Deep Learning Based Intelligent Spectrum Sensing in Cognitive Radio Networks

Spectrum sensing is pivotal in cognitive radio (CR), a burgeoning technology for optimizing radio spectrum utilization. Traditional spectrum sensing techniques like energy detection, matching filter, and cyclic stationary detection have been proposed, which rely on prior knowledge and models. These techniques suffer from challenging issues such as missed detection and false alarms, which impede the effective utilization of the spectrum. Inaccurate assumptions or limited knowledge can hinder detection. To tackle these challenging issues, we propose a novel deep learning-oriented spectrum sensing (DLoSS) technique and highlight the use of deep neural networks (DNNs) for cooperative spectrum sensing (CSS) model. Specifically, we propose a "DLSpectSenNet," a DLoSS-based model, utilizes structural information from incoming modulated signals for spectrum sensing. Particularly, we combine convolutional neural network (CNN) and long-short-term memory (LSTM) network in series, extracting hidden spatial information and temporal data, respectively. The simulation results using the RadioML2016.10b dataset, show the proposed DLSpectSenNet's improved detection performance, especially under low SNR conditions, surpassing traditional cooperative algorithms. It outperforms previous models, enabling improved spectrum detection.