Multi-Feature Fusion Based Recognition and Relevance Analysis of Propagation Scenes for High-Speed Railway Channels

Abstract

This paper proposes a multi-feature fusion based propagation scene recognition model for high-speed railway (HSR) channels and presents the channel relevance analysis of HSR scenes. Extensive field measurement data in typical HSR scenes, including rural, station, suburban and multi-link scenes, are collected with the assist of railway long-term evolution (LTE) networks. The datasets of space-time-frequency channel features, involving Ricean K-factor, root mean square delay spread, Doppler spread, and angle spread, are generated for the model training and testing as well as the relevance analysis. The proposed model merges a weighted score fusion scheme into the deep neural network (DNN) in order to adaptively determine the optimal weights for each feature stream. This weighted score fusion based DNN model is implemented and evaluated in terms of accuracy, confusion matrix, F-score, and receiver operating characteristic (ROC) curve, which exhibits better performance than other machine learning models like random forest, support vector machine (SVM), $k$ -nearest neighbor (KNN), and weighted KNN. In addition, the channel relevance of HSR scenes is analyzed from perspectives of high-dimensional distribution distance and joint correlation of multiple features. Two metrics, Wasserstein distance and correlation matrix collinearity, are used in the analysis. Statistical results are provided, which reveals the relatively strong channel relevance between the multi-link and suburban scenes.