: ZigBee simulation parameters. Application Control Respiration Temperature ExG1  Data Rate 200 bits/0.5 s 200 bits/0.5 s 200 bits/1 s 800 bits/0.2 s  Table 4.2: ZigBee application traffic parameters. Figure 4.1 depicts the average reception ratio for including ExG sensor nodes and Figure 4.2 for excluding ExG sensor nodes. Both of the figures exhibit a trend toward smaller average reception ratios when there are more WBASNs in the system. This is as expected since more WBASNs (more sensor nodes) are competing for the limited bandwidth. Average reception ratio is largest, i.e., nearly 100%, when WBASNs are directly connected to the gateway and tends to be stable when WBASNs are further away. As in multi-hop cases, routers have to compete with upstream and downstream peers, resulting in collisions and packet drops; on the other hand, more hops in the middle does not have a negative effect as throughput reaches the network bandwidth and every WBASN has a steady data rate. Comparing Figure 4.1 and Figure 4.2, it is clear to see that high-data-rate sensors (ExG sensors) degrade performance when there is more than one WBASN competing for the bandwidth. About 20% to 40% more packets are dropped when ExG sensors are included in each WBASN.  4.1.3  Throughput  Complementing the average reception ratio, Figure 4.3 and Figure 4.4 depict the throughputs for the above two WBASN setups. When WBASNs are directly connected to the gateway, like a star network, almost all packets reach the destination, exhibiting a throughput up to 20 Kbps. However, when one more hop is involved between the gateway and WBASN, network throughput degrades to 5 Kbps (Figure 4.3) and 1.5 - 3 Kbps (Figure 4.4). The throughputs of both scenarios tend to be stable. These observations comply with those for average reception ratio. For multi-hop scenarios, we note there is only a minor difference in throughput when different numbers of WBASN are in connection. This is likely due to the limitation of throughput for the last hop of routes, i.e., the routers one-hop away from the gateway, because they experience more severe competition in 52  4.1. Performance Evaluations of ZigBee for WBASN  Figure 4.1: Average reception ratio for WBASNs with ExG sensors.  bandwidth.  Figure 4.3: Throughput for WBASNs with ExG sensors.  53  4.1. Performance Evaluations of ZigBee for WBASN  Figure 4.2: Average reception ratio for WBASNs without ExG sensorss.  Figure 4.4: Throughput for WBASNs without ExG sensors.  4.1.4  Latency  In this subsection, we examine the average delay of the ZigBee network. Simulation results are shown in Figure 4.5 and Figure 4.6. 54  4.1. Performance Evaluations of ZigBee for WBASN  Figure 4.5: Delay for WBASNs with ExG sensors.  Figure 4.6: Delay for WBASNs without ExG sensors.  The results show an average delay from 0.02 to 0.25 second, which is acceptable to most WBASN applications. In both figures a larger number of hops results in a longer delay, which is expected due to the time taken for each router. More WBASNs in connection in the system generally exhibit 55  4.1. Performance Evaluations of ZigBee for WBASN larger average delays but the difference is only prominent between scenarios with one WBASN and multiple WBASNs. Also, when comparing the two figures, we note there is little difference for the same WBASN and multi-hop setup. These are due to the disconnections of WBASNs and consequently the loss of packets. More WBASNs with higher data rates result in frequent disconnections of WBASNs and huge packet loss. These dropped packets, spend much less time on route compared to those successfully delivered packets, contribute to smaller averaged delay.  4.1.5  Fairness  After investigating the overall system performance, in this subsection, we evaluate the fairness across the network. Without any QoS protocols implemented in the ZigBee network, it is important to see whether there is any bias toward certain sensor nodes in one WBASN or for certain WBASNs in a system. Specifically, we compare the average reception ratio and throughput of each sensor node for each network setup. Two scenarios are studied, both of which have four WBASNs in the network. One has all WBASNs positioned two hops from the gateway; the other one set of four WBASNs at different distances, i.e., one to four hops respectively, from the gateway. In order to better evaluate the fairness under a situation where the whole network performs relatively well, ExG sensors are not included in these simulations (otherwise, in the case of a four-WBASN two-hop deployment, overall network reception ratio is near 20% as shown in Figure 4.1 and Figure 4.2). Figure 4.7 to Figure 4.10 illustrate the results.  Figure 4.7: Average reception ratio for sensors of WBASNs positioned two hops away.  56  4.1. Performance Evaluations of ZigBee for WBASN  Figure 4.8: Average reception ratio for sensors of WBASNs positioned different numbers of hops away.  Figure 4.9: Throughput for sensors of WBASNs positioned two hops away.  57  4.1. Performance Evaluations of ZigBee for WBASN  Figure 4.10: Throughput for sensors of WBASNs positioned different numbers of hops away.  We observe in the above four figures, that the same kinds of sensors from different WBASNs perform comparably when all WBASNs are positioned the same distance away from the gateway. Control, respiration and temperature sensors show a similar average reception ratio (temperature sensors have a slightly smaller average reception ratio), which is larger than that of the blood sensors. This is explained by the longer application data payload blood sensors have, which leads to more collisions and packet loss. While having almost the same average reception ratio, temperature sensors are noted to have half of the throughput, which can be traced back to the less application data rate generated. Comparing the figures (Figure 4.7 to Figure 4.10), it is clear that all sensors from a one-hop WBASN exhibit better performance with 20% to 30% more average reception ratio and 30 to 150 bps more throughput. The same kind of sensors from different WBASNs with varying number of hops (two to four hops) exhibit the same performance as the other scenario.  