Deep Learning Utilization to Control Uplink Power of a Cell-Free 5G Wireless System

This thesis investigates the application of deep reinforcement learning (DRL), specifically the Soft Actor-Critic (SAC) model, for uplink power control in cell-free Massive MIMO (mMIMO) systems by conducting detailed simulations; it compares the SAC-based DRL model's performance with traditional power control algorithms—Max-Min, Max-Prod, and Max-Sum-Rate—across metrics like spectral efficiency, area throughput, uplink user equipment power, and SINR. The development and optimization of the SAC model highlight its efficacy in complex, dynamic environments through the strategic balance of exploration and exploitation using entropy. Results, visualized through cumulative distribution functions and heat maps, show the superiority of the DRL models over the traditional algorithms in SINR improvements near the access points, elevating low SINR and improving the total system throughput, suggesting more efficient power allocation and interference management. The findings underscore the potential of DRL in enhancing network performance, paving the way for further exploration into advanced machine-learning techniques for optimizing wireless communications.