Multiagent DRL-Based Dynamic Beam Hopping and Resource Allocation in LEO Satellite Communication Systems

Efficient utilization of limited beam resources in low Earth orbit (LEO) satellite systems remains a significant challenge due to the uneven distribution and temporal variation of ground traffic demand. Multi-beam satellites can enhance system capacity through dynamic beam hopping (BH), but conventional BH schemes often rely on fixed, periodic patterns that fail to adapt to real-time traffic fluctuations. To address this, we propose a dynamic BH and resource allocation framework that incorporates precoding to mitigate co-channel interference. We formulate a joint optimization problem encompassing beam pattern design, bandwidth allocation, power control, and precoding, aiming to maximize the minimum traffic satisfaction across all ground cells. To solve this, we adopt a multi-agent proximal policy optimization (MAPPO) approach, reformulating the problem as a multi-agent Markov decision process (MDP) and applying centralized training with decentralized execution. Simulation results demonstrate that our method dynamically adjusts beam patterns and resource allocation in response to non uniform traffic demands, achieving superior traffic satisfaction and outperforming existing benchmark algorithms.