PILA: Sub-Meter Localization Using CSI from Commodity Wi-Fi Devices

Zengshan Tian, Ze Li, Mu Zhou, Yue Jin, Zipeng WuSensors 2016 · DOI

TL;DR

This paper introduces PILA, a novel indoor localization system that achieves sub-meter accuracy using commodity Wi-Fi devices. By leveraging Channel State Information (CSI) to estimate Angle-of-Arrival (AOA) and Received Signal Strength (RSS), PILA overcomes the limitations of existing methods that often require specialized hardware or extensive calibration.

What Problem Does This Paper Address?

Accurate indoor localization is crucial for many Location-Based Services (LBSs), but Global Navigation Satellite Systems (GNSS) fail indoors. While Wi-Fi localization offers a promising alternative due to widespread device adoption, existing methods have drawbacks. AOA-based systems can achieve high accuracy (around 0.4m) but typically require multiple antennas on Access Points (APs), leading to high costs and complex hardware modifications. RSS-based systems are easier to deploy but suffer from lower accuracy (around 3m). This paper aims to bridge this gap by developing a high-accuracy indoor localization system that utilizes readily available commodity Wi-Fi devices and their CSI, minimizing hardware requirements and calibration effort.

Methods

The PILA system employs a multi-stage approach:

  • CSI-based AOA and TOA Estimation: The core innovation lies in using Channel State Information (CSI) from commodity Wi-Fi devices (specifically, Intel 5300 NICs with three antennas) to estimate the Angle-of-Arrival (AOA) and Time-of-Arrival (TOA) of Wi-Fi signals. This is achieved by leveraging the Orthogonal Frequency Division Multiplexing (OFDM) properties of Wi-Fi signals. To combat the interference from multipath effects, a Two-Dimensional Spatial Smoothing (2D-SS) technique is applied to the CSI covariance matrix.
  • Direct Signal Path Identification: Identifying the direct signal path (Line-of-Sight or LOS) is critical for accurate localization, as multipath signals can distort AOA and TOA measurements. PILA addresses this by:
  • Packet Detection Delay (PDD) Handling: Recognizing that PDD introduces a consistent offset in TOA measurements, a phase sanitization algorithm is proposed to mitigate its impact. This algorithm optimizes a linear fitting model for the CSI phase to remove the PDD-induced variance.
  • Gaussian Means Clustering: After AOA and TOA are estimated (and phase-sanitized), Gaussian means clustering is used to group the detected signal paths. Clusters with smaller AOA and TOA variance are considered candidates for the direct signal path.
  • Likelihood Assignment: A probabilistic approach assigns a likelihood value to each candidate path based on the number of peak points and the variance of AOA and TOA within its cluster. The path with the highest likelihood is identified as the direct signal path.
  • Target Localization: The system integrates the estimated AOA and RSS measurements. To handle the non-convex nature of the localization problem, a Second-order Cone Programming (SOCP) relaxation approach is employed to transform it into a convex problem, which can then be solved efficiently.

Datasets

The paper evaluates the proposed system using a custom testbed in a typical lab environment. The dataset consists of CSI collected from four Wi-Fi APs (equipped with Intel 5300 NICs) and a target device transmitting Wi-Fi signals. The locations of the APs were measured accurately using a laser range finder. No pre-existing public datasets are mentioned; the evaluation is based on data collected in their controlled experimental setup.

Key Insights and Ideas

  • CSI as a Rich Source for Localization: The paper demonstrates that CSI, readily available in commodity Wi-Fi devices, can be effectively used to extract precise AOA and TOA information, circumventing the need for specialized hardware.
  • Mitigating Multipath Effects: The combination of 2D-SS for AOA/TOA estimation and phase sanitization for PDD handling is crucial for robustly identifying direct signal paths in complex indoor environments.
  • Probabilistic Direct Path Identification: Using Gaussian clustering and a likelihood assignment based on AOA/TOA variance provides a more robust method for identifying the direct path compared to simply selecting the path with the highest spectrum value.
  • SOCP for Convex Localization: Transforming the non-convex localization problem into a convex one using SOCP relaxation is a standard but effective technique that ensures a reliable and efficient solution.

Key Results

  • Sub-meter Localization Accuracy: The proposed PILA system achieves a median localization error of 0.7 m, with a 68% error of 1 m. This is significantly better than systems relying solely on RSS (around 3m) or conventional AOA methods.
  • Improved AOA Estimation: The AOA estimation approach achieves a median angle error within 5°, outperforming the conventional MUSIC algorithm.
  • High Direct Signal Path Identification Precision: The direct signal path identification method effectively reduces the probability of large localization errors.
  • Robustness to Height Differences: The system demonstrates robustness to height differences between APs and the target, with additional errors being relatively small.
  • Scalability with APs and Packets: Increasing the number of APs improves accuracy, and using more signal packets slightly enhances localization precision.

Limitations and Future Work

  • Height Difference Impact: While the system is robust to height differences, the paper notes that increasing height differences between APs and the target does increase the AOA estimation error. Further investigation into mitigating this effect could be beneficial.
  • Environmental Variations: Although the paper claims robustness to environmental changes, the evaluation is conducted in a controlled lab environment. Real-world deployment in more dynamic environments might reveal further challenges.
  • Future Work: The authors suggest that their system, leveraging OFDM properties, can be easily applied to future 5G communication systems, hinting at potential extensions and applications in next-generation wireless technologies.

Summary updated

BibTeX
@article{tian2016pila,
  author = {Zengshan Tian and Ze Li and Mu Zhou and Yue Jin and Zipeng Wu},
  title = {{PILA: Sub-Meter Localization Using CSI from Commodity Wi-Fi Devices}},
  journal = {Sensors},
  year = {2016},
  doi = {10.3390/s16101664},
}
Figure