Energy-Aware Q-Learning for 5G O-RAN Control

Open Radio Access Network (O-RAN) allows parts of the wireless network to be controlled through the Near-Real-Time RAN Intelligent Controller (Near-RT RIC). This creates an opportunity to use lightweight machine-learning methods for radio resource control. Reinforcement learning has already been studied for resource allocation in 5G networks. Recent work has also explored energy-aware learning for 5G-RAN and O-RAN systems. In this work, we study a simple question: Can a small reinforcement-learning controller reduce radio power consumption without reducing the amount of traffic that the network can serve? We study this problem in a simulated 5G O-RAN downlink with a fixed total transmit-power budget of 1 W.