Lightning: A fast and lightweight acoustic localization protocol using lowend wireless microsensors'

1
LightningA fast and lightweight acoustic
localization protocol using low-end wireless
micro-sensors.

• Qixin Wang, Rong Zheng, Ajay Tirumala, Xue Liu
  and Lui Sha.

2
Outline of the presentation

• Demand
• Observations and Solution Heuristics
• Protocol Details
• Theorems and Experiment Results
• Demo Video
• Conclusion

3
Demand

• Want a fast, deterministic (i.e.O(1) response
  time) acoustic event localization scheme.
• Fits low-end wireless micro-sensor networking.
• Proximity localization, i.e. electing the closest
  sensor, is good enough.

4
Observations

• Sound sources are often directional and of
  unknown intensities. This implies
• Intensity-based localization is not desirable.
• Time-Of-Arrival (TOA) is a more reliable metric.
  When sensors are densely deployed, Line-Of-Sight
  to the closest sensor can usually be guaranteed,
  therefore earliest TOA ?? closest sensor.
• Radio (RF) wave travels much faster than acoustic
  wave.
• When a lightning strikes, people see the
  lightning before hearing the rumbling of the
  thunder.

5
Solution Heuristic I

• When sound reaches the closest sensor, the
  closest sensor should immediately announce the
  event (via RF broadcast) to all other sensors and
  suppress them even before they hear the sound.

6
Observation

• Immediate Data Packet (DP) RF broadcast is not
  practical, because of collisions.
• To think several sensors, all almost the same
  distance to the sound source, try to broadcast
  data packets at almost the same time.

7
Solution Heuristic II

• Do NOT use data packet broadcast, broadcast RF
  burst instead.RF burst is not susceptible to
  overlapping.

8
Observation

• If there are multiple closest sensors, there can
  be multiple election winners.
• How to guarantee every time there is only one
  winner?

9
Solution Heuristics III

• Color the sensors, to differentiate RF burst
  duration, to break ties.
• It is proven, with regular sensor layout and
  proper coloring, it is guaranteed to always elect
  one winner sensor.
• (To be included in our upcoming publications)
  Empirically, even with random sensor layout and
  without coloring, the number of winners is still
  well limited.

10
Remark

• Heuristics I, II and III lead to the design of
  Basic Lightning Protocol.

11
Observation

• Energy cost is a concern.
• Currently, a sensor has to have RF on all the
  time to listen to possible RF bursts. Remember a
  sensor to be suppressed receives RF burst before
  hears the sound.
• How to have RF module sleep during most of the
  time and only be turned on when there is an
  acoustic event?

12
Solution Heuristics IV

• RF Sleep during usual time.
• When hears a sound, turn on RF and RF listen for
  ?defer sec, to make sure all other sensors that
  can hear the sound have turned on their radios.
  Then carry out the same procedure as Basic
  Lightning Protocol.
• Equivalent to the sound takes place ?defer sec
  later in real-world, and Basic Lightning Protocol
  is deployed.

13
Remark

• Heuristics IV leads to the design of
  Energy-Efficient Lightning Protocol.

14
Basic Lightning Protocol

• Regular sensor layout with certain coloring
• All sensors are initially in RF-listen mode
• Beep recognized
• broadcast RF burst without backoff for iTburst.
• Listen for RF bursts for Tb.
• No other burst recognized elected
• Other burst recognized supressed
• During RF listen
• RF burst recognized
• Enter supressed mode
• Reenter RF listen mode after basic timer expires.

RF burst for i.Tburst
RF Listen
Beep recognized
Timer expires
Burst recognized
Supressed Set reset timer
Burst recognized
Post burst listen
Timer expires
Elected set reset timer
No burst recognized
15
Energy-efficient lightning protocol
RF listen for ?defer

• All sensors initially in RF sleeping
• When beep is recognized
• Listen for RF bursts but defer bursting for
  ?defer
• If no burst is recognized, transmit RF burst
  without backoff,for iTburst sec.
• Post burst RF listen
• No other burst recognized elected
• Other burst recognized suppressed
• If burst is recognized in RF listen enter
  suppressed mode
• When reset timer expires, return to RF Sleeping
  mode

RF Sleeping
Beep recognized
Timer expires
Burst recognized
Supressed Set reset timer
RF burst for i.Tburst
Burst recognized
Timer expires
Post burst listen
Elected set reset timer
No burst recognized
16
Key Properties

• Theorem 1,2 Both Basic and Energy-Efficient
  Lightning Protocol elect a unique winner with
  deterministic localization distance error.
• Corollary 1,2 Both have a short and O(1) time
  bound for the election.

17
Experiment results

• Implemented on U.C. Berkeley MICA Motes
• Directional Sound Source

18
Experiment results

• Comparable, or even better accuracy than ideal
  (no pkt loss) data packet based localization.

19
Experiment Results

• Fast in the sense that the Upper Bound of
  Lightning Protocol Election Delay is shorter than
  the Lower Bound of Data Packet Election Protocol

20
Experiment Results

• Deterministic Election Delay, while data packet
  localizations election delay are random due to
  MAC contention.

21
Experiment Results

• O(1) broadcasts
• Theoretical Guaranteed to be lt 4.
• Experiment Never more than 2.

22
Demo video (Qixin Wang)
URL http//www-rtsl.cs.uiuc.edu/papers/LightningD
emo.html See reference 13
23
Conclusion

• Fast and Deterministic
• O(1) election delay.
• O(1) number of broadcasts.
• Lightweight
• only involves simple comparison (ltgt) operation
• no clock synchronization needed
• simple network stack.
• Comparable, or even better localization accuracy
  than ideal data packet scheme.
• Robust
• Immune to RF broadcast overlapping
• Handles directional sound source
• Energy Efficient
• only turns on RF module when there is an acoustic
  event.

24
Thank you!