Multipath Model Order Selection for Non-Line Of Sight Radar Localization in Urban Environment

In an urban environment, the Non-Line of Sight (NLOS) target position can be determined by exploiting reflections on surrounding building, for instance using Matched Subspace Filter (MSF). However, it has been shown that the MSF output exhibits strong localization ambiguities when different positions share similar paths measurements. This ambiguity phenomenon is all the more exacerbated since the zones in the research domain are generally not illuminated by the same number of paths. In this paper, the well-known Bayesian Information Criterion (BIC) is considered to tackle this problem. Besides, we adopt a multipath selection procedure to select the relevant model for each position under test based on the Orthogonal Least Squares (OLS) sparse approximation algorithm and combine it with BIC as a stopping rule. These solutions, applied on both simulated and experimental data, show better localization results compared to the classical localization scheme.