Performance Comparison of NLOS Detection Methods in UWB

Abstract

For indoor positioning, it is important to accurately calculate inter-node distances, in which identifying whether the channel environment is line-of-sight (LOS) or non-LOS (NLOS) condition is critical. The traditional methods for NLOS detection often use extracting features of the channel environment. However, machine learning has recently known to make it possible to identify the channel environment more accurately than traditional methods. Therefore, we compare the performance of feature extraction-based SVM model for NLOS detection and CNN model based on imaging algorithms. Experiments show that CNN classifiers provide higher classification accuracy than SVM classifiers. In addition, it shows that applying imaging algorithms to data further improves the performance of CNN classifiers.