Robust LMedS-Based WLS and Tukey-Based EKF Algorithms Under LOS/NLOS Mixture Conditions

Chee-Hyun Park, Joon-Hyuk ChangIEEE Access 2019 · DOI

Abstract

Non-Line-of-Sight (NLOS) propagation is a significant challenge in wireless positioning, often introducing large biases that degrade the accuracy of standard estimators. This paper explores three robust algorithms—LMedS-WLS, LMedS-SSUT WLS, and a Tukey-based EKF—designed to maintain high precision in mixed LOS/NLOS environments. By integrating the Least Median of Squares (LMedS) for outlier identification with refined statistical filters, these methods achieve positioning accuracy near the theoretical Cramér-Rao Lower Bound (CRLB).

Introduction

In wireless localization, the ideal condition is Line-of-Sight (LOS), where signals travel directly from a source to a sensor. However, real-world environments like "urban canyons" or indoor spaces frequently encounter NLOS conditions, where signals reflect off obstacles. This adds a positive bias to distance measurements, causing traditional algorithms like Least Squares (LS) to fail. While robust statistical methods exist, they often struggle to balance computational efficiency with high accuracy. This research addresses this gap by proposing a two-stage approach: robust outlier detection followed by high-precision refinement.

Methodology

The proposed framework relies on a "Detection and Refinement" strategy to handle up to 50% outlier contamination.

LMedS-Based Weighted Least Squares (LMedS-WLS)

The first stage uses the LMedS algorithm to find an initial solution by minimizing the median of squared residuals. This provides a robust starting point that is immune to a high percentage of outliers. Following this, a 3-σ edit rule is applied to identify and replace measurements that deviate significantly from the LMedS estimate. Finally, a two-step WLS algorithm refines the position estimate.

LMedS-SSUT WLS

To further improve accuracy, the authors incorporate the Spherical Simplex Unscented Transform (SSUT). Unlike standard Taylor series expansions used in many WLS implementations, SSUT more accurately propagates the mean and variance through nonlinear functions. This leads to a more optimal weighting matrix, enhancing the precision of the final estimate while maintaining computational efficiency.

Tukey-Based Extended Kalman Filter (EKF)

For tracking applications, the authors developed an EKF that utilizes Tukey’s biweight risk function in the innovation step. Standard filters often "smear" the error from outliers across the state estimate. Tukey’s function, being "redescending," completely ignores measurements with residuals exceeding a specific threshold. This "hard-line" approach ensures that extreme NLOS errors do not corrupt the tracking performance.

Performance Evaluation

The algorithms were tested using numerical simulations in a 20 x 20 meter environment with nine sensors. The noise was modeled as a two-mode Gaussian mixture to represent the combination of LOS and NLOS signals.

The results indicate that the proposed methods significantly outperform benchmarks:

  • Accuracy: The LMedS-WLS and Tukey-EKF achieved a Root Mean Square Error (RMSE) of approximately 0.09m, compared to 0.15m for existing robust methods.
  • Stability: The positioning error remained nearly constant even as the magnitude of the NLOS bias increased, demonstrating the effectiveness of the outlier rejection mechanism.
  • Optimality: Performance metrics were shown to be very close to the CRLB, suggesting that the algorithms are performing near the physical limits of the system.

Conclusion

This research demonstrates that combining robust outlier detection with sophisticated refinement techniques can effectively mitigate the impact of NLOS errors in wireless positioning. The use of LMedS as a "gatekeeper" and Tukey’s function as a robust filter allows for near-optimal performance in challenging environments. Future work will focus on adapting these algorithms for real-time use in massive sensor networks and validating them with experimental hardware data.

Summary updated

BibTeX
@article{park2019robust,
  author = {Chee-Hyun Park and Joon-Hyuk Chang},
  title = {{Robust LMedS-Based WLS and Tukey-Based EKF Algorithms Under LOS/NLOS Mixture Conditions}},
  journal = {IEEE Access},
  year = {2019},
  doi = {10.1109/access.2019.2946376},
}
Figure