In this letter, a dynamic channel reservation (DCR) strategy based on deep Q network (DQN) is proposed for multi-service low earth orbit (LEO) satellite communication system. We develop a novel modeling method to represent DCR problem of multiple services as a reinforcement learning (RL) task. Based on this model, we calculate the influence of current channel allocation results on future environment and take it as one of the factors for channel allocation decision. Moreover, a corresponding neural network is designed as a decision evaluator to provide an end to end mapping of decision to its value, which effectively avoids the influence of artificial preconditions. Simulation results show that the proposed strategy can improve the overall quality of service (QOS) of the system.

中文翻译：

---

多业务LEO卫星通信系统中基于DQN算法的动态信道预留策略

在这封信中，提出了一种基于深度Q网络（DQN）的动态信道预留（DCR）策略，用于多业务低地球轨道（LEO）卫星通信系统。我们开发了一种新颖的建模方法，可以将多种服务的DCR问题表示为强化学习（RL）任务。在此模型的基础上，我们计算了当前信道分配结果对未来环境的影响，并将其作为决策信道分配的因素之一。此外，将相应的神经网络设计为决策评估器，以提供决策与其价值的端到端映射，从而有效避免了人工前提的影响。仿真结果表明，该策略可以提高系统的整体服务质量。