MADRL-Based Joint Power Allocation and Hybrid Beamforming of mmWave NOMA Systems

This letter addresses joint hybrid beamforming (HBF) and power allocation (PA) in millimeter-wave (mmWave) non-orthogonal multiple access (NOMA) systems. We consider downlink transmission, with HBF at the base station (BS), and each user is equipped with a single antenna. User clustering at the BS is achived using a K-means clustering algorithm based on channel correlation statistics. To leverage the strengths of Multi-Agent Deep Reinforcement Learning (MADRL), we develop a new MADRL-based algorithm that jointly optimizes the HBF matrix and PA for each user. Simulations verify that the proposed solution outperforms other state-of-the-art schemes in perfect and imperfect Channel State Information (CSI) cases.