Spectrum Sensing for Cognitive Radio Based on Feature Extraction and Deep Learning

Abstract

In cognitive radio, spectrum sensing is used to determine whether the primary user is using the spectrum based on the signal received on a specific frequency band, thereby determining whether the secondary user can use the spectrum. The main problem faced by spectrum sensing is how to identify the existence of the primary signal under the condition of low signal-to-noise ratio (SNR). Compared with traditional technologies, deep learning methods can identify the features of input data more efficiently and accurately. Based on convolutional neural network (CNN), This paper regard spectrum sensing as a binary classification problem. In the method we proposed, different features of received are extracted, and a dataset of feature matrices obtained under different SNRs is constructed for the training of the CNN network. Experiment results show that under the condition of low signal-to-noise ratio, the performance of our method is improved compared with the traditional method, and the combination of different features can improve the sensing accuracy.