561 likes | 724 Views
Node Localization in Wireless Sensor Networks. Radu Stoleru Computer Science University of Virginia. Motivation. Sensor. Radio link. Sensor. Sensor. Sensor. Credits: OTB News. Motivation. Credits: Matt Welsh (Harvard). Motivation. Credits: www.visitingdc.com. Enabling Technology?.
E N D
Node Localization in Wireless Sensor Networks Radu Stoleru Computer Science University of Virginia
Motivation Sensor Radio link Sensor Sensor Sensor Credits: OTB News
Motivation Credits: Matt Welsh (Harvard)
Motivation Credits: www.visitingdc.com
Enabling Technology? Mica Dot Telos Mica2 Pluto ETag Mica Mote EmberNet DSYS25 XYZ BT Node
Existing WSN Applications Military (VigilNet, Virginia) Volcano Monitoring (Harvard) Structural Monitoring (Berkeley) Credits: OTB News, www.visitingdc.com. Matt Welsh (Harward)
How to Localize Nodes? • Why node localization is a Hard problem? WiFi Hotspots Outdoor GPS Expensive,Precise Cheap, Inaccurate Range-Based absolute distance/angle estimates Range-Free use only connectivity and proximity
Outline Contributions • Node Localization in WSN • Event-Based Localization • Asymmetric Architecture • Image-Based Passive Localization • Asymmetric Architecture • Implementations, Real Evaluations • Robust Localization Framework • Proximity to Deployment Area, Deployment Knowledge • Integrated WSN Systems • Programming Abstractions Spotlight (SenSys ’05) StarDust (SenSys ’06) They work too! EmNetS ‘07 SECON ’04, EmNetS ‘04 Mobisys ‘04, RTAS ‘05, INFOCOM ‘06, TOSN ‘06 ICDCS ‘04
Concept of Operation (X1, Y1, R1) at T1 (X2, Y2, R2) at T2 (X2, Y2, R2) (X1, Y1, R1)
Questions • How to efficiently distribute events in the network? (trade-off: energy spent for event creation vs. accuracy) • How to deal with sensing channel noise? • How to deal with unrecoverable errors in the sensing channel? Can we do better than loosing everything? • What is the range of the system? • How long does it take to localize a network?
System Architecture Generates Events Timestamp Events Report Timestamps Compute Location Disseminate Location Asymmetric Architecture Event Distribution Function (EDF): the core of Spotlight We propose 3 Functions
0 L Point Scan EDF Light Spot Sensor s
Area Cover EDF Event Detection Errors?
Area Cover EDF Event Detection Errors? Use ECC, e.g. Hamming(7,4) dddpdpp Does the code placement matter? What if we can not even detect an error?
Area Cover EDF Code placement strategies Code placement, No ECC, 1-bit error Code placement, with ECC, 3-bit error
Design Analysis • Execution Cost assuming: • All nodes in a square area, with length D • N events / unit time generated by the Spotlight device • r is tolerable localization error Spotlight Power r2N D r N D2 N / 2
Implementation • mSpotlight System • Spotlight Device: • Projector • Laptop • Mica2 motes • Short range (10-20m) • Versatile, it generates: • Point Scan • Line Scan • Area Cover
Implementation • Spotlight System • Spotlight Device: • Telescope Mount • Diode Laser • Laptop • XSM motes • Long range (>1000m) • It generates: • Point Scan • Line Scan
Spotlight - Point Scan EDF Localization of an area 100x40 m2
Corner-Cube Retroreflector • Problem: Sensor Nodes have limited power supply. Hence, impossible to shine like a star.
Concept of Operation ID ? ID ? ID ? ID ? ID ? ID ? ID ? We need to match node IDs to locations!
System Architecture • Take two pictures, one without light and one with light • Image Processing to identify the positions of the light-spots • Label relaxation algorithm to associate the positions with node Asymmetric Architecture
Image Processing • A football stadium where we deploy 6 sensor nodes in a 3x2 grid. The distance between the lighting device and the sensor nodes is approximately 500ft. Deployment area Without illumination With illumination
Image Processing • Each pixel P is described by an RGB Value • The light reflected back by CCR has highest intensity. • Edge Detection is applied to identify the location of light spot Difference EdgeDetection
System Architecture ID matching turns out to be a very difficult problem! Isomorphism V’ G(L,E) G’(V’,E’)
Mapping: Positions ↔ Nodes • Idea I: Use constraints • Color constraints • Connectivity constraints • Spatial Deployment constraints • Temporal constraints • Idea II: Combine different constraints • Idea III: Label relaxation algorithm
Color Constraints • Use color to map positions to nodes. • Two extreme Cases • Each node has different CCR color (unique mapping). • Only one color available (no constraints)
Connectivity Constraints ? • A pair of nodes must be located near each other if they can communicate with each other.
Connectivity Constraints Radio Model: N nodes in square of size L; k = # unidirectional radio connections; R = radio range N = 26 L = 60ft K = 180 R = 25ft
Spatial Deployment Constraints • Drop sensor nodes at different locations • Wind affects the size of landing area • x-y displacement follows Rayleigh distribution A B A B Landing area Probabilistic Mapping Unique Mapping
Time Constraints • Drop sensor nodes at different times • Two extreme cases • Drop one by one. Take a picture after each drop. • Drop all at the same time. No constraints.
Mapping: Positions ↔ Nodes • Idea I: Use constraints • Color constraints • Connectivity constraints • Spatial Deployment constraints • Temporal constraints • Idea II: Combine different constraints • Idea III: Label relaxation algorithm
Hybrid constraints? • Color & Connectivity: Introduction of the label relaxation algorithm ni can not be 8 or 1 nj can not be 3 Fact: ni is connected with nj Support
Label relaxation algorithm denotes the probability node ni has label λk denotes thesupportfor label λkof node ni The label relaxation algorithm is iterative is a normalization factor to ensure S is the number of iterations
Example of Supports • Connectivity constraint supports according to the link quality ( number of packets received) • Time constraints supports • Color constraint supports • Space constraints supports according to displacement distance
Experiments • Outdoor deployment of 26 nodes in a 120X60 ft2 area • Investigate hybrid solution: connectivity and color constraints Credits: Shan Lin
Coloring Space Size Larger color space means dramatically improved performance Colors randomly assigned to nodes.
Color Uniqueness It is better to assign one color to multiple nodes. Colors randomly assigned to nodes
Localization Time Improved time (w.r.t. Color) Improved accuracy (w.r.t. Connectivity) Assume 50 colors available.
Error Tolerance Tolerate 10% false negatives, for double of localization error) False Negatives: influence of not identifying nodes in the picture
StarDust Summary • Accuracy • 5 feet error under 12 colors with 91% correctness in mapping • Flexible framework • Allows for novel constraints => higher accuracy • Range • 1500 ~4500 ft (clearness of the atmosphere) • Time • 52 seconds (our case) for color and connectivity. • Milliseconds with unique colors. • Cost • less than $0.1 per node