SLIDE 1 mD-Track: Leveraging Multi-Dimensionality for Passive Indoor Wi-Fi Tracking
Yaxiong Xie*☆, Jie Xiong☨, Mo Li*, Kyle Jamieson☆
*Nanyang Technological University
☨University of Massachusetts Amherst ☆Princeton University
SLIDE 2 Wi-Fi Sensing
ØWi-Fi sensing enables diverse applications
Sender Receiver Human
Human tracking HCI interface Respiration monitoring Proximity advertising
SLIDE 3
Why Wi-Fi sensing is possible?
SLIDE 4
The signal changes with target movements
Transmitter Propagation channel Receiver
SLIDE 5
The signal changes with target movements
Transmitter Propagation channel Receiver
SLIDE 6
The signal changes with target movements
Transmitter Propagation channel Receiver
SLIDE 7 Target reflections
Wall Mobile devices Obstacles Wi-Fi AP
ØThe target-reflected signal carries the information about the target
Human target 1 Human target 2
SLIDE 8 A fundamental task in wireless sensing
Retrieve the target-reflected signal which contains the context information
8
SLIDE 9 Challenge
Ø The reflected signals from multiple targets are mixed together with direct path and other reflection signals Ø It’s challenging to separate the signals due to limited channel bandwidth (40MHz) and limited number (3- 4) of antennas at the Wi-Fi access point
9
SLIDE 10
Separating two signals
Resolvability presents us the capability of separating two signals
SLIDE 11 Spatial domain separation
30 60 90 120 150 180
Ø Signals can be separated in spatial domain with angle-of-arrival (AoA) information Ø AoA resolvability is determined by the number of antennas in the array Ø More antennas result in higher resolvability
SLIDE 12 Time domain separation
Ø Signals can also be separated in time domain with time-of-flight (ToF) information Ø ToF resolvability is determined by the signal channel bandwidth Ø Wider bandwidth results in higher resolvability
time ToF 1
Signal from phone Signal from chair
ToF 2
SLIDE 13
The Wi-Fi sensing performance is fundamentally limited by signal resolvability
SLIDE 14
How to improve the resolvability of Wi- Fi signals without any hardware modifications?
SLIDE 15 The power of multi-dimensionality
Transmitter Receiver
𝝔𝟐 𝝔𝟑 𝝊 𝟐 𝝊 𝟑
Time of Flight Angle of Arrival
𝑻𝟐 𝑻𝟑 𝑻𝟐 𝑻𝟑
Two signals are close in time domain but far away in spatial domain
SLIDE 16 The power of multi-dimensionality
Transmitter Receiver
𝝔𝟐 𝝔𝟑 𝝊 𝟐 𝝊 𝟑
Time of Flight Angle of Arrival
𝑻𝟐 𝑻𝟑 𝑻𝟐 𝑻𝟑
Two signals are close in spatial domain but far away in time domain
SLIDE 17 The power of multi-dimensionality
Transmitter Receiver
𝝔𝟐 𝝊 𝟐
Time of Flight Angle of Arrival
𝑻𝟐 𝑻𝟐
Doppler Shift
𝜹
Two signals close in both spatial and time domain can be separated in frequency domain (Doppler shift)
𝑻𝟑 𝑻𝟑
SLIDE 18
The power of multi-dimensionality
Multi-dimensional information helps separate close-by signals !
SLIDE 19
Combining information from multi-dimensions
AoA, AoD, ToF and Doppler shift
for signal separation
SLIDE 20
What can be achieved?
Ø More accurate sensing Ø Multi-target sensing (a well known challenge for passive sensing) Ø Larger sensing range
SLIDE 21
Joint estimation for a single path signal
Please refer to our paper for more details!
SLIDE 22 Joint estimation for multiple signals
Single-path signal Multiple signals
Signal Noise Signal 1 Signal 2 Noise
SLIDE 23 Joint estimation for multi-path signals
Signal
Successive Interference Cancellation
Noise
Single path 𝒏- dimensional estimator Signal parameter 𝝔, 𝝌, 𝝊, 𝜹
SLIDE 24 Joint estimation for multi-path signals
Signal
Successive Interference Cancellation
Noise
Single path 𝒏- dimensional estimator Signal parameter 𝝔, 𝝌, 𝝊, 𝜹
Reconstructed signal Reconstruction
SLIDE 25 Joint estimation for multi-path signals
Signal
Successive Interference Cancellation
Noise Reconstructed signal
SLIDE 26 Joint estimation for multi-path signals
Successive Interference Cancellation
Noise Signal
Single path 𝒏- dimensional estimator Signal parameter 𝝔, 𝝌, 𝝊, 𝜹
SLIDE 27 Reduce the computational load
ØComputation overhead
- EM algorithm adopted is computational intensive
- We apply Generalized EM algorithm (SAGE) to reduce the computational
- verhead
Please refer to our paper for more details!
SLIDE 28
Performance Evaluation
SLIDE 29 Testbed
ØWe implement our system on both WARP and commodity Wi-Fi AP
- WARP supports 8 antennas
- WPJ558 and TP-Link 4300 Wi-Fi AP are equipped with 3 antennas
SLIDE 30
Resolvability
ØWe achieve 20 times higher resolvability compared to the state-of-arts MUSIC and SpotFi algorithm
SLIDE 31 Passive localization
mD-Track 4D achieves a median error of 0.28m with 8 antennas and 40MHz bandwidth
WARP Commercial Wi-Fi AP
mD-Track 4D achieves a median error of 0.48m with 3 antennas and 40MHz bandwidth
SLIDE 32
Motion tracking
ØTwo target moving at the same time We can separate the movements of two close-by fingers!
SLIDE 33
Conclusions
ØMD-Track addresses one fundamental problem in wireless sensing ØMD-Track is not limited to localization but can be applied to a large range of sensing applications ØMD-Track can be employed for sensing with other wireless technologies such as RFID, acoustic and mmWave
SLIDE 34
