Measurement confidence in a real sensor network S ebastien Cevey - - PowerPoint PPT Presentation

Introduction Materials Experiments Results Conclusion

Swarm Intelligence Project

Measurement confidence in a real sensor network

S´ ebastien Cevey Sandro Stucki 9 February 2006 / EPFL

Introduction Materials Experiments Results Conclusion

Overview

Introduction Materials Experiments Results Conclusion

Introduction Materials Experiments Results Conclusion

Problem

• Sensor networks efficient at gathering data over a wide area
• Nodes (AKA motes) may encounter failures :
• Erroneous sensing
• Transmission packet loss
• System crash
• Need a way to assess the quality of acquired data

Introduction Materials Experiments Results Conclusion

Goal

Exploit the distributed nature of sensor networks :

• Use neighbors’ measurements to reinforce confidence
• Distribute computation of result value

= ⇒ Improve quality of measurements for various network layouts

Introduction Materials Experiments Results Conclusion

Materials

Hardware MICAz motes : light sensor, radio System TinyOS : component-based Language NesC : asynchroneous, C-like Base station Megabase client in Java

Introduction Materials Experiments Results Conclusion

Method : state-based controller

ADC WARMUP SYNC LISTEN BCST RESULT BCST SLEEP INIT after 1st send delay after warmup delay after received sync msg after last send + sleep delay after 2nd send delay immediately

• Initial synchronization (sync message from Megabase)
• Two broadcasts (timer triggered) : measurement, evaluation
• Inactive period (sleep)

Introduction Materials Experiments Results Conclusion

Method : linear result computation

Measurement vector : − → d Evaluation matrix : M Result vector : − → r = M · − → d Example evaluation matrix (Gaussian filter) : M =          6 1 · · · 1 1 6 1 1 6 . . . ... . . . 6 1 1 · · · 1 6         

Introduction Materials Experiments Results Conclusion

Experiment 1 : Distributed Filtering

• Rotating lamp on circle of motes
• Measurements every two cycles (4 seconds)

after 2 cycles 2 3 4 5 6 7 8 9 10 1

Experiment 1A Identity matrix as evaluation matrix Experiment 1B Gaussian filter of size 3

Introduction Materials Experiments Results Conclusion

Experiment 2 : Sensing under Real Conditions

BC217 BC225 BC221 5/6 3/4 7/8 9/10 BC294.6 (Corridor) Station Base 1/2

• Real environment (rooms in BC)
• Measurements every minute during 24 hours
• Considering both pair motes (weight 2) and two neighboring

pairs (weight 1)

Introduction Materials Experiments Results Conclusion

Results for Experiment 1

• Light travelling wave visible in both experiments
• Mote 3 not working in Experiment 1A
• Results smoother in Experiment 1B (Gaussian filter)

Introduction Materials Experiments Results Conclusion

Results for Experiment 2 : Measurements of all motes

400 500 600 700 800 900 1000 200 400 600 800 1000 1200 Result Time (min) Experiment 2 (altered) Mote 1 Mote 2 Mote 3 Mote 4 Mote 5 Mote 6 Mote 7 Mote 8 Mote 9 Mote 10

Light evolution matches closely with environment (daylight, lamps), expected results.

Introduction Materials Experiments Results Conclusion

Results for Experiment 2 : Measurements vs. Results comparison for a pair of motes

1000 2000 3000 4000 5000 6000 7000 8000 200 400 600 800 1000 1200 1400 Result Measurement Time (min) Experiment 2 (results vs measurements) Mote 7 results Mote 8 results Mote 7 measurements Mote 8 measurements

Clean measurements, noisy computed results (due to packet loss).

Introduction Materials Experiments Results Conclusion

Conclusion

• Flexible system adaptable to varied network layouts using

the evaluation matrix.

• Disappointing results due to bad handling of packet loss

(easy to improve).

• Interesting insight in distributed computations, as needed

in large-scale sensor networks.