A Machine Learning approach for routing in satellite Mega-Constellations

Current packet-switched networks, and Internet in particular, are experiencing explosive growth in traffic, mainly due to the rapid development of new communication technologies and the associated applications. Existing network routing policies might not be sophisticated enough to cope with the ever-changing network conditions resulting from huge traffic growth, opening to possible inefficiencies. Deep learning, as a recent turning point in the area of Machine Learning, has received significant research attention in numerous areas of Artificial Intelligence, such as Computer Vision, Natural Language Processing, Autonomous Driving, Robotics Process Automation and so forth. Machine learning applications in network-related areas is relatively recent but it appears to represent an innovative approach to conFigure and manage networks in a more intelligent, efficient and autonomous way. On the other hand, we are witnessing as well the huge investments on satellite-based networks for global broadband Internet access, called mega-constellations, leveraging thousands of satellites in LEO orbit and expected to process Tbit/s. The aim of this paper is to describe the adoption of an innovative Machine Learning-driven and Data-driven approach applied to the routing of packets in such future generation of satellite mega constellations, enhancing services and applications QoS. In this work, we demonstrated the effectiveness of a Deep Learning based routing approach in comparison with the commonly used network routing approach of the Shortest Path, in a preliminary performance evaluation with few satellite nodes. The proposed approach can be applied to mega constellation satellite networks such as Starlink by SpaceX and conceptually applicable also to other (also terrestrial-only) networks.