Deep belief network for automated modulation classification in cognitive radio

In this paper, we propose low-complexity binarized deep belief network (DBN) based deep learning approach along with noise resilient spectral correlation function as a feature characterization mechanism for automated modulation classification (AMC). Through simulation results, we have shown the detection accuracy of the proposed method is above than 90% when the channel SNR ≥ 0 dB and classification accuracy remains more than 85% for all the considered modulation schemes in multi-path fading channels with SNR 0 dB. Furthermore, as shown in our simulation results, the performance of our proposed binarized DBN based method is comparable to the regular DBN for SCF pattern based AMC. We consider 4FSK, 16QAM, BPSK, QPSK and OFDM with BPSK subcarrier modulations in our simulations.