Estimating TOA Reliability With Variational Autoencoders

Radio frequency (RF)-based localization yields centimeter-accurate positions under mild propagation conditions. However, conditions predominant in indoor environments (e.g. industrial production) are often challenging as signal blockage, diffraction and dense multipath lead to errors the time of flight (TOF) estimation hence a degraded accuracy. A major topic high-precision RF-based is identification such anomalous signals that negatively affect performance, mitigate introduced by them. As error characteristics depend on environment, data-driven approaches have shown be promising. there trade-off bad generalization need for an extensive time-consuming recording training data associated with it. We propose use generative deep learning models out-of-distribution detection based channel impulse responses (CIRs). Variational Autoencoder (VAE) predict anomaly score TOF-based Ultra-wideband (UWB) system. Our experiments show VAE trained only line-of-sight (LOS) generalizes well new detects non-line-of-sight CIRs accuracy 85%. also integrating our into extended Kalman filter (EKF) improves tracking performance over 25%.