Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Scheduling for On-Board Federated Learning with Satellite Clusters

Nasrin Razmi, Bho Matthiesen, Armin Dekorsy, Petar Popovski

TL;DR

This work tackles federated learning over satellite constellations with intermittent ground-station connectivity. It introduces a two-tier scheduling framework—Global Update (GU) at the GS and Cluster Update (CU) within each orbit cluster—that exploits both per-satellite visibility and the cumulative orbit-level visibility to time global updates and local training. GU selects the earliest feasible update times $t_n$ using per-cluster demand and finish times, while CU adapts the number of local epochs $I_{n,p}$ according to the available time in each time slot, coordinating parameter exchange via intra-orbit ISL. Experiments on a Walker Delta constellation with CIFAR-10 show the scheduler improves test accuracy and reduces wall-clock time compared to non-scheduled baselines, highlighting the practical impact for robust on-board FL in space missions.

Abstract

Mega-constellations of small satellites have evolved into a source of massive amount of valuable data. To manage this data efficiently, on-board federated learning (FL) enables satellites to train a machine learning (ML) model collaboratively without having to share the raw data. This paper introduces a scheme for scheduling on-board FL for constellations connected with intra-orbit inter-satellite links. The proposed scheme utilizes the predictable visibility pattern between satellites and ground station (GS), both at the individual satellite level and cumulatively within the entire orbit, to mitigate intermittent connectivity and best use of available time. To this end, two distinct schedulers are employed: one for coordinating the FL procedures among orbits, and the other for controlling those within each orbit. These two schedulers cooperatively determine the appropriate time to perform global updates in GS and then allocate suitable duration to satellites within each orbit for local training, proportional to usable time until next global update. This scheme leads to improved test accuracy within a shorter time.

Scheduling for On-Board Federated Learning with Satellite Clusters

TL;DR

This work tackles federated learning over satellite constellations with intermittent ground-station connectivity. It introduces a two-tier scheduling framework—Global Update (GU) at the GS and Cluster Update (CU) within each orbit cluster—that exploits both per-satellite visibility and the cumulative orbit-level visibility to time global updates and local training. GU selects the earliest feasible update times using per-cluster demand and finish times, while CU adapts the number of local epochs according to the available time in each time slot, coordinating parameter exchange via intra-orbit ISL. Experiments on a Walker Delta constellation with CIFAR-10 show the scheduler improves test accuracy and reduces wall-clock time compared to non-scheduled baselines, highlighting the practical impact for robust on-board FL in space missions.

Abstract

Mega-constellations of small satellites have evolved into a source of massive amount of valuable data. To manage this data efficiently, on-board federated learning (FL) enables satellites to train a machine learning (ML) model collaboratively without having to share the raw data. This paper introduces a scheme for scheduling on-board FL for constellations connected with intra-orbit inter-satellite links. The proposed scheme utilizes the predictable visibility pattern between satellites and ground station (GS), both at the individual satellite level and cumulatively within the entire orbit, to mitigate intermittent connectivity and best use of available time. To this end, two distinct schedulers are employed: one for coordinating the FL procedures among orbits, and the other for controlling those within each orbit. These two schedulers cooperatively determine the appropriate time to perform global updates in GS and then allocate suitable duration to satellites within each orbit for local training, proportional to usable time until next global update. This scheme leads to improved test accuracy within a shorter time.
Paper Structure (11 sections, 12 equations, 6 figures, 1 algorithm)

This paper contains 11 sections, 12 equations, 6 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model
  3. Constellation Configuration
  4. Cumulative Visibility Pattern
  5. Communication Model
  6. Learning Model
  7. Scheduling scheme
  8. Global Update (GU) Scheduler
  9. Cluster Update (CU) Scheduler
  10. Performance Evaluation
  11. Conclusion

Figures (6)

  • Figure 1: A satellite constellation with $P=5$ orbital planes and $K=40$ satellites. The number of equidistant satellites in all orbits is $K_p=8$, $\forall p \in \mathcal{P} = \{ 1, \dots, 5 \}$.
  • Figure 2: GS located in Bremen, Germany
  • Figure 3: GS located in São Paulo, Brazil
  • Figure 5: The time instant for the $n$-th global iteration, $t_{n}$, is calculated by the gu scheduler and along with the global model parameters, $\bm{w}^{n-1}$, are transmitted to the cu schedulers.
  • Figure 6: Comparison of the learning duration between the scheduled and unscheduled schemes in the Walker Delta constellation with five orbits, each containing eight satellites as considered in \ref{['fig: Bremen']}.
  • ...and 1 more figures