September 16,2003 MobiCom'03 University of Virginia 1 Range-Free Localization Schemes in Large Scale Sensor Networks Tian He Chengdu Huang Brian. M. Blum John A. Stankovic Tarek F. Abdelzaher Department of Computer Science, University of Virginia
September 16,2003 MobiCom'03 University of Virginia 2 Outline Problem Statement State of the Art Motivation & Contribution A.P.I.T. Algorithm Details Evaluation Conclusion
September 16,2003 MobiCom'03 University of Virginia 3 Problem Statement Localization Problem: –How nodes discover their geographic positions in 2D or 3D space? Target Systems: –Static large scale sensor networks or one with a low mobility Goal: –An affordable solution suitable for large-scale deployment with a precision sufficient for many sensor applications.
September 16,2003 MobiCom'03 University of Virginia 4 State of the Art (1) Range-based Fine-grained localizations –TOA (Time of Arrival ): GPS –TDOA (Time Difference of Arrival): MIT Cricket & UCLA AHLos –AOA (Angle of Arrive ): Aviation System and Rutgers APS –RSSI (Receive Signal Strength Indicator) : Microsoft RADAR and UW SpotOn Required Expensive hardware Limited working range ( Dense anchor requirement) Log-normal model doesn ’ t hold well in practice [D. Ganesan]
September 16,2003 MobiCom'03 University of Virginia 5 State of the Art (2) Range-Free Coarse-grained localization –USC/ISI Centroid localization –Rutgers DV-Hop Localization –MIT Amorphous Localization –AT&T Active Badge Simple hardware/ Less accuracy
September 16,2003 MobiCom'03 University of Virginia 6 Motivation High precision in sensor network localization is overkill for a lot of applications. Large scale deployment require cost-effective solutions. Routing Delivery Ratio Entity Tracking Time Under different localization Error ( % Radio Range)
September 16,2003 MobiCom'03 University of Virginia 7 Contributions A novel range-free algorithm with enhanced performance under irregular radio patterns and random node placement with a much smaller overhead than flooding based solutions The first to provide a realistic and detailed quantitative comparison of existing range-free algorithms. First investigation into the effect of localization accuracy on application performance
September 16,2003 MobiCom'03 University of Virginia 8 Overview of APIT Algorithm APIT employs a novel area-based approach. Anchors divide terrain into triangular regions A node’s presence inside or outside of these triangular regions allows a node to narrow the area in which it can potentially reside. The method to do so is called Approximate Point In Triangle Test (APIT). Out IN
September 16,2003 MobiCom'03 University of Virginia 9 APIT Main Algorithm Pseudo Code: Receive beacons (X i,Y i ) from N anchors N anchors form triangles. For ( each triangle T i Є ){ InsideSet Point-In-Triangle-Test (T i ) } Position = COG ( ∩Ti InsideSet); For each node Anchor Beaconing Individual APIT Test Triangle Aggregation Center of Gravity Estim.