4.1.6  Discussions  To conclude, we note that the one-hop WBASN has obvious privilege over multi-hop WBASNs in a ZigBee network. The number of WBASNs in the network plays an important role in determining the average reception ratio, and the application data rate is crucial in determining throughput. There is no QoS guarantee in ZigBee for WBASN sensors during random deployment. However, to realize our W3ECG system and ensure reliability, a QoS framework is necessary.  58  4.2. Performance Evaluations of Proposed QoS Provisioning Platform  4.2  Performance Evaluations of Proposed QoS Provisioning Platform  The performance of our proposed admission control and scheduling algorithms is evaluated by extensive computer simulations. The OPNET 14.5 modeler with an open source implementation of the IEEE 802.15.4 beacon-enabled model is used [45]. We implemented our algorithms based on this model. Functions for association request and response were added to the model, and reserved fields in these command packets were used for passing admission requests and responses. The FCFS GTS scheduling scheme was modified according to Algorithm 3.  4.2.1  Simulation Setup  According to our service differentiation method, three categories of traffic are introduced to the network. The number of flows in each traffic category and the respective QoS vectors are specified in Table 4.3. These values come from the responses to the IEEE 802.15.6 Call for Applications [25], and our WBASN system implementation experience. Particularly, AC traffic is representative of traffic generated by a device such as pace maker, heart rate monitor, or fall detection sensor. CD traffic is a representation of traffic generated by a device such as a pulse oximeter, glucose sensor, blood pressure sensor, respiration sensor, or a watch or cell phone working as a remote controller. Packet arrivals for AC and CD traffic follow the exponential distribution with the average packet rate as specified in Table 4.3. RT traffic is representative of data generated by ECG/EEG and motion sensors. We assume that each IEEE 802.15.4 end-device is dedicated to only one sensor/actuator type, and thus it generates only a particular traffic type. Priority  Delay  AC  0.1 s  AC  0.2 s  CD  0.5 s  RT RT  0.2 s Uniform (0.5-1.0s) 0.5 s  RT  Packet Rate /Sampling Rate Exponential (mean: 0.01 Hz) Exponential (mean: 0.1 Hz) Exponential (mean: 1 Hz) 250 Hz 2 Hz  Payload Size/ Resolution 64 bits  No. of Traffic  64 bits  1  16 bits  10  12 bits 12 bits  6 6  20 Hz  36 bits  6  1  Table 4.3: QoS application traffic parameters. Following, we evaluate the number of admitted traffic flows for each category, the overall network throughput, the energy consumption, the time constraint compliance ratio, and the average latency. Evaluation of our simulation results reveals that varying value of α has an insignificant effect on the final results due to a limited number of AC and CD traffic flows simulated. This is realistic in a typical WBASN setup, while it is less applicable to scenarios where CAP is easily saturated. 59  4.2. Performance Evaluations of Proposed QoS Provisioning Platform In the following results we show the performances with varying β only (for α = 0.8, 0.9, 1.0, the simulations exhibit the same result).  4.2.2  Number of Admitted Traffic Flows  Figure 4.11 shows the number of admitted traffic flows vs. β (ranging from 0.1 to 1). We observe that for all values of β, all AC/CD traffic can be admitted. In comparison, small values of β (0.1, 0.2) results in a low admission rate for RT traffic. This is expected because AC and CD traffic flows are well within the available bandwidth, while admission decision for RT traffic is largely dependent on β. When β is too small, a larger bandwidth is reserved for AC and CD traffic, which has limited benefits because of the relatively low bandwidth requirements of AC/CD flows in the scenario under consideration.  Figure 4.11: Number of admitted traffic flows in QoS framework simulation.  4.2.3  Throughput  However, if we look at the network throughput in Figure 4.12, scenarios with a larger number of admitted RT traffic do not necessarily exhibit better performance in terms of throughput. In this Figure, the overall throughput of the network is shown together with the separate throughput of CAP and CFP. As designed, RT traffic flows in CFP contribute the most to the overall throughput (above 95%). While the CAP traffic generates the same throughput in all cases (around 200 bps), CFP throughput increases and then saturates (at around 22 Kbps) when β exceeds 0.4. Since β is a rough estimation of the CFP utilization, these results indicate a relatively low utilization value. 60  4.2. Performance Evaluations of Proposed QoS Provisioning Platform This is because of the IFS needed, and the effect of restricting traffic flows to time slots. This effect is more severe when each time slot (slot duration), determined by the Superframe Scheduler, is relatively small. Nevertheless, the low utilization mirrors former research results [52][31], although a different definition for utilization exists in this thesis as explained in Chapter 2.  Figure 4.12: Throughput in QoS framework simulation.  