NLOS Mitigation for UWB Localization Based on Sparse Pseudo-Input Gaussian Process

Abstract

Ultra-wideband technology has found promising application in high accuracy localization due to its high time resolution and through-wall propagation properties. However, its performance seriously degrades in non-line-of-sight (NLOS) scenario. Gaussian Process (GP) regression is the state-of-the-art machine learning approach that addresses this issue. But it is too complex in its original form. This paper proposes a novel NLOS mitigation method based on Sparse Pseudo-input Gaussian Process (SPGP) with low complexity. In contrast to conventional approaches which perform NLOS identification first, this approach directly mitigates the bias of both LOS and NLOS conditions. Monte-Carlo simulations demonstrate that with much less (very sparse) training data, SPGP achieves performance comparable to GP regression.