A Distributed Deep Reinforcement Learning Framework for Power-Adaptive UAV Communication

In recent years, energy-efficient Unmanned Aerial Vehicles (UAV) communication has become a serious challenge due to the necessity for reliable coverage, fair user distribution, and minimal interference in dynamic environments. The significant growth in Deep Reinforcement Learning (DRL) has enabled UAVs to learn optimal control policies through interaction. However, existing models struggle to balance interference, energy use, and fairness effectively. Therefore, this research proposes a Multi-Agent Proximal Policy Optimization with Signal-to-Interference-plus-Noise Ratio (SINR)-Awareness and Adaptive Power Control (MAPPO-SAC) model. The projected framework employs distributed optimization where several UAVs autonomously learn to regulate their movement and transmission power based on SINR feedback. Initially, the UAVs explore the 3D environment and adaptively regulate power levels for interference-aware coverage. Further, a reward function guides learning by balancing coverage, fairness, and energy consumption. Additionally, a clipped update mechanism stabilizes training. Experimental results establish that proposed MAPPO-SAC significantly outperforms baseline models in terms of coverage ratio (93.5%), and average energy consumption (67.8joules) in dynamic UAV communication scenarios.