Position Location using Radio Fingerprints in Wireless Networks Prashant Krishnamurthy Graduate Program in Telecom & Networking
Agenda Introduction Radio Fingerprints What Industry is Doing Research Conclusions
Introduction Why Position Location? Location based services Driving directions, concierge services, etc. E-911 mandate 100m at least 67% of the time and within 300m at least 95% of the time Metrics Accuracy (e.g., 100m) Precision (e.g., 67%) Delay Coverage Capacity
Algorithms for Position Location Association to a Point of Access (POA) Time or Time Difference of Arrival (TOA/TDOA) Other distance or range based schemes Angle or Direction of Arrival (AOA/DOA) Radio Fingerprinting
Algorithms Again Tx-A Tx-B.. Rx Tx-1 Tx-2 (a) Cell-ID based position location Rx (c) TDOA based position location Tx-3 Tx-1 Tx-2 (b) TOA based position location Rx Rx (d) AOA based position location Tx-3.. Tx-1 Tx-2
Remarks (1) Cell-ID (POA) 43% of the time, a MS may associate itself with a base station that is NOT closest to it Poor accuracy - 800m in NY area TOA/TDOA approaches Several standards in cellular networks Provide reasonable accuracy
Remarks (2) AOA/DOA Techniques Many cells use omnidirectional antennas 120 o antennas have large beamwidths to accurately estimate directions Not part of any standard
Radio Fingerprinting: Idea Fingerprint at grid location r 2 Decision Boundary AP2 AP1 RSS from AP 2 Estimated Location RSS from AP 1 r 1 Access Point Grid Point Fingerprint Sample RSS vector Idea for WiFi (with some measurements) was first published by researchers from Microsoft
What makes up a Fingerprint? Any unique characteristic that differentiates location Common to use RSS from multiple base stations or access points Others: Signal-to-Interference, time delays, cell- IDs seen, etc. Match observed sample with entries in database to estimate location Exact matches are unlikely - errors
Why Fingerprinting? Multipath propagation Impacts error with TOA/TDOA and AOA techniques Beneficial in the case of fingerprinting Software only approach No new hardware, spectrum, or sensing technologies outside of what already exists Improved time to fix Lower power consumption (compared to GPS)
Why Not Fingerprinting? Database of fingerprints is laborious to create Unclear how much information needs to be stored Too much or too little? Censored data Database may have to be regularly updated New cells, change in environment, etc. Self-healing?
Fingerprinting in Cellular- Only Networks Comparison with Assisted-GPS in mix of indoor and outdoor test points Blind trial in New York City and Toronto by operators Polaris Wireless judged the best Accuracy Precision < 50m 74%, 69% < 100m 91%, 90% < 150m 99%, 96% < 200, 300m 100% Source: M. J. Feuerstein, "Urban and Indoor Location using Pattern Matching of Wireless Network Measurements," Invited Workshop on Opportunistic RF Localization for Next Generation Wireless Devices, Worcester Polytechnic Institute, June 2008.
Hybrid Positioning in Cellular-Only Networks Combining techniques improves accuracy and precision WLS = Wireless Location Signature Source: M. J. Feuerstein, "Urban and Indoor Location using Pattern Matching of Wireless Network Measurements," Invited Workshop on Opportunistic RF Localization for Next Generation Wireless Devices, Worcester Polytechnic Institute, June 2008.
Using WiFi with Cellular and GPS Approach made popular by Skyhook and iphone XPS - Hybrid Positioning WPS - WiFi Positioning Available for Windows Mobile devices as well Without Skyhook With Skyhook Source: Skyhook Wireless
Combining WiFi with GPS and Cellular Why? Over 50 million WiFi APs deployed 26 million in the US Downtown area - average of 10-18 APs detected in any location Ideal to use SSIDs & RSS as the radio fingerprints Use only 2 GPS satellites with the radio fingerprints Source: F. Alizadeh, Opportunistic vs Hybrid positioning, Invited Workshop on Opportunistic RF Localization for Next Generation Wireless Devices, Worcester Polytechnic Institute, June 2008.
Performance of Skyhook s WPS HTC Tilt Manhattan & San Francisco 60 outdoor and 40 indoor points Indoor Outdoor 50% Prec 95% Prec Better accuracy than GPS Better coverage than GPS Source: F. Alizadeh, Opportunistic vs Hybrid positioning, Invited Workshop on Opportunistic RF Localization for Next Generation Wireless Devices, Worcester Polytechnic Institute, June 2008.
WPS/XPS Performance Technique WPS XPS 50% Prec. 68 m 44 m Power Consumption (lower estimate) One Shot 50% Tracking, 50% one-shot 95% Prec. 117 m 97 m Recent work at Skyhook Prototype results Source: F. Alizadeh, Opportunistic vs Hybrid positioning, Invited Workshop on Opportunistic RF Localization for Next Generation Wireless Devices, Worcester Polytechnic Institute, June 2008.
Research Goals Better understanding for Indoor-Only environments with WiFi Can we develop a model that can predict the accuracy and precision? Factors: Number of access points, path-loss exponent, variability of RSS, how close should grid points be Can we use the model to develop guidelines for system deployment?
Challenges and Highlights Lacked measurement data We took extensive measurements in the IS Building and Hillman Library Variability of the RSS is not Gaussian Even if it is assumed to be Gaussian, the constellation of fingerprints is highly irregular Employed concept of neighborhood graphs to improve model s precision We observe clustering - actually good news for system deployment
People Two Ph.D. students Kamol Kaemerungsi (2004) Nattapong Swangmuang (2008) Future? Impact of censored fingerprint data? Extension to ad hoc/sensor networks? Impact of/on dynamic spectrum access?
Thank You!