Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OptiVote: Non-Coherent FSO Over-the-Air Majority Vote for Communication-Efficient Distributed Federated Learning in Space Data Centers

Anbang Zhang, Chenyuan Feng, Wai Ho Mow, Jia Ye, Shuaishuai Guo, Geyong Min, Tony Q. S. Quek

TL;DR

OptiVote addresses the challenge of training distributed models across space data centers by enabling robust federated learning over inter-satellite free-space optical links. It introduces a non-coherent AirComp framework that aggregates sign-based gradients using majority voting and pulse-position modulation, eliminating the need for phase synchronization and instantaneous CSI. A CSI-free, importance-aware dynamic power control scheme mitigates aggregation bias due to heterogeneous channels, while a theoretical analysis provides error bounds and convergence guarantees for non-convex objectives. Simulations on IID and Non-IID data demonstrate faster convergence and higher accuracy than baselines, highlighting OptiVote's potential for scalable, resilient in-orbit intelligence under stringent communication constraints.

Abstract

The rapid deployment of mega-constellations is driving the long-term vision of space data centers (SDCs), where interconnected satellites form in-orbit distributed computing and learning infrastructures. Enabling distributed federated learning in such systems is challenging because iterative training requires frequent aggregation over inter-satellite links that are bandwidth- and energy-constrained, and the link conditions can be highly dynamic. In this work, we exploit over-the-air computation (AirComp) as an in-network aggregation primitive. However, conventional coherent AirComp relies on stringent phase alignment, which is difficult to maintain in space environments due to satellite jitter and Doppler effects. To overcome this limitation, we propose OptiVote, a robust and communication-efficient non-coherent free-space optical (FSO) AirComp framework for federated learning toward Space Data Centers. OptiVote integrates sign stochastic gradient descent (signSGD) with a majority-vote (MV) aggregation principle and pulse-position modulation (PPM), where each satellite conveys local gradient signs by activating orthogonal PPM time slots. The aggregation node performs MV detection via non-coherent energy accumulation, transforming phase-sensitive field superposition into phase-agnostic optical intensity combining, thereby eliminating the need for precise phase synchronization and improving resilience under dynamic impairments. To mitigate aggregation bias induced by heterogeneous FSO channels, we further develop an importance-aware, channel state information (CSI)-free dynamic power control scheme that balances received energies without additional signaling. We provide theoretical analysis by characterizing the aggregate error probability under statistical FSO channels and establishing convergence guarantees for non-convex objectives.

OptiVote: Non-Coherent FSO Over-the-Air Majority Vote for Communication-Efficient Distributed Federated Learning in Space Data Centers

TL;DR

OptiVote addresses the challenge of training distributed models across space data centers by enabling robust federated learning over inter-satellite free-space optical links. It introduces a non-coherent AirComp framework that aggregates sign-based gradients using majority voting and pulse-position modulation, eliminating the need for phase synchronization and instantaneous CSI. A CSI-free, importance-aware dynamic power control scheme mitigates aggregation bias due to heterogeneous channels, while a theoretical analysis provides error bounds and convergence guarantees for non-convex objectives. Simulations on IID and Non-IID data demonstrate faster convergence and higher accuracy than baselines, highlighting OptiVote's potential for scalable, resilient in-orbit intelligence under stringent communication constraints.

Abstract

The rapid deployment of mega-constellations is driving the long-term vision of space data centers (SDCs), where interconnected satellites form in-orbit distributed computing and learning infrastructures. Enabling distributed federated learning in such systems is challenging because iterative training requires frequent aggregation over inter-satellite links that are bandwidth- and energy-constrained, and the link conditions can be highly dynamic. In this work, we exploit over-the-air computation (AirComp) as an in-network aggregation primitive. However, conventional coherent AirComp relies on stringent phase alignment, which is difficult to maintain in space environments due to satellite jitter and Doppler effects. To overcome this limitation, we propose OptiVote, a robust and communication-efficient non-coherent free-space optical (FSO) AirComp framework for federated learning toward Space Data Centers. OptiVote integrates sign stochastic gradient descent (signSGD) with a majority-vote (MV) aggregation principle and pulse-position modulation (PPM), where each satellite conveys local gradient signs by activating orthogonal PPM time slots. The aggregation node performs MV detection via non-coherent energy accumulation, transforming phase-sensitive field superposition into phase-agnostic optical intensity combining, thereby eliminating the need for precise phase synchronization and improving resilience under dynamic impairments. To mitigate aggregation bias induced by heterogeneous FSO channels, we further develop an importance-aware, channel state information (CSI)-free dynamic power control scheme that balances received energies without additional signaling. We provide theoretical analysis by characterizing the aggregate error probability under statistical FSO channels and establishing convergence guarantees for non-convex objectives.
Paper Structure (21 sections, 57 equations, 5 figures)

This paper contains 21 sections, 57 equations, 5 figures.

Table of Contents

  1. Introduction
  2. Minority follows Majority: AirComp-based Distributed Strategy
  3. Distributed Federated Learning Model
  4. SignSGD With Majority Vote
  5. Free-space Optics Link Integrated Aircomp
  6. OptiVote: Phase-Less Signal Aggregation and Importance-Aware Power Design
  7. Transmitter Design - Importance-Aware PPM-MV
  8. Receiver Design - Non-coherent Energy Detection
  9. Error Probability Analysis and Convergence Rate Performance
  10. Reliable Assumptions on Theoretical Analysis
  11. Received Signal Power of MV
  12. Bit Error Probability Analysis
  13. Convergence Rate over Fading Channel
  14. Simulation Results
  15. Experimental Setup and Dataset
  16. ...and 6 more sections

Figures (5)

  • Figure 1: Transceiver design on distributed Learning via non-coherent over-the-air computation based on importance-aware majority vote strategy.
  • Figure 2: Transceiver design on distributed Learning via non-coherent over-the-air computation based on importance-aware majority vote strategy.
  • Figure 3: IID versus non-IID data considered for the detailed numerical analyses. Satellite nodes are randomly distributed inside the sphere, ranging $500\sim2000$ KM. (a): All space nodes have the same data samples for 10 different digits on their locations. (b): The available digits at the space nodes change based on their locations in space.
  • Figure 4: MNIST test accuracy versus communication rounds in the considered AS-space-nodes FL network over inter-satellite FSO uplinks. We compare FedAvg-AirComp, OBMA without TCI, OBDA with TCI, OptiVote, and FSK-MV under (a) IID and (b) Non-IID data partitions.
  • Figure 5: MNIST training loss versus communication rounds under the same experimental setting as Fig. 4. We report the convergence behavior of FedAvg-AirComp, OBMA without TCI, OBDA with TCI, OptiVote, and FSK-MV for (a) IID and (b) Non-IID data partitions, illustrating both the convergence speed and the stability of the learning process over inter-satellite FSO aggregation.

Theorems & Definitions (4)

  • Proof 1
  • Proof 2
  • Proof 3
  • Proof 4