A Double Extended Kalman Filter Algorithm for Weakening Non-Line-of-Sight Errors in Complex Indoor Environments Based on Ultra-Wideband Technology

Sheng Xu, Qianyun Liu, Min Lin, Qing Wang, Kaile ChenSensors 2025 · DOI

TL;DR

This paper introduces a novel Double Extended Kalman Filter (DEKF) algorithm to combat Non-Line-of-Sight (NLOS) errors in Ultra-Wideband (UWB) indoor positioning. By employing a cascaded structure with a residual classification and covariance adjustment (RCCA) module, the DEKF significantly improves tracking accuracy in challenging environments where direct signal paths are frequently obstructed.

What Problem Does This Paper Address?

Accurate indoor positioning is crucial for a wide range of applications, from robotics and augmented reality to logistics and asset tracking. Ultra-Wideband (UWB) technology offers high precision, but its performance is severely degraded by multipath effects, particularly Non-Line-of-Sight (NLOS) conditions. In NLOS scenarios, UWB signals are reflected, refracted, or diffracted, leading to significant measurement errors that can cause traditional filtering algorithms to diverge or produce inaccurate results. Existing methods often struggle to reliably distinguish between Line-of-Sight (LOS) and NLOS measurements or require prior knowledge of NLOS conditions, which is not always available in dynamic indoor environments. This paper aims to address the challenge of maintaining high tracking accuracy in complex indoor environments with frequent and prolonged NLOS occurrences.

Methods

The core of the proposed solution is a Double Extended Kalman Filter (DEKF), which consists of two cascaded Kalman filters.

  1. First Stage: Kalman Filter with Residual Classification and Covariance Adjustment (RCCA)
    • This stage focuses on processing the raw UWB Time-of-Arrival (TOA) measurements.
  2. The state variables here are the distances to the anchors and their rates of change.
  3. The novel Residual Classification and Covariance Adjustment (RCCA) module is applied. It calculates the difference (residual) between predicted and measured distances.
  4. Based on the magnitude of these residuals, the algorithm classifies the measurement quality and dynamically adjusts the process noise covariance (Q) and measurement noise covariance (R) matrices of the first KF. This adjustment helps to mitigate the impact of NLOS errors by down-weighting unreliable measurements and up-weighting more trustworthy ones.
  5. The output of this stage is a smoothed distance state and its associated covariance.
  6. Second Stage: Extended Kalman Filter (EKF)
    • The smoothed distance information from the first KF is then used as the "observation" for the second EKF.
  7. This second EKF estimates the target's physical state, including its position, velocity, and acceleration (using Constant Velocity (CV) or Constant Acceleration (CA) models).
  8. By using the refined distance estimates from the first stage, the second EKF receives a more stable and accurate measurement input, leading to improved overall tracking performance.

The key innovation lies in the RCCA module, which adaptively adjusts the filter's noise covariances based on measurement residuals without explicit LOS/NLOS detection, making it robust to varying environmental conditions.

Datasets

The paper does not introduce new datasets. Instead, it relies on:

  • Simulated data: The proposed DEKF algorithm is evaluated using simulated data generated in MATLAB. The simulations involve both Constant Velocity (CV) and Constant Acceleration (CA) motion models for the target. Various NLOS scenarios (S1-S4) are created by defining different probabilities of NLOS occurrences for the five UWB anchors, mimicking diverse environmental conditions.
  • Experimental data: An experimental setup was built in an office environment (approximately 46 m²) using four UWB anchors and one target. This real-world environment was designed to include factors that induce NLOS conditions, such as people and equipment, and even metal sheets were used to block signals.

