The proliferation of mobile devices and web-based applications has led to a significant increase in demand for wireless data exchange within the Internet of Things (IoT). Consequently, a growing need to focus on energy efficiency and spectrum management arises, especially with the increasing entry of IoT products into the market. Despite a utilization rate of only 15-85% of the spectrum allocated to licensed users, access to the spectrum remains limited and costly. To address this issue, cognitive radio technology has emerged as a potential solution to reduce spectrum frequency waste and overcome existing limitations. Cognitive radio networks enable the reuse of allocated frequency bands; one widely adopted model is the interweave model, wherein secondary users scan the frequency spectrum before transmitting data. If primary users are not actively utilizing the channel, secondary users are granted access to transmit their data. However, this approach introduces the challenge of potential interference with primary users during the transmission of secondary users. In this context, machine learning and deep learning, both subsets of artificial intelligence, have demonstrated superiority over traditional methods in spectrum detection and primary user detection. The inherent capability of machine learning algorithms and neural networks to automatically and accurately extract hidden information from data renders them more intelligent and flexible, allowing for improved performance with increasing data volumes and diverse user scenarios. This research focuses on employing deep neural structures, configuring relevant network layers, and applying appropriate processing techniques to achieve comparable results to prior studies. The investigation centers on three distinct structures based on machine learning to enable intelligent spectrum detection in cognitive radio networks. The proposed deep neural structure effectively extracts signal energy and cyclic prefix similarities for efficient Orthogonal Frequency Division Multiplexing (OFDM) signal detection without the need for prior knowledge. Furthermore, it enables the extraction of various describable and indescribable features of the OFDM signal. Various metrics, such as the confusion matrix, area under the curve, loss value, Receiver Operating Characteristic (ROC) curve, and detection probability for different signal-to-noise ratios, were employed to eva‎luate the performance of the proposed structures. The results indicated significant performance improvement in the third phase due to the collaboration between weak and robust learners compared to previous methods. In conclusion, this study demonstrates the potential of employing machine learning and deep neural structures for intelligent spectrum detection in cognitive radio networks. The findings highlight the effectiveness of these techniques in mitigating spectrum frequency waste and improving overall network performance, thus contributing to the advancement of wireless communication systems within the IoT paradigm.

مهدي مهدوي

محمدرضا احمدزاده

حسين سعيدي , نغمه سادات مويديان