Source-linked AI summary
mD-Track: Leveraging Multi-Dimensionality in Passive Indoor Wi-Fi Tracking
Yaxiong Xie, Jie Xiong, Mo Li, Kyle Jamieson
TL;DR
Passive Wi-Fi tracking is limited by multipath interference and by the resolution available from antenna count and channel bandwidth. mD-Track jointly estimates multiple signal dimensions, refines paths iteratively, and bounds computation for real-time operation. Experiments show improved passive localization, motion tracking, and signal estimation using a single transmitter-receiver pair.
Problem
Weak target reflections overlap with direct-path and clutter signals, while conventional path resolution is limited when paths have similar parameters.
Method
mD-Track fuses multiple signal dimensions, iteratively refines and subtracts reconstructed paths, and uses multidimensional channel sampling with tractable estimation.
Results
mD-Track demonstrates improved passive multi-target localization and motion tracking, with 3.5× accuracy improvements over SpotFi in parameter estimation and passive localization.
Takeaways & Limitations
The system supports passive localization and motion tracking with a single transmitter-receiver pair and has been demonstrated on WARP and commercial off-the-shelf Wi-Fi hardware.
Takeaways & Limitations
The design assumes perfect transceivers without phase offsets across radio chains, justified through phase calibration.
Abstract
from arXiv · showhide
Wi-Fi localization and tracking face accuracy limitations dictated by antenna count (for angle-of-arrival methods) and frequency bandwidth (for time-of-arrival methods). This paper presents mD-Track a device-free Wi-Fi tracking system capable of jointly fusing information from as many dimensions as possible to overcome the resolution limit of each individual dimension. Through a novel path separation algorithm, mD-Track can resolve multipath at a much finer-grained resolution, isolating signals reflected off targets of interest. mD-Track can localize human passively at a high accuracy with just a single Wi-Fi transceiver pair. mD-Track also introduces novel methods to greatly streamline its estimation algorithms, achieving real-time operation. We implement mD-Track on both WARP and cheap off-the-shelf commodity Wi-Fi hardware and evaluate its performance in different indoor environments.
1 INTRODUCTION
Passive motion tracking enables applications such as security surveillance, elderly care, and retail business. Wi-Fi systems are promising because Wi-Fi access points are pervasively available.
- Passive tracking can support security surveillance, elderly care, and retail business applications.
- Wi-Fi-based systems are particularly promising because access points are pervasively available.
AoA
Passive Wi-Fi tracking must resolve weak target reflections from strong direct paths and clutter, while conventional AoA and ToF methods face dimension-specific resolution limits. mD-Track addresses this by jointly exploiting multiple signal dimensions and refining paths iteratively, with experiments reporting improved estimation and localization accuracy.
- Weak target reflections are superimposed with stronger direct-path and clutter reflections, making accurate path identification difficult.
- AoA and ToF methods struggle when multiple paths have similar respective parameters, because their paths cannot be reliably resolved.
- AoA resolvability depends on antenna count, whereas ToF resolvability depends on transmission bandwidth, making hardware or bandwidth expansion costly or constrained.
- mD-Track jointly fuses ToF, AoA, and Doppler shift or AoD to separate paths without changing any individual dimension’s resolution limit.Signals unresolved in one dimension can be separated using differences in another dimension.
- mD-Track combines all available signal parameters, iteratively re-estimates and subtracts paths, and uses bounded computation for real-time operation.The iterative refinement addresses interference from strong direct paths, while the computation design targets the exponential cost of joint estimation.
- 3.5× accuracy improvements over SpotFi were demonstrated in parameter estimation and passive localization, while adding Doppler yielded approximately 3× improvement.
2 THE WIRELESS CHANNEL
The wireless channel is represented through multidimensional path parameters describing propagation, direction, motion, and attenuation. mD-Track samples these dimensions across transmit and receive antennas, subcarriers, and time to estimate signal paths.
