부산대학교 도서관

In multibeam satellites, transmit power is a limited resource shared among beams, and allocating higher power to certain beams may cause more interference with others. Moreover, beam placement is the issue of determining the center position of each beam and potentially leads to interbeam interference depending on the locations of ground nodes. Therefore, in this study, we investigate the joint optimization problem of beam placement and transmit power to maximize the sum spectral efficiency in multibeam low-Earth-orbit satellite communication systems, taking into account the interbeam interference and user distribution. We solve the optimization problem using the gradient ascent method and quadratic transform, and then propose an optimization-based algorithm that iteratively searches for the beam center positions and transmit power levels. To reduce the complexity of this iterative algorithm, we present a deep neural network (DNN) architecture and a training method for approximating optimal solutions, and propose a deep-learning (DL)-based algorithm that quickly infers the optimal values using the pretrained DNN. Simulation results show that the two proposed algorithms have a clear tradeoff in performance between the spectral efficiency and the computation time. In particular, the DL-based algorithm achieves 3% lower spectral efficiency than the optimization-based algorithm; however, the computation time can be significantly reduced. Furthermore, both schemes achieve at least 10% higher spectral efficiency than the benchmark schemes without joint optimization by optimally adjusting both the beam center position and transmit power according to the node distribution and satellite environments.