Hierarchical learning for interference management in multi-user LEO satellite networks

In low Earth orbit (LEO) satellite networks, multiple satellites contend for limited frequency resources when they provide downlink services to ground users, necessitating efficient interference management. Particularly when there are multiple LEO service providers that do not explicitly exchange messages, satellites should learn about per-channel per-user interference. The problem is very challenging due to high learning complexity increasing with user population and time-varying interference caused by satellite orbiting. By exploiting reinforced learning (RL) techniques, we develop a low-complexity learning scheme that effectively allocate resources in respond to time-varying interference in multi-user multi-channel LEO satellite networks. The proposed scheme employs a hierarchical structure that aggregates information, reducing the complexity substantially, and enables the learning during short contact time. We demonstrate through simulations that our proposed scheme improves the sample efficiency and enhances throughput performance through successful interference management.