LAUSR

We investigate effective interference management for Low Earth Orbit (LEO) satellite networks that provide downlink services to ground users and share the same frequency spectrum range. Since there are multi-group LEO satellites with different constellation orbits, the ground users will experience time-varying interference due to the overlapping of main/side lobes of the satellite beams, which becomes even more challenging when the interfering satellites cannot communicate directly. To address the problem, we consider two LEO satellite groups that provide communication service in the same ground area, while competing for communication resources. We develop solutions that maximize the throughput and manage the time-varying interference under a certain level, without explicit message exchanges between the satellite groups. By exploiting statistical learning and deep reinforcement learning techniques, we develop learning-based resource allocation schemes and evaluate their performance through extensive simulations. We show their effectiveness under different reward settings and different interference managements, and demonstrate that a Deep Q-Network (DQN)-based scheme can achieve the close-to-optimal performance.