Enhanced Vehicular Localization under Non-Line-of-Sight Multipath Propagation

Sung Il Choi, Hong Ki Kim, Sang Hyun Lee

Abstract: This paper develops localization strategies of non-line-of-sight (NLoS) hidden vehicles using Bayesian inference techniques and study an enhanced convergence property via statistical physics based analysis methods. NLoS vehicular localization is achieved with the measurement of multi-path channel propagation information, such as angle-of-arrival (AoA), angle-of-departure (AoD) and time-of-arrival (ToA) by characterizing their geometric relationship in a probabilistic inference task. An efficient solution is derived using generalized approximate message passing (GAMP), while convergence dynamics of GAMP has been reported as a critical issue with this approach. This paper develops a novel approach called generalized message passing (GMP) that revolves the convergence problem via compensation based on Onsager terms of the interaction model. Numerical results verify that the developed algorithm stabilizes the localization performance.