4.2.4  Energy Consumption  In order to quantify the energy needed by our QoS platform, the power model of MicaZ [8] has been chosen to compute the energy consumption of the WBASN network over one minute of simulation time. The result is presented in Figure 4.13. We note that when β varies from 0.1 to 0.3, energy consumption increases due to more admitted RT traffic flows in the network. This is congruent with the above discussions. We also observe that when β increases from 0.3 to 1, the consumed energy drops. This is because of the higher efficiency in the transmissions of the application-layer’s data payload. When β is relatively smaller, admitted RT traffic flows are assigned larger GTS lengths. After sending out the queued application data within the time constraint, there will still be time left in the same GTS allocation. As a result of our greedy approach, application data arriving at a high rate is sent out immediately, which tremendously increases MAC and physical layer overhead, as well as the number of IFS involved. Compared to the typical size of a burst (12-36 bits), the MAC and physical layer overhead and IFS period, taken together, is more than three times its size (about 120 bits).  61  4.2. Performance Evaluations of Proposed QoS Provisioning Platform  Figure 4.13: Energy consumption in QoS framework simulation.  4.2.5  Constraint Compliance Ratio  Figure 4.14 illustrates the results of meeting the time constraints of the simulated traffic sources. For CAP traffic from AC and CD sources, the result is perfect. Regardless of β, CAP traffic shows a 100% time constraint compliance ratio. CFP traffic from RT sources also experiences good performance with the time constraint compliance ratio decreasing from around 99.9% to 99.2% when β increases from 0.1 to 1. Although more RT traffic flows are admitted as β gets larger, they could not be scheduled, or had to compete for the limited number of slots in CFP. This leads to relatively more traffic flows missing their time constraints.  62  4.3. Performance Evaluations of The W3ECG System  Figure 4.14: Constraint compliance ratio in QoS framework simulation.  4.2.6  Latency  Besides meeting time constraints with a very good ratio, the network serves all traffic flows with a low average queuing delay. In this paper, queuing delay is defined as the difference between the time a packet is generated by the higher layer application, and the time it gets transmitted or dropped. For packets in CAP that require ACKs, the queuing delay takes into account the delay while waiting for ACK packets. We observe in Figure 4.15 that CAP traffic experiences an average of about 0.02 s latency in the queue, while that for RT traffic varies between 0.05 s to 0.1 s. The slight differences for different β values can be explained by the varying superframe structure because of the changing β value. By observing that traffic flows have similar average queuing delay when β ranges from 0.5 to 0.6, or from 0.7 to 1, we support our claim for reduced energy consumption observed in Figure 4.15.  4.3  Performance Evaluations of The W3ECG System  The evaluation of the W3ECG system consists of three parts. First, we used a commercial patient monitor to study and verify the effects of putting two input electrodes close to each other (2-inch distance). Secondly, we simulated the signals that would appear on the three pads if we were able to place them on the locations specified in Table 3.1. Lastly, we collected data from our manufactured pads and assessed our hardware and software implementations 1 . 1  All the experimental studies have been conducted on the author’s body.  63  4.3. Performance Evaluations of The W3ECG System  Figure 4.15: Latency in QoS framework simulation.  4.3.1  Experimental Study Using Commercial Patient Monitor  A commercial patient monitor (Vital-Guard 450C) has been employed to study the features of a typical ECG graph. The focuses are the P, Q, R, S and T waves, P-R, S-T and Q-T intervals, and the changes of these features when two input electrodes are moved closer and placed at specific locations (Table 3.1). The analog output of the patient monitor is connected to a data acquisition board (Measurement Computing USB-1408FS), which is connected to a personal computer to record the data. Unfortunately, Vital-Guard 450C provides only one output channel (lead). So recordings must be done one by one. Figure 3.7 shows the placement of electrodes RA, LA and LL for a 12-lead ECG system, and the three sets of electrodes for the W3ECG system. The patient monitor was first connected to RA, LA and LL to record signals for the standard leads, (I, II, III), and it was then connected to each set of three electrodes for our wireless pads. Figure 4.16 and Figure 4.17 show the corresponding graphs of recordings.  64  4.3. Performance Evaluations of The W3ECG System  Voltage (V)  Voltage (V)  Voltage (V)  Lead I 2 1 0 0  0.5  1  1.5  2 Lead II  2.5  3  3.5  4  0  0.5  1  1.5  2 Lead III  2.5  3  3.5  4  0  0.5  1  1.5  2 Time (s)  2.5  3  3.5  4  2 1 0  2 1 0  Figure 4.16: ECG signals of Lead I, II and III from patient monitor.  Voltage (V)  Voltage (V)  Voltage (V)  Pad 1 Location 2 1 0 0  0.5  1  1.5  2 2.5 Pad 2 Location  3  3.5  4  0  0.5  1  1.5  2 2.5 Pad 3 Location  3  3.5  4  0  0.5  1  1.5  3  3.5  4  2 1 0  2 1 0 2 Time (s)  2.5  Figure 4.17: ECG signals of Pad 1, Pad 2 and Pad 3 locations from patient monitor.  