Distributed Deep Learning Approach for Seamless Handover Management in Non-Terrestrial Networks

Satellite communications play a crucial role in extending next-generation technologies to areas beyond the reach of terrestrial networks. The deployment of Low Earth Orbit satellite constellations aims to provide uninterrupted Internet connectivity to users on the ground, regardless of location or time. However, integrating Non-Terrestrial Networks (NTN) with existing terrestrial systems presents significant challenges, particularly in managing handover strategies due to the rapid movement of satellites. In this study, we propose a Distributed Multi-Agent Deep Q-learning method to optimize handover decisions and dynamically select satellites, ensuring continuous connectivity and efficient resource management. Our solution employs an adaptive resource allocation strategy that adjusts based on user demand and network conditions, maintaining consistent Quality of service levels. We validate and evaluate this approach using a satellite network simulator, demonstrating its robustness and effectiveness. Through comprehensive sensitivity analysis and comparisons with alternative methodologies, our approach is shown to significantly reduce both the number of handovers and the blocking rate. Moreover, by dynamically adapting to fluctuating user demands, our method enhances resource utilization, improving the overall performance of NTN communications and ensuring seamless service continuity.