Intelligent Spectrum Sharing Allocation and Decision-Making Based on Deep Learning in HSR NOMA–T2X Networks

Abstract

In high-speed railway (HSR) scenarios, the high mobility of trains introduces unique challenges, such as rapid channel variations and significant Doppler shifts. These challenges, together with the coexistence of multiple communication modes and the explosive growth of service demands, render spectrum resource scarcity and multi-mode communication coordination as pressing issues that must be considered. To address these issues, this paper introduces non-orthogonal multiple access (NOMA) technology in train-to-everything (T2X) scenarios to improve spectrum utilization, and proposes a NOMA-T2X communication framework, in which the train-to-ground (T2G) communication mode (Mode 1) uses a NOMA approach to collaboratively share the spectrum with the train-to-train (T2T) communication mode (Mode 2). Unlike existing studies that usually only consider a single time slot and ignore interzone handover, this paper studies a multi-slot global problem for T2X heterogeneous demand networks, including interzone handover. A collaborative multiple-actor critic (CMAC) algorithm based on deep learning is proposed, which achieves intelligent spectrum sharing as well as power allocation and handover decision-making through the collaborative joint training of multiple actor networks. Simulation results show that the proposed algorithm achieves the global optimization goal while taking into account interzone handover issue in HSR fast-moving scenarios, and compared with baseline schemes, it achieves performance improvements of no less than 17.1% and 7.9% in the framework level and algorithm level, respectively. The proposed algorithm is not limited to HSR scenarios, it also presents a novel approach for solving long-term optimization problems that combine 0-1 programming with continuous variables.