Transformer-based NLoS detection in UWB localization systems

Abstract

Ultra-Wide-Band (UWB) communication is recognized as one of the most promising technologies for indoor localization due to its ability to capture a high-resolution channel impulse response (CIR) at the receiver and penetrate walls and other obstacles. In UWB-based localization systems, a distance between the target object and anchor nodes is estimated by measuring the travel times of UWB signals. In indoor environments, numerous obstacles that block the Line-of-Sight (LoS) path between the target object and anchor nodes induce errors in the estimation of the time of signal flight. As a result, localization accuracy deteriorates. One of the techniques to address this issue is to discard range measurements of non-LoS (NLoS) anchors or to handle them differently when calculating the target position. In this paper, we apply Transformer deep learning model to detect NLoS channel conditions from the raw CIR data. The model has been trained and evaluated on a realistic dataset collected in an industrial environment. The results show performance improvement over the state-of-the-art convolutional neural network model.