FILA: Fine-grained Indoor Localization

Transcription

1 IEEE 2012 INFOCOM FILA: Fine-grained Indoor Localization Kaishun Wu, Jiang Xiao, Youwen Yi, Min Gao, Lionel M. Ni Hong Kong University of Science and Technology March 29 th, 2012

2 Outline Introduction Motivation System Design Performance Evaluation Conclusions 2

3 Indoor Location-based Services Assert Tracking Security Goals: To fast locate objects To obtain high accuracy To minimize deployment costs Emergence Response Healthcare Social Network 3 3

4 RF-based Indoor Localization Techniques Radio Frequency (RF) is the frequency that the radio signals are carried and transmitted from the antenna. Existing RF-based Indoor Localization Techniques Cost Techniques System Accuracy Cost GPS Accuracy SnapTrack >5m Medium WLAN GSM GSM Fingerprinting >2.5m Medium Ultrasonic Criktet 4*4 Sq Ft (100%) High Infrared Active Badge 5-10m Medium UWB Ubisense 15cm High RFID LANDMARC 2m Low Sensors RIPS 3cm High WLAN RADAR, Horus 3-5m; 2m Low Highest! Lowest! 4

5 WLAN Advantages Dense-deployed APs Prevalent WiFi-enabled devices Low cost and easy for implementation 5

6 WLAN RSSI-based Indoor Localization Techniques RSSI Received Signal Strength Indicator (RSSI) a measurement of the power present in a received radio signal Radio propagation model: distance=f(rssi) Free space path loss model Free Space Path loss AP RSSI Distance WiFi-enabled device 6

7 Related Work RADAR [INFOCOM 00], Horus [Mobisys 05] Fingerprint: signal strength radio map Accuracy: 3m for 50%, 0.7m for 50% Wideband Powerline Positioning [UbiComp 08] Apply wideband frequency to mitigate the time variance. Accuracy: reduce accuracy degradation over time Indoor localization without pain [MobiCom 10] All based on RSSI Radio propagation model based Accuracy: 2m for 50% 7

8 Is RSSI a reliable indicator? 8

9 Observation RSSI value is a packet-level estimator Multipath Path loss Average the signal power over a packet. RSSI is easily varied by multipath. Constructive Destructive RSSI is not reliable! 9

10 Could we find a reliable metric to improve indoor localization? 10

11 Key Insight Orthogonal Frequency Division Multiplexing IEEE a/g/n leverage OFDM to provide high throughput In OFDM, a channel is orthogonally divided into multiple sub channels, namely subcarriers Data is transmitted in parallel on multiple subcarriers that overlap in frequency Transmitter Receiver Data in Data out Modulation Modulation IFFT FFT 1 st Subcarrier 2 nd Subcarrier 3 rd Subcarrier D/A A/D Baseband O FDM signal Baseband O FDM signal 11

12 Key Insight Channel State Information Transmitter H Channel N Receiver Data in Encoder X x + Y Decoder Data out In OFDM system, the received signal over multiple subcarriers is Y = H X + N (X transmit signal, N noise) H=Y/X -- Channel State Information (CSI) H=he jw (h: amplitude, w: phrase) CSI is the channel response at the receiver in frequency domain 12

13 Key Insight. RSSI CSI CSIs RSSI Receiver Packet 2.4GHz S/P FFT antenna Baseband RSSI estimates the channel in packet level. CSI estimates the channel in subcarrier level. [1] Only a single amplitude Vector with amplitude and phase [1]D Halperin and et al., Predicable Packet Delivery from Wireless Channel Measurements, in SIGCOMM,

14 So compared to RSSI, CSI Is Fine Grained metric full of frequency domain information! We expect to exploit such information to obtain a reliable indicator for location. 14

15 Scope Motivation RSSI is inaccurate and not reliable CSI is fine grain information Approach Replace RSSI with CSI Design a FILA system Goal Improve the indoor localization performance 15

