Neural Network-Based Multi-DOA Tracking for High Speed Railway Communication Systems

Abstract

Along with the rapid development of high-speed railway (HSR), it is of vital importance to explore new technologies to meet the stringent requirements on wireless transmission latency and reliability in such high mobility scenarios. To address this, fast beamforming schemes with precise direction-of-arrival (DOA) can achieve effective channel gain and improve interference mitigation. In this paper, we propose a novel radial-basis function neural network (RBFNN)-based method for the multi-DOA tracking, in which the RBFNN is trained by the location-based prior information. Consequently, the DOA tracking problem is transformed into an RBFNN with trained input/output pairs. The effectiveness of the proposed scheme in terms of complexity and accuracy is verified by simulation results. Finally, the impact of the shapes of the array on estimation accuracy is also discussed.