Deep Learning for Ultra-Wideband Indoor Positioning

Abstract

In recent years, the Ultra-wideband (UWB) system has been investigated for indoor localization and navigation by academia and industry. However, the UWB localization accuracy deteriorates when the signal propagates under severe non-line-of-sight (NLoS) conditions. We use two deep learning network models, the long short-term memory (LSTM) network and deep neural network (DNN), to analyze five different UWB signal features. The five features are received signal strength indication (RSSI), time of arrival (ToA), time difference of arrival (TDoA), first path (FP) amplitude from channel impulse response (CIR), and metric Mc (the ratio of the first path amplitude to peak amplitude). Then, we combine the five features into six different datasets for our deep learning models. Based on the prediction accuracy of the deep learning models for each combined feature, we propose a weighted indoor positioning (WIP) algorithm. The experiment results show that the WIP algorithm has better positioning accuracy than baseline works.