- A signal path is characterized by ToF, AoA, AoD, Doppler shift, and complex attenuation.
- ToF describes propagation time and has resolution inversely proportional to channel bandwidth.
- AoA and AoD describe arrival and departure directions, with their resolutions determined by receive and transmit antenna counts, respectively.
- Doppler shift captures frequency changes caused by transmitter, receiver, or reflector movement, with finer resolution from longer observation intervals.
- mD-Track uses linear transmit and receive antenna arrays with N and M elements, respectively.
3 PARAMETER ESTIMATION
mD-Track estimates multiple wireless-signal dimensions jointly and separates multipath through iterative reconstruction and cancellation. Its sequential estimator combines antenna, temporal, and frequency information while refining path parameters toward convergence.
- 3.1 Multi-dimensional estimator: The receiver estimates channels from the 802.11n preamble by decoupling simultaneously transmitted signals and removing known cyclic time-delay phase shifts.The mapping matrix is known, enabling channel recovery through multiplication by its conjugate.
- 3.1 Multi-dimensional estimator: mD-Track’s four-dimensional estimator jointly processes AoA, AoD, Doppler, and ToF for each wireless propagation path.The estimator combines individual modules in a two-antenna transmitter and receiver configuration.
- 3.1 Multi-dimensional estimator: AoA, AoD, Doppler, and ToF are estimated sequentially: antenna beamforming precedes Doppler removal, followed by correlation with a delayed transmit signal.The processing order exploits Doppler’s equal effect across antennas, preserving phase differences used by AoA and AoD estimation.
- 3.2 Resolving multiple paths: mD-Track separates multiple paths by estimating the strongest signal, reconstructing and canceling it, then iterating on the residual until all paths are estimated.The number of paths is obtained during this process, and the final residual estimates background noise.
- 3.2 Resolving multiple paths: After three iterations, reconstructed stronger and weaker signals produce z-function peaks much closer to the true path parameters than initial estimates.Initial interference and energy leakage can substantially distort weaker-path estimation, motivating iterative refinement.
- 3.2 Resolving multiple paths: The iterative refinement process belongs to the EM family, so convergence is guaranteed, although improper initialization may lead to a local rather than global maximum.The paper reports that experiments empirically examine the cancellation-based initialization.
4 CHANNEL MEASUREMENT
mD-Track estimates channel parameters by sampling multiple domains jointly, while correcting hardware-induced uncertainty that affects angle, ToF, and frequency-shift estimates.
- Multiple channel parameters require sampling across corresponding spatial, frequency, and time domains.AoA and AoD use spatial sampling, ToF uses frequency sampling, and frequency shift uses time sampling.
- A four-dimensional channel measurement matrix has size N × M × F × T, and submatrices estimate parameter subsets.N and M count transmitting and receiving antennas, F counts subcarriers, and T counts time samples.
- Phase offsets across radio chains must be eliminated because they affect AoA and AoD estimation.The offsets are constant across time and frequency and can be measured using a coaxial cable connection.
- SFO and STO make absolute ToF difficult, but relative ToF remains accurate because their time delay is shared across paths.In a controlled experiment, absolute ToF had a median error of 13 ns, while relative ToF had a median error of 0.48 ns.
- mD-Track calibrates reflection-path ToF using the direct path as a reference and corrects the measured offset.The direct path is identified by its shortest ToF and largest amplitude, then its measured-to-ground-truth ToF difference calibrates other paths.
- CFO and Doppler are indistinguishable to the estimator as frequency shifts, but CFO affects all multipaths whereas Doppler affects only mobile paths.CFO is typically hundreds of hertz, while human-induced Doppler is only a few hertz.
5 IMPLEMENTATION
mD-Track runs its core estimation algorithms on a backend workstation while WARP boards and commodity Wi-Fi routers collect channel measurements.
- The core mD-Track estimation algorithms run on a desktop workstation backend.
- WARP v3 boards collect time-domain IQ samples and send extracted LTF data to the server through an Ethernet-connected PC.
