Power Allocation in Cell-Free mmWave Massive MIMO: Using Deep Deterministic Policy Gradient

This paper studies the weighted sum-spectral efficiency (SE) power allocation problem in cell-free (CF) mmWave massive multiple-input multiple-output (MIMO) with mobile user equipments (UEs). This is a non-convex problem that needs to be solved within a time constraint of the channel state information (CSI) variation. Although several heuristic methods, e.g., the weighted minimum mean square error (WMMSE) algorithm, have been developed to solve this problem, these methods entail considerable computational complexity therefore hardly meet the time constraint. To address this issue, we propose a deep reinforcement learning (DRL) solution based on the deep deterministic policy gradient (DDPG) method. It only needs several layers of neural network to perform the power allocation. The numerical results, obtained from a particular 3GPP scenario, show that the proposed DDPG method outperforms the existing algorithms.