Satellite MEC with Federated Learning: Architectures, Technologies and Challenges

Abstract

Satellite communication has made great progress in recent years since it is characterized by wide information coverage and can support diverse types of users, which beneficially fulfills the demand of beyond 5G communications. Besides, mobile edge computing (MEC) technologies energize the edge devices with computational abilities to deal with the majority of training tasks without having to upload to the cloud server, which substantially enhances a system's efficiency. In satellite MEC, the raw data of edge users vested in different owners cannot be allowed to be shared, considering data privacy requirements. To address this, federated learning (FL) architecture can be applied to satellite MEC where only parameters and model updates can be transmitted, which avoids the interaction of raw data from diverse sources. In this article, we construct a FL-based satellite MEC architecture, followed by introducing its key techniques in the aspects of resource management and multi-modal data fusion. Furthermore, we study the data privacy and security protection on the FL-aided satellite MEC relying on a blockchain framework. Finally, we portray the challenges of FL-aided satellite MEC systems.