65  4.3. Performance Evaluations of The W3ECG System We observe clear QRS complexes in both Figure 4.16 and Figure 4.17 for every lead/pad, although with different magnitudes. Compared to waveforms of Leads I, II and III, those of Pad 1 and Pad 3 present less changes during P-wave and T-wave durations. But Pad 2 exhibits a similar waveform (P wave, T wave and QRS complex) to Lead III. Recall that Lead III is the differential signal between LL and LA electrodes. These two electrodes correspond to a lead vector, which can be very close to that of Pad 2, considering its placement orientation. This explains why the recorded signals from Lead III and Pad 2 are almost the same in shape. However, the parallelity in placement orientation does not always indicates the similar lead vector if we note Lead I and Pad 1 also share same orientation. We also note from the above figures that the amplitude of the ECG waveform is less dependent on the distance between two input electrodes, but more relies on the locations of these electrodes. As Pad 3 is placed a little farther away from the heart, compared to Pad 1 and Pad 2, it experiences weak differential potential values. This can be also explained as it corresponds to a shorter lead vector. However, the length of the lead vector is irrespective of the distance between pads.  4.3.2  Computer Simulations Based on Practical EASI Data Set  In this subsection, we evaluate W3ECG’s performance with the pad placement positions chosen in Table 3.1. ECG data collected by Dalhousie’s BSM system for four patients [18] are examined. Acquired ECG data is from patients with acute myocardial infarction. Because there were no pads actually placed on these locations (Table 3.1), the signals that would appear on the three pads if we were able to place them are calculated by computer, based on how the differential signal is obtained on each bipolar lead. We pick recorded potential values for nodes 170 and 127, 178 and 177, as well as 109 and 107 from [18]. Substraction was then done between potential values of the two nodes in each pair, for each sampling time. These three time series of differential potential values were then applied to Equation 3.14 to synthesis the simulated 12-lead ECG signals. We compare the generated 12-lead ECG signals with the directly obtained version by evaluating the cross-correlation coefficients. Assuming the directly obtained signal is Vi (t) (i indicates the 12 different leads), and the synthesized signal is Vi (t), then the cross-correlation coefficient ri can be represented as in Equation 4.1, where Vi and Vi are means of Vi (t) and Vi (t) respectively. t (Vi (t)  ri = (  − Vi )(Vi (t) − Vi )  2 t (Vi (t) − Vi ) (  (4.1)  2 t (Vi (t) − Vi ))  There are 12 coefficients for each patient. In order to prove the effectiveness of W3ECG’s approach, these coefficients are drawn in the figures (Figure 4.18 to Figure 4.21) with those obtained by using the EASI approach, for each patient.  66  4.3. Performance Evaluations of The W3ECG System  Figure 4.18: Comparison of cross-correlation coefficients between W3ECG system and EASI system for Patient 1.  Figure 4.19: Comparison of cross-correlation coefficients between W3ECG system and EASI system for Patient 2.  67  4.3. Performance Evaluations of The W3ECG System  Figure 4.20: Comparison of cross-correlation coefficients between W3ECG system and EASI system for Patient 3.  Figure 4.21: Comparison of cross-correlation coefficients between W3ECG system and EASI system for Patient 4.  From the above comparisons, we observe a good performance for our proposed W3ECG system for patient 2 and 4, but larger discrepancies for patient 1 (lead V4) and 3 (lead III). This coincides with those of the EASI system. We note for the EASI system, very small coefficients appear (0.3134 for lead V4 of patient 1 and -0.0850 for lead III of patient 3). However, other leads of patient 1 and 68  4.3. Performance Evaluations of The W3ECG System 3 show high cross correlation in the EASI system. In comparison, having very small coefficients for some leads in the W3ECG system seems to have a negative effect on the other lead’s performance, as we observe that lead II and aVF for patient 1 and 3 also experience small coefficients. This can be explained by the linear dependence our three pads may have on each other. We also compare the waveforms of original 12-lead signals and generated signals directly for patient 4. Figure 4.22 shows the two versions of signals with respect to time. It gives an idea of how close the waveforms appear for different coefficient values. We note the signals for Lead III for both versions look quite similar in the QRS complex duration, although in Figure 4.21 we have a relatively smaller cross-correlation coefficient (0.7806). The similarity between original and generated waveforms also applies to Lead aVL, which has the second least cross-correlation coefficient (0.7983).  69  4.3. Performance Evaluations of The W3ECG System  Figure 4.22: Comparison between synthesized 12-lead ECG signals and directly obtained versions.  4.3.3  Deployments of W3ECG and Experimental Studies  Above, we showed the ECG graphs for our designated three pad locations, however, they were recorded by a commercial patient monitor. In this section, we present the signals collected from a fully running W3ECG, in which pads are coordinated by a wireless system.  70  4.3. Performance Evaluations of The W3ECG System Effects of The Feedback Block As discussed in the pad design section of Chapter 3, the feedback block of the front-end circuit has two effects: (1) it sets the subject’s body to the appropriate voltage level, and (2) it attenuates the 50 / 60 Hz noise on the body. These two are both crucial in getting meaningful signal from the body. If the body is not set at the right voltage level, the amplifiers in the separation block and common-mode noise reduction block will saturate. Without the feedback signal from the circuit, we observe an ECG signal severely corrupted by the 50 / 60 Hz noise. The signal-to-noise ratio (SNR) decreases from 60 dB (with feedback) to 30 - 40 dB (without feedback). Figure 4.23 and Figure 4.24 illustrate the ECG signal obtained without using the feedback electrode in time domain and frequency domain respectively. They are compared to a signal obtained using the same pad, but with all three electrodes. Two Electrodes 1.5 Voltage (V)  1 0.5 0 −0.5 −1 −1.5  0  0.5  1  1.5  2  2.5  3  3.5  4  3  3.5  4  Three Electrodes 1.5 Voltage (V)  1 0.5 0 −0.5 −1 −1.5  0  0.5  1  1.5  2 Time (s)  2.5  Figure 4.23: Time domain of ECG graph corrupted by 50/60 Hz noise.  