Dynamic-Beamforming-and-Intelligent-UAV-Handover-in-6G-RAN-Using-DRL

The need to enhance coverage and capacity in response to unpredictable challenges faced by telecommunication operators has become a key topic of research. Ground-based stations often struggle to provide sufficient coverage and capacity when users move between different coverage areas, especially due to obstacles that weaken the signal strength. To address this, we utilize Unmanned Aerial Vehicles (UAVs)  that receive directives from a controller network via the x2 interface to assist users in blind spots. The controller utilizes the Cumulative Shapley Additive Explanation (CSHAP) to select the suitable UAV for serving users, while the UAV employs the Legendre polynomials to achieve antenna beamforming. The challenge is modeled as a Markov Decision Process, with the Deep Deterministic Policy Gradient (DDPG) and the  Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithms used to solve it. Additionally, we incorporate a Least Recently Used memory (LRUM)  mechanism to optimize the Prioritized Experience Replay (PER) sampling process, which enhances the training of DDPG, in solving the UAV problem. Extensive simulations have been performed, achieving a 5\\% increase in handover success rate, a 10\\% rise in data rate, and a 15\\% decrease in latency compared to the DDPG, PPO  and SAC baselines.