1. Evaluation of Body Sensor Network Platforms A Design Space and Benchmarking Analysis Ayan Banerjee, Sandeep K.S. Gupta IMPACT Lab, Arizona State University Sidharth Nabar, Radha Poovendran Network Security Lab (NSL), University of Washington, Seattle

2. Outline Background and Motivation Proposed Evaluation Framework – Design Space Determination Design Coordinates, Metrics and Benchmarking – Design Space Exploration Case Study Conclusion and Future Work 2

3. Pervasive Health Monitoring Components Used Miniature sensors, gateway device Services Enabled Continuous, remote patient monitoring: No time & space restrictions Utilize wearable and in-vivo medical sensors Reduced medical errors Early detection of ailments and actuation through automated health data analysis Nano-scale Blood Glucose level detector @ UIUC Temperature Tele-sensor @ Oak Ridge National Lab L:ifeshirt @ Vivometrics Applications Sports Health Management Home-based care Computer Assisted Rehabilitation Medical Facility Management Principal enabling technology: Body Sensor Networks (BSNs) 3

4. BSN node v3 Imote2 TelosB Shimmer Radio – CC2420+ miniaturized chip antenna Processor – MSP430 Size – 26mm x 16mm x 2mm Radio – CC2420 (802.15.4) + RN-42 (Bluetooth) Processor – MSP430 Size – 53mm x 32mm x 15mm Radio – CC2420 + Inverted-F antenna Processor – MSP430 Size – 65mm x 31mm x 6mm Radio – CC2420 + Surface mount antenna Processor – Intel Xscale Size – 36mm x 48mm x 9mm BSN Platforms Components: Microprocessor, radio, onboard memory, power supply interface, etc. Applications: Used for BSN research experiments and clinical trials. Diversity in available platforms- How to choose? 4

5. Job Hiring Example EMPLOYER Knows job requirements Defines candidate qualifications Reviews multiple candidates Checks performance Selects most suitable candidate CANDIDATES Provide resume Understand market requirements and peer competition Acquire new skills to improve Standard evaluation method and well-understood performance metrics enable candidate selection 5

6. Lack of standard evaluation method and performance metrics BSN Platform Selection Uses BSN platforms for research or clinical trials Knows application requirements Need to map to platform specifications Need to quantify platform performance BSN PLATFORM USERPLATFORM DESIGNERS Provide datasheet Need to improve design based on new, emerging applications Need to objectively compare performance of multiple platforms 6

7. Main Idea Common set of parameters Desirable parameter values Select platform satisfying most/all constraints Available Platforms Application Requirements 7

8. Research Challenges Mapping diverse platforms to common evaluation ground – Design Coordinate: A feature of the BSN platform that determines its performance, e.g. Available Memory. – Design Space: The space defined by the design coordinates Quantify design coordinates, performance parameters – Evaluation Metrics, e.g. kB of RAM, W of power consumed Measure performance in real application scenarios – Develop benchmarks based on BSN applications Search design space for suitable platform 8

9. Goal and Contributions Goal: Evaluation Framework for BSN platforms Contributions: – Identify design coordinates for BSN platforms – Map a given platform to a point in the design space: Use metrics to quantify design coordinates, define two new metrics for BSN platforms. Develop BSN-specific benchmarking suite: BSNBench – Provide a method to search the design space 9

10. SET OF APPLICATION REQUIREMENTS CONSTRAINTS ON DESIGN COORDINATES ELIMINATE PLATFORMS VIOLATING CONSTRAINTS DESIGN SPACE EXPLORATION DESIGN COORDINATES EVALUATION METRICS BSNBENCH, DATASHEET, MODELS DESIGN SPACE DETERMINATION Proposed Evaluation Framework Set of BSN Platforms Most suitable BSN Platform for application PROPOSED EVALUATION FRAMEWORK 10 Application Requirements

11. Example: Design Space Determination Set of available platforms Design Coordinates Average Radio Power Consumption Form Factor On-board data memory mWVolume (mm 3 )kB of RAM Metrics BSNBenchDatasheet Evaluation Method P2P2 P1P1 P3P3 P4P4 11

12. P2P2 Example: Design Space Exploration RAM Availability (KB) Form Factor (mm 3 ) Average Radio Power Consumption (mW) P1P1 P3P3 P4P4 Design coordinate axis Constraints on design coordinates Sensor platforms Appropriate region in the design space 12

13. Design Coordinates Based on typical BSN application requirements Decompose platform functionality into individual modules DESIGN COORDINATES COMPUTATION WIRELESS COMMUNICATION ENERGY SOURCE PHYSICAL ASPECTS (Processor, data and program memory, signal processing capability) (Radio reliability, average power consumption, interoperability) (Battery, energy scavenging support) (Form factor, thermal safety, sensor integration) 13

14. Evaluation Metrics Use suitable metrics to quantify design coordinates: Some traditional metrics are independent of the target applications. e.g. MIPS, MIPS/W Consider BSN application characteristics to develop more suitable metrics. – For example, processor speed measured in units of samples processed per second Design CoordinateMetric Radio ReliabilityOn-body PDR [1] BatteryCapacity (mAh), size (mm 3 ) 14 [1] A. Natarajan, B. Silva, K. Yap, and M. Motani. To hop or not to hop: Network architecture for body sensor networks. In IEEE SECON, 2009.

