A Deep Reinforcement Learning Approach for Power Allocation in 5G Homogeneous Network Scenario

This paper presents a novel approach to optimizing power allocation within 5G homogeneous networks using deep reinforcement learning (DRL). The proposed method involves a two-stage training paradigm, beginning with offline deep Q learning (DQL) to train a dueling double deep Q network (D3QN) in a simulated environment. Subsequent online fine-tuning with real-world data further enhances the D3QN's performance. Simulation results demonstrate that the D3QN outperforms existing DQL-trained networks, achieving superior averaged sum rates and showcasing heightened adaptability across various user densities. This approach offers a promising solution for efficiently managing power allocation in wireless cellular networks, particularly within the context of 5G homogeneous network scenarios.