RESOURCE ALLOCATION OPTIMIZATION IN BEAM-HOPPING MULTI-BEAM LEO SATELLITE SYSTEMS

This Master's thesis considers a LEO satellite communicating with users on Earth through a beam-hopping (BH) scheme. The satellite houses a planar array antenna, with a fixed number of array unities (Nx by Ny). We consider a snapshot of users distributed over the service area and we intend to optimize the available resources to satisfy the overall user demand as effectively as possible. Each user, or group of users, is characterized by a certain value of demanded traffic, or aggregated traffic; for sake of simplicity, we refer to both of the subjects with the words user and traffic. We are going to perform the optimization for several time windows in order to analyze the performances of the optimization without focusing to a specific case. We are calling Di the demanding traffic for user i, Ci the capacity of the link for that user, and Ri the residual bitrate from the previous time windows. All these three factors are combined in the cost function, whose primary objective is to serve all the users in the shortest possible time. The resources we have to optimize are the available bandwidth B, power P, and the antenna unities in the antenna array: we need to assign a proper fraction of them to each user (Bi, Pi and Ni). In the first sections of the work is presented the physical explanation of the satellite and communication elements that are involved, then follows their mathematical modeling. Another relevant preliminary part regards the modeling of the users in the space, their needs and their position with respect to the satellite: each user suffers different interfering conditions, can experience a different communication delay and can be served for a different amount of time. When all the elements required for our analysis are presented, we integrate them together in order to obtain the final mathematical formulation of the problem. The resulting optimization problem is an NP-hard (non-deterministic polynomial time) mixed-integer nonlinear program (MINLP). We therefore employ SCIP, which is suitable for this class of optimization problems. However, solving large instances of the original formulation is computationally prohibitive. We consequently introduce a series of heuristic steps to reduce the problem size to one that can be handled efficiently by the solver. The differences between the optimum result and the heuristic is then discussed.