Non-Line-of-Sight Identification Based on Unsupervised Machine Learning in Ultra Wideband Systems

Abstract

Identi cation of line-of-sight (LOS) and non-line-of-sight (NLOS) propagation conditions is very useful in ultra wideband localization systems. In the identi cation, supervised machine learning is often used, but it requires exorbitant efforts to maintain and label the LOS and NLOS database. In this paper, we apply unsupervised machine learning approach called ``expectation maximization for Gaussian mixture models'' to classify LOS and NLOS components. The key advantage of applying unsupervised machine learning is that it does not require any rigorous and explicit labeling of the database at a certain location. The simulation results demonstrate that by using the proposed algorithm, LOS and NLOS signals can be classi ed with 86.50% correct rate, 12.70% false negative, and 0.8% false positive rate. We also compare the proposed algorithm with the existing cutting-edge supervised machine learning algorithms in terms of computational complexity and signals' classi cation performance.