Deep Learning for Beam Hopping in Multibeam Satellite Systems

Data-driven approaches, e.g., deep learning (DL),have been widely studied in terrestrial wireless communications fields, proving the benefits and potentials of such techniques. In comparison, DL for satellite networks is studied to a limited extent in the literature. In this paper, we develop a DL assisted approach to facilitate efficient beam hopping (BH) in multibeam satellite systems. BH is adopted to provide a high level of flexibility to manage irregular and time variant traffic requests in the satellite coverage area. Conventional iterative optimization approaches and typical data-driven techniques may have their respective limitations in achieving timely and satisfactory performance. We herein explore a combined learning-and-optimization approach to provide a fast, feasible, and near-optimal solution for BH scheduling. Numerical study shows that in the proposed solution, the learning component is able to largely accelerate the procedure of BH pattern selection and allocation, while the optimization component can guarantee the solution's feasibility and improve the overall performance.