Dynamic%20Clustering%20for%20Acoustic%20Target%20Tracking%20in%20Wireless%20Sensor%20Network

1
Dynamic Clustering forAcoustic Target Tracking
inWireless Sensor Network

• Wei-Peng Chen, Jennifer C. Hou, Lui Sha

Presented by Ray Lam Oct 23, 2004
2
Outline

• Introduction to sensor network
• Technical background for the system
• The dynamic clustering algorithm
• Limitations of the system
• Conclusion

3
Sensor Network

• Nodes in the network
• Sensor to sense physical environment
• On-board processing, limited capability
• Wireless communication
• Limited power from batteries

4
The Network

• The network
• 2 kinds of nodes source and sink
• Wireless network
• Berkeley motes use CSMA MAC
• Ad-hoc type
• Multi-hop routing
• Nodes sleep periodically

5
Data Dissemination

• Some research questions
• How to coordinate sensors?
• How to route data?
• How to do in-network data fusion?
• What to do with congestion?
• How to do the above efficiently
• in terms of energy?
• in terms of time?
• We need distributed solutions

6
The Acoustic Target Tracking System
7
Energy-based Localization

• Signal strength decreases exponentially with
  propagation distance

received signal strength in the ith sensor
strength of an acoustic signal from the target
target position yet to be determined known
position of the ith sensor attenuation
coefficient white Gaussian noise
8
Energy-based Localization

• With a pair of energy readings
• Target is closer to sensor i than to sensor j

j
i
9
Energy-based Localization

• Voronoi diagram
• 2-D space divided into Voronoi cells
• V(pi) Voronoi cell containing node pi
• V(pi) contains all points closer to pi than to
  any other pj
• ri larger than all neighbors readings only if
  target in V(pi)

10
Network Characteristics

• Network structure 2-layer hierarchy
• Static backbone of sparse cluster heads
• Dense sensors for detecting targets
• Radio transmission range 2 signal detection
  range
• Ensure 1 cluster at a time
• Ensure nodes in a cluster hear each other directly

11
The Dynamic Clustering Algorithm

• 4 component mechanisms
• Initial distance calibration and tabulation
• Cluster head (CH) volunteering
• Sensor replying
• Reporting of tracking results

12
Idea of the Algorithm

• Objective minimize messages sent in the network
  and avoid collisions
• Given an energy reading, estimate distance from
  target
• Using Voronoi diagram, estimate probability that
  target is in my Voronoi cell
• In CH volunteering and sensor replying process
• Nodes with high probability speak quickly
• When you hear a higher energy reading from
  others, you give up speaking

13
Initial Distance Calibration and Tabulation

• Each sensor to know 2-D coordinates of all other
  sensors in its transmission range
• Each CH constructs a Voronoi diagram for
  neighboring CHs
• Each sensor (including CH) constructs a Voronoi
  diagram for neighboring sensors

14
Initial Distance Calibration and Tabulation

• Each CHi pre-computes for different d
• Target on the circle centered at CHi with radius
  d
• conditional probability that target
  locates within V(CHi) given d
• 3 cases

15
Three Cases

• d lt radius of inner circle
• d gt radius of outer circle
• In between
• Take sample points on the circle
• Check location of each point
• Estimate as of sample points inside
  V(CHi) / total of sample points

16
Initial Distance Calibration and Tabulation

• Sensors do similarly
• Each sensor Sj pre-computes for
  different
• ri energy reading from CHi
• rj energy reading of Sj
• conditional probability that
  target locates in V(Sj) given

17
CH Volunteering

• Distributed election algorithm
• CH closest to target should be elected
• Solicitation packet
• Request to form cluster and volunteer to be the
  cluster head
• Contains signal signature
• Contains signal strength detected by CH (CHi)

18
CH Volunteering

• Random delay-based broadcast mechanism
• CHi detects a signal, estimates d, checks
• Sets a back-off timer with back-off time
• CHi does not broadcast solicitation packet until
  timer expires
• If during back-off, hears other solicitation
  packets with higher energy readings, gives up
  volunteering

19
Sensor Replying

• Sensor Sj receives a solicitation packet
• Matches signal signature with buffered data
• Upon a match, calculates signal strength rj
• Attempts to send a reply using similar
  delay-based mechanism

20
Sensor Replying

• Random delay-based broadcast mechanism
• Calculates , checks
• Sets back-off timer with back-off time
• If during back-off, hears other reply packets,
  records the sensor that reports largest signal
  strength
• When timer expires, sends reply packet if
• rj higher than all others energy readings or
• Sj is a Voronoi neighbor of the sensor that
  reports the largest signal strength

21
Reporting Tracking Results

• CH receives replies from sensors
• Sufficient number of replies
• A reply from Sj with largest signal strength
• Replies from all Sjs Voronoi neighbors
• Takes location of Sj as location of target
• Sends result to sink through static backbone

22
Limitations

• Limited application space
• Not applicable to general monitoring applications
  without target
• Signals must attenuate with propagation distance
• 1 cluster for 1 signal
• Signals may come simultaneously
• Multiple clusters may form simultaneously causing
  more collisions

23
Limitations

• Energy inefficiency
• Radio transmission range 2 signal detection
  range
• Can be improved by considering multi-hop routing
• Signals at any position must be detected by at
  lease 1 CH
• Tradeoff of sensor density and energy efficiency

24
Conclusion

• Data dissemination in sensor network
• Dynamic clustering triggered per signal
• More research on
• Collision behavior between clusters
• Multi-hop routing
• Time efficient data dissemination

25
Discussion

• The End
• Thank you for coming!