Research Paper Review

1. Research Paper Review Presented by Shiraz Saleem Mobile Computing Fall 2007

2. FINE-GRAINED NETWORK TIME SYNCHRONIZATION USING REFERENCE BROADCASTS By Jeremy Elson, Lewis Girod and Deborah Estrin 5th Symposium on Operating Systems Design and Implementation (OSDI – 2002)

3. References • Time Synchronization in Wireless Sensor Networks, by Jeremy Elson and Deborah Estrin • Time Synchronization and Calibration in Wireless Sensor Networks, by Kay Romer,Phillip Blum, Lennart Meier • Applied Statistics and Probability for Engineers, by Douglas C. Montogomery, George C. Rungner Research Paper Review

4. Agenda • Background • Need for synchronization • Sources of error • Traditional methods • Problem Statement and Motivation • Key Ideas and Results • Basic algorithm • Estimating Skew • Initial Results • Comparison with NTP • Extension to Multi-hop networks • Contributions • Improvement Oppurtunity • Conclusion • Relevance to Project Research Paper Review

5. Background Why synchronize time in sensor nets? • Data fusion • Eg:-Forest Fires • Fine-grained radio scheduling • Localization • Temporal coordination of sleep periods Research Paper Review

6. Use existing methods? • Network Time Protocol(NTP) • Ubiquitous in Internet • GPS • High precision anywhere • High-stability oscillators • Rubidium, Cesium Research Paper Review

7. Limitation of Sensor Networks • Fewer resources • Energy, Network bandwidth • No infrastructure available • Master NTP server? • Cost and size factor • GPS receiver or an expensive oscillator on cheap mote? Research Paper Review

8. Sources of Synchronization Error • Send Time • O.S delays • Transfer message to communication module • Access Time • Waiting to access the transmit channel • Propagation Time • Receive Time • Same as send time • What if you can timestamp message arrival at interrrupt handler? Research Paper Review

9. Traditional Synchronization • Client request time to server and server responds.(NTP) • Calculate one way latency to account for message delay. Problem? • Send and Access time are non-deterministic. • Causes variance in message delays between sender and receiver. Research Paper Review

10. Problem Statement and Motivation A timing synchronization method which has the following properties, given the constraints of sensor networks • Good precision • Wide scope and availability • Energy Efficient • Low Cost Research Paper Review

11. Proposed solution: Reference Broadcast Synchronization • RBS synchronizes a set of receivers with each other • A Node periodically send beaconmessages to their neighbors • Receiver records the local time as soon as the message arrives • Compare local clock values when the message was received. Research Paper Review

12. Critical Path Reduced Traditional critical path: From time the sender reads its clock, to when the receiver reads its clock RBS: Only sensitive to the differences in receive time and propagation delay Research Paper Review

13. Assumptions • Propagation delay = 0 • Expect receiver error to be deterministic by recording interrupt time at which packet arrives • Validation • Experiment was done on berkley motes. • GPIO pin high on packet arrival and recorded time of reception using a logic analyzer. • Inter-receiver phase offset appeared to follow a Gaussian distribution. Research Paper Review

14. Gaussian parameters  = 0,  = 11.1 Receiver Determinism Research Paper Review

15. Single Hop RBS without correcting for skew • Receiver error follows a well behaved distribution • Can help to reduce error statistically by sending multiple pulses over time and averaging • A node broadcasts m reference beacons • Each of n receivers records local time of each m refs • Exchange: Offset[i,j] = 1/m S (Tj,k – Ti,k) Research Paper Review

16. Estimation of Clock Skew • Oscillator characteristics • Accuracy • Stability • Due to skew, clocks tend to drift apart • Compensation: instead of averaging phase offsets, perform least-squares linear regression • Fitting a line assumes that frequency is stable • Assume high short-term frequency stability • Ignore data more than a few minutes old Research Paper Review

17. Implementation on Berkley Motes Results sec Research Paper Review

18. .…Contd • Point (0,0) marks the first pulse • Receivers synchronized, no clock skew • Clock skew increases as time increases • Best fit line-Line that minimizes RMS error • Reject outliers that contribute most to the total error. • y intercept= Initial Phase Offest, slope = Clock Skew • Residual Error = RMS error = Indication of the quality of fit, 11.2 sec in this experiment. Research Paper Review

19. Second Implementation Results Sync Pulses Dddrigrift Estimate Drift Estimate Test Pulses Synchronization of clocks on PC104-compatible single-boardcomputer using a Mote as NIC. Sync pulses send for 2 minutes, points near x = 200 are reference pulses, which show a synchronization error of 7.4µsec 60 seconds after the last sync pulse. Research Paper Review

20. Comparison with NTP • Comparison performed on commodity hardware • IPAQs running Linux • 802.11 wireless Ethernet adapters • 3 different synchronization schemes • RBS • NTP • NTP-Offset • Correction for phase error • Two traffic scenarios • Light load • Heavy load Research Paper Review

21. Results Matrix Research Paper Review

22. Analysis • RBS performs eight times better than NTP at light network load • The performance against NTP was even better at heavy traffic load • The slight degradation in performance at heavy network load (2 secs) was due to the loss of broadcast packets Research Paper Review

23. Post-facto Synchronization • Power up radios when an event of interest occurs • Synchronize with neighbors • Use least square regression to extrapolate backwards • RBS thus helps to conserve power in sensors – Sleep periods Research Paper Review

24. Mutli-Hop Synchronization • A and B periodically send sync pulses • Node 4’s unique position allows it to relate clocks from one cluster to the other Research Paper Review

25. Time Routing in Multi-Hop Networks • Physical topology easily converted into logical topology • links represent possible clock conversions • Use shortest path search to find a time route • Edges can be weighted by error estimates, RMS Research Paper Review

26. Average 4 hop-error=3.68(+-)2.57sec Synchronization error grows slowly Average n-hop path approximately σ*n0.5 Performance of Multi-hop BBS Research Paper Review

27. Contributions • Largest sources of synchronization error, send & access time are removed from the critical path • Allows for post-facto synchronization and thereby conserves power on sensors • Useful for sensor networks where an absolute time referernce may not be available Research Paper Review

28. Improvement Oppurtunity • Dynamic election of the nodes to act as beacon senders • Quantifying precision v/s bandwidth used for beacons • Convergence time • Energy efficient scheme to synchronize beacon nodes Research Paper Review

29. Conclusion • sec precision • Tested on various platforms • Outperforms NTP • Slow decay of precision in multi-hop networks • Applicable to wired networks also • Needs a broadcast medium to work. • Scalabilty issues in multi-hop environments • Performance not tested when nodes are mobile Research Paper Review

30. Relevance to Project “Finding Interaction Using Sound” • Accurate time intervals for which an interaction is detetected must be recorded by sensors. • Agglomeration of this ‘Interaction Data’ to find the parties who are interacting • Requires sensors to have common sense of time Research Paper Review

31. Thank You! Research Paper Review