Decimeter-Level Localization with a Single WiFi Access Point

Transcription

1 Decimeter-Level Localization with a Single WiFi Access Point Presented By: Bashima Islam Indoor Localization Smart Home Occupancy Geo Fencing Device to Device Location 1

2 Previous Work 10 cm Accuracy Commodity Chipset & Sensors Multiple Access Point Goal Single WiFi Access Point Commodity Chipset Only (No Sensors) Decimeter Level Accuracy 2

3 Techniques to Achieve Goal Absolute Time of Flight Emulate Wideband Radio Measuring Time of Flight Figure 1: WiFi Bands: 3

4 Measuring Time of Flight the wireless cha h = ae j2πf τ al magnitude, f τ = h mod 1 2πf f quation gives us the time-o s: h = 2πf τ mod 2π Exploit Relation Between Time of Flight and Phase of Wireless Channel Measuring Time of Flight Chinese Reminder Theorem τ Figure 2: Measuring Time-of-Flight: Consider a wirele 4

5 Challenges Packet Detection Delay Multipath Effect Frequency Hopping Phase Offset Packet Delay Detection receivers detect the presence of a packet based on the energy of its first few time samples packet detection delays are often an order-of-magnitude larger than time-of-flight goal is to derive the true channel hi from the measured channel hi takes benefit of OFDM subcarrier The measured channel at subcarrier-0 does not experience packet detection delay, i.e., it is identical in phase to the true channel at subcarrier 0 5

6 Packet Delay Detection, h i,k = 2πf i,k τ mod 2π, he time-of-flight. i,k = 2π(f i,k f i,0 )δ i, e packet detection delay. us, the total measured channel phase at subcarrie h i,k =( h i,k + i,k ) mod 2π =( 2πf i,k τ 2π(f i,k f i,0 )δ i ) mod 2π tice from the above equation that the second Notice from the above equation that i,k = 2π(f i,k f i,0 )δ i = 0 at k = 0. Packet Delay Detection WiFi transmitters do not send data in subcarrier 0. Interpolate the measured channel phase across all the subcarriers to estimate phase at subcarrier zero. 6

7 Combating Multipath p h i,0 = a k e j2πf i,0τ k, for i = 1,..., n k=1, we need to disentangle these different paths Inverse Discrete Fourier Transformation non- contiguous Figure 1: WiFi Bands: Combating Multipath Inverting Non-uniform Discrete Fourier Transform does not have a single closed-form solution adds constraint to the inverse-ndft optimization Inverting provides the time- of-flight on all paths. the direct path is the shortest 7

8 Resolving Phase Offset Frequency hopping causes a random phase offset in the measured channel phase-locked loop (PLL) generates the center frequency for transmitter and receiver starts at random initial phase Carrier Frequency Offset occurs due to small differences in the carrier frequency of the transmitting and receiving radio Resolving Phase Offset band can be written as: j(f tx f rx i,0 )t+j(φtx i,0 φrx i,0 ) CSI rx i,0(t) = h i,0 e i,0 ow do we remove the phase and frequen CSI tx rx tx j(f i,0(t) = h i,0 e i,0 fi,0 )t+j(φrx i,0 φtx i,0 ) wireless the channel, channel h, as in equations follows: 11 an h 2 i,0 = CSI rx i,0(t)csi tx i,0(t) nder how Chronos measure 8

9 Resolving Phase Offset Forward and reverse channels cannot be measured at exactly the same t. can be resolved by averaging over several packets. Delays in the hardware pre-calibrated Standard Fourier Transform properties dictate that a minimum separation of Δf uses CSI measurements at center frequencies Compute Distance and Location d = Absolute TOF * Velocity of Light d 9

10 Compute Distance and Location refines the distance measurements by utilizing geometric constraints, imposed by the relative locations of the antennas on the access point and the client formulates a quadratic optimization problem Computing Distance and Location CL l cl i,j. d ij a i d i j etwe os b l ap ij. e sep AP i,j d ij d i j < l ap ii Chronos between 10

11 Testbed Lab Testbed: Result CDF LOS NLOS Time Error (in ns) (a) Time of Flight median errors in time-of-flight estimation are 0.47 ns and 0.69 ns Fraction of packets Propagation Delay Pa Dete De Delay (in ns (c) Packet Detection D Figure 5: Accuracy in Time of Flight: (a) The CDF of error in time-of-flight between two devices in Line of S Non-Line of Sight (NLOS). (b) Representative multipath profiles. (c) Histograms of time-of-flight and packet detecti of measuring sub-nanosecond time-of-flight between a pair of commodity WiFi devices. Method: We run our experiments in the testbed in Fig. 4. In each experiment, we randomly pick a location for the AP. We then randomly pick a client location that is within 15 meter from the AP. We experiment with both Chronos in the above experiments. Fig. 5( sentative multipath profiles in line-of-sight environments. We note that both profiles ar the profile in multipath environments havi nant peaks. Across all experiments, 11 the me dominant peaks in the multipath profiles is

