Abstract:Space AI has become increasingly important and sometimes even necessary for government, businesses, and society. An active research topic under this mission is integrating federated learning (FL) with satellite communications (SatCom) so that numerous low Earth orbit (LEO) satellites can collaboratively train a machine learning model. However, the special communication environment of SatCom leads to a very slow FL training process up to days and weeks. This paper proposes NomaFedHAP, a novel FL-SatCom approach tailored to LEO satellites, that (1) utilizes high-altitude platforms (HAPs) as distributed parameter servers (PS) to enhance satellite visibility, and (2) introduces non-orthogonal multiple access (NOMA) into LEO to enable fast and bandwidth-efficient model transmissions. In addition, NomaFedHAP includes (3) a new communication topology that exploits HAPs to bridge satellites among different orbits to mitigate the Doppler shift, and (4) a new FL model aggregation scheme that optimally balances models between different orbits and shells. Moreover, we (5) derive a closed-form expression of the outage probability for satellites in near and far shells, as well as for the entire system. Our extensive simulations have validated the mathematical analysis and demonstrated the superior performance of NomaFedHAP in achieving fast and efficient FL model convergence with high accuracy as compared to the state-of-the-art.

What problem does this paper attempt to address?

The paper primarily addresses the communication efficiency issues faced by Federated Learning (FL) in the integration of Low Earth Orbit (LEO) satellite networks and High Altitude Platforms (HAPs). Specifically, the paper aims to solve the following key problems: 1. **Addressing the slow training problem of Federated Learning in satellite communications**: Due to the unique environment of satellite communications, traditional Federated Learning methods encounter very slow training processes in LEO satellite networks, which may take days or even weeks to complete a full training cycle. 2. **Improving communication efficiency using Non-Orthogonal Multiple Access (NOMA)**: A new Federated Learning framework, NomaFedHAP, is proposed. This framework combines NOMA technology, allowing LEO satellites on different orbital shells to concurrently transmit models, thereby significantly improving bandwidth utilization and model transmission speed. 3. **Introducing High Altitude Platforms as distributed parameter servers**: To enhance satellite visibility and reduce the impact of Doppler shifts, the scheme utilizes multiple HAPs as distributed parameter servers. These platforms collaborate to improve the overall communication efficiency of the system. 4. **Optimizing the model aggregation strategy**: A new model aggregation scheme is proposed, which balances the number of models between different orbits, avoids skewed global models, and ensures the quality of the overall model. 5. **Improving the communication topology**: A new communication topology is designed, where HAPs not only serve as parameter servers but also act as relay nodes for satellites between different orbits, effectively mitigating the Doppler shift problem. In summary, the goal of the paper is to significantly enhance the training speed and efficiency of Federated Learning in LEO satellite networks through a series of technological innovations, including the introduction of NOMA, the utilization of HAPs, and improved model aggregation and communication strategies, thereby achieving faster and more efficient model convergence.

Communication-Efficient Federated Learning for LEO Satellite Networks Integrated with HAPs Using Hybrid NOMA-OFDM

Non-orthogonal Multiple Access Assisted Federated Learning for UAV Swarms: an Approach of Latency Minimization

FedLEO: An Offloading-Assisted Decentralized Federated Learning Framework for Low Earth Orbit Satellite Networks

AsyncFLEO: Asynchronous Federated Learning for LEO Satellite Constellations with High-Altitude Platforms

Optimizing Federated Learning in LEO Satellite Constellations via Intra-Plane Model Propagation and Sink Satellite Scheduling

FedSN: A Federated Learning Framework over Heterogeneous LEO Satellite Networks

Satellite Based Computing Networks with Federated Learning

DFedSat: Communication-Efficient and Robust Decentralized Federated Learning for LEO Satellite Constellations

FedSpace: An Efficient Federated Learning Framework at Satellites and Ground Stations

Federated Learning Via Non-Orthogonal Multiple Access for UAV-Assisted Internet of Things

Federated Learning Via Nonorthogonal Multiple Access for UAV-Assisted Internet of Things

SatFed: A Resource-Efficient LEO Satellite-Assisted Heterogeneous Federated Learning Framework

Energy-Efficient Hierarchical Collaborative Learning over LEO Satellite Constellations

A Hybrid NOMA-OMA Scheme for Inter-plane Intersatellite Communications in Massive LEO Constellations

Secure and Efficient Federated Learning in LEO Constellations using Decentralized Key Generation and On-Orbit Model Aggregation

Cooperative Federated Learning over Ground-to-Satellite Integrated Networks: Joint Local Computation and Data Offloading

On-board Federated Learning for Satellite Clusters with Inter-Satellite Links

Deep Reinforcement Learning-Based NOMA-Aided Slotted ALOHA for LEO Satellite IoT Networks

Stitching Satellites to the Edge: Pervasive and Efficient Federated LEO Satellite Learning

Satellite Federated Edge Learning: Architecture Design and Convergence Analysis

Over-the-Air Federated Learning in Satellite systems