- Commodity access points use Atheros Wi-Fi hardware and customized OpenWRT software to enable CSI collection.
6 EVALUATION
The evaluation measures mD-Track’s resolvability, parameter-estimation accuracy, passive localization, multi-target tracking, and computational cost across WARP and commodity Wi-Fi hardware. Across these settings, adding signal dimensions improves accuracy and enables resolving multiple motions without extra hardware overhead.
- 6.1.1 Resolvability: mD-Track outperforms 2D SpotFi in resolving signals with the same dimensionality through iterative interference cancellation and path refinement.Increasing dimensionality further improves resolvability significantly.
- 6.1.2 Estimation Accuracy: With three antennas, mD-Track achieves median direct-path AoA errors of 4.4° in 4D and 6.2° in 2D, versus 13.4° and 17.1° for SpotFi and MUSIC.For reflection paths, the corresponding mD-Track errors are 5.6° and 7.3°, compared with 16.9° and 24.2° for SpotFi and MUSIC.
- 6.2 Passive Localization: With eight antennas and 40 MHz bandwidth, mD-Track 3D and 4D achieve median localization errors of 0.36 m and 0.28 m, versus 1.56 m for SpotFi.With only three antennas, mD-Track 3D and 4D achieve median errors of 0.67 m and 0.48 m.
- 6.2 Passive Localization: At fixed bandwidth and antenna count, 4D mD-Track achieves a 0.28 m median error, compared with 0.36 m for 3D and 1.16 m for 2D.Increasing antennas or bandwidth also improves performance, but increasing dimensionality does not incur extra hardware overhead.
- 6.2 Passive Localization: mD-Track localizes two nearby people with a median error of 0.51 m and four targets with median errors of 0.47 m at 3 m separation and 0.94 m at 0.5 m separation.Doppler identifies moving paths, while ToF, AoA, and AoD locate multiple motions simultaneously.
- 6.3 Computational Complexity: mD-Track runs faster than SpotFi because it performs multiple one-dimensional searches instead of SpotFi’s multi-dimensional search.The evaluation varies dimensionality and angle and time step sizes when comparing computational complexity.
7 RELATED WORK
Related work spans active and passive indoor localization and motion tracking using RSSI, AoA, ToF, Doppler, acoustic, optical, and camera-based approaches. The paper positions mD-Track as a more extensible multidimensional estimator that also provides location-related parameters for each resolvable multipath.
- Indoor active localization: RSSI-based indoor localization is coarse, while AoA and ToF systems improve accuracy but remain vulnerable to multipath and limited path resolvability.AoA resolution is constrained by antenna count, while ToF systems use channel combining or hopping to obtain wider virtual bandwidth.
- Passive localization and motion tracking: Passive Wi-Fi localization systems based on RSSI or CSI signatures often rely on high-density deployments, limiting their practicality for large-scale deployment.Other passive approaches include acoustic systems and Wi-Vi’s signal-nulling technique.
- Passive localization and motion tracking: WiSee and mD-Track both estimate small Doppler shifts using long observation intervals over multiple received signal samples.mD-Track additionally estimates AoA, AoD, and ToF for every resolvable multipath.
- Passive localization and motion tracking: Several prior systems estimate multiple parameters but use fixed-dimensional designs that are not easily generalized to additional signal dimensions.Spatial smoothing in some prior systems also increases computational complexity significantly.
8 CONCLUSION
mD-Track incorporates information from as many signal dimensions as possible to improve passive wireless sensing accuracy in multipath environments. Experiments demonstrate improved passive multi-target localization and motion tracking, along with support for gesture recognition and Wi-Fi imaging.
- mD-Track incorporates information from as many signal dimensions as possible to improve passive wireless sensing accuracy in multipath environments.
- Experiments demonstrate greatly improved performance for passive multi-target localization and motion tracking.
- mD-Track supports applications including gesture recognition and Wi-Fi imaging.