Supervised Machine Learning Algorithms for LOS/NLOS Classification in Ultra-Wide-Band Wireless Channel

Abstract

Ultra-Wide-Band (UWB) channels have attracted attention due to its potential in constructing accurate indoor positioning system (IPS) such as residential environment. Although promising, UWB channels experience some problems inherent to indoor environments particularly when in both the transmitting and the receiving antenna there is no a direct line of sight which producing an estimation error on the positioning accuracy. Thus, in this paper, it is evaluated and compared the application of different machine learning algorithms (MLA) to UWB channel classification in order to detect if a UWB channel is a line of sight (LOS) or a not line of sight (NLOS) channel. The MLA are capable to work on different characteristics which improves channel classification. The attained results prove the validity of the different MLA in order to classify between LOS/NLOS channels where the best performing classifier was obtained by the Gradient Boosting on the set of simulated UWB channel realizations.