Automatic Modulation Recognition Of Communication Signals Based On Deep Learning

With the rapid upgrading of communication technology,the modulation types of communication signals are becoming more and more abundant.In non-cooperative communication system,the communication receiver cannot know many parameters of the signal in advance to obtain its modulation mode,so Automatic Modulation Recognition technology(AMR)is required to detect the modulation type of the signals.AMR technology is of great significance in cognitive radio,electronic countermeasure,spectrum monitoring,etc.In recent years,,deep learning has been widely used.And this thesis proposes two AMR methods of communication signals based on deep learning.The main research contents of this thesis are as follows:Most of the existing AMR techniques only use single modal of the signals and fail to complement the different feature information of the signal.Thus,an AMR algorithm using multiple modes of signals is proposed,which aims to fuse the multimodal information and make the network learn the joint features.First,the in-phase and quadrature components of the original signals are converted to the expression of amplitude and phase.Second,this thesis studies the time-frequency analysis methods of signals,compares the differences of spectrogram images in different methods,and determines the time-frequency analysis scheme,so as to extract the time-frequency features of the signals.Finally,according to the extracted features,a matching parallel network is designed for feature learning,and the extracted deep features are fused to identify the modulation type of the signals.The experimental results suggest that the proposed method achieves complementary gains of multimodal information,and effectively improve the signal recognition rate.In the current constellation-based AMR tasks,high-order modulation types are not easy to distinguish under low signal-to-noise ratio.In this thesis,an AMR technology based on improved Transformer is proposed,which learns the multiple modal features of signals in association and further enhances the fusion ability of modal features.Firstly,a constellation smoothing filtering algorithm is presented,which maps the modulation signals used in 5G communication standard directly to the constellation matrix,and combines the amplitude and phase characteristics of signals as inputs to the network.Then an improved Transformer network is built to learn the two features interactively.The results indicate that the model has super feature fusion ability and can improve the recognition accuracy of high-order modulation types under low signal-tonoise ratio,which supplys an effective method for Automatic Modulation Recognition of communication signals.