Segal’s Law

1. Segal’s Law A man with a watch knows what time it is. A man with two watches is never sure.

2. A Decentralized Frame Synchronization of a TDMA-based Wireless Sensor Network Fasika Assegei Advisors: Frits van der Wateren – Chess B.V. dr.ir. Peter Smulders – TU/e http://www.chess.nl

3. This research is …. • Conducted at Chess B.V. in Haarlem , the Netherlands • Part of MyriaNed project • DevLab project • a project to build a functional WSN for research on protocols, power management, programming models, and security. What is MyriaNed ?

4. Outline of presentation • Synchronization in Wireless Sensor Networks • What is wrong with Median algorithm ? • Proposed algorithms • Energy and its conservation • Simulation Results • Conclusion and future work

5. Why synchronization? • Having the same notion of time TDMA Slots NTP Data Integration

6. What now ? • Isn’t this a solved problem by now ??? • NTP, MACs with sync built in (802.11), time broadcasts (GPS, WWVB), high-stability oscillators (Rubidium, Cesium) • New problems arise in Wireless Sensor Networks • Important assumptions no longer hold • (fewer resources -- such as energy, good connectivity, infrastructure, size, and cost -- are available) • Sensor apps have stronger requirements • (…but we have to do better than the Internet anyway)

7. Wireless Sensor Networks: Why different ? • Energy limitation • Each node has a limited battery life • Dynamic nature of the network, as well as inaccessibility • Mobility and interference • Diverse applications • Relative and absolute reference • Cost of node • Cheap node

8. Synchronization Methods- Previously • Centralized synchronization • Central reference time • Global or Relative • Receiver-Receiver • RBS ( Receiver Broadcast Synchronization)

9. Synchronization Methods- Previously • Decentralized synchronization • Estimation of the other’s clock time

10. What is NEW here ? • Decentralized synchronization • Relative time references • No time stamping • Low message overhead and no sync message • Primarily using estimation (biased and unbiased) • Energy efficient • Able to use as low energy as possible Unnecessary synchronization wastes energy, and insufficient synchronization leads to poor performance.

11. Definitions • Phase error • Time difference between the clocks • Frequency error • The difference in the rates of the clocks • Clock cycle (clk) • The time between adjacent pulses of the oscillator • Wake-up time • The time that the node starts to listen

12. Sources of Error • Oscillator characteristics: • Accuracy: Difference between ideal frequency and actual frequency of the oscillator. • Stability: Tendency of the oscillator to stay at the same frequency over time. • Network and System Parameters • Receive and Transmit Delay: Time duration between message generation and network injection. • Propagation Delay: Time to travel from sender to receiver. • Access Delay: Time to access the channel.

13. Clock drift The nodes clock time is thus bounded as : where ρ is the maximum clock drift.

14. Synchronization algorithms Synchronization Parameter Space: (max error, lifetime, scope, convergence, stability)

15. Synchronization frequency The period in which the network can stay synchronized without the application of the synchronization algorithm.

16. Mathematical Model What is f ???

17. Median as a method for synchronization • Very simple method. • Calculates the Median of the phase errors • Adjusts the offset in the next wakeup time of the node. • With the simplicity, comes the price…… • Not stable in a highly dynamic networks.

18. What is wrong with Median ? Got message from Node 3 Calculating offset Got message from Node 10

19. A WSN scenario for Median contd. Node 9 drifted from its neighbors

20. Weighted measurements Weight selection : two steps in determining the weight factor where Offset calculation:

21. Weighted Measurements Contd. The larger the phase error is, the lower the value it is assigned.

22. Non Linear Least Squares Contd. Non Linear Least Squares - Curve: Set of data points : Squares of error : where: Iteration of parameters: where:

23. Non Linear Least Squares approach Upper bound is the synchronization error to be permitted….

24. Discrete time kalman filter • Kalman filter • estimates a process by using a form of feedback control • the filter estimates the process state at some time then obtains feedback in the form of (noisy) measurements. • Predict equations • Update equations

25. Kalman Filter Contd.

26. Reducing the duty cycle …….. For a performance improvement: A decrease in the duty cycle:

27. Simulation setup Input Discrete Event Simulation Mixim Octave Output

28. Abstraction

29. Simulation results

30. Simulation results Contd.

31. Simulation results Contd.

32. Simulation results Contd.

33. Simulation results

34. Simulation results Contd.

35. Simulation results Contd.

36. Simulation results Contd.

37. Computation vs. communication energy cost 900 mAh • Tradeoff? • Directly comparing computation/communication energy cost not possible but: put them into perspective! • Energy ratio of “listening for 1 clk ” vs. “computing one instruction per cycle”: In the range of 5 to 10 times • Try to compute instead of communicate whenever possible Communication is more expensive than computing.

38. Energy Consumption Battery life of a MyriaNode vs Guard time

39. Conclusion • A decentralized synchronization for a TDMA-based WSN algorithm is proposed using KF, WM and NLLS. • WM and NLLS have a better tolerance to the dynamic nature of Wireless Sensor Network. • KF performs the best in all synchronization space, both in static as well as dynamic environments. • Median is still the choice in static environment, when considered in all aspects (energy and performance). • Reducing the communication cost and increasing cost of computing is worthy bait.

40. Future work • Software power minimization techniques to reduce the power consumption of the algorithms. • Implementation of the algorithms on the MyriaNode and further investigation for perfection. • Additional tools for frequency error minimization, using the available resources. • More advanced estimation techniques and low power implementation.

41. Thank you!

42. The frequency of the clock is given as : a is the aging factor fe is the environmental factor (temperature..) fr is the noise instability fo is the nominal frequency What causes the clock to drift ? where :

43. Lower bound of synchronization • where n is the number of nodes in the network. • This means that synchronization is not only a local property, in the sense that the clock skew between two nodes depends not only on the distance between the nodes • but also on the size of the network.

44. What is synchronization ? • Having the same time of reference and count the same time [either global (UTC) or local (relative)] • Operate a system in unison • Same notion of time We are synchronized!!!!

45. Wireless Sensor Networks • Wireless network of low cost sensors. • Nodes communicate by broadcasting. • Multihop communication needed in order to reach far-away destinations.