September 16,2003 MobiCom'03 University of Virginia 10 Point-In-Triangle-Test For three anchors with known positions: A(a x,a y ), B(b x,b y ), C(c x,c y ), determine whether a point M with an unknown position is inside triangle ∆ABC or not. B(b x,b y ) C(c x,c y ), A(a x,a y ) M
September 16,2003 MobiCom'03 University of Virginia 11 Perfect P.I.T Theory If there exists a direction in which M is departure from points A, B, and C simultaneously, then M is outside of ∆ABC. Otherwise, M is inside ∆ABC. Require approximation for practical use –Nodes can’t move, how to recognize direction of departure –Exhaustive test on all directions is impractical
September 16,2003 MobiCom'03 University of Virginia 12 Departure Test Recognize directions of departure via neighbor exchange 1.Receiving Power Comparison ( the solution we adopt) 2.Smoothed Hop Distance Comparison ( Nagpal 1999 MIT) Experimental Result from Berkeley Experiment Result from UVA
September 16,2003 MobiCom'03 University of Virginia 13 A.P.I.T. Test Approximation: Test only directions towards neighbors –Error in individual test exists, however is relatively small and can be masked by APIT aggregation. APIT(A,B,C,M) = IN APIT(A,B,C,M) = OUT
September 16,2003 MobiCom'03 University of Virginia 14 APIT Aggregation Aggregation provides a good accuracy, even results by individual tests are coarse and error prone. With a density 10 nodes/circle, Average 92% A.P.I.T Test is correct Average 8% A.P.I.T Test is wrong Localization Simulation example Grid-Based Aggregation High Possibility area Low possibility area
September 16,2003 MobiCom'03 University of Virginia 15 Evaluation (1) Comparison with state-of-the art solutions –USC/ISI Centroid localization by N.Bulusu and J. Heidemann 2000 –Rutgers DV-Hop Localization by D.Niculescu and B. Nath 2003 –MIT Amorphous Localization by R. Nagpal 2003 Centroid DV-Hop (online)/ Amorphous (offline)
September 16,2003 MobiCom'03 University of Virginia 16 Evaluation (2) Radio Model: Continuous Radio Variation Model. –Degree of Irregularity (DOI ) is defined as maximum radio range variation per unit degree change in the direction of radio propagation DOI =0 DOI = 0.05 DOI = 0.2 α
September 16,2003 MobiCom'03 University of Virginia 17 Simulation Setup Setup –1000 by 1000m area –2000 ~ 4000 nodes ( random or uniform placement ) –10 to 30 anchors ( random or uniform placement ) –Node density: 6 ~ 20 node/ radio range –Anchor percentage 0.5~2% –90% confidence intervals are within in 5~10% of the mean Metrics –Localization Estimation Error ( normalized to units of radio range) –Communication Overhead in terms of #message
September 16,2003 MobiCom'03 University of Virginia 18 Error Reduction by Increasing #Anchors AH=10~28,ND = 8, ANR = 10, DOI = 0 Placement = UniformPlacement = Random
September 16,2003 MobiCom'03 University of Virginia 19 Error Reduction by Increasing Node Density AH=16, Uniform, AP = 0.6%~2%, ANR = 10 DOI=0.1DOI=0.2
September 16,2003 MobiCom'03 University of Virginia 20 Error Under Varying DOI ND = 8, AH=16, AP = 2%, ANR = 10 Placement = UniformPlacement = Random
September 16,2003 MobiCom'03 University of Virginia 21 Communication Overhead Centroid and APIT –Long beacons DV-Hop and Amorphous –Short beacons Assume: 1 long beacon = Range 2 short beacons = 100 short beacons APIT > Centroid –Neighborhood information exchange DV-Hop > Amorphous –Online HopSize estimation ANR=10, AH = 16, DOI = 0.1, Uniform
September 16,2003 MobiCom'03 University of Virginia 22 Performance Summary CentroidDV-HopAmorphousAPIT Accuracy FairGood Node Density >0>8 >6 Anchor >10>8 >10 ANR >0 >3 DOI Good FairGood GPSError Good FairGood Overhead SmallestLargestLargeSmall
September 16,2003 MobiCom'03 University of Virginia 23 Hermes UVA NEST Demo EnviroTrack Real-Time Routing QoS Scheduling Data Aggregation Lazy Binding MAC Sensing Coverage APIT Localization Mote Test Bed
September 16,2003 MobiCom'03 University of Virginia 24 Conclusions Range-free schemes are cost-effective solutions for large scale sensor networks. Through a robust aggregation, APIT performs best with irregular radio patterns and random node placements APIT performs well with a low communication overhead( e.g instead of 25,000 msgs)
September 16,2003 MobiCom'03 University of Virginia 25 Questions? Thanks
September 16,2003 MobiCom'03 University of Virginia 26 Error Case Since the number of neighbors is limited, an exhaustive test on every direction is impossible. – InToOut Error can happen when M is near the edge of the triangle –OutToIn Error can happen with irregular placement of neighbors PIT = IN while APIT = OUT PIT = OUT while APIT = IN
September 16,2003 MobiCom'03 University of Virginia 27 Empirical Study on APIT Approximation Percentage of error due to APIT approximation is relatively small (e.g. 14% in the worst case, 8% when density is 10) More important, Errors can be masked by APIT aggregation. APIT Error under Varying Node Densities