A Survey on Satellite Networks with Federated Learning to Analyze Data or Manage Resource
Junsuk Oh, Dong-Hyun Lee, Thanh Phung Truong, Donghyeon Hur, Seonghun Hong, Sungrae Cho · 2025
Satellite networks are critical infrastructures for next-generation networks. Federated learning (FL) has emerged as an innovative approach to address two fundamental challenges in these networks: analyzing massive volumes of generated data and managing dynamic resources in complex, distributed, and heterogeneous environments. FL enables decentralized training of machine learning models, reducing the need for bandwidth-intensive raw data transmission and preserving privacy. In addition, FL supports scalable and adaptive resource management, enabling efficient traffic offloading, task distribution, and link scheduling across satellite mega-constellations and space-air-ground integrated networks. In this work, we survey the state-of-the-art FL for satellite networks from the perspective of two use cases: data analysis and resource management. Then, we explore synchronous and asynchronous approaches related to the FL process and highlight their advantages, limitations, and applications. The insights provided aim to guide key methodologies and future directions in leveraging FL to build intelligent, scalable, and efficient satellite mega-constellation networks.