Lower-Power Distributed Event Detection in WSNs

1. Lower-Power Distributed Event Detection in WSNs Y. Zhu, Y. Liu, M. Ni, Z. Zhang Presented by Shan Gao

2. Event Detection • Detect events which may occur momentarily. • Two essential properties • Physical events are usually persistent, which can last for seconds or even longer. • Many applications accept a certain detection delay. • Duty cycling is a fundamental approach to conserving energy in WSNs.

3. Duty Cycle: δ = τon / τcycle • To conserve energy, • shorten τon • τon = τwarmup+ τsensing+ τprocessing+ τcommunication • Usually τon is fixed, tens of milliseconds • Lengthen τcycle • Problem • A longer τcycle leads to a longer delay and lower detectability.

4. Tradeoff • Longer τcycle • Longer delay • Lower detectability • Longer network lifetime • τcycle ≥ τevent • Detectability is 100% • τcycle < τevent • Some events possibly won’t be detected. • Users • Network lifetime   Delay & detectability  τcycle

5. RIW • Random independent wakeup • Simple but low efficient • The sensors which are close to each other may wakeup at about the same time due to the lack of awareness about their neighborhood.

6. CAS • Coordinated Wakeup Scheduling • Fully localized algorithm • The sensors reside closely should separate their wakeups as much as possible such that the detection delay can be minimized. • Stage 1: Initialization • Stage 2: Sensors wakeup and detect events at the time determined in Stage 1.

7. distributed scheduling coordination • Each sensor need to cooperate with neighbors to determine their wakeup time. • Tasks: • Identify neighbors, which are sensors within CR (cooperative range), 0 < CR <= 2Rs • Determine wakeup time, meanwhile reduce the detection delay.

8. Phase 1: Each sensor randomly picks up a wakeup time and broadcasts it to its neighbors. • Phase 2: Each sensor does multiple rounds of Aggressive Wakeup Adjustment. • For keeping synchronization, in each round, only one sensor can broadcast its adjustment and its neighbors update their own schedule table according to this adjustment.

9. Backoff technique • After using AWA to calculate the adjustment and waiting for a random time, one sensor broadcast this adjustment in a UPDT message. • Upon receiving this UPDT message, if the sender is in the receivers neighbor list, the receiver cancel sending its own adjustment request and update its own wakeup time table. • Multiple rounds of AWA is necessary to reach a reasonable schedule plan. • Usually each sensor need to send 2 UPDT messages successfully at least.

10. Aggressive Wakeup Adjustment • A sensor need to determine • whether it should adjust its wakeup time • and what the new wakeup is • For each sensor, it is desirable to evenly distribute the wakeups of its cooperative neighbors and itself over τcycle. • Wakeup Separation: t1, t2, t3 t1 t2 t3 t1 0 s1 s2 s3 τcycle

11. s1 • Variance: difference between wakeup separations • If the new variance is decreased, s1 broadcasts a UPDT message. t1 t1 t2 t3 t3 t1 t1 0 0 s2 s2 s3 s3 s1 τcycle τcycle

12. Performance Evaluation • CR

13. Τcycle = 10