Research On Communication Signal Modulation Recognition Technology Based On Deep Learnin

Automatic modulation recognition(AMR)plays an important role in wireless communication systems.In recent years,deep learning technology develops rapidly,and AMR based on deep learning is an important research direction in the field of communication.But nowadays the electromagnetic environment is more and more complex,the existing AMR methods face many challenges.Therefore,this paper carries out the research on communication signal AMR methods based on deep learning.The main work contents are as follows:(1)In order to solve the problem of low recognition accuracy in existing AMR models under low signal-to-noise ratio(SNR)conditions,this paper proposes an AMR model based on adaptive denoising residual neural network(ADRes Net).At the input end of the model,the communication signal is first preprocessed to extract the in-phase,quadrature,amplitude,phase,and frequency components of the complex baseband signal as combined feature inputs to the model.The adaptive denoising module of this model combines the soft threshold function and the improved channel attention mechanism to filter the combined features.Then,the residual neural network(Res Net)and Inception structure are combined to form the Inception-Res Net network to extract the deep feature of the denoised feature map.Finally,the full connection layer is used to achieve modulation recognition.The simulation results show that the proposed model performs high recognition accuracy under the condition of low SNR.(2)In order to solve the problems of complex structure and large parameters in existing AMR models based on deep learning,this paper combines convolutional neural network(CNN)and gated recurrent unit(GRU)to propose a spatiotemporal multichannel convolutional and gated recurrent unit deep neural network(MCGDNN)for AMR.This network utilizes three spatial feature extraction branches that integrate complex convolution and one-dimensional convolution to extract complex and single channel features of in-phase and quadrature(IQ)signals,respectively.The Inception structure and group convolution are used to reduce the parameter quantity.In addition,MCGDNN also integrates GRU to further extract the spatiotemporal features of the signal.The simulation results show that the proposed MCGDNN has good recognition performance while having a lower number of parameters.(3)In order to solve the problem that the deep learning models are difficult to carry out AMR effectively under the condition of a small number of labeled samples,this paper proposes a few-shot learning AMR model based on hybrid attention prototype network(HAPN).This model takes the time-frequency images in the form of Episodes as the input,and then uses the feature extraction module equipped with a hybrid attention mechanism to map the samples with a small number of tags and the samples to be recognized into a unified metric space.The measurement module determines the modulation type of the sample to be recognized by comparing the distance between the sample to be recognized and the prototype of different modulation types.The simulation results show that the proposed model performs high recognition accuracy under the condition of small samples.