NLOS identification using parallel deep learning model and time-frequency information in UWB-based positioning system

Abstract

Ultra wide band (UWB) radio positioning technology is widely used in indoor high-precision positioning scenes, while obstacles in the wireless signal propagation path will cause the non-line of sight (NLOS) propagation of UWB signal and the reduction of positioning reliability. In this paper, an efficient NLOS identification scheme based on multiple input learning (MIL) neural network model with channel impulse response (CIR) and time-frequency diagram of CIR (TFDOCIR) is proposed by direct detection in UWB positioning system. It is experimentally demonstrated that the average NLOS identification accuracies reach 86.82%, 92.53%, 91.61%, 92.91%, 92.02% corresponding to five different obstacles including wooden door, concrete wall, metal plate, human body and glass window, respectively. Additionally, the overall NLOS identification accuracy achieves 91.74%. Through the proposed NLOS identification scheme with weight least squares (WLS), the indoor UWB-based positioning tests are performed with the average error 7.35 cm, thereby proving its ability of ranging error preventation.