Wireless Channel Modeling Based on Machine Learning in the High-speed Railway Scenario

Abstract

In the high-speed railway (HSR) scenario, there is a problem of poor generalization ability caused by the overfitting phenomenon through wireless channel modeling. In this paper, a precise modeling method based on machine learning is proposed to address this issue under the wireless channel of HSR. According to the K-means clustering algorithm, the typical values of the K-factor in a multitude of scenarios are obtained through Rician K-factor clustering in the wireless channel fading model. Founded on the theory of model evaluation and selection in machine learning, the measured values of path loss are fitted by least squares regression using the cross-validation method. Then, we obtain the formulas of the path loss models and the expected generalization errors of the models in different scenarios. The mathematical relationship of the ergodic capacity depending on the Rician K-factor, the signal-noise ratio of the receiver, and the path loss generalization error are established after the ergodic capacity analysis of the HSR wireless channel. Finally, the lower boundary curves of the ergodic capacity are obtained by simulation experiments in different scenarios.