12 Result Power Power 4 LOS Multipath Time (in ns) (b) Multipath Profiles Fraction of packets Propagation Delay Packet In both cases, we observe Detection that the leftmost peaks in both profiles correspond Delay to the true location of the source. Delay (in ns) (c) Packet Detection Delay racy in Time of Flight: (a) The CDF of error in time-of-flight between two devices in Line of Sight (LOS) and ht (NLOS). (b) Representative multipath profiles. (c) Histograms of time-of-flight and packet detection delay. sub-nanosecond time-of-flight between a dity WiFi devices. e run our experiments in the testbed in experiment, we randomly pick a location then randomly pick a client location that is r from the AP. We experiment with both d non-line-of-sight settings. We perform ts using a 10 ASUS EEPC netbook as a inkpad W300 Laptop emulating a WiFi AP th devices are equipped Result with the 3-antenna set. The antennas are placed at the corner of hich results an average antenna spacing of hinkpad AP and 12cm for the ASUS0.08 OS client. bove setup, we have run 400 localization 0.06 r different AP-client pairs. For each pair, s channel hopping protocol. We compute 0.04 ht ultipath between each transmit antenna0.02 and ree measure the ground-truth location using of architectural drawings of our building LM50 laser Time distance (in ns) measurement tool [1], s distances (b) Multipath up toprofiles 50 m with an accuracy of ound truth time-of-flight is the ground truth d by the speed of light. Fraction of packets Chronos in the above experiments. Fig. 5(b) plots representative multipath profiles in line-of-sight and multipath environments. We note that both profiles are sparse, with the profile in multipath environments having five dominant peaks. Across all experiments, the mean number of dominant peaks in the multipath profiles is 5.05 on average, with standard deviation 1.95 indicating that they are indeed sparse. As expected, the profile in line-of-sight has even fewer dominant peaks than the profile in multipath settings. In both cases, we observe that the leftmost peaks in both profiles correspond to the true location of the source. Further, we observe that the peaks in both profiles are sharp due to two reasons: 1) Chronos effectively spans a large bandwidth that includes all WiFi frequency bands, leading to Packet high time resolution; 2) Chronos s resolution is further Detection improved by exploiting sparsity that focuses on Delay of 177 ns across experiments. retrieving the sparse dominant peaks at much higher resolution, as opposed to all peaks. Propagation Delay observes a median packet detection delay Packet 200 Detection 300 Delay Results: Past work on WiFi time Delay (in ns) measurement and/or synchronization cannot measure the (c) Packettime-of-flight Detection Delayof a packet separately from its detection de- in Line [38]. of([35] Sight measures (LOS) and the distribution of detection de- CDF of error in time-of-flight between two deviceslay ltipath profiles. (c) Histograms of time-of-flight and packet lays detection but not the delay. detection delay of a particular packet.) t Results: We first evaluate Chronos s In contrast, Chronos has a novel way for separating the me-of-flight. between a Fig. Chronos 5(a) in depicts the above the experiments. CDF detection Fig. 5(b) delay plots fromrepre- sentative in multipath line-of-sight profiles set-in line-of-sight understand and the multipath importance of this capability for the suc- the time-of-flight. We would like to -flight of the signal -line-of-sight. testbed in We environments. observe that We note the metime-of-flight that both cess profiles of Chronos. are sparse, Thus, with we use the measurements from the k a location estimation the profile are in multipath 0.47 ns and environments above experiments having five dominant peaks. Across all experiments, the mean number of to compare time-of-flight in indoor environments against packet detection delay. ation that is tively. These results show that Chronos t with both dominant peaks in the multipath profiles is 5.05 on average, with standard deviation 1.95 indicating that they omise of computing time-of-flight at subcuracy. To put this in perspective, consider delay and time-of-flight across experiments. Chronos ob- Fig. 5(c) depicts histograms of both packet detection We perform etbook as a are indeed sparse. As expected, the profile in line-of-sight 8], a state-of-the-art system for time synourcesync achieves 95 th fewer dominant peaks than the profile in multi- serves a median packet detection delay of 177 ns across g a WiFi AP has even percentile syn- experiments. We emphasize two key observations: (1) 12

13 Result 6: Ranging Accuracy: Plots error in distance CDF LOS NLOS Locali ation Error(m) Localization Accuracy: Plots CDF of localiz Achievement Median Distance Error 14.1 cm to 20.7 cm Localization Error 65 cm to 98 cm 13

14 Achievements Real World Smart Home Occupancy 94.3% WiFi Geo Fencing 97% Personal Drone 4.2 cm Comments WiFi hopping implementation will be hard in vast already deployed access points. Estimating large gap between 2.4 GHz and 5GHz can be errorneous. 14

15 Thank you 15