An Optimal Satellite Antenna Profile Using Reinforcement Learning

This paper addresses a detailed procedure to generate an optimal satellite antenna profile. The goal of antenna profile is to provide a sequence of commands for antenna movements such that the antenna directs as many ground station as possible under some constraints. The main task in generating the antenna profile is to schedule the antenna movements taking account of satellite orbit and attitude at all time points, given a mission trajectory. To generate the antenna profile, it is necessary to transform the direction of antenna from the antenna body frame to the satellite body frame and from the satellite body frame to the earth-centered fixed frame. For an optimal tracking of ground station, we generate a maneuvering sequence of azimuth and elevation angles of the antenna considering the projected beamwidth of the antenna on the ground, the off-pointing boundary, and the pointing errors. An optimal maneuvering sequence is generated by reinforcement learning (RL), which is an optimization search algorithm based on penalties and rewards obtained iteratively as episode increases. Through numerical simulations and with actual satellite data, the effectiveness of using RL is illustrated.