Satellite Edge Computing for Mobile Multimedia Communications: A Multi-agent Federated Reinforcement Learning Approach

The rapid expansion of satellite mega-constellations has highlighted the potential of satellite edge computing as a promising solution for mobile multimedia communications. While reinforcement learning has been explored in satellite communication systems, significant challenges remain, including high latency and limited resources. This study addresses these challenges by focusing on the joint optimization of communication, computing, and caching resources in satellite edge computing to support mobile multimedia applications. A mixed-integer nonlinear programming (MINLP) problem is formulated with the objective of minimizing the total delay experienced by mobile users, subject to multidimensional resource capacity constraints, which are NP-hard and computationally intractable to solve in polynomial time. To address this complexity, we propose a multi-agent federated reinforcement learning (MAFRL) approach as an efficient solution. In this framework, each satellite operates as an autonomous learning agent equipped with an actor-critic network structure. The proposed MAFRL method demonstrates superior performance, achieving lower delays compared to all baseline approaches. It effectively optimizes delay-sensitive mobile multimedia communications by minimizing total delay and improving task-offloading ratios. To the best of the authors’ knowledge, this study is the first to introduce an MAFRL-based approach for resource allocation in satellite edge computing, marking a significant contribution to the field.