Power Allocation for Millimeter-Wave Railway Systems with Multi-Agent Deep Reinforcement Learning

Railway is evolving into the new era of smart railway. Unfortunately, the challenge of obtaining accurate instantaneous channel state information in high-speed railway (HSR) scenario makes it difficult to apply conventional power allocation schemes. In this paper, we propose an innovative experience-driven power allocation algorithm which is capable of learning power decisions from its own experience instead of the accurate mathematical model, just like one person learns one new skill, e.g. driving. To be specific, with the purpose of maximizing the achievable sum rate, we first formulate a joint hybrid beamforming and power allocation problem based on the millimeter-wave HSR channel model. Then, both at the transmitters (TXs) and receivers (RXs), we obtain the solution of beamforming design. Finally, experience-driven power allocation algorithm with multi-agent deep reinforcement learning is proposed. The numerical results indicate that the spectral efficiency of proposed algorithm significantly outperforms the existing state-of-the-art schemes.