15. SPSW (Samples Processed per Second per Watt) is defined as: Captures tradeoff between processor speed and power consumption. “Processing a sample” is application-specific. For example, a platform motes calculates mean of 1000 data samples in 100 ms and consume 25 mW power. Then, SPSW = 1000/(100 X 25) = 400 samples/mJ SPSW Metric for Processor (Time taken) (Power consumed) (No. of Samples Processed) SPSW = 15

16. EPC Metric for Radio Radio power consumption depends on radio specifications as well as duty cycle. EPC (Expected Power Consumption) is defined as: For example, if radio transmits for 5% time, with power draw 10 mW and is in SLEEP state for remaining 95% time, with power 0.1 mW, EPC = 0.05 * 10 + 0.95 * 0.1 = 0.595 mW Fraction of time in state S * Power consumed in state S EPC = ∑ all states 16

17. BSNBench: A BSN-specific benchmark Key Observation: In spite of diversity in BSN applications, some basic tasks are common. Type of benchmark: Microbenchmark Composition: – Data Operations (Statistics, Differential Encoding) – Signal Processing (FFT, Peak detection) – Radio Communication (Duty-cycled handshake) – Sensor Interface (Sensed Data Query) Implemented in TinyOS 2.0 17

18. CONSTRAINTS ON DESIGN COORDINATES ELIMINATE PLATFORMS VIOLATING CONSTRAINTS PRIORITIZE DESIGN COORDINATES DESIGN SPACE EXPLORATION Evaluation Framework Workflow DESIGN COORDINATES EVALUATION METRICS BSNBENCH, DATASHEET, MODELS DESIGN SPACE DETERMINATION 18

19. Design Space Exploration Constraints on Design Coordinates Application Requirements Define subspace of design space Set of suitable platforms Prioritize design coordinates Identify most suitable platform 19

20. Case Study We consider two typical BSN applications: 1.Continuous Glucose Monitoring (CGM): -Long term monitoring application -Sensor measures the blood glucose level and transmits this data to a gateway device. 2.Epileptic Seizure Detection (ESD): -Detect onset of epileptic seizures using an ECG sensor. -Perform peak detection on ECG signal to calculate RR intervals. -Intervals are converted to FFT coefficients and sent to the gateway device. 20

21. Case Study: CGM Set of platforms: TelosB, Mica2, Imote2 and BSN v3 Constraints on EPC and Form Factor coordinates Form Factor (mm 3 ) EPC (W) iMote2 TelosB Mica2 BSN node v3 35.62 mW (50 X 50 X 50) 21

22. Case Study: ESD Constraints on signal processing capability and communication reliability (on-body PDR) On-body PDR 128 – point FFT Imote2 TelosB Mica2 BSN node v3 256 - point FFT 0.7 22 Available RAM

23. Conclusion and Future Work Conclusion – Proposed design space approach for evaluation framework – Identified design coordinates for BSN platforms – Developed novel, BSN-specific metrics – Proposed benchmarking suite for BSNs Future work – Extend BSNBench by including data privacy tasks – Complex objective functions in design space exploration. – Extend set of design coordinates: fault tolerance, etc. – Explore metrics for human centric evaluation 23

24. Thank You! Questions and Comments?

25. BSNBench: Composition SectionTaskExample BSN Applications DATA OPERATIONS StatisticsGlucose Monitoring Out-of-RangePosture monitoring (accelerometer) Differential EncodingTemperature Recording SIGNAL PROCESSING Fast Fourier TransformElectromyography (EMG) analysis FIR filteringMotion analysis, De-noising data Peak detectionECG analysis RADIO COMMUNICATION Duty-cycled handshakeAll wireless BSN applications Reliable communication Applications with on-body gateway device SENSOR INTERFACESensed Data QueryAll sensing applications

26. References [1] http://ubimon.doc.ic.ac.uk/bsn/a1875.htmlhttp://ubimon.doc.ic.ac.uk/bsn/a1875.html [2] K. Lorincz, B. Kuris, S. Ayer, S. Patel, P. Bonato, and M. Welsh. Wearable wireless sensor network to assess clinical status in patients with neurological disorders. In Proceedings of the 6th international conference on Information processing in sensor networks. ACM, 2007. [3] C. Park, J. Liu, and P. Chou. Eco: an ultra-compact low-power wireless sensor node for real-time motion monitoring. In IPSN 2005., pages 398–403. [4] M. Hempstead, M. Welsh, and D. Brooks. TinyBench: The case for a standardized benchmark suite for TinyOS based wireless sensor network devices. 2004. [5] L. Nazhandali, M. Minuth, and T. Austin. SenseBench: toward an accurate evaluation of sensor network processors. In Workload Characterization Symposium, 2005. Proceedings of the IEEE International, pages 197–203 [6] A. Natarajan, B. Silva, K. Yap, and M. Motani. To hop or not to hop: Network architecture for body sensor networks. In IEEE SECON, 2009.