Multi-Agent DDPG-Based Joint Beam Hopping and Resource Allocation for Cognitive GEO-LEO Satellite Networks

This letter presents a multi-agent deep reinforcement learning-based framework for joint beam hopping and resource allocation in the secondary LEO system within a cognitive GEO-LEO satellite network. Beam hopping is adopted to flexibly steer limited resources toward areas with high and dynamic traffic demand, improving spectral efficiency and load balancing. A centralized critic multi-agent deep deterministic policy gradient approach is proposed to optimize beam hopping patterns, bandwidth and power allocation, aiming to enhance system throughput and reduce service delay imbalance across LEO cells. Simulation results show that the proposed method outperforms benchmark approaches, achieving approximately 45% higher throughput and around 37% lower service delay under varying traffic and power conditions.