26
Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada), Yu Liu (University of New Orleans, USA), Jianping Pan (University of Victoria, Canada) Infocom 2012 MengLin, 2012 1

Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

Embed Size (px)

Citation preview

Page 1: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

1

Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor

Networks

Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada), Yu Liu (University of New Orleans, USA), Jianping Pan (University

of Victoria, Canada)

Infocom 2012

MengLin, 2012

Page 2: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

2

Outline

• Introduction• Clock skew measurements• AMKF Clock skew estimation• Environment-aware clock synchronization• Performance study• Conclusions

Page 3: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

3

Introduction

• Clock synchronization is important for network systems such as scheduling

• Improper to do synchronization frequently in WSN due to dynamic and unpredictable environment, ex: sync failure and overhead

• Estimate clock skew accurately can prolong synchronization period while suffering from temperature variance

Page 4: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

4

Introduction

• It is full of noise to measure clock skew directly so they use temperature measurements to assist clock skew estimation, called additional information aided multi-model Kalman filter (AMKF) algorithm, then using this to propose an environment-aware clock synchronization (EACS) scheme to dynamically compensate clock skew

Page 5: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

5

Terminology

• Clock offset is the difference between the time reported by two or more clocks

• Clock skew is the differential coefficient of the clock offset, the tick duration difference

Page 6: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

6

Constant Environment

Page 7: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

7

Dynamic Environment

25 40 10 40 10 25

25 40 10 40 10 25

Page 8: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

8

Outline

• Introduction• Clock skew measurements• AMKF Clock skew estimation• Environment-aware clock synchronization• Performance study• Conclusions

Page 9: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

9

Measurement Setting

• One laptop• Two Mica2 nodes– One senses temperature and send to the other one– The other one sends timestamp containing the

temperature information to laptop through UART

• Use heater to increase and a fan to reduce the environment temperature

Page 10: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

10

Measurement Results and Analysis

• Clock skew– Constant with noise in stable environment– Dynamic as temperature changes

• Hybrid two-model

: the changing rate of clock skew

: the sampling interval

: the processing noise

• Hard to choose model

Page 11: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

11

Measurement Results and Analysis

• Use temperature info to assist choosing model• Calculate the relationship between clock skew

and temperature

• Only -0.505 when using raw measurement data

• Reach -0.973 after using the moving average to smooth measurement results

Page 12: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

12

Outline

• Introduction• Clock skew measurements• AMKF clock skew estimation• Environment-aware clock synchronization• Performance study• Conclusions

Page 13: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

13

What is AMKF

• AMKF is an additional information aided multi-model Kalman filter, which can estimate the model probability for one process using the model probability of another related process of which the model probability is easier to obtain

• Different from the traditional approaches, where the decision is based on the estimated process itself only.

Page 14: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

14

Why Choose AMKF?

• Clock skew is not a stationary random process• Their previous work utilized an IMM Kalman filter

to tackle the model uncertainty in clock skew– Combine and use the weighted sum of several filters’

output as system output– Use Markov chain with preset transition probabilities– In each iteration, every model processes the

measurement data and likelihood function independently

• The measurement noise for temperature is much smaller than that for clock skew

Page 15: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

15

Outline

• Introduction• Clock skew measurements• AMKF Clock skew estimation• Environment-aware clock synchronization• Performance study• Conclusions

Page 16: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

16

How to Do EACS?

• Build a clock skew table containing clock skew w.r.t temperature for each node

• Use a local variable to indicate the accumulated clock offset in n-th time slot since the previous sync.

: instantaneous clock skew based on table

: the temperature sampling rate

• If the clock offset exceeds a given threshold, then

- • If the current temp. is beyond table range, then re-sync

and estimate clock skew as a new entry of table

Page 17: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

17

Outline

• Introduction• Clock skew measurements• AMKF Clock skew estimation• Environment-aware clock synchronization• Performance study• Conclusions

Page 18: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

18

Evaluation of Model Determination

• Based on measurement results to generate simulation traces• Temp. 20 ~ 50; Clock skew 20 ~ 40

Temperature measurements and the probability of constant velocity model

Page 19: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

19

Clock Skew Estimation

The estimation results

Page 20: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

20

Error Estimation

• Root mean square error (RMSE)• Cramér–Rao bound– Determine the lower bound on the variance of

estimators which can be used to indicate the estimation accuracy in this paper

RMSE of clock skew estimation

Page 21: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

21

Performance Evaluation by Simulation

• Below 2ms during whole 8000 s simulation

Simulation setting Clock offset with compensation schemes

Page 22: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

22

Performance Evaluation by Experiment

• Below 8ms over the 7200 s test

Verification trace Verification results

Page 23: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

23

Outline

• Introduction• Clock skew measurements• AMKF Clock skew estimation• Environment-aware clock synchronization• Performance study• Conclusions

Page 24: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

24

Conclusion

• Based on AMKF, using EACS to conduct clock synchronization and compensation, which prolongs the synchronization period

• Good article organization• Too less analysis in real test• The scenario for comparison between simulation

and experiment is different

Page 25: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

25

Q&A

Thanks you!

Page 26: Environment-Aware Clock Skew Estimation and Synchronization for Wireless Sensor Networks Zhe Yang (UVic, Canada), Lin Cai (University of Victoria, Canada),

26

Reference

• Kalman filter– http://en.wikipedia.org/wiki/Kalman_filter

• Stationary process– http://en.wikipedia.org/wiki/Stationary_process– http://cnx.org/content/m10684/latest/