Synchronized 3-Pad ECG Signals Figure 3.8 shows the snapshot of the W3ECG server GUI with ECG signals from 3 pads. From the figure, we observe good synchronization across three pads over a period of 10 seconds. If we compare Figure 3.8 with Figure 4.17, we note that the signals obtained from patient monitor are very similar to the signals obtained from the pads. This proves that our ECG front71  4.3. Performance Evaluations of The W3ECG System  Two Electrodes Magnitude (dB)  0 −20 −40 −60 −80 −100  −50  0  50  100  50  100  Three Electrodes Magnitude (dB)  0 −20 −40 −60 −80 −100  −50  0 Frequency (Hz)  Figure 4.24: Frequency domain of ECG graph corrupted by 50/60 Hz noise.  end circuit design is comparable with the patient monitor. However, we do observe regular pulses added to each ECG signal of W3ECG as noises, which is due to the large current draw of radio chip during transmission. If we calculate the cross-correlation coefficients between each of the three signals shown in Figure 3.8, we obtain the results as shown in Table 4.4. We note that signals from Pad 1 and Pad 3 are almost independent, while signals from Pad 1 and Pad 2, and signals from Pad 2 and Pad 3 are partially dependent on each other. These can be explained by the location of placements that Pad 1 and Pad 3 are placed relatively far away from each other, while Pad 2 is in the middle. The cross-correlation coefficients also show from another perspective that three pads are necessary as the signals are providing different information of the heart’s electronic activities.  Pad 1 Pad 2 Pad 3  Pad 1 1.0000 0.6786 0.0602  Pad 2 0.6786 1.0000 0.5470  Pad 3 0.0602 0.5470 1.0000  Table 4.4: Cross-correlation coefficients between each of the three signals obtained from W3ECG.  72  4.3. Performance Evaluations of The W3ECG System Synthesized 12-lead ECG Signals Following we compare the synthesized 12-lead ECG signals from W3ECG to original 12-lead ECG signals. The original 12-lead ECG signals are obtained one lead at a time, by using the same pad as those in W3ECG. The pad is connected to electrodes placed at the estimated 12-lead locations using wires. The reason for this setup is to achieve synchronization between W3ECG pads and the reference pad using the existing system. The locations where electrodes have been placed for standard 12 leads are summarized in Table 4.5 Leads Lead I Lead II Lead III Lead aVR Lead aVL Lead aVF Lead V1 Lead V2 Lead V3 Lead V4 Lead V5 Lead V6  Positive Electrode between 69 and 70 between 343 and 344 between 343 and 344 between 59 and 61 between 69 and 70 between 343 and 344 169 171 173 216 between 217 and 218 219  Negative Electrode between 59 and 61 between 59 and 61 between 69 and 70 235 212 64 212 212 212 212 212 212  Reference Electrode between 343 and 344 between 69 and 70 between 59 and 61 between 343 and 344 between 343 and 344 212 between 343 and 344 between 343 and 344 between 343 and 344 between 343 and 344 between 343 and 344 between 343 and 344  Table 4.5: Placement of electrodes for standard 12 leads. Figure 4.25 presents the cross-correlation coefficients (calculated using Equation 4.1) for observed and synthesized 12-lead signals obtained from the setup described above. Figure 4.26 to Figure 4.37 show the waveforms of two versions for each lead.  73  4.3. Performance Evaluations of The W3ECG System  Figure 4.25: Coefficients for cross-correlation between observed and synthesized 12-lead signals.  74  4.3. Performance Evaluations of The W3ECG System  Lead I  0.6  1  0.5  0.8  0.4  0.6  0.3 0.4 0.2 0.2 0.1 0 0 −0.2  −0.1 −0.2  −0.4  −0.3  −0.6  −0.4  0  0.2  0.4  0.6  0.8  1  Synthesized Lead I: Voltage (V)  Observed Lead I: Voltage (V)  1.2  −0.8  Time (s)  Figure 4.26: Comparison between observed and synthesized Lead I signals.  Lead II 6 1.4  Observed Lead II: Voltage (V)  1  4  0.8  3  0.6 2 0.4 0.2  1  0  0  Synthesized Lead II: Voltage (V)  5  1.2  −0.2 −1 −0.4  0  0.2  0.4  0.6  0.8  1  Time (s)  Figure 4.27: Comparison between observed and synthesized Lead II signals.  75  4.3. Performance Evaluations of The W3ECG System  6  1  5  0.8  4  0.6  3  0.4  2  0.2 1 0  Synthesized Lead III: Voltage (V)  Observed Lead III: Voltage (V)  Lead III 1.2  0 −0.2 −1 −0.4  0  0.2  0.4  0.6  0.8  1  Time (s)  Figure 4.28: Comparison between observed and synthesized Lead III signals.  1  0.2  0.5  0 0 −0.2 −0.5 −0.4 −1  −0.6  −1.5  −0.8 −1  −2  −1.2  Synthesized Lead aVR: Voltage (V)  Observed Lead aVR: Voltage (V)  Lead aVR 0.4  −2.5  −1.4 −3 −1.6  0  0.2  0.4  0.6  0.8  1  Time (s)  Figure 4.29: Comparison between observed and synthesized Lead aVR signals.  76  4.3. Performance Evaluations of The W3ECG System  Lead aVL 1 0.3  Observed Lead aVL: Voltage (V)  0.1 0 0 −0.1  −0.5  −0.2 −0.3  −1  −0.4 −1.5  −0.5  Synthesized Lead aVL: Voltage (V)  0.5  0.2  −0.6 −2  −0.7 −0.8  0  0.2  0.4  0.6  0.8  1  Time (s)  Figure 4.30: Comparison between observed and synthesized Lead aVL signals.  Lead aVF 1.4 5  4  1 0.8  3  0.6 2  0.4 0.2  1  0 0  Synthesized Lead aVF: Voltage (V)  Observed Lead aVF: Voltage (V)  1.2  −0.2 −1  −0.4 0  0.2  0.4  0.6  0.8  1  Time (s)  Figure 4.31: Comparison between observed and synthesized Lead aVF signals.  