Skip to content

Toward Scalable SDN for LEO Mega-Constellations: A Graph Learning Approach

Home / Papers / Toward Scalable SDN for LEO Mega-Constellations: A Graph Learning Approach

Toward Scalable SDN for LEO Mega-Constellations: A Graph Learning Approach

Authors: Sivaram Krishnan, Bassel Al Homssi, Zhouyou Gu, Jihong Park, Sung-Min Oh, Jinho Choi Date: 2026-04-30 Paper ID: openalex:2604.27478

Summary

This paper addresses the scalability challenges of software-defined networking (SDN) in large-scale low Earth orbit (LEO) satellite constellations. The authors introduce a hierarchical architecture that utilizes Graph Koopman Autoencoders (GKAE) to model and forecast complex, nonlinear spatio-temporal network dynamics within simplified linear subspaces. By aggregating these shell-level predictions at a central controller, the framework achieves efficient, coordinated network control. Experimental results on Starlink demonstrate superior spatial compression and temporal forecasting performance with a reduced model footprint compared to existing baselines.

Key Contributions

  • Proposes a hierarchical SDN framework for LEO mega-constellations to mitigate network management bottlenecks.
  • Introduces the Graph Koopman Autoencoder (GKAE) to model constellation spatio-temporal dynamics in a reduced linear subspace.
  • Achieves 42.8% improvement in spatial compression and 10.81% improvement in temporal forecasting accuracy on Starlink constellation simulations.

Open Questions & Future Work

Key Concepts

  • graph-koopman-autoencoder: A GNN-based autoencoder that utilizes Koopman theory to project non-linear spatio-temporal dynamics into a linear subspace for scalable forecasting.

Archivist Review

I approved the Graph Koopman Autoencoder as it represents a sophisticated, reusable approach to dimensionality reduction in temporal graph signals. I also approved the SDN controller placement question, as it highlights a concrete trade-off between control hierarchy and physical hardware limitations. Other candidates were rejected due to overly broad scope or lack of specific research focus.

Approved Concepts

  • Graph Koopman Autoencoder: It provides a novel, reusable mechanism for compressing non-linear, high-dimensional spatio-temporal graph dynamics into a linear subspace for efficient forecasting.

Approved Open Questions

  • SDN Controller Placement Trade-offs: This is a critical architectural bottleneck for scaling SDN to massive, highly dynamic constellations, as it directly impacts control latency and network robustness.

Rejected Candidates

  • [open_question] Heterogeneous Graphs in 6G (heterogeneous-graphs-6g-ntn) - low_impact: The proposal is a broad vision for 6G integration rather than a specific, well-defined research bottleneck.

Metadata & Links

url
https://arxiv.org/abs/2604.27478
paper_id
2604.27478
paper_source
openalex
domain
time-series
tags
time-series-forecastinggraph-neural-networkssdnsatellite-networks
architectures
datasets
concept_slugs
graph-koopman-autoencoder
dataset_slugs
skill
TimeSeriesSkill
created_at
2026-05-03T06:03:14Z