Joint power control, beamforming, and sleep-mode selection for energy-efficient cell-free networks using surrogate machine learning models

In recent years, sleep-mode or access point (AP) on-off switch techniques have attracted significant attention for reducing the energy consumption of cell-free massive MIMO systems. In this context, this work considers the problem of finding the smallest subset of active access points (APs) needed to satisfy minimum quality-of-service requirements while considering the optimal configuration of uplink transmit powers and (potentially distributed) beamformers. To address this challenging problem, we judiciously combine novel fixed-point methods for jointly optimal power control and distributed beamforming design with a global optimization framework based on surrogate machine-learning models. Numerical results show that our proposed on-off switch technique can achieve a significantly higher reduction in total APs power consumption than a baseline that does not jointly configure the uplink transmit powers and the beamformers.