Cooperative Spectrum Sensing: A Multi-Channel Deep Learning Method

In this paper, we studied the cooperative spectrum sensing (CSS) problem based on deep learning. The current CSS methods based on deep learning only extract general features of signal samples, without considering the specific details of real and imaginary parts of the signal samples. In order to fully extract features of the signal samples, we propose a multi-channel deep neural network (MCDNN) for CSS, the proposed method firstly utilizes three neural network branches to separately extract the features from real part, imaginary part, combination of real and imaginary parts of signal samples. Simulation results indicate that the detection performance of this method is better than other detection methods, especially in low signal-to-noise ratio (SNR) situations. Additionally, this method is robust to time-varying non-uniform received signals and noise power.