Key Insights and Ideas

  • Adaptive Covariance Adjustment is Key: The RCCA module's ability to dynamically adjust noise covariances based on measurement residuals is crucial. It allows the filter to adapt to changing signal conditions (LOS vs. NLOS) without needing explicit classification, which is often difficult in practice.
  • Cascaded Filters for Robustness: Using a two-stage filtering approach (KF for refined measurements, EKF for state estimation) effectively separates the problem of measurement error mitigation from state estimation, leading to improved overall performance.
  • NLOS Tolerance: The DEKF demonstrates remarkable resilience to prolonged NLOS periods, maintaining accuracy where other methods might fail or diverge. This is a significant advantage for real-world applications in complex indoor spaces.
  • No Prior LOS/NLOS Knowledge Needed: A major benefit is that the algorithm does not require prior knowledge or explicit detection of LOS/NLOS conditions, simplifying its deployment and making it more practical.

Key Results

  • Superior Accuracy: In both simulated and experimental tests, the proposed DEKF algorithm consistently outperformed existing methods like EKF, IMED-KF, and M-REKF, especially in mixed LOS/NLOS environments.
  • Centimeter-Level Accuracy: The DEKF achieved centimeter-level accuracy in LOS environments (within 4 cm) and maintained good accuracy in NLOS environments (within 10 cm).
  • Robustness in Severe NLOS: Even in challenging scenarios with a high proportion of NLOS measurements (e.g., 4-NLOS), the DEKF remained stable and achieved significantly lower average RMSEs compared to other algorithms. For instance, in the CA model with 4-NLOS, the DEKF had an average RMSE of 0.26 m, while EKF's average RMSE was 53.56 m.
  • Faster Convergence: Simulations indicated that the DEKF converges at a faster speed than its counterparts.

Limitations and Future Work

  • Computational Complexity: While the paper doesn't explicitly detail computational load, cascaded filters and adaptive adjustments can increase computational requirements. However, the authors claim it performs well in real-time scenarios.
  • Model Assumptions: The performance is evaluated under CV and CA models. Its effectiveness with more complex motion models might require further investigation.
  • Anchor Placement: While not a direct limitation of the algorithm, optimal anchor placement is always critical for UWB systems. The paper uses a fixed set of anchors for experiments.

Future work could explore extending this approach to other wireless technologies or investigating more advanced motion models. Further research into optimizing the RCCA parameters for even greater adaptability could also be beneficial.

Summary updated

Abstract

In complex indoor environments, target tracking performance is impacted by non-line-of sight (NLOS) noises and other measurement errors. In order to fix NLOS errors, a double extended Kalman filter (DEKF) algorithm is proposed, which refers to a kind of cascaded structure composed of two Kalman filters. In the proposed algorithm, the first filter is a classic Kalman filter (KF) and the second is an extended Kalman filter (EKF). Time of arrival (TOA) measurements collected by multiple stationary ultra-wideband (UWB) sensors are used. The residual errors between the measured TOA and that of the first KF are predicted, and the covariance of the first KF is adjusted correspondingly. Then, we use the estimated distance state of the first KF as a measurement vector for the second EKF in order to obtain a smoother observation. One of the advantages of the proposed algorithm is that it is able to perform target tracking with good accuracy even without or with only one LOS TOA measurement for a period of time without prior information about the NLOS noise, which may be difficult to obtain in practical applications. Another advantage is that the accuracy does not greatly decrease when NLOS noises exist for a long period of time. Finally, the proposed DEKF can maintain the high-precision positioning characteristics in both the constant velocity (CV) model and the constant acceleration (CA) model in the LOS/NLOS environment. Our simulation and experimental results show that the proposed algorithm performs much better than other algorithms in SOTA, particularly in severe mixed LOS/NLOS environments.
BibTeX
@inproceedings{xu2025double,
  author = {Sheng Xu and Qianyun Liu and Min Lin and Qing Wang and Kaile Chen},
  title = {{A Double Extended Kalman Filter Algorithm for Weakening Non-Line-of-Sight Errors in Complex Indoor Environments Based on Ultra-Wideband Technology}},
  booktitle = {Sensors},
  year = {2025},
  doi = {10.3390/s25030740},
}
Figure