using FM radio signals Andrei Popleteev Advisors: Oscar Mayora - PowerPoint PPT Presentation

Indoor positioning using FM radio signals Andrei Popleteev Advisors: Oscar Mayora Venet Osmani

Outline • Introduction • State of the art • Proposed approach • FM localization – With local transmitters (FM L ) – With broadcasting stations (FM B ) • Conclusion 21 April 2011 2

Indoor localization • Ambient intelligence • Assisted daily living • Activity recognition • Behavior analysis • Object tracking 21 April 2011 3

Indoor localization • GPS does not work indoors. • Specialized systems are expensive. • Systems based on cellular networks: – Good coverage – Low accuracy • Wi-Fi is the de-facto standard, but – Limited coverage – High power consumption 21 April 2011 4

Indoor localization: FM radio • FM radio addresses these issues, and provides: – High coverage – Long battery life – Good accuracy 21 April 2011 5

Source: fmscan.org 21 April 2011 6

FM-enabled mobile devices 21 April 2011 7

Power consumption 1400 Power consumption, mW 1200 1200 1000 800 600 400 300 200 50 15 0 Wi-Fi Wi-Fi FM receiver FM receiver (constantly (power saving) (with RDS) (without RDS) active) Wi-Fi data from [Anand et al. 2005] 21 April 2011 8 FM data from Si4703 and TDA7088 datasheets

State of the art 21 April 2011 9

State of the art: FM localization • There are few works on FM positioning. • All of them consider only outdoor scenarios. • Achieved accuracy: – 2005: 8 km with 50% probability (Krumm et al.) – 2009: 20 m with 67% probability (Fang et al.) There are no results for indoors performance of FM localization. 21 April 2011 10

State of the art: Summary Technology Accuracy Coverage Battery life System costs Wi-Fi Medium Low Low Low Cellular Low Medium Low Low UWB High Low High High FM (outdoor) Low High High Low FM (indoor) ? The Gap 21 April 2011 11

Localization methods • Proximity-based • Direction-based • Time-based • Based on signal properties – Propagation modeling – Fingerprinting Used in this work 21 April 2011 12

Fingerprinting Includes two phases: • Calibration : creation of a database matching signal strength samples with the location. • Positioning : comparing the observed signal properties to those in the database. 21 April 2011 13

Proposed approach 21 April 2011 14

FM radio signal sources • Short-range FM transmitters – Off-the-shelf devices – No licensing required – Can transmit arbitrary sound • Broadcasting FM stations – Zero cost for localization – Worldwide coverage • Both signal sources have been used in this thesis 21 April 2011 15

Experimental setup UBiNT lab Create-Net 6 m 12 m 21 April 2011 16

FM L : positioning using local transmitters 21 April 2011 17

FM L : positioning using local transmitters  FM L performance FM L vs. Wi-Fi • Orientation analysis • Accuracy degradation • 21 April 2011 18

FM L positioning • Suitable signal features for fingerprinting: – Received signal strength (RSS) – Audio signal-to-noise ratio (SNR) – Stereo channel separation (SCS) 21 April 2011 19

Signal properties vs. distance SNR SCS RSSI 1.0 Normalized value 0.8 0.6 0.4 0.2 0.0 0 1 2 3 4 5 6 7 8 Distance from transmitter, meters Receiver: Brando USB FM radio 21 April 2011 20

FM L positioning performance 1.0 0.8 Confidence 0.6 SNR SCS 0.4 RSSI 0.2 Baseline 0.0 0 1 2 3 4 5 6 7 8 Error distance, meters Receiver: Brando USB FM radio; grid: 1 m. 21 April 2011 21

FM L positioning accuracy (RSSI) kNN Gaussian Processes 1 1 0.9 0.9 2.65m @ 95% 3.88m @ 95% 0.8 0.8 0.7 0.7 Confidence Confidence 0.6 0.6 0.5 0.5 0.93m @ 50% 0.97m @ 50% 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0 0 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 Error, meters Error, meters Receiver: HTC Artemis; grid: 0.5 m. 21 April 2011 22

FM L : positioning using local transmitters FM L positioning •  FM L vs. Wi-Fi Orientation analysis • Accuracy degradation • 21 April 2011 23

FM L versus Wi-Fi Gaussian Processes kNN 1 1 0.9 0.9 0.8 0.8 0.7 0.7 Confidence Confidence 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0 0 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 Error, meters Error, meters FM Wi-Fi Receiver: HTC Artemis; grid: 1 m. FM RSSI granularity reduced to ensure a fair comparison. 21 April 2011 24

FM L : positioning using local transmitters FM L positioning • FM L vs. Wi-Fi •  Orientation analysis Accuracy degradation • 21 April 2011 25

