• Proposed an FL-based computation offloading algorithm for an LEO satellite edge computing system (LSECS) that combines federated learning and mobile edge computing.
• Achieves weighted sum-rate maximization for downlink and uplink cellular networks with full-duplex base stations and half-duplex users while ensuring user data privacy.
• Utilizes deep reinforcement learning for joint sub-channel assignment and power allocation, exploiting global search capabilities.
• Simulation results demonstrate superior weighted sum-rate performance and excellent convergence compared with baseline algorithms.