330 likes | 476 Views
Distributed localiza tion in wireless sensor networks. Koen Langendoen Niels Reijers Delft University of Technology The Netherlands. Technology trend. S mall integrated devices Smaller, cheaper, more powerful PDAs, mobile phones Many opportunities, and research areas Power management
E N D
Distributed localization in wireless sensor networks Koen Langendoen Niels Reijers Delft University of Technology The Netherlands
Technology trend • Small integrated devices • Smaller, cheaper, more powerful • PDAs, mobile phones • Many opportunities, and research areas • Power management • Distributed algorithms
Wireless sensor networks • Wireless sensor node • power supply • sensors • embedded processor • wireless link • Many, cheap sensors • wireless easy to install • intelligent collaboration • low-power long lifetime
Possible applications • Fire rescue • breadcrumbs • exit path • hazard detection • Environmental monitoring • detecting forest fires • Monitoring bulk goods (potatoes) • mix sensors with goods • temperature, humidity
Required technologies • Efficient data routing • ad-hoc network • one or more ‘datasinks’ • In-network data processing • large amounts of raw data • limited power and bandwidth • Node localization
Ad-hoc localization • Many nodes (> 100) • NO infrastructure • NO central processing • Sparse anchor nodes • known position • Other nodesdetermine position using this data • Distance measurement
Ad-hoc localization • 2D, static node positions • Several different algorithms have been proposed • 3 will be compared • Simulations on DAS2 supercomputer
Main result no ‘one size fits all’ Best algorithm depends on: • error in range measurement (range variance) • connectivity (number of neighbours) • network topology • node capabilities • application requirements
Three-phase approach • Determine distance to anchor nodes (communication) • Establish position estimates (computation) • Iteratively refine positions using additional range measurements (both)
Phase 1: Distance to anchor • Three algorithms • Sum-dist [Savvides et al.] • DV-Hop [Niculescu et al., Savarese et al.] • Euclidean [Niculescu et al.] • anchors flood network with their known position
6 6 4 5 5 A: 5 B: 6+4 = 10 C: 5+6+4 = 15 Phase 1:Sum-dist Anchors • flood network with known position Nodes • add hop distances • require range measurement B C A
A-B: 12 3 2 4 1 2 3 1 2 1 4 1 2 2 3 Phase 1: DV-hop Anchors • flood network with known position • flood network with avg hop distance Nodes • count #hops to anchors • multiply with avg hop distance 3 hops B avg hop: 4 C A
Phase 1:Euclidean Anchors • flood network with known position Nodes • determine distance by • range measurement • geometric calculation • require range measurement B C A
A-G = 8 Phase 1:Euclidean (2) D • Wanted: Distance A-G G E Using AEGF: A-G = 8 ...or 3 F Using AEGD: A-G = 8 ...or 0.5 A
Phase 1:Euclidean (3) • Needs high connectivity • Error prone (selecting wrong distance) • Perfect accuracy possible B D G E C F A
Phase 1:Comparison • Range measurement • Very accurate: Euclidean • Reasonable: Sum-dist • None / very bad: DV-hop
Phase 2:Determining position • Two algorithms: • Lateration • very common • local triangulation • solve [Ax=b] • Min-max [Savvides et al.] B C A
Phase 2:Min-max • Using range to anchors to determine a bounding box • Use center of box as position estimate B C A
A problem with Min-max Very sensitive to anchor placement
Phase 1 + 2 combined Euclidean: very sensitive to both range variance and connectivity
There is a tradeoff between coverage and error Error and coverage
Phases 1 and 2 • Position error usually 30% of the radio range or higher • Range measurements between nodes only used to determine anchor distance • Can we do better?
Phase 3: Iterative refinement • obtain initial position (phases 1 and 2) • broadcast my position • iteratively refine position using: • ranges to direct neighbours • their initial positions
Broadcast new position Phase 3:Iterative refinement • Initial estimate • Receive neighbour positions • Local lateration A
Conclusion • No ‘one size fits all’ • Refinement needs better coverage to be useful • Lots of room for improvement in all phases • Details in Tech Report PDS-2002-03 (http://pds.twi.tudelft.nl/reports/2002/PDS-2002-003)
What is wrong? • Bad topology • identical hop-TERRAIN positions • twins • Error propagation • rapid infection of complete network • – hop – triangulate – hop – triangulate –
Confidence weights • Weight input for triangulation (wAx = wb) • Initialization • anchors 1.0 • twins, identical hops 0 • others 0.1 • Triangulation • large residue 0 • small residue avg of input confidences