Improved UWB-based indoor positioning system via NLOS classification and error mitigation

Abstract

Non-Line-of-Sight (NLOS) conditions in indoor positioning systems significantly degrade positioning accuracy. Although Ultra-Wideband (UWB) technology is renowned for its high precision in Line-of-Sight (LOS) environments, under NLOS conditions, positioning errors typically exceed 30 cm. To address this issue, we propose a method for identifying and classifying NLOS signals based on Support Vector Machine Recursive Feature Elimination (SVM-RFE). We extract multiple features from the UWB Channel Impulse Response (CIR) and perform correlation analysis using the Pearson Correlation Coefficient (PCC) to select the most discriminative features via the SVM-RFE algorithm. The classification results are then utilized within an Adaptive Robust Extended Kalman Filter (AREKF) to establish an error model for mitigation. The proposed method was evaluated using both a public dataset from Ghent University and a locally collected dataset. On the public dataset, the SVM-RFE algorithm achieved classification accuracies of 97.6% in the hallway environment and 96.6% in the office environment, outperforming transfer learning (TL) deep neural networks (DNNs) tested on the same dataset. To further validate the robustness of the algorithm, experiments on the locally collected office dataset demonstrated a classification accuracy of 97.2%. In terms of distance measurement error mitigation, the proposed AREKF algorithm reduced errors at the 95th percentile by 70% and 75% in two different environments compared to transfer learning methods on the same public dataset. When tested on the locally collected dataset, the positioning error of the AREKF was significantly lower than that of other mainstream algorithms, highlighting the practical advantages of the proposed method.