Deep reinforcement learning has recorded remarkable performance in diverse application areas of artificial intelligence: pattern recognition, robotics, object segmentation, recommendation-system, and gaming. In recent times, the applicability of deep learning to telecommunication technology is gradually attracting a lot of attention, especially in spectrum sensing, a core component in cognitive radio. The traditional approaches to spectrum sensing are heavily prone to noise uncertainty and often rely on either complete or partial prior knowledge of the primary users. An alternative method that can curb the aforementioned problem is deep reinforcement learning, which integrates several layers of neural networks for extracting and learning features automatically from a given data. Hence, we survey and propose a theoretical hypothetic model formulation of deep reinforcement learning as an effective method for creating a cooperative spectrum sensing model that can overcome the limitations of traditional spectrum sensing methods, which are often prone to low sensing precision. Also, the study provides an overview of past, current, and future advances in cognitive radio networks. The discussion herein will be of interest to a wide range of audiences in telecommunication and artificial intelligence.

深度强化学习在人工智能的各种应用领域中表现出了卓越的性能：模式识别，机器人技术，对象分割，推荐系统和游戏。近年来，深度学习在电信技术中的适用性逐渐引起人们的广泛关注，尤其是在频谱感知方面，认知无线电的核心组件。频谱感测的传统方法在很大程度上容易产生噪声不确定性，并且通常依赖于主要用户的全部或部分先验知识。可以解决上述问题的另一种方法是深度强化学习，该方法集成了多层神经网络，用于从给定数据中自动提取和学习特征。因此，我们调查并提出了深度强化学习的理论假设模型公式，将其作为创建协作频谱感知模型的有效方法，该模型可以克服传统频谱感知方法的局限性，而后者往往容易导致感知精度低。此外，该研究还概述了认知无线电网络的过去，现在和将来的发展。本文的讨论将引起电信和人工智能领域的广泛受众的兴趣。