Modulation Recognition Method of Communication Signal Based on CNN-LSTM Dual Channel

Considering the problem of low recognition rate and a large number of parameters in end-to-end modulation recognition methods based on deep learning, a CNN-LSTM dual channel(DCNN-LSTM) modulation recognition method based on the combination of Convolutional Neural Network (CNN) and Long Short-Term Memory(LSTM) network is proposed in this paper. Firstly, the proposed method extracts the frequency domain features of the IQ signal using one-dimensional convolution in the CNN channel and the time-domain features of the transformed AP data of the IQ signal in the LSTM channel. Then, fusing the features of the two channels by the Concatenate operation and convolution layer. Finally, the method uses Global Average Pooling (GAP) instead of Flatten, which integrates the feature information of the vectors without additional parameters. Experimentally, the number of parameters of the proposed method is respectively reduced by 98.4%, 77.8%, and 88.9% compared with CNN, LSTM, and SCRNN, while the recognition rate is respectively improved by 9.7%, 3.3%, and 0.4%. The method proposed in this paper has theoretical reference value for the research field of communication modulation signal recognition, and engineering reference significance for the study of the intelligent classification of spatial signals in complex environments.