The detection of primary user signals is essential for optimum utilization of a spectrum by secondary users in cognitive radio (CR). The conventional spectrum sensing schemes have the problem of missed detection/false alarm, which hampers the proper utilization of spectrum. Spectrum sensing through deep learning minimizes the margin of error in the detection of the free spectrum. This research provides an insight into using a deep neural network for spectrum sensing. A deep learning based model, “DLSenseNet”, is proposed, which exploits structural information of received modulated signals for spectrum sensing. The experiments were performed using RadioML2016.10b dataset and the outcome was studied. It was found that “DLSenseNet” provides better spectrum detection than other sensing models.

No abstract

No abstract

Cognitive radio has been proposed to improve spectrum utilization in wireless communication. Spectrum sensing is an essential component of cognitive radio. The traditional methods of spectrum sensing are based on feature extraction of a received signal at a given point. The development in artificial intelligence and deep learning have given an opportunity to improve the accuracy of spectrum sensing by using cooperative spectrum sensing and analyzing the radio scene. This research proposed a hybrid model of convolution and recurrent neural network for spectrum sensing. The research further enhances the accuracy of sensing for low SNR signals through transfer learning. The results of modelling show improvement in spectrum sensing using CNN-RNN compared to other models studied in this field. The complexity of an algorithm is analyzed to show an improvement in the performance of the algorithm.

No abstract