QoS aware Network Slicing for bandwidth optimisation on 5G NTN systems using Reinforcement Learning

Kurzfassung

Non-Terrestrial Networks (NTNs) are becoming an integral part of 5G and beyond communication systems. The realisation of the diverse use cases expected to be supported in the near future necessitates the deployment of Low Earth Orbit (LEO) satellites equipped with Inter-Satellite Links (ISLs). Hence, ISLs must be capable of handling traffic demands belonging to different Quality of Service (QoS) classes. In this regard, network slicing plays a pivotal role in allocating resources on the satellites. The principal resources on the satellite can be classified as computational, communication and storage resources. In this work, we focus on communication resources, particularly the ISLs across inter- and intra-orbital planes. We propose a Deep Reinforcement Learning algorithm for efficiently allocating bandwidth on the ISLs. Three QoS classes with distinct priorities are considered. The main goal is to distribute the total bandwidth among the slices to enable each slice to accommodate traffic spikes. Results demonstrate that the proposed approach prevents the over-utilisation of resources allocated to each slice, in accordance with the priority of each traffic class.