Spectrum allocation method for millimeter-wave train-ground communication in high-speed rail based on multi-agent attention

Abstract

With the advancement of high-speed railways toward intelligent systems, a large number of IoT devices have been deployed in both onboard and trackside systems. The resulting surge in data transmission has intensified competition for spectrum resources, thereby significantly increasing the demand for train-ground communication systems with high capacity, low latency, and strong interference resilience.The millimeter wave (mmWave) frequency band provides a large bandwidth to support massive data transmission from IoT devices. Aiming at addressing the issues of low network capacity, high interference, and low spectral efficiency in mmWave train-ground communication systems under 5G-R for high-speed railways, we propose a multi-agent attention mechanism for mmWave spectrum allocation in train-ground communication. First, we analyzed the spectrum requirements of mmWave BS and onboard MRS, constructed a spectrum resource allocation model with the optimization objective of maximizing system network capacity, and transformed it into a Markov decision process (MDP) model. Next, considering the need for coordinated spectrum allocation and interference suppression between mmWave BS and MRS, we develop a resource optimization strategy using the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm. Specifically, we combine multi head attention mechanism to improve the Critic network of MADDPG algorithm. This enhancement enables coordinated global–local strategy optimization through attention weight computation, thereby improving decision-making efficiency. Simulation results demonstrate that compared to existing methods, our algorithm achieves superior spectrum allocation performance, significantly increases network capacity while reducing interference levels, and meets the spectrum requirements of HSR communication systems.