Improving UWB Multipath Error Correction in TDoA Systems Using TWR-Derived Models

Abstract

Ultra-wideband (UWB) localization systems commonly rely on Two-Way Ranging (TWR) or Time Difference of Arrival (TDoA) techniques, each associated with its strengths and limitations. While TWR offers accurate ranging without clock synchronization, it requires frequent packet exchanges. In contrast, TDoA enables scalable, low-latency localization but suffers from synchronization and multipath-induced errors. Although many error correction approaches have been proposed for TWR, scientific work on error correction for TDoA is lacking. In this work, we propose a hybrid learning-based approach that leverages data collected from TWR systems to enhance time of arrival (ToA) error correction in TDoA-based localization. The motivation behind this strategy is: (1) labeled TWR datasets are more readily available in the scientific community, and (2) obtaining accurate error labels in real-life conditions for TWR is significantly easier than for TDoA. To this end, we train a convolutional neural network (CNN) using channel impulse response (CIR)s and their corresponding ranging error labels from a TWR system. Once trained, the model is applied directly, without retraining or fine-tuning to CIRs captured in a TDoA setup. Experimental results demonstrate that existing error correction approaches, such as anchor node selection, fail when encountering strong multipath environments, with 11.6% of positions not being calculated due to the removal of too many CIRs. In contrast, our approach enables significant gains in positioning accuracy in dense multipath environments, where only a limited number of signals are received.