MADDPG-Based Power Allocation Algorithm for Network-Assisted Full-Duplex Cell-Free MmWave Massive MIMO Systems with DAC Quantization

Abstract

Network-assisted full-duplex (NAFD) systems reduce the cross-link interference (CLI) by dividing the remote antenna unit (RAU) into the transmitting RAU (T-RAU) and receiving RAU (R-RAU), keeping them geographically separated and flexibly utilizing duplex modes, which further improves the system performance. The NAFD cell-free millimeter-wave (mmWave) massive multiple-input multiple-output (MIMO) systems with digital-to-analog converter (DAC) quantization is investigated in this paper. We propose an optimization problem of jointly power allocation of the T-RAUs and uplink users to maximize the weighted uplink and downlink sum rate, in which bidirectional power constraints need to be satisfied. To handle this intractable problem, we further apply a deep reinforcement learning algorithm based on multi-agent deep deterministic policy gradient (MADDPG) instead of the conventional convex optimization approach. The simulation results verify the convergence of the proposed MADDPG-based algorithm, explore the learning performance of each agent, analyze the impact of DAC quantization on NAFD cell-free mmWave massive MIMO systems, and compare the performance of the MADDPG-based algorithm in static and dynamic environments.