Communication Signal Modulation Recognition Based On Deep Learning

In modern communication,as a non cooperative party,it is often necessary to correctly identify the modulation mode of the signal when the prior knowledge is almost zero,so as to facilitate the subsequent demodulation and analysis.Therefore,modulation recognition technology plays an indispensable role in spectrum sensing,electronic countermeasures and communication reconnaissance.However,the traditional modulation recognition process requires manual extraction of signal features,which is cumbersome and uncertain.It can not adapt to the increasingly complex communication environment,and also can not meet the efficiency requirements of modern communication.In order to avoid many disadvantages brought by manual processing,more and more scholars begin to use deep learning method to extract features directly from the original data of communication signals,which ensures the real-time and robustness of automatic modulation recognition of communication signals.Compared with several common machine learning methods based on expert features,deep learning method has higher recognition accuracy,stronger generalization ability and higher speed.The main research content of this paper is to use the excellent deep learning model to build a classification discriminator which can quickly and effectively identify the modulation mode of communication signal.Aiming at the recognition of 11 kinds of communication signals in complex communication environment,this paper proposes a deep learning network classifier of communication signal modulation recognition,which combines Dense Convolutional Network(Dense Net)and Residual Neural Network(Res Net).After the classification and comparison of Dense Net and Res Net with different structures,the internal structures of the two networks are determined.By comparing the network recognition performance under different combination modes,the basic model of connecting Dense Net and Res Net in series is established.On this basis,in order to further improve the performance of the overall network model for communication signal modulation recognition,a Long Short-Time Memory(LSTM)structure is added to the back end of the series model,so that the network can extract the time series characteristics of the sample data.At the same time,Convolution Block Attention Module(CBAM)is added into dense connection structure and residual connection structure,which improves the performance of the whole network model for communication signal modulation recognition.By using the new network to recognize 11 kinds of communication signal samples under different signal-to-noise ratios,it is found that the Dense Net + Res Net + LSTM(DRL)model has excellent recognition effect for ten kinds of signals other than WBFM signals when the SNR is above 0d B.Compared with CNN + LSTM network,DRL network has obvious advantages in the recognition performance of QAM16 and QAM64 signals,and the overall recognition rate of the former is about 6.4% higher than that of the latter.All these show that the new network structure proposed in this paper is superior in modulation recognition of communication signals.