Research On Modulation Recognition Technology Based On Convolution Neural Network And Signal Cyclostationarity

Signal modulation recognition has important application value in both military and civil applications,and it is a hot research direction in the field of communication.The research results of modulation recognition emerge in an endless stream,among which improving the recognition rate under low SNR is a difficult problem.In recent years,with the rise of artificial intelligence,more and more scholars have applied deep learning technology to the field of modulation recognition.Convolutional neural network(CNN)is one of the commonly used models for deep learning.On the other hand,most of the communication signals have cyclostationarity,which can improve the anti noise ability of the algorithm.Therefore,this paper studies the modulation recognition method based on convolutional neural network and signal cyclostationary characteristics in order to improve the recognition rate in low SNR environment.Modulation recognition generally includes signal preprocessing,feature extraction and classification.In the preprocessing part of this paper,cyclic spectrum estimation is used to generate cyclic spectrum feature map,and in the classification part,Support Vector Machine(SVM)classifier is used.This paper mainly improves the algorithm for feature extraction,and the specific work content is as follows:1.A modulation recognition algorithm based on CNN+t-SNE is proposed.In the feature extraction module,CNN is used to extract the cyclic spectrum features,and t-distributed Stochastic Neighbor Embedding is used to reduce the dimension.The CNN+t-SNE feature extractor is designed,and the final extracted features are input to the nonlinear SVM for classification.The experimental simulation shows that the recognition rate is 96% when the SNR is-2d B.Compared with other algorithms,the recognition rate is further improved,which proves the superiority of the algorithm.2.A modulation recognition algorithm based on HOG+CNN+t-SNE is proposed.Use Histogram of Oriented Gradient(HOG)and CNN to extract input features,and use feature cascade method for feature fusion,combined with t-SNE to form HOG+CNN+t-SNE feature extractor,and finally SVM is used for classification.According to the experimental simulation,the algorithm based on HOG+CNN+t-SNE has a certain improvement in the recognition rate under low signal-to-noise ratio compared with the method based on CNN+t-SNE.3.A modulation recognition algorithm based on ResNet+PCA is proposed.The structure of CNN network is improved,and the residual network(ResNet)is designed to extract features.Combined with principal component analysis(PCA),the ResNet+PCA feature extractor is designed.The algorithm has faster calculation speed and better classification performance while maintaining high recognition rate.The algorithms proposed in this paper not only has good signal recognition accuracy,but also has strong robustness in different channel environments.It shows good performance and has practical significance.