A Novel Hierarchical Federated Edge Learning Framework in Satellite-Terrestrial Assisted Networks

On-board federated learning based on dense Low Earth Orbit satellite constellations can meet the data privacy requirements of users in the coverage of non-terrestrial networks. However, traditional satellite-terrestrial assisted federated learning may encounter challenges due to limited satellite resources. To solve the problem, a satellite-terrestrial assisted hierarchical federated edge learning (STA-HFEL) framework is established in this paper. By leveraging well-endowed cloud servers for processing, inter-satellite links, predictability in satellite positioning, and partial aggregation, substantial reductions in training duration and communication costs are achieved. Furthermore, we define a problem within the STA-HFEL framework that involves optimizing the allocation of computation and communication resources for device users to attain overall cost minimization. To address this challenge, we introduce a resource allocation algorithm that operates effectively. Extensive performance evaluations demonstrate that the potential of STA-HFEL as a cost-efficient and privacy-preserving approach for machine learning tasks across distributed remote environments.