Research on LOS/NLOS signal classification method based on CNN and CSI

In this paper, we propose a novel method for classifying Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) signals using Convolutional Neural Networks (CNN) and Channel State Information (CSI). The process involves removing pilot and guard carriers, extracting real, imaginary, and amplitude components, normalizing the data, padding it to a uniform length of 256, and reshaping it into 16x16 matrices to form RGB images. Experimental results show that our model outperforms traditional methods such as threshold-based recognition and SVM classification in various scenarios, demonstrating superior accuracy, precision, and recall. This highlights the efficiency, feasibility, and robustness of our approach, significantly enhancing classification accuracy and generalization.