Research On Signal Modulation Recognition Method Based On Deep Learning

Signal modulation recognition refers to the recognition of the modulation mode according to the time domain or frequency domain characteristics of the signal after receiving the unknown modulation signal,which is a key step in signal demodulation and of great significance.However,nowadays the channel environment is more and more complex and the signal modulation methods are increasing gradually.Traditional artificial modulation recognition method because of equipment and personnel requirements are relatively high and has long been abandoned,the method based on likelihood decision theory is complete but complexity is too high,the method based on feature extraction,extraction of the characteristics of good or bad for a greater influence on the accuracy,robustness is poorer,difficult to popularize,so urgently need a kind of efficient and robust alternatives.In recent years,the rapid development of Machine Learning(ML)and the advantages of Deep Learning(DL)in the field of image and speech have led many scholars to propose the application of Deep Learning in the field of signal modulation recognition.In this paper,the deep learning knowledge is used to build the end-to-end network model,and the signal modulation pattern is identified by the input of IQ road sampling signals in the public data set,which is an automatic modulation recognition method without any prior knowledge.Based on the characteristics of data samples,this paper draws lessons from the deep learning Network model commonly used in the image processing field,and proposes a method combining Residual Network(ResNet)and Gate Recurrent Unit(GRU),namely ResNet_GRU.This method can pass the residual network and gating cycle unit respectively to extract the characteristics of spatial and temporal signal to the unknown signal recognition,compared with CNN_LSTM(convolution neural network and the length of the memory network)method,this article will be one of the two-dimensional convolution replaced with one-dimensional convolution,can be in almost the same number of cases,increase the depth of the network.Simulation results show that the proposed method can effectively identify the unknown modulation signal,and when the SNR is greater than 6dB,the accuracy can reach about 85%,and it is better than the existing algorithms,such as the machine learning method based on feature extraction and the CNN_LSTM method.Finally,based on the simulation to calculate the confusion matrix,ResNet_GRU study analyzed the limitations of,found that identification error occurs mainly on the individual modulation mode,so this paper proposes a hierarchical task network model,the different ways of modulation recognition task group,set up the depth of the hierarchical learning neural network,the recognition difficulty of low and high difficulty of grouping to improve recognition accuracy.Simulation results show that when the SNR of this model is greater than 0dB,the recognition accuracy can reach more than 90%.