The emergence of mega-constellations of interconnected satellites has a major impact on the integration of cellular wireless and non-terrestrial networks (NTN). This paper studies the problem of running a federated learning (FL) algorithm within a low Earth orbit (LEO) constellation of satellites connected with intra-orbit inter-satellite links (ISL). Satellites apply on-board machine learning and transmit the local parameters to the parameter server (PS). The main contribution is a novel approach to enhance FL in satellite constellations using intra-orbit ISLs. The key idea is to rely on the predictability of satellite visits to create a system design in which ISLs mitigate the impact of intermittent connectivity and transmit the aggregated parameters to the PS. We first devise a synchronous FL, which is then extended towards an asynchronous FL for the case of sparse satellite visits to the PS. An efficient use of the satellite resources is attained by sparsification-based compression the aggregated parameters of each orbit before forwarding to the PS. Performance is evaluated in terms of accuracy and the required size of data to be transmitted. The numerical results indicate a faster convergence rate of the presented approach compared with the state-of-the-art FL on satellite constellations.

翻译：相互连接的卫星巨型星座的出现对蜂窝无线与非地面网络（NTN）的融合产生了重大影响。本文研究在低地球轨道（LEO）卫星星座中运行联邦学习（FL）算法的问题，该星座通过轨道内星间链路（ISL）连接。卫星应用星载机器学习，并将本地参数传输至参数服务器（PS）。主要贡献在于提出了一种利用轨道内ISL增强卫星星座中联邦学习的新方法。关键思路是借助卫星访问的可预测性设计系统架构，使ISL能够缓解间歇性连接的影响，并将聚合后的参数传输至PS。我们首先设计了同步联邦学习，随后针对卫星访问PS稀疏的情况将其扩展为异步联邦学习。通过基于稀疏化的压缩方法，在将每个轨道的聚合参数转发至PS之前实现卫星资源的高效利用。性能评估基于准确率和所需传输数据量。数值结果表明，与现有卫星星座联邦学习技术相比，所提方法收敛速度更快。