Explainable Deep Reinforcement Learning Based Intelligent Service Provisioning in UAV-Assisted 6G RAN

Unmanned aerial vehicles (UAVs) have emerged as a promising solution to enhance the coverage and capacity of radio access networks in response to the surging and emergency usage demands in hot and blind spots. Adaptive resource allocation is crucial for deploying UAVs to achieve desirable performance, which encounters tough challenges due to environmental complexity, QoE sensitivity, and resource scarcity. To address this issue, we present an explainable deep reinforcement learning scheme to provide intelligent service provisioning. In which the Cumulative Shapley Additive Explanation is utilized to supply state information in determining the serving UAV for users, while Legendre polynomials-based dynamic antenna beam-forming leverages the UAV resource allocation. UAV resource allocation and handover scheduling are optimized using the Deep Deterministic Policy gradient (DDPG) and Twin Delayed Deep Deterministic Policy Gradient algorithms, respectively. A Least Recently Used Memory mechanism is incorporated to optimize the Prioritized Experience Replay sampling process, which enhances the training of DDPG. Extensive simulations have been performed, and the comparative results demonstrate the outstanding performance of the presented scheme compared to the DDPG, PPO, and SAC baselines.