Reinforcement Learning during ISL-assisted Feeder Link Switch-Over in 6G Non-Terrestrial Networks

The transition to the Sixth-Generation (6G) era brings new difficulties and possibilities for Non-Terrestrial Networks (NTNs), notably in terms of seamless, high-speed, and ubiquitous connectivity. In this context, Low-Earth Orbit (LEO) satellite-based NTNs can ensure global connectivity and low latencies. However, the rapid mobility of LEO satellites around the Earth leads to frequent Feeder Link Switch-Over (FLSO) between the LEO satellite and the NTN Gateways (NTN-GWs) on the ground. Inter-Satellite Links (ISLs) between LEO satellites have been introduced during FLSO to reduce the FLSO time when the NTN-GWs are positioned far from each other. In this paper, we propose to exploit Artificial Intelligence, specifically Reinforcement Learning (RL) techniques, to select the best inter-satellite path by dynamically learning from the satellite environment (i.e., state) and choosing the correct action that maximizes the reward in order to increase throughput, reduce latency, and avoid packet losses during the ISL-assisted FLSO procedure. Finally, the convergence analysis of the exploited RL techniques is done in terms of reward and satellite path goodness, whereas the system-level performance is assessed in terms of throughput and latency when delivering enhanced Mobile Broadband (eMBB) services in 6G NTN systems.