TL;DR
This paper introduces an Improved Strong Tracking Cubature Kalman Filter (ISTCKF) to enhance the accuracy and stability of Ultra-Wideband (UWB) positioning systems. By intelligently reconstructing observations and adaptively identifying Non-Line-of-Sight (NLOS) errors, the ISTCKF significantly reduces positioning errors, outperforming existing methods.
What Problem Does This Paper Address?
Wireless positioning systems, particularly those using Ultra-Wideband (UWB) technology, are crucial for the Internet of Things. While UWB offers high accuracy in ideal Line-of-Sight (LOS) conditions, its performance degrades significantly in real-world scenarios due to Non-Line-of-Sight (NLOS) errors. These errors, caused by signal reflections, refractions, and diffractions from obstacles, are a major challenge, leading to reduced trustworthiness of measurements and potential filter divergence. Existing methods for handling NLOS errors often struggle with real-time performance, applicability in complex environments, or require extensive training data. This paper aims to address these limitations by developing a robust and adaptive filtering approach for UWB positioning in the presence of NLOS errors.
Methods
The core of this paper is the proposed Improved Strong Tracking Cubature Kalman Filter (ISTCKF). The approach involves several key steps:
- Observation Reconstruction: The algorithm first reconstructs the original UWB observations. This is done by combining positioning results from a predictive dynamics model (based on Cubature Kalman Filter - CKF) and a least squares method. This reconstruction aims to mitigate observation noise and initial NLOS errors.
- NLOS Identification: The reconstructed observations are then used to identify NLOS signals. This is achieved by analyzing the statistical differences between LOS and NLOS observation noise. The paper proposes a two-stage NLOS identification process (primary and secondary) to improve robustness.
- Adaptive Fading Factor: Based on the NLOS identification results, the algorithm adaptively adjusts a "fading factor." This factor is crucial for the strong tracking aspect of the filter. In LOS conditions, it focuses on mitigating predictive dynamics model errors, while in NLOS conditions, it emphasizes historical observations to counter NLOS errors.
- Improved Cholesky Decomposition: To enhance filter stability, the paper replaces the standard Cholesky decomposition with Singular Value Decomposition (SVD) when dealing with covariance matrices. This prevents potential issues with non-positive definite matrices that can arise in filtering processes.
The ISTCKF builds upon the standard Cubature Kalman Filter (CKF) by incorporating these novel NLOS handling and adaptive filtering mechanisms.
Datasets
The paper utilizes both simulated and real-world experimental data for evaluation.
- Simulated Data: The simulation scenario involves four anchor nodes deployed in a 20m x 20m area. Tags move along a predefined trajectory, and simulated LOS Gaussian noise and continuous pedestrian NLOS errors (modeled as uniform distribution U(90 cm, 120 cm)) are added to the observations. The simulation considers scenarios with 1, 2, 3, and 4 NLOS observations occurring simultaneously.
- Field Experiment Data: A DWM1000 UWB positioning module was used for field experiments. The setup involved four anchor nodes and a tag moving around columns in a hall, intentionally creating NLOS interference. The tag followed a preset trajectory, and positioning was performed every 0.2 seconds.
Key Insights and Ideas
- Reconstruction as a Preprocessing Step: Reconstructing observations before filtering is a powerful way to pre-process noisy and erroneous data, making subsequent filtering more effective.
- Statistical Differences for NLOS Detection: Leveraging the distinct statistical properties of LOS and NLOS noise is a robust way to identify problematic observations without relying on complex geometric constraints or extensive prior knowledge.
- Adaptive Fading for Targeted Error Mitigation: The idea of using different fading factors for different error sources (dynamics vs. NLOS) is key to achieving balanced performance. It allows the filter to dynamically prioritize which error to combat based on the current environmental conditions.
- SVD for Robustness: Replacing Cholesky decomposition with SVD is a subtle but important improvement for ensuring the numerical stability of the Kalman filter, especially in dynamic or noisy environments.
Key Results
The proposed ISTCKF algorithm demonstrated significant improvements in positioning accuracy and stability:
- Positioning Accuracy: The ISTCKF reduced positioning errors by 55.2% compared to STCKF, 32.3% compared to ACKF, and 28.9% compared to RSTCKF in simulation.
- NLOS Error Mitigation: The algorithm effectively mitigates NLOS errors, as evidenced by the reduced positioning errors in scenarios with multiple NLOS observations.
- Field Experiment Validation: In field experiments, the ISTCKF achieved an average positioning error of 9.86 cm, outperforming STCKF (22.04 cm), ACKF (14.57 cm), and RSTCKF (13.88 cm).
- Stability: The velocity component of the filtered positioning results from ISTCKF remained relatively stable, indicating enhanced robustness, especially during smooth tag movements.
Limitations and Future Work
- Misjudgment Thresholds: The paper acknowledges that the current NLOS identification scheme makes a "simple judgment" on the bounds between observation noise and NLOS errors or prediction dynamics model errors. This can lead to occasional misjudgments of observation trustworthiness, potentially affecting accuracy.
- Future Work Suggestion: The authors suggest that future work could involve a more comprehensive consideration of thresholds for observation trustworthiness, enabling adaptive localization and more accurate threshold determination. This could lead to further improvements in robustness and accuracy.
Summary updated