1. Precise Indoor Localization using PHY Layer Information Aditya Dhakal

2. Localization To be able to locate the user to a certain area. Many methods exists for localization. Global Positioning System, Triangulation, dead-reckoning, guessing etc.? Indoor localization? Still a challenge.

3. What are the Challenges? GPS (Outdoor Localization) It can be accurate to 5 meter radius and still functional Signal is hard to get indoors Might not be precise for indoor use Most other localization methods are even worse in terms of accuracy

4. Super Market Layout

5. Existing Systems Cricket: utilizes ultrasound/Radio-based infrastructure installed on ceilings to measure position very accurately. Horus: Utilizes signal strength coming from multiple APs of 802.11 Wireless LAN UnLoc: Dead reckoning combined with land marking system.

6. PinLoc Precise indoor localization Utilizes detailed physical (PHY) layer information Multipath signals components arrive in a given location with distinct phase and magnitude

8. PinLoc The distinct value of phases and magnitude aggregated over multiple OFDM sub-carriers in 802.11 can provide a finger print of a location. Gathering data over all possible location in room can make a map that can be used to locate user.

9. PinLoc

10. Background of the Technique How is information transmitted in modern digital radios using OFDM. Y(f) = H(f)X(f) Where Y(f) is received symbol, X(f) is transmitted symbol and vector H is called channel frequency response (CFR)

11. Background of the Technique CFR changes entirely once transmitter or a receiver moves more than a fraction of a wavelength. (12 cm for WiFi radio) CFR experiences channel fading due to changes in the environment at different time-scales

12. Hypotheses 1.The CFRs at each location look random but exhibit a statistical structure. 2.The “size” of the location (over which the CFR structure is defined and preserved) is small. 3.The CFR structure of a give location is different from structures of all other locations.

13. Experiments to Verify Hypotheses The CFRs at each location appear random but actually exhibit a statistical structure over time.

14. Experiments to Verify Hypotheses

16. The process of Clustering

17. Statistical Structure of CFR Temporal Stability of cluster: - The clusters ought to be stable to be able to be used in localization

18. Statistical Structure of CFR

19. Size of the Location WiFi has wavelength of 12cm. CFR cross-correlation drifts apart with increasing distance, and is quite low even above 2cm. However, PinLoc collects multiple fingerprints from around 1m x 1m spot.

20. Uniqueness of CFR Structure

21. PinLoc Architecture

22. Data Sanitization

23. CFR Clustering K-means is done with K=10 Clusters with smaller weight than certain cutoff is dropped Dropping small clusters don’t affect the performance

24. CFR Classification First PinLoc computes macro-location based on WiFi SSIDs. Shortlist spot and put them in Candidate Set. Compute distance between packet P sent by certain AP and spots in the candidate sets. The likely spot would have minimum distance.

25. War Driving Way to collect date from many locations for supervised learning. In experiment a Roomba robot is used to get data from 2cm x 2cm locations. Collect CFR and then cluster them Doesn’t need to be every possible location

26. Accuracy 89% accuracy in test location 7% false positive across 50 locations At least 3 Aps to get reasonable accuracy

27. Limitations Antenna’s Orientation Height and 3D war-driving Phone mobility Dependency on Particular hardware cards

28. Related Work RF signal based: – Horus and LEASE utilize RSSI to create location fingerprints Time Based: – Utilizes time delays to estimate distance between wireless transmit-receiver. GPS etc. Angle of Arrival based: – Use of multiple antennas to find angle of which signal arrives. Employs geometric or signal phase relationship.