77  4.3. Performance Evaluations of The W3ECG System  Lead V1 0 0.2 −0.1  Observed Lead V1: Voltage (V)  −0.2 −0.3  −0.4  −0.4  −0.6  −0.5  −0.8  −0.6  −1  −0.7  −1.2  −0.8  −1.4  −0.9  −1.6  −1  0  0.2  0.4  0.6  0.8  1  Synthesized Lead V1: Voltage (V)  0 −0.2  −1.8  Time (s)  Figure 4.32: Comparison between observed and synthesized Lead V1 signals.  Lead V2 0.2  Observed Lead V2: Voltage (V)  0 0 −0.1 −0.2  −0.5  −0.3 −1 −0.4 −0.5  −1.5  −0.6 −2  −0.7 −0.8  Synthesized Lead V2: Voltage (V)  0.5  0.1  0  0.2  0.4  0.6  0.8  1  Time (s)  Figure 4.33: Comparison between observed and synthesized Lead V2 signals.  78  4.3. Performance Evaluations of The W3ECG System  Lead V3 0.7 1.5  Observed Lead V3: Voltage (V)  0.5 1 0.4 0.3 0.5  0.2 0.1  0  0 −0.1 −0.2  Synthesized Lead V3: Voltage (V)  0.6  −0.5 0  0.2  0.4  0.6  0.8  1  Time (s)  Figure 4.34: Comparison between observed and synthesized Lead V3 signals.  Lead V4 4  1  3.5 3 0.6  2.5 2  0.4 1.5 0.2  1 0.5  0  Synthesized Lead V4: Voltage (V)  Observed Lead V4: Voltage (V)  0.8  0 −0.2  −0.5 0  0.2  0.4  0.6  0.8  1  Time (s)  Figure 4.35: Comparison between observed and synthesized Lead V4 signals.  79  4.3. Performance Evaluations of The W3ECG System  Lead V5 4  1  3.5  Observed Lead V5: Voltage (V)  3 0.6  2.5 2  0.4  1.5 0.2  1 0.5  0  0  Synthesized Lead V5: Voltage (V)  0.8  −0.2 −0.5 −0.4  −1 0  0.2  0.4  0.6  0.8  1  Time (s)  Figure 4.36: Comparison between observed and synthesized Lead V5 signals.  Lead V6 1.2 3 2.5 0.8 2 0.6 1.5 0.4 1 0.2 0.5 0  Synthesized Lead V6: Voltage (V)  Observed Lead V6: Voltage (V)  1  0  −0.2  −0.5  −0.4 0  0.2  0.4  0.6  0.8  1  Time (s)  Figure 4.37: Comparison between observed and synthesized Lead V6 signals.  80  4.3. Performance Evaluations of The W3ECG System From the comparisons, we have the following observations. • Synthesized 12-lead signals exhibit almost the same QRS complexes as those of the observed versions, except Lead V2. • The high correlations during QRS-complex duration make the corresponding 9 cross-correlation coefficients higher than 0.75 (Figure 4.25), 1 coefficient between 0.5 and 0.6, 2 coefficients below 5.5. • Two versions of signals show more similarities in the P-R segments, but large differences in S-T segments. • Synthesized 12-lead signals have amplified magnitudes, 1 to 4 times of those of the observed 12-lead signals.  4.3.4  Discussions  After examining the above simulation and experimental results of W3ECG, we deem that the present W3ECG implementation performs very well in terms of cross-correlation coefficients. It can be used for diagnosis of diseases which can be detected by examining the QRS complexes and/or P-R segments of 12-lead ECG waveforms. However, we do have observed small cross-correlations for certain leads during either simulation or experimental study. And some of the waveforms show relatively large discrepancies, especially during S-T segments. This discrepancies can be crucial when the lost information is critical for real-time diagnose of certain acute diseases or those which are life-threatening; however, when we come to the long-term ambulatory monitoring of ECG, with the intent to detect chronic diseases, the small amount of information lost may not be vital [11][13][34]. Recall that VCG system, EASI system, and our proposed W3ECG system are all based on the hypothesis that electrical heart activity can be represented by a stationary dipole in a threedimensional space. This hypothesis can be simplistic in reality. In other words, a model to describe the heart activity in higher dimensional space can be developed in the future. Correspondingly, an interesting work would be to implement a wireless n-pad ECG system (n = 4, 5, 6......) to represent the cardiac activity in the n-dimensional space. Obviously, the benefits of wireless pads are prominent only when n is small.  81  Chapter 5  Conclusion As a complement to existing wireless technologies, WBAN plays a very important role in humancentric communications. Its advancements have been the result of interdisciplinary research and development. In this thesis, we have provided a comprehensive review and outlook of this promising field through the survey of enabling technologies and pioneer projects. Based on the survey of QoS work in WBAN, we proposed a QoS provisioning framework for WBAN traffic employing the IEEE 802.15.4 superframe structure in the beacon-enabled mode. A method of prioritizing WBAN traffic has been proposed, along with algorithms for admission control and scheduling. Utilizing both the CAP and CFP mechanisms, the proposed QoS framework can better differentiate WBAN application streams and serve periodic traffic more directly through contention-free multiple-access methods. Computer simulations reveal that our scheme is able to guarantee a 100% time constraint compliance ratio for traffic in contention access periods, and a 99.2% to 99.9% constraint compliance ratio for traffic in contention-free periods, while still admitting and accommodating tens of WBASN traffic streams. The bandwidth utilization and power efficiency of the proposed framework can be further improved by reasonably arranging RT traffic flows in CAPs. During our framework design, we have kept maximum compliance to the existing standard to minimize engineering efforts needed for implementation. Currently, we are working on realizing the Admissions Controller, Superframe Scheduler and CFP Scheduler on a TelosB platfrom. Utilizing the proposed QoS framework, we have proposed a novel wireless ECG system: the W3ECG. W3ECG furthers the pad design idea of the single-pad approach. With the goal of designing a wireless ECG system to render caregivers standard 12-lead ECG information, we have introduced another two pads compared to wireless single-pad ECG systems to gain spatial variety during