Beam Hopping and Power Allocation of LEO Multi-Satellite Communication Network Based on Multi-Agent DQN Algorithm

Low Earth Orbit (LEO) satellite communication systems have recently been considered the key point for achieving seamless coverage and high-speed Internet services, with beam hopping (BH) technology being one of the core to achieve this goal. However, the rapid movement of LEO satellites results in regional and time-varying traffic demands. How to meet the time-varying and differentiated needs of ground users with limited beam resources efficientlty is a critical issue that current LEO satellite communication systems face. Additionally, overlapping coverage areas among multiple LEO satellites often lead to significant intra-band interference due to the full-frequency reuse bandwidth allocation scheme commonly used in existing LEO communication systems. This paper proposes a multi-agent Deep Q Network (DQN) deep reinforcement learning algorithm based on interference avoidance, where each agent is responsible for only beam illumination or power allocation of a single beam. The agents collaborate by sharing global states and rewards to schedule beam resources while decreasing beam-to-beam interference. Simulation results indicate that the proposed method outperforms traditional BH algorithm by effectively increasing system throughput under interference isolation.