UWB NLOS identification with feature combination selection based on genetic algorithm

Abstract

Non-line-of-sight (NLOS) identification is very important for accurate localization based on ultra-wide band (UWB) system. One of the most widely used approach for NLOS detection is based on machine learning algorithms with features extracted from the channel impulse response (CIR). Features, such as kurtosis, mean excess delay, root mean delay, energy and rise time are discussed in a lot of papers. Other features, like signal to noise ratio, form factor and crest factor etc. are barely discussed but they are also very useful parameters for NLOS detection. In this paper 18 useful features are discussed in total. The support vector machine (SVM) is used for the identification of the NLOS condition. Since the identification accuracy does not always improve with an increase in the number of used features, in this paper the best feature combination is selected based on genetic algorithm. By reducing the used features, not only the accuracy improves, but also the computation complexity is reduced. The experimental results show that, the RMS delay, maximal amplitude, received signal energy, distance between MS and BS, peak to start of the received pulses time delay are the optimal combination leading to best accuracy.