PAGE CONTENTS
Objectives
Design and development of machine learning algorithms for network management in optical satellite constellation networks. Experimental comparison in simulations against solutions with classical network optimisation techniques and evaluate the performance under varied scenarios with respect to their throughput, latency, resilience and adherence to Quality of Service (QoS) measures.
Benefits
The project supports the management of optical satellite constellations by introducing machine learning–based techniques for the two use cases of selecting ground-satellite feeder links and dynamic routing, with a strong focus on Quality of Service (QoS) support. Instead of optimising solely for throughput, the approach enables more balanced network operation by considering latency, reliability, and service differentiation in dynamic conditions.
A key aspect is the alignment with emerging standards, in particular the ESTOL framework, ensuring that the developed solutions are compatible with future interoperable optical network architectures.
The modular simulation and evaluation environment allows systematic assessment of network behaviour under realistic scenarios, including varying traffic demands and link conditions. This supports informed design decisions and reduces development risks for future constellation deployments.
Features
The product consists of a modular simulation and optimisation framework for optical satellite networks. It includes components for orbital propagation, visibility analysis, topology generation, and network simulation, each implemented as independent modules with well-defined interfaces.
Machine learning algorithms are integrated into the network management layer to optimise routing and resource allocation based on dynamic network conditions. The framework supports the comparison of these approaches against classical optimisation techniques using consistent metrics such as throughput, latency, and jitter.
Intermediate results are stored in standardised formats, enabling reproducibility and efficient reuse across experiments. The system supports both synthetic and externally provided orbit data, as well as detailed modelling of atmospheric effects on optical links.
Challenges
Improve already existing state-of-the art techniques for network optimisation by using machine learning, bringing an improvement on link capacity utilisation of at least 50%, and demonstrated under a testbed capable of simulating a representative scenario running these new algorithms.
System Architecture
- The system is designed as a modular architecture, consisting of independent stages: Orbital propagation, visibility calculation, topology management and the network simulation.
- All stages are independent executables and the results are saved in a defined file format, allowing the re-use of intermediate results in experiments, as well as full reproducibility of all results.
- The orbit propagation module performs a numerical propagation of Keplerian orbits, but also allows external inputs for orbit replaying.
- The visibility stage includes a detailed simulation of the effect of atmospheric conditions and turbulence on the optical connections between ground stations and feeder links.
- The network simulation consists of a flow-based network model on which packet-level statistics are generated stochastically.
- All stages are implemented in Python, but the modular design allows to switch out components in case it becomes necessary for performance reasons.
Plan
The project has the following milestones:
- Definition of use cases and scenarios
- System Requirements Review
- Preliminary Design Review
- Critical Design Review
- Implementation and Verification
- Acceptance
- Final Review
Current Status
The project has passed critical design review on the 16 January 2026 and is now in the implementation phase. A prototype is available and initial experiments are underway.
Related Links
Companies
