Location Centric Distributed Computation and Signal Processing

1
Location Centric Distributed Computation and
Signal Processing

• University of Wisconsin, Madison
• Acknowledgements
• DARPA SensIT Contract F30602-00-2-0055

2
Team Members

• Faculty
• P. Ramanathan
• A. Sayeed

• Y.-H. Hu
• K. K. Saluja

• Students
• K.-C. Wang
• T.-L. Chin
• T. Clouqueur

• A. Ashraf
• A. DCosta
• M. Duarte

• D. Li
• X. Sheng
• V. Phipatansuphorn

3
Sensor Network Characteristics

• Commands/queries are typically issued to a
  geographic region instead of specific nodes, such
  as
• Average temperature in a given region
• Unidentified object counts in a given region
• Specific object tracking in a given region
• Only devices in the specified geographic region
  need to participate in executing a command/query
• Solution Location-Centric Computing

4
Location-Centric Computing?

• All nodes are aware of own location (GPS).
• However, geographic regions are the only
  addressable entities.
• Regions play the traditional role of a node.
• Regions are created before commands are issued.
• Each node participates in activities of regions
  it belongs to.
• Embedded manager region coordinates intra-region
  activities.

5
Implementation

• Application Programming Interface
• UW-API
• Set of communication primitives tailored for
  location centric information exchange
• Networking Support
• UW-Routing
• Location-aware routing scheme for wireless ad
  hoc sensor networks

6
UW-API

• Data exchange primitives
• Affinity with standard message passing interface
  for distributed computing
• E.g. the well-known MPI 1.1 library
• Regional communication services
• Send, Receive, Reduce, Barrier
• Administrative primitives
• Create_region and delete_region

7
UW-API

• Example SN_Send
• Sends a message from a node to all nodes in the
  addressed region
• Used to send commands and data
• Example SN_Reduce
• Aggregates data within a region.
• Aggregates data as min, max, average, sum, .
• Collect results in embedded manager region.

8
UW-Routing

• A location-aware on-demand routing protocol
• Each node maintains a routing table of paired
  destination region, next hop entries.
• Routing entry is created on demand using
  RouteRequest (RREQ) and RouteReply (RREP).
• RREQ and RREP flood in a limited scope similar to
  Location-aided Routing Vaidya .

9
Inter-Region and Intra-Region Routing for Send

• Message sent from a source node to a region.

• Message flooded to all nodes in the region.

10
Collaborative Target Tracking

• Create regions at possible entry points.
• Nodes in the created regions collaborate to
  detect any entering targets.

11
Collaborative Target Tracking

• When a target is detected, nodes in the region
  start localization and tracking.
• Tracking results are used to estimate future
  target locations.
• Additional regions are created in possible target
  locations.

12
Target Detection

• At each node Energy Detector
• Dynamic noise level estimation.
• Target detected if received energy exceeds
  estimated noise level.
• Constant false alarm rate maintained.

• Within a region Decision fusion
• Nodal detection decisions sent to manager nodes.
• Manager nodes fuse nodal decisions into regional
  decision.
• Different fusion wrights for different modalities.

13
Target Classification

• Within a region
• Nodal classification decisions sent to manager
  nodes.
• Manager nodes fuse nodal decisions into regional
  decision.
• Target type with most votes becomes winner.

• At each node
• Maximum Likelihood classifier using Gaussian
  model for the classes
• Training done using SITEX02 data
• Acoustic modality only
• Decisions as AAV, DW, HMMWV, or Unknown

14
Target Localization

• PIR Modality
• Estimate target to be at the projection from a
  detecting node onto the road.
• PIR Localizations occur infrequently as compared
  to acoustic.
• PIR localizations are less susceptible to noise.

• Acoustic Modality
• Energy based localization
• Let Ei be the energy reading at sensor node i
  located at ri.
• Estimate target to be at a location where

15
Target Tracking and Prediction

• Tracker is based on Kalman Filter.
• Tracker assigns different weights to acoustic and
  PIR localization estimates.
• Tracker provides feedback to acoustic based
  localization in terms of better refined search
  area.
• Tracker predicts location of target in the near
  future.
• Tracker predictions are used to create and
  activate additional regions for possible target
  detection.

16
UW-Senware
17
Case Studies on Two Testbeds

• Single vehicle (AAV, DW, HMMWV) traversing
  SITEX02 sensor field
• SITEX02 timeseries collected in 29 Palms, CA. on
  Nov 14, 2002
• Two vehicles (AAV and DW) crossing each other in
  Waltham sensor field
• Synthetic timeseries based on SITEX02 data
• Two vehicle (AAV and DW) meeting each other and
  turning back in Waltham sensor field
• Synthetic timeseries based on SITEX02 data

18
Control vs Payload Messages (AAV run SITEX02)
19
In-Region vs Out-of-Region Messages (AAV run
SITEX02)
20
Packets Injected vs Forwarded (AAV run SITEX02)
21
Per Node Bandwidth Consumed (AAV run SITEX02)
22
Time Trace of Messages Sent/Forwarded (AAV
SITEX02)
23
Target Detection(AAV SITEX02)
24
Target Classification(AAV SITEX02)
25
Target Tracking (AAV SITEX02)
26
Publications (1 of 2)

• R. Brooks, P. Ramanathan, and A. Sayeed,
  Distributed target classification and tracking
  in sensor networks, Submitted to IEEE
  Proceedings.
• P. Ramanathan, K. K. Saluja, and Y.-H. Hu,
  Collaborative sensor signal processing for
  target detection, localization, and tracking, To
  appear in Army Sciences Conferecens, December
  2002.
• T. Clouqueur, V. Phipatansuphorn, P. Ramanathan,
  and K. K. Saluja, Sensor deployment strategy for
  target detection, Workshop on Sensor Networks
  and Applications, September 2002.
• A. DCosta, Y.-H. Hu, and A. M. Sayeed,
  Classification of targets using multiple sensing
  modalities in distributed micro-sensor networks,
  Submitted for publication.
• V. Phipatansuphorn and P. Ramanathan,
  Vulnerability of sensor networks to unauthorized
  traversal and monitoring, Submitted to IEEE
  Transactions on Computers, April 2002.

27
Publications (2 of 2)

• T. Clouqueur, K. K. Saluja, and P. Ramanathan,
  Fault tolerance in collaborative sensor networks
  for target detection, Submitted to IEEE
  Transactions on Computers, April 2002.
• D. Li and Y.-H. Hu, Energy based collaborative
  source localization using acoustic microsensor
  array, Submitted for publication, February 2002.
• D. Li, K. Wong, Y.-H. Hu, and A. Sayeed,
  Detection, classification, and tracking of
  targets, IEEE Signal Processing Magazine, March
  2002.
• K.-C. Wang and P. Ramanathan, Multiuser receiver
  aware multicast in CDMA-based multihop wireless
  networks, in Proceedings of Mobihoc, pp.
  291-294, October 2001.
• T. Clouqueur, P. Ramanathan, K. K. Saluja, and
  K.-C. Wang, Value-fusion versus decision-fusion
  for fault-tolerance in collaborative target
  detection in sensor networks, in Proceedings of
  Fourth International Conference on Information
  Fusion, August 2001.