CNN Based Wideband Spectrum Occupancy Status Identification for Cognitive Radios

Artificial intelligence has been viewed as a promising solution for cognitive radio (CR) in recent years. In this paper, we address the task of cooperative wideband spectrum sensing (CWSS) on the basis of machine learning (ML) techniques in cognitive radio networks (CRN). Specifically, a classification method based on deep learning is proposed for wideband spectrum occupancy status identification in the CRN. The proposed method comprises four phases: data collecting, data preprocessing, training, and wideband spectrum occupancy status identification. In this regard, the energy matrix measured at multiple secondary users (SUs) in all subbands is treated as a feature matrix and fed into the convolutional neural network (CNN), serving as a multiple-category classifier, to determine the spectrum occupancy pattern. It has been confirmed in simulations that the CNN can achieve fast detection while maintaining satisfactory performance compared with some of the traditional CWSS methods and typical ML-based CWSS algorithms.