For 2-step random access of 5G-and-beyond low earth orbit (LEO) satellite communication system, the large coverage of satellite causes the serious collisions on MsgA channel which reduce the access success probability seriously, and introduces the problem of high dimensionality. To maximize the access success probability of the system, we propose a dynamic MsgA channel allocation (MCA) strategy. The strategy is realized by flexibly pre-configuring the channels mapping relationship on MsgA via the perception of historical access information. To mitigate the curse of dimensionality for the large action space, we establish an improved twin delayed deep deterministic policy gradient (TD3) based deep reinforcement learning (DRL) model for the strategy. Simulation result shows that the proposed strategy can increase the access success probability of the system by 39.12%.

Spectrum sharing is a pivotal technology for integrated satellite-maritime network to improve spectrum efficiency. However, the majority of existing spectrum sharing approaches relied on time-frequency multiplexing to circumvent or suppress interference, and often overlooked the problem of incomplete observability and interference present in heterogeneous systems. This limitation hindered overall system throughput improvements. In this paper, an intelligent spectrum sharing strategy for integrated satellite-maritime heterogeneous mobile networks is firstly proposed. This strategy aims to operate co-frequency communication transmissions within the interference constraints, as derived from protocols and standards of International Telecommunication Union (ITU), to maximize the overall system throughput and spectral efficiency. Furthermore, considering the limitation that the satellite cannot fully observe channel statuses, we model the spectrum sharing process as a Partially Observable Markov Decision Process (POMDP). We employ an algorithm architecture that combines Dueling Deep Q-learning Network (DQN) and Double DQN to address this challenge, which enables satellites to efficiently evaluate the value of all channel allocation actions, thereby enhancing decision-making accuracy and algorithm convergence speed. Simulation results validate the superiority of our proposed strategy over compared methods in terms of throughput and stability.

Agile-satellite mission planning is a crucial issue in the construction of satellite constellations. The large scale of remote sensing missions and the high complexity of constraints in agile-satellite mission planning pose challenges in the search for an optimal solution. To tackle the issue, a dynamic destroy deep-reinforcement learning (D3RL) model is designed to facilitate subsequent optimization operations via adaptive destruction to the existing solutions. Specifically, we first perform a clustering and embedding operation to reconstruct tasks into a clustering graph, thereby improving data utilization. Secondly, the D3RL model is established based on graph attention networks (GATs) to enhance the search efficiency for optimal solutions. Moreover, we present two applications of the D3RL model for intensive scenes: the deep-reinforcement learning (DRL) method and the D3RL-based large-neighborhood search method (DRL-LNS). Experimental simulation results illustrate that the D3RL-based approaches outperform the competition in terms of solutions’ quality and computational efficiency, particularly in more challenging large-scale scenarios. DRL-LNS outperforms ALNS with an average scheduling rate improvement of approximately 11% in Area instances. In contrast, the DRL approach performs better in World scenarios, with an average scheduling rate that is around 8% higher than that of ALNS.