Precision Beamforming for SAGIN Networks: A Reinforcement Learning Approach to Combat UAV Hovering Instability

In this study, we explore a novel approach to beamforming in a Space-Air-Ground Integrated Network (SAGIN) by utilizing a single unmanned aerial vehicle (UAV) equipped with a mounted antenna array. The UAV dynamically adjusts its beamforming weights in real-time to address the challenges posed by UAV hovering, which can impair the beam's performance. We propose a model-free reinforcement learning (RL) framework, integrated with a neural network, to predict the optimal beamforming weights and enhance the received signal-to-noise ratio (SNR). Traditional RL methods face difficulties with continuous action spaces, as they require explicit representation and updating of Q-values for each possible action, which is infeasible in complex scenarios like UAV beamforming. Our approach leverages deep reinforcement learning (DRL) to learn high-level decision-making strategies, demonstrating significant improvements in beamforming efficiency and energy utilization. The DRL framework showcases remarkable performance in mitigating beam distortion and optimizing SNR, thus advancing the state-of-the-art in UAV-assisted mmWave communications within SAGIN networks.