Research On Modulation Recognition Technology Based On Convolutional Neural Network

Modulation identification is an indispensable technology in software radio system.Modulation recognition technology plays an important role in parameter estimation and demodulation and decoding,so in recent years,many researchers have devoted themselves to the research of Modulation recognition technology.Dedicated to the identification of more types of recognition,simple algorithm.At present,modulation recognition can be divided into two kinds according to feature extraction: one is based on deep learning,the other is based on traditional algorithm.Because the traditional feature parameters need to design the corresponding algorithm to compute the feature of the signal,which increases the complexity of the algorithm.Therefore,the use of deep learning to allow neural networks automatically extract features.With the application of deep learning in modulation recognition,the features of signals are gradually obtained by convolution between network layers.However,there are still some problems to be solved.Therefore,this paper will do further research and analysis based on the choice of network model and the anti-frequency deviation of network model,the main contents are as follows:Firstly,in view of the complexity of traditional feature parameter classifier,this paper presents a modulation recognition technique combining feature parameter with Three-layer Full Connected Neural Network Classifier(TFCNNC).According to the needs of TFCNNC input layer,the corresponding feature parameter values are designed and sent to TFCNNC for classification and recognition.Secondly,for the spectra generated by high order spectra,it is impossible to compute the features contained in the spectra by effective algorithms,and an algorithm combining Deep Sparse Auto Encoder Network(DSAEN)with high order spectra is proposed to identify the modulation mode of signals.The simulation results show that the DSAEN can be trained by inputting the high order spectrum into the DSAEN,and the DSAEN network can be used for feature extraction and signal recognition.At the same time,the network also has a certain ability to resist frequency offset.Finally,for DSAEN signal needs to be processed by high order spectrum,the modulation signal recognition method based on Convolutional Neural Network(CNN)is studied in this paper.The simulation results show that the CNN network has the ability to resist frequency offset,and the CNN network model can reduce the steps of signal processing.DSAEN and CNN training time is long,the identification signal type is few,difficult to integrate engineering application and so on.This paper presents the application of Mobile-Net neural network to modulation recognition.The simulation results show that Mobile-Net network model has the ability to resist frequency offset.The simulation results show that the Mobile-Net network model can identify the modulation modes of multiple signals under certain SNR.