Research On Satellite Signal Modulation Mode Recognition Technology Based On Deep Learning

Compared with terrestrial mobile communication,satellite communication has the characteristics of long transmission distance and is not limited by terrain,and is used in many scenarios,communication signal modulation mode identification is the basis of signal confirmation,interference identification and interference suppression,and is an important component of electronic countermeasures.By judging the modulation method used by the communication signal,the parameters selected for modulation are estimated.Faced with the complexity of satellite communication channels and the low signal-tonoise ratio caused by transmission paths,how to improve the recognition rate of modulation methods poses severe challenges to research in this field.Aiming at the problem of low recognition rate of commonly used modulation method recognition method based on deep learning,this paper proposes a modulation method recognition method based on dual-channel hybrid network model.Firstly,the signal features are initially extracted through two layers of convolutional layers,and then the spatiotemporal features of the signal are jointly extracted by parallel branches composed of residual network and long short-term memory network,and finally the signal classification is realized by the two-layer deep neural network,and then the modulation mode recognition is completed;In addition,the public dataset is selected and the simulation results of the method are compared and analyzed from the three dimensions of convolution kernel size,network layer number and number of residual blocks,and the results show that when the SNR is greater than-2d B,the average recognition accuracy of the method for 11 different signals can reach 91.56%,which is improved to varying degrees compared with other deep learning methods.Furthermore,in order to verify the applicability of the proposed method to satellite communication signals,the satellite modulation signal dataset is selected,and the simulation results of the proposed method are compared and analyzed from the four dimensions of convolution kernel size,network layer,number of residual blocks and input data length,and the results show that the proposed new method can be successfully applied to the modulation and recognition of satellite signals and has good performance.