Range-Only UWB NLOS Identification Outside the Convex Hull in Dynamic Environments

Abstract

This letter addresses non-line-of-sight (NLOS) identification for ultra-wideband (UWB) localization in scenarios where a tag is located outside the anchor network’s convex hull. In these regions, conventional identification methods using least-squares trilateration (LST) subsets suffer from extreme instability due to high geometric dilution of precision (GDOP). This letter proposes a pairwise intersection kernel density estimation (PI-KDE) algorithm to mitigate this effect. Unlike traditional clustering that relies on biased LST estimates, PI-KDE constructs a candidate point set directly from the geometric intersections of anchor pairs. By bypassing the ill-conditioned trilateration process, PI-KDE suppresses GDOP-induced error amplification and employs kernel density estimation (KDE) to robustly identify NLOS anchors. Experimental results demonstrate that PI-KDE significantly improves identification accuracy outside the convex hull compared to state-of-the-art methods.