Effect of orientation • Human body influences the signal distribution by reflecting and attenuating radio waves. • This might impact the localization accuracy. – It does for Wi-Fi. – Does it for FM? 21 April 2011 26

Effect of orientation • Four datasets collected, one for each direction. • “All FM” – accuracy when Confidence all four datasets are utilized. • Other graphs - accuracy within each dataset. Error distance, meters  User direction has no significant effect on FM localization accuracy. 21 April 2011 27

Recognition of orientation • Is it possible to 60 detect the orientation 50 using FM RSS Probability, % 40 fingerprints? 30  No, the result 20 is random. 10 0 Exact Adjacent Opposite Recognized orientation 21 April 2011 28

FM L : positioning using local transmitters FM L positioning • FM L vs. Wi-Fi • Orientation analysis •  Accuracy degradation 21 April 2011 29

What if… 21 April 2011 30

Signal strength distribution RSSI Before: coordinate R SS I Now: coordinate 21 April 2011 31

Accuracy degradation • Signal fingerprints change with time due to: – Furniture layout – Air temperature and humidity – Hardware temperature • These fluctuations affect the accuracy. • The solution : periodic recalibration – Requires personnel or additional hardware – Is tedious and expensive 21 April 2011 32

Spontaneous recalibration • Recalibration performed automatically when the device position is known: – In a cradle – On a nightstand – Connected to a wall charger • No additional hardware required • Transparent for the user 21 April 2011 33

Effect of recalibration 1 0.9 0.8 Confidence 0.7 0.6 0.5 0.4 0.3 Original 0.2 Degraded 0.1 Recalibrated 0 0 1 2 3 4 5 6 7 Error, meters 21 April 2011 34

FM B : positioning using broadcasting FM stations 21 April 2011 35

FM B : positioning using broadcasting FM stations  FM B performance FM B vs. Wi-Fi and GSM • Signal stability and people’s presence • Power consumption • 21 April 2011 36

FM B experiments • Performed in the same 12x6 m testbed (with slightly changed layout). • 76 active FM stations detected. • 3 local FM transmitters for comparison. • KNN classifier, leave-one-out evaluation. 21 April 2011 37

FM B localization performance 1.0 0.9 4.71m @ 95% 0.8 0.7 Confidence 0.6 0.5 0.91m @ 50% 0.4 0.3 FMb (all beacons) 0.2 FML (3 beacons) 0.1 Baseline 0.0 0 1 2 3 4 5 6 7 8 9 Error distance, meters 21 April 2011 38

FM station selection • More stations in fingerprint result in: – More accurate localization, but – Higher computational load – Longer scanning times • Do all the stations contribute equally? • Is there a trade-off between the number of stations and localization performance? 21 April 2011 39

Station selection methods • Naïve approach: select stations with – strongest signals; – weakest signals. • Alternative approach: select the stations which vary the most across the test points. 21 April 2011 40

Station selection methods 6 Strongest 5 Weakest Highest diversity Median error, meters 4 3 2 1 0 1 7 13 19 25 31 37 43 49 55 61 67 73 Fingerprint width (number of stations) 21 April 2011 41

FM B with 10% of stations 1.0 0.9 0.8 0.7 Confidence 0.6 0.5 0.4 0.3 FMb (all beacons) 0.2 FMb (7 beacons) 0.1 Baseline 0.0 0 1 2 3 4 5 6 7 8 9 Error distance, meters 21 April 2011 42

FM B : positioning using broadcasting FM stations FM B performance •  FM B vs. Wi-Fi and GSM Signal stability and people’s presence • Power consumption • 21 April 2011 43

FM B versus Wi-Fi 1.0 0.9 0.8 0.7 Confidence 0.6 0.5 0.4 0.3 FMb (all beacons) 0.2 Wi-Fi (all beacons) 0.1 Baseline 0.0 0 1 2 3 4 5 6 7 8 9 Error distance, meters 76 FM and 17 distinct Wi-Fi beacons 21 April 2011 44

FM B versus GSM 1.0 0.9 0.8 0.7 Confidence 0.6 0.5 0.4 FMb (all beacons) 0.3 FMb (7 beacons) 0.2 GSM (all beacons) 0.1 Baseline 0.0 0 1 2 3 4 5 6 7 8 9 Error distance, meters 15 distinct GSM beacons 21 April 2011 45 GSM RSSI acquired with HTC Artemis

FM B localization: Summary Localization accuracy for different technologies (in meters) measured in the same conditions. Confidence FM B Wi-Fi GSM FM B (7 stations) 50% 0.9 1.6 3.1 1.3 67% 1.3 1.9 4.2 2.1 90% 3.4 3.5 6.2 4.0 95% 4.7 4.0 9.1 4.9 21 April 2011 46