Back to search

Georgia Institute of Technology

Learning-Based Load Balancing for Low Earth Orbit Satellite Networks

Abstract

dc:description.abstract

Low Earth Orbit (LEO) satellite constellations—such as Starlink and OneWeb—are reshaping the global connectivity landscape by offering low-latency, high-throughput internet access to remote and underserved regions. These systems are often built around bent-pipe architectures, where satellites do not perform any onboard routing or inter-satellite forwarding. Instead, all user traffic is relayed through ground gateways, regardless of the destination. Many of these satellites also employ transparent payloads, meaning they forward radio signals without demodulating or processing them digitally onboard. While this design simplifies satellite hardware and aligns well with existing deployment practices, it introduces coordination challenges, particularly in assigning cells and gateways under ynamic visibility and mobility constraints. As satellites move rapidly across their orbital paths, visibility relationships between satellites, ground cells, and gateways change continuously. Making real-time decisions about which satellite should serve which ground cell, and through which gateway to route the data, requires accounting for factors such as gateway backhaul constraints, non-uniform user demand, and shifting network topology. Although there is increasing interest in leveraging inter-satellite links (ISLs) for multi-hop connectivity, bent-pipe architectures remain attractive in many settings, including those where minimizing latency or preserving simplicity is a priority. This thesis addresses the coordination problem specifically in the context of bent-pipe LEO satellite networks. It presents a learning-based framework that enables satellites to make localized assignment decisions without requiring global state. The satellite-ground network is modeled as a dynamic heterogeneous graph, where satellites, gateways, and ground cells are treated as distinct node types. A Graph Neural Network (GNN) is trained to support structure-aware inference based on satellite visibility and system constraints. Simulation results show that the proposed framework ensures full ground-cell coverage and significantly improves load balancing compared to heuristic baselines. It increases gateway utilization, adapts to changing network conditions, and remains robust under partial infrastructure failures. By combining structural generalization with role-specific coordination policies, this work contributes a scalable and resilient solution for resource management in the transparent satellite context.

Degree

thesis:*
Level thesis:degree_level
Masters
Department dc:contributor.department
Electrical and Computer Engineering
Grantor dc:publisher
Georgia Institute of Technology
Year dc:date.issued
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Jayarajan, Aruna
Advisor dc:contributor.advisor
  • Sundaresan, Karthikeyan
Committee members dc:contributor.committeemember
  • Durgin, Gregory D.
  • Blough, Douglas M.

Subjects

dc:subject × 6

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/78731

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Jayarajan, Aruna. Learning-Based Load Balancing for Low Earth Orbit Satellite Networks. Masters thesis, Georgia Institute of Technology, 2025. https://hdl.handle.net/1853/78731