Smart Beamforming for High Mobility Millimeter-Wave Train-to-Infrastructure Networks: A Machine Learning Approach

The evolution of wireless communication systems is undergoing a transformative shift with the integration of Artificial Intelligence (AI). In the era of high-mobility millimeter-wave (mmWave) train-to-infrastructure (T2I) communication systems, the dynamic nature of the environment poses unique challenges for traditional beamforming approaches. Therefore, the demand for robust and adaptive beamforming solutions is paramount. This paper introduces a novel machine learning (ML)-driven beamforming solution tailored for predicting pairs of Three-dimensional (3D) beams at the receiver and transceiver sides in both Line-of-Sight (LoS) and Non-Line-of-Sight (NLoS) scenarios. The proposed approach addresses the specific challenges posed by mmWave frequencies, train mobility, and diverse propagation and environmental conditions. The methodology of our work integrates the collection of a comprehensive dataset capturing the environmental conditions and encompassing the different characteristics of the train movement. To ensure accurate and timely predictions of 3D beam pairs, we carefully develop and compare various multi-class supervised machine learning classification algorithms. Experimental evaluations conducted in LoS and NLoS scenarios showcase the superior performance of the proposed beamforming technique. Our approach excels in accurately predicting 3D beams with negligible training overhead.