A Nested DRL-Based Method for Power Allocation and AP Sleep Control in Cell-Free Massive MIMO Systems

Cell-free massive MIMO (CFmMIMO) systems are increasingly recognized as a key technology to enhance 6G network performance. The large-scale coverage of access points (APs) and irrational power allocation leads to low energy efficiency and suboptimal Quality of Service (QoS). This paper proposes a nested deep reinforcement learning (DRL)-based optimization method utilizing novel double nested actor-critic (AC) frameworks to optimize power allocation and AP sleep control, which efficiently addresses the complex decision-making processes involved in dynamic environments. Extensive simulations results show that our proposed method enhances the peak energy efficiency in CFmMIMO systems by 19.6% and achieves a better trade-off between QoS and power consumption than the state-of-the-art and heuristic alternatives.