UWB TDoA Error Correction Using Transformers: Patching and Positional Encoding Strategies

Abstract

UWB TDoA localization accuracy degrades in industrial non-line-of-sight (NLOS) environments, where traditional methods of excluding NLOS links are often infeasible and degrade geometric precision. To address these limitations, we propose a novel position correction method using a transformer encoder. The model directly processes raw channel impulse responses (CIRs) from all available anchors by first partitioning them into patches. These patches are converted into tokens and combined with novel spatial positional encodings before the transformer learns their complex interdependencies to compute a final position correction. We analyze multiple patching and encoding strategies to evaluate their impact on performance and scalability. Based on experiments on real-world UWB measurements, our approach can provide accuracies of up to 0.39 m in a complex environment consisting of (almost) only NLOS signals, which is an improvement of 73.6% compared to the TDOA baseline.