Radio Signal Automatic Modulation Classification based on Deep Learning and Expert Features

Automatic modulation classification (AMC) becomes more and more important in the electronic reconnaissance. Recently, lots of researchers focus on deep learning architecture based AMC approach but the recognition rate of WBFM and QAM is less than desirable. In this paper, we proposed a joint AMC model of two expert features and CNN-LSTM networks. Before entering the deep learning network, the un-classified signal is first detected whether WBFM or not by the maximum of zero-center normalization amplitude spectrum density. Then the signal which is not WBFM will be inputted to the CNN-LSTM network, while QAM16 and QAM64 are regarded as the same class here. Finally, Haar-wavelet transform crest searching is used to classify QAM16 and QAM64. Compared with former CNN-LSTM architecture, the results of the experiment and deduction show the average recognition rate of the proposed model is increased by 11.5% at 10 dB SNR.