Research on Kernel Functions of SVM for Line-of-sight Identification in Vehicle-to-Vehicle MIMO System

Generally, propagation channels show significantly different characteristics for line-of-sight (LOS) and non-line-of-sight (NLOS) conditions. Due to their good performance in classification problems, support vector machines (SVM) have been widely used for NLOS identification of propagation channels. In this paper, we investigate the impact of different kernel functions on the accuracy of SVM-based NLOS identification and validate the performance based on measured channel data. We find that a Gaussian kernel reduces the mis-identification rate by a factor 4 compared to a linear kernel, and also outperforms polynomial and sigmoid kernels.