The integration of cognitive radio-based Internet of Things devices in 5G network environments for smart city applications necessitates effective spectrum management. The critical aspect of spectrum management lies in making appropriate spectrum decisions for selecting idle channels that meet the quality of service requirements of secondary users while avoiding harmful interference with primary users (PUs). This article addresses the challenges by proposing an 8-state-based discrete-time Markov chain model to analyze the busy and idle times of PUs in CRNs. By leveraging this model, expressions for the traffic state and channel state belief vector are derived under imperfect sensing conditions. Additionally, a deep neural network (DNN)-based spectrum decision algorithm is introduced to optimize spectral resource utilization, considering spatial and temporal availability and energy-saving aspects in cognitive packet transmission. Our analytical and numerical evaluations demonstrate the superiority of the DNN-based algorithm over traditional methods, showcasing improved spectral resource utilization.

在智慧城市应用的 5G 网络环境中集成基于认知无线电的物联网设备需要有效的频谱管理。频谱管理的关键在于做出适当的频谱决策，以选择满足次要用户服务质量要求的空闲信道，同时避免对主要用户（PU）的有害干扰。本文通过提出基于 8 状态的离散时间马尔可夫链模型来分析 CRN 中 PU 的繁忙和空闲时间来解决这一挑战。通过利用该模型，在不完善的传感条件下导出了交通状态和信道状态置信向量的表达式。此外，引入基于深度神经网络（DNN）的频谱决策算法来优化频谱资源利用，同时考虑认知数据包传输中的空间和时间可用性以及节能方面。我们的分析和数值评估证明了基于 DNN 的算法相对于传统方法的优越性，展示了光谱资源利用率的提高。