Feature Selection for Real-Time NLOS Identification and Mitigation for Body-Mounted UWB Transceivers

Ultra-wideband (UWB) is a popular technology for indoor positioning systems (IPSs) due to its robust signaling in harsh environments, through-wall propagation, and high-resolution ranging. However, the propagation of wireless signals in indoor environments is affected by nonline-of-sight (NLOS) conditions, which can lead to positively biased distance estimates. Several research works were carried out in this area, but the NLOS effect caused by the human body shadowing on the ranging performance is not properly addressed in the literature. In this paper, a commercial UWB transceiver is used to assess the impact of the human body shadowing on the ranging accuracy. A set of real-time features is combined with different machine learning algorithms for NLOS identification and mitigation. With a subset of four features, a 0.97 F1-score was obtained for NLOS identification. For NLOS mitigation, with a subset of three features, the residual error follows a Gaussian distribution, has a mean error close to zero, and an STD of 0.67 m.