recording. As only wireless pads are deployed on the body, a W3ECG system experiences limited artifacts or noises introduced by wires and provides patient comfort. It also provides conventional 12-lead ECG signals, which most caregivers are trained to diagnose. The criteria for evaluating pad placements have been discussed and one example placement has been provided. Performance of our selected pad placement positions shows comparable result with respect to an EASI lead system. The results show a prominent potential of W3ECG’s wide application for ambulatory long-term ECG monitoring. In addition, we have been able to manufacture the frontend ECG circuit, and combine it with IEEE 802.15.4 radio platform. Software packages for the server and pad have also been developed to make a fully running W3ECG possible. We intend to further our research on better pad positions and the evaluation of our proposed system over a large pool of patients whenever the data are available. Our long term goal is to develop a wireless multiple-pad system that works in such a way that the placement of the pads is not 82  Chapter 5. Conclusion critical. In the current W3ECG, the positions of pads are important, but not unique. Intelligence can be provided for instructing the placement of pads in real-time by locating pads and computing the corresponding transformation matrix in real-time, i.e., if the computed transformation matrix is not acceptable, suggestions for how to move the pads would be provided in real-time.  83  Bibliography [1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci. Wireless sensor networks: A survey. Computer Networks, 38:393 – 422, 2002. [2] ZigBee Alliance. ZigBee Specification. [3] Bluetooth SIG. Specification of The Bluetooth System, 2009. [4] H. Cao, V.C.M. Leung, C. Chow, and H.C.B. Chan. Enabling technologies for wireless body area networks: A survey and outlook. IEEE Communications Magazine, 47:84 – 93, 2009. [5] H. Cao, X. Liang, I. Balasingham, and V.C.M. Leung. Performance analysis of zigbee technology for wireless body area sensor networks. In Advanced Sensor Integration Technology, 2009. [6] CDC. Centers for Disease Control and Prevention. http://www.cdc.gov/nchs/fastats/ bodymeas.htm. [7] L. Cheng, X. Zhang, and A.G. Bourgeois. Gts allocation scheme revisited. Electronics Letters, 43:1005 – 1006, 2007. [8] Crossbow Technology. http://www.xbow.com/. [9] K. Y. Yazdandoost et al. Channel model for body area network (ban). Technical Report IEEE P802.15-08-0780-08-0006, IEEE, 2009. [10] E. Farella, A. Pieracci, L. Benini, L. Rocchi, and A. Acquaviva. Interfacing human and computer with wireless body area sensor networks: the wimoca solution. Multimedia Tools Appl., 38:337 – 363, 2008. [11] D. Q. Feild, C. L. Feldman, and B. M. Horacek. Improved easi coefficients: Their derivation, values, and performance. Journal of Electrocardiology, 35:23 – 33, 2002. [12] R. Fensli, E. Gunnarson, and T. Gundersen. A wearable ecg-recording system for continuous arrhythmia monitoring in a wireless tele-home-care situation. In Proceedings of 18th IEEE Symposium on Computer-Based Medical Systems, 2005. [13] R. Fensli, E. Gunnarson, and T. Gundersen. A wearable ecg-recording system for continuous arrhythmia monitoring in a wireless tele-home-care situation. In CBMS ’05: Proceedings of the 18th IEEE Symposium on Computer-Based Medical Systems, 2005.  84  Chapter 5. Bibliography [14] D. D. Finlay, C. D. Nugent, J. G. Kellett, M. P. Donnelly, P. J. McCullagh, and N. D. Black. Synthesising the 12-lead electrocardiogram: Trends and challenges. European Journal of Internal Medicine, 18(8):566 – 570, 2007. [15] D. D Finlay, C. D Nugent, P. J McCullagh, and N. D Black. Mining for diagnostic information in body surface potential maps: A comparison of feature selection techniques. BioMedical Engineering OnLine, 4, 2005. [16] E. Frank. General theory of heart-vector projection. Circulation Research, 2:258 – 270, 1954. [17] T. Gao, T. Massey, L. Selavo, D. Crawford, B. Chen, K. Lorincz, V. Shnayder, L. Hauenstein, F. Dabiri, J. Jeng, A. Chanmugam, D. White, M. Sarrafzadeh, and M. Welsh. The advanced health and disaster aid network: A light-weight wireless medical system for triage. Biomedical Circuits and Systems, IEEE Transactions on, 1:203 – 216, 2007. [18] A. L. Goldberger, L. A. N. Amaral, L. Glass, J. M. Hausdorff, P. Ch. Ivanov, R. G. Mark, J. E. Mietus, G. B. Moody, C.-K. Peng, and H. E. Stanley.  PhysioBank, Phys-  ioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation, 101(23):e215–e220, 2000 (June 13). Circulation Electronic Pages: http://circ.ahajournals.org/cgi/content/full/101/23/e215. [19] P. S. Hall and Y. Hao, editors. Antennas and Propagation for Body-Centric Wireless Communications. Artech House Publishers, 2006. [20] J.H. Hauer. Tkn15.4: An ieee 802.15.4 mac implementation for tinyos 2. Technical Report TKN-08-003, Technical University Berlin, 2009. [21] R. Hoekema, G. J. H. Uijen, and A. van Oosterom. On selecting a body surface mapping procedure. Journal of Electrocardiology, 32(2):93 – 101, 1999. [22] B. M. Horacek, J. W. Warren, D. Q. Feild, and C. L. Feldman. Statistical and deterministic approaches to designing transformations of electrocardiographic leads. Journal of Electrocariology, 35:41 – 52, 2002. [23] B. M. Horacek, J. W. Warren, C. J. Penney, R. S. MacLeod, L. M. Title, M. J. Gardner, and C. L. Feldman. Optimal electrocardiographic leads for detecting acute myocardial ischemia. Journal of Electrocardiology, 34:97 – 111, 2001. [24] Y. Huang, A. Pang, and H. Hung. Adaptive gts allocation scheme for ieee 802.15.4. IEEE Transactions on Parallel and Distributed Systems, 19:641 – 651, 2008. [25] IEEE 802.15 WPAN Task Group 6 (TG6). IEEE 802.15 WPAN Task Group 6 (TG6) Body Area Networks. http://www.ieee802.org/15/pub/TG6.html. [26] IEEE Computer Society. 