Presenter: Dragos Niculescu

1

• Presenter Dragos Niculescu
• PI Badri Nath
• Site Visit Briefing
• July 2001
• http//www.cs.rutgers.edu/dataman/webdust
• badri_at_cs.rutgers.edu
• Co-PIs Tomasz Imielinski, Rich Martin

2
ad hoc positioning system(APS)

• the problem
• ad hoc deployed nodes should know their location
• sensor networks label reported information
• ad hoc networks help in routing
• why not use GPS? because of
• line of sight
• foliage
• urban canyon
• form factor
• precision
• battery life

3
related work

• centralized solution Berkeley
• positioning using a grid infrastructure UCLA
• CRICKET uses radio and ultrasound MIT
• RADAR premapping of the radio properties of the
  region Microsoft
• positioning relative to a chosen node - EPFL

4
motivation

• We aim for a positioning method that would
• provide global coordinates
• require no additional infrastructure
• allow disconnected regions to function
  independently
• have low overhead for mobility
• provide accuracy comparable to node radio range

5
approach basic idea

(x, y, d)
landmark
6
approach

• a few nodes(landmarks) know their position
• other nodes infer ranges to at least three
  landmarks
• to estimate distances to neighbors use
• signal strength measurement
• hop count
• a hybrid between GPS and DV routing
• DV distances are propagated hop by hop
• GPS each node estimates its own location
• each landmark is treated independently at each
  node
• may use different methods to propagate distance

7
dv-hop propagation method

• standard DV propagation
• insensitive to errors
• each node maintains a table by running standard
  DV
• each landmark
• floods its (x,y) coordinate to its neighbors
• estimates the size of a hop(true dist)/( of
  hops)
• and floods it into the network
• each node
• uses the estimate from the closest landmark
• multiply its hop distances by the estimate

8
dv-distance propagation method

• DV propagation using travel distance, in meters
• sensitive to errors
• each node maintains a table with travel distances
  
• each landmark
• floods its (x,y) coordinate to its neighbors
• computes a correction(true dist.)/(measured
  dist.)
• floods it into the network
• each node
• uses the correction from the closest landmark
• multiply its distances by the correction

9
euclidean propagation method

• node A
• measures distances to immediate neighbors B and C
  
• learns distance BC from either B or C,
• or, possibly infers it by mapping all its
  neighbors
• B and C know their euclidean distances to
  landmark D
• A has to find the diagonal AD

10
simulation scenario isotropic

11
positioning error for regular nodes
dv-distance

12
positioning error for regular nodes euclidean

13
simulation scenario nonisotropic

14
positioning error for regular nodes
dv-distance

15
positioning error for regular nodes euclidean

16
simulation results

• locations obtained are usable for geodesic
  routing
• accuracy of 5-100 relative to radio range
• better accuracy with more landmarks
• "euclidean"
• works well with anisotropic topologies
• higher signalling cost
• predictable performance
• 1 flooding
• "dv-distance", "dv-hop"
• works well with isotropic topologies
• low signalling cost
• 2 floodings

17
conclusions

• APS DV GPS
• distributed
• no infrastructure necessary
• recomputing only for moving nodes
• propagation methods
• "dv-hop"
• "dv-distance"
• "euclidean"
• results
• good accuracy (5-100)
• signalling-accuracy tradeoff

18
status of work/future plans

• Task1 (Completed July 01) APS
• Task2 (year 1) Mobile APS
• some nodes move have to recompute positions
• landmarks move provide more precision
• Task3 (year 2) Query Aggregation
  Methods
• large number of similar, overlapping queries
• how to aggregate them to minimize communication

19
information

• webdust
• http//www.cs.rutgers.edu/dataman/webdust
• Pointers to papers
• http//www.cs.rutgers.edu/dataman/papers/aps.ps -
  admitted into GLOBECOM 2001
• http//www.cs.rutgers.edu/dnicules/research/aps/a
  ps_pres.pdf - slides
• E-mail address
• dnicules_at_cs.rutgers.edu

This research work was supported in part by
DARPA under contract number N-66600-00-1-8953