Description

Objective: Design and development of software for network management in optical satellite constellation networks with machine learning techniques. The performance shall be benchmarked against solutions with classical network optimisation techniques.Targeted Improvements: 50% improvement in capacity utilisation while delivering guaranteed quality of service (QoS). Description:Satellite constellation networks with optical ISLs bring about the possibility of high-throughput and low-latency global space networks. However, management of such networks has been identified to be a major problem due to high number of satellites and high target optical transmission rates. The heterogeneous statistical nature of the input traffic (i.e., quasi constant bit rate vs burst mode) may justify the presence of both circuit and packet switching on different wavelengths in such networks. Circuit switching is not limited to wavelength switching but can also be implemented by multiplexing/demultiplexing transport network circuits (e.g., OTN circuits) at different transmission rates of the hierarchy supported by the corresponding standard. Especially in multi-layer constellations with Non-GEO and GEO satellites, the first problem is building the network topology by establishing ISLs among satellites. This is followed bydeciding on end-to-end routes for traffic streams with different QoS classes, assigning wavelengths for single-hop and multi-hop connections, and establishing transport network circuits at suitable and valid transmission rates on each link. There is rich literature on design and optimisation of networks in the terrestrial domain. However, the problem is complicated in satellite constellation networks, or hybrid terrestrial-satellite networks, due to several reasons. First, the input traffic matrix to such a satellite constellation network is global and time-varying. In addition, the network topology is large and time-varying, albeit deterministic, dueto satellite mobility. Finally, the limited resources (space, weight, and power) on board the satellite limit the number and the configuration of wavelengths that can be assigned to each ISL/GSL/SGL and transport network circuit transmission rates that can be activated in each wavelength. Solutions from terrestrial fibre-optic and 5G networks must be tailored for optical satellite constellation networks. In the open literature, machine learning techniques have been considered in the past for terrestrial network design andoptimisation. In addition, machine learning techniques in network management exist that monitor network performance and input traffic characteristics before dynamically reconfiguring routes, wavelengths, and circuits in a closed-loop fashion. In this activity, network management software shall be developed for optimum topology, routing, wavelength, and circuit assignment in optical satellite constellations. The activity shall develop a benchmark solution together with improvements using machine learning. The resulting software is aimed as reference implementation for future pre-developments for real deployments.