802.15.4-2006 IEEE Standard for Information Technology- Telecommunications and Information Exchange Between Systems- Local and Metropolitan Area 85  Chapter 5. Bibliography Networks- Specific Requirements Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (WPANs). [27] IMEC. http://www.imec.be/. [28] I. Jantunen, H. Laine, P. Huuskonen, D. Trossen, and V. Ermolov. Smart sensor architecture for mobile-terminal-centric ambient intelligence. Sensors and Actuators A: Physical, 142:352 – 360, 2008. [29] B. Khaddoumi, H. Rix, O. Meste, M. Fereniec, and R. Maniewski. Body surface ecg signal shape dispersion. Biomedical Engineering, IEEE Transactions on, 53:2491 – 2500, 2006. [30] L. Ko and Z. Chou. A novel multi-beacon superframe structure with greedy gts allocation for ieee 802.15.4 wireless pans. In Proceedings of WCNC, 2007. [31] A. Koubaa, M. Alves, and E. Tovar. An implicit gts allocation mechanism in ieee 802.15.4 for time-sensitive wireless sensor networks: theory and practice. Real-Time Systems, 39:169 – 204, 2008. [32] A. Kurs, A. Karalis, R. Moffatt, J. D. Joannopoulos, P. Fisher, and M. Soljacic. Wireless power transfer via strongly coupled magnetic resonances. Science, 317:83 – 86, 2007. [33] P. Kuryloski, A. Giani, R. Giannantonio, K. Gilani, R. Gravina, V.P. Seppa, E. Seto, V. Shia, C. Wang, P. Yan, A. Yang, J. Hyttinen, S. S. Sastry, S. Wicker, and R. Bajcsy. Dexternet: An open platform for heterogeneous body sensor networks and its applications. Technical report, EECS Department, University of California, Berkeley, Dec 2008. [34] C. Levkov. Derived 12 channel electrocardiogram from 4 channel holter electrocardiogram. In XI International Scientific and Applied Conference, 2002. [35] G. E. Mailloux and R. M. Gulraajani. Theoretical evaluation of the mcfee and frank vectorcardiographic lead systems using a numerical inhomogeneous torso model. Biomedical Engineering, IEEE Transactions on, 29:322 – 332, 1982. [36] A. Milenkovic, C. Otto, and E. Jovanov. Wireless sensor networks for personal health monitoring: Issues and an implementation. Comput. Commun., 29:13 – 14, 2006. [37] M.C. Munshi, X. Xu, X. Zou, E. Soetiono, C. S. Teo, and Y. Lian. Wireless ecg plaster for body sensor network. In Proceedings of the 5th International Workshop on Wearable and Implantable Body Sensor Networks (BSN 2008) in conjunction with the 5th International Summer School and Symposium on Medical Devices and Biosensors (ISSS-MDBS 2008), 2008. [38] C. Na, Y. Yang, and A. Mishra. An optimal gts scheduling algorithm for time-sensitive transactions in ieee 802.15.4 networks. Computer Networks, 52:2543 – 2557, 2008. [39] OPNET. http://www.opnet.com/.  86  [40] Kaveh Pahlavan and Prashant Krishnamurthy. Principles of Wireless Networks: A Unified Approach. Prentice Hall PTR, 2001. [41] P. Park, C. Fischione, and K. H. Johansson. Performance analysis of gts allocation in beacon enabled ieee 802.15.4. In Proc. of Sixth Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks (IEEE SECON 09), 2009. [42] A. Pentland. Healthwear: Medical technology becomes wearable. Computer, 37:42 – 49, 2004. [43] V. Shnayder, B. Chen, K. Lorinczd, T. R. F. Fulford-Jones, and M. Welsh. Sensor networks for medical care. In SenSys ’05: Proceedings of the 3rd international conference on Embedded networked sensor systems, 2005. [44] Takizawa, K. Aoyagi, and R. T. Kohno. Channel modeling and performance evaluation of uwbbased wireless body area networks. In Communications, 2009. ICC ’09. IEEE International Conference on, 2009. [45] ART-WiSe Team. OpenSource Toolset for IEEE 802.15.4 and ZigBee. http://www.open-zb. net/. [46] Texas Instruments. http://www.ti.com/. [47] Texas Instruments. 2.4 GHz IEEE 802.15.4 / ZigBee-Ready RF Transceiver Datasheet, 2007. [48] U.S. Census Bureau. http://www.census.gov/. [49] J. G. Webster, editor. Medical Instrumentation: Application and Design. Wiley, 3rd edition, 1997. [50] P. J. Xu, H. Zhang, and X. M. Tao. Textile-structured electrodes for electrocardiogram. Textile Progress, 40:183 – 213, 2008. [51] G.Z. Yang, editor. Body Sensor Networks. Springer, 2006. [52] S. Yoo, D. Kim, M. Pham, Y. Doh, E. Choi, and J. Huh. Scheduling support for guaranteed time services in ieee 802.15.4 low rate wpan. In Proceedings of the 11th IEEE International Conference on Embedded and Real-Time Computing Systems and Applications, RTCSA05, 2005. [53] Y. P. Zhang, L. Bin, and C. Qi. Characterization of on-human-body uwb radio propagation channel. Microwave and Optical Technology Letters, 9:1365 – 1371, 2007. [54] G. Zhou, J. Lu, C. Wan, M. D. Yarvis, and J. A. Stankovic. Bodyqos: Adaptive and radioagnostic qos for body sensor networks. In 27th IEEE Conference on Computer Communications, Infocom, 2008.  87  Appendix A  Statement of Co-Authorship I am the first author and the main contributor of all the published papers  2 3 4  and submitted  paper that are of the main contribution to this thesis. My contributions to these papers were: • identification and design of research program • performing the research • data analyses • manuscript preparation  2  H. Cao, V.C.M. Leung, C. Chow and H.C.B. Chan, ”Enabling Technologies for Wireless Body Area Networks: A Survey and Outlook”, IEEE Communications Magazine, vol. 47, no. 12, pp. 84-93, Dec. 2009. 3 H. Cao, S. Gonz´lez-Valenzuela and V.C.M. Leung, ”Employing IEEE 802.15.4 for Quality of Service Provisioning in Wireless Body Area Sensor Networks”, in Proc. AINA 2010, Perth, Australia, Apr. 2010. 4 H. Cao, X. Liang, I. Balasingham and V.C.M. Leung, ”Performance Analysis of ZigBee Technology for Wireless Body Area Sensor Networks”, in Proc. ASIT 2009, Niagara Falls, ON, Sep. 2009.  88