Comparative Study of VPE-Driven CNN Models for Radio Wave Propagation Modeling in Tunnels

Abstract

Radio wave propagation modeling in railway environments is of fundamental importance in designing reliable train communication systems. In recent years, many machine learning (ML) techniques have been applied to accelerate the modeling process. In particular, convolutional neural networks (CNNs) have presented a superior performance in extracting features and reconstructing field distribution. However, the relevant literature is still missing a comprehensive study on CNN architecture design and the performance of different CNN models. In this article, we compare the performance of nine different CNNs, including recently developed advanced CNN techniques, for radio wave propagation modeling in tunnels. Each model is assessed in three different size variants to examine the effect of model complexity on performance. The CNN model is driven by a vector parabolic equation (VPE) channel simulator based on super-resolution. In addition, we investigate the performance of hybridizing various CNN architectures and present a CNN design roadmap for radio wave propagation modeling in tunnels. Besides, the proposed models are validated against measurement campaigns in two realistic tunnels.