Research On Modulation Classification Method Based On Deep Learning

In the era of information explosion,with the rapid increase of communication data and the increasingly complex communication environment,modulation classification has gradually become an important research topic in communication systems,which aims to detect and analyze the modulation type of signals.However,as the modulation types of signals are becoming more and more diverse,the traditional modulation classification methods can no longer be adapted,so the deep neural network(DNN),which is widely popular in recent years,is introduced,and it has powerful classification functions and autonomous learning ability to accurately extract the characteristic features of signals.Based on this,this paper focuses on analyzing the performance of signal modulation classification based on deep learning,studying the effect of signal modulation classification method under asymmetric multichannel collaboration and the effect of modulation classification method combining two-channel convolutional neural network(CNN)-long and short-term memory network(LSTM)and residual network(ResNet)on signal modulation classification.The main research is as follows:(1)Research on modulation classification methods for asymmetric multi-channel cooperation.In order to make up for the problems that the traditional single-channel network structure has limited extracted features and cannot fully extract the signal modulation type features,resulting in less than ideal classification results,an asymmetric multi-channel structure is first designed,in which the depth of network layers and parameters of each channel are not exactly the same,making the extracted features more diverse and representative;then the features extracted from multiple channels are fused,so that the highly correlated features are clustered together,and the irrelevant non-critical features are weakened,thereby improving the classification accuracy of signal modulation types.(2)Research on modulation classification method based on the cooperation of dualchannel CNN-LSTM and residual network.In order to make up for the shortcomings such as CNN only considers spatial features and LSTM only considers time series information,a two-channel structure is first established,one channel uses CNN and one channel uses LSTM;then the fusion layer is used to fuse the features extracted from the two channels to achieve feature extraction of both time series and spatial information,which makes the captured features more comprehensive.Finally,considering that ResNet is easy to optimize and can improve the accuracy by increasing the network depth,a modulation classification method based on the cooperation of dual-channel CNN-LSTM and ResNet is proposed by combining the above two parts.Finally,according to the above two different modulation classification models,several groups of comparative experiments were carried out,which verified the effectiveness of the asymmetric multi-channel structure and the modulation classification method based on the cooperation of two-channel CNN-LSTM and ResNet.The thesis has45 figures,6 tables and 83 references.