DNN-Based Distributed Downlink Power Control in User-Centric Cell-Free Massive MIMO Systems

This paper studies the distributed downlink power control problem in user-centric cell-free systems, where each user equipment (UE) is served by a subset of access points (APs) to reduce fronthaul requirements and computational complexity at the AP side. We propose a distributed downlink power control method based on a deep neural network (DNN) to enhance the long-term downlink data rate across the entire network. This method relies solely on locally collected large-scale fading information as the DNN input. Also, the method can adapt to dynamic scenarios with varying numbers of associated UEs while meeting real-time power control requirements. Simulation results demonstrate that the proposed scheme outperforms benchmark approaches in terms of network performance.