Beam-Hopping Based Resource Allocation for Multi-Beam Satellite Communication Networks

In this paper, a beam resource scheduling algorithm employing deep reinforcement learning is developed, leveraging the decision-making capabilities of reinforcement learning alongside the perceptual strengths of deep learning. The beam allocation problem is conceptualized as a Markov decision process, where the state is defined by the number and delay of data packets, and the action is specified as the beam allocation, with associated rewards defined accordingly. Utilizing a convolutional neural network, the algorithm effectively extracts pertinent features. The primary optimization goals are to minimize the latency of real-time services, maximize the throughput for non-real-time services, and ensure equitable distribution of beam resources. The deep reinforcement learning algorithm refines the dynamic beam scheduling strategy through extensive training involving numerous state matrix inputs. Experimental results underscore that the algorithm not only enhances the fairness of beam allocation but also significantly reduces data packet transmission delays and boosts the overall throughput of the communication system.