16 Outline Introduction Motivation System Design Performance Evaluation Conclusions 16

17 Cross Layer Architecture AP2 d2 AP1 d1 Tx Network layer Cross layer AP Location Information (2) Process CSI CSI eff (2) Distance Calculator + (3) Locate Rx AP3 (1) Collect CSI Pysical layer Channel Estimation Rx OFDM Demodulator OFDM Decoder Normal Data 17

18 Design Approach CSI Collection Process CSI and Distance Estimation Location Determination 18

19 Approach (1 st Step) CSI Collection Process CSI Location The first step is to collect the subcarriers CSI which divided into 30 groups on the received baseband in WLAN. Location Determination System Wireless API Operating System Device Driver Hardware Wireless card 19

20 Approach (2 nd Step) CSI Collection Process CSI Location Two processing mechanisms: #1 Time-domain Multipath Mitigation #2 Frequency-domain Fading Compensation Distance Estimation 20

21 Channel Response Amplitude Approach (2 nd Step) CSI Collection Process CSI Location Time-domain Multipath Mitigation The received signal is the combination of multiple reflections with LOS signal If bandwidth is wider than coherence bandwidth, the reflections will be resolvable. The bandwidth of n is 20MHz, that provides the capability of the receiver to resolve the different reflections in the channel h=ifft(csi) Time delay 21

22 Receive Power(dBm) Approach (2 nd Step) CSI Collection Process CSI Location Frequency-domain Fading Compensation When the space between two subscarriers is larger than coherence bandwidth, they are fading independently Exploit the frequency diversity of CSI to eliminate small-scale fading We define effective CSI as the weighting average among all subcarriers CSI eff = 1 K K k=1 f k f 0 CSI k, k [ 15,15] Subcarrier Index 22

23 Approach (2 nd Step) CSI Collection Process CSI Location Distance Determination Refined model: distance= f(csi eff ) Initialized δ and n d = 1 4π c f 0 CSI eff 2 δ 1 n Choose a CSIeff dataset corresponding to a distance, and then train δ and n δ: environment factor n: path loss fading exponent KNN algorithm Use δ,n to verify the CSIeff dataset of other distances Other distances are in conformity with the training δ and n 23

24 Approach (3 rd Step) CSI Collection Process CSI Location Obtain the coordinates of the APs. Calculate the distance between object and the APs. Apply the trilateration method to locate object. d 1 = x 1 x y 1 y 0 2 d 2 = x 2 x y 2 y 0 2 d 3 = x 3 x y 3 y 0 2 AP2 AP1 AP3 So, we can determine the location of the! 24

25 Outline Introduction Motivation System Design Performance Evaluation Conclusions 25

26 Experimental Setup Hardware Intel WiFi Link 5300, n router iwl5300 Router Software Linux kernel, Matlab, Python 26

27 Implementation (4 Scenarios) Chamber Lab Lecture Hall 3m Χ 4m 5m Χ 8m Corridor 20m Χ 25m 27

28 Evaluation Metric Temporal stability Accuracy 28

29 CSI eff amplitude Relation between CSI and Distance CSIeff amplitude Exponential Fitting Distance (meters) 29

30 Temporal Stability 30 30

31 Accuracy of Distance Estimation 31 31

32 Location Accuracy in Lab 32 For over 90% of data points, the localization error < 1m For over 50% of data points, the localization error < 0.5m 32

33 Location Accuracy in Lecture Hall For over 90% of data points, the localization error < 1.8m For over 50% of data points, the localization error < 1.2m 33

34 Location Accuracy in Corridor 34 For over 90% of data points, the localization error < 2m For over 50% of data points, the localization error < 1.2m 34

35 Outline Introduction Motivation System Design Performance Evaluation Conclusions 35

36 Conclusions 36 We use fine gained PHY information (CSI) in OFDM-based WLANs to improve indoor localization performance. We design FILA, a fine grained cross layer localization system leveraging CSI based on existing WLAN standards. Experiments with commercial NICs in different scenarios show that FILA can achieve significantly accuracy gain comparing with corresponding RSSI methods. 36

37 Thanks. Questions? PhD Hong Kong University of Sci.& Tech. 37