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

TL;DR

This work tackles the slow convergence of federated learning over sporadic, short satellite visibility in LEO networks. It introduces NomaFedHAP, a synchronous FL framework that leverages high-altitude platforms as distributed parameter servers and integrates NOMA-OFDM to enable concurrent, bandwidth-efficient model transmissions, while a new satellite-HAP topology mitigates Doppler effects and a balanced aggregation scheme prevents orbit bias. The authors derive closed-form outage probabilities for NS/FS and the overall system, and validate the approach through extensive simulations and real satellite imagery, showing an order-of-magnitude speedup in convergence and higher accuracy compared to state-of-the-art FL-SatCom methods. The results demonstrate practical feasibility for rapid, scalable FL in space-based networks, with strong performance under both IID and non-IID data and realistic communications conditions.

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.

Figures (10)

• Figure 1: An FL system in the context of satellite constellations, using multiple HAPs as parameter servers. For the sake of clarity, only two orbit shells are shown.
• Figure 2: Illustration of an FL-LEO system.
• Figure 3: A simulated visibility pattern over two days of six LEO satellites located in three different shells, each containing two satellites positioned at an altitude of 500km, 1250km, and 2000km, respectively. The $\mathcal{PS}$ is a GS located in Rolla, Missouri, USA. In the figure, $k_{i,j}$ represents the i-th satellite in the j-th orbit.
• Figure 4: NomaFedHAP with orbital shells 1,2,...$N$ and a single HAP as $\mathcal{PS}$ for illustration. Only visible satellites are drawn so all the shown links are LoS links.
• Figure 5: Illustration of the proposed model propagation algorithm. (a) The source HAP $\mathpzc{h}_1$ forwards the global model $\boldsymbol{w}^\beta$ to the sink HAP $\mathpzc{h}_4$; (b) Visible satellites are represented by blue, and invisible satellites by black. Colored curved arrows show the propagation of models (global model and sub-orbital model) from $k_2 \rightarrow k_4$, $k_4 \rightarrow k_7$, $k_7 \rightarrow k_{10}$, $k_{10} \rightarrow k_{11}$, $k_{11} \rightarrow k_{2}$. (c) The sink HAP $\mathpzc{h}_4$ propagates the sub-orbital model $w_k{^\beta}$ to the source HAP $\mathpzc{h}_1$ along the reverse pathway.

Theorems & Definitions (5)

• Theorem 1
• Lemma 1: Results of single-step SGD
• Lemma 2: Bounding the variance
• Lemma 3: Bound $\boldsymbol{{w}}^{\beta}_{k}$ divergence
• proof