| At present,communication technology is developing rapidly,and new modulation methods are emerging one after another.Regardless of whether it is civilian or military,the receiving end needs strong support of modulation recognition technology in the process of signal processing such as demodulation.Therefore,the study of modulation recognition technology is of vital importance.In the traditional modulation recognition algorithm,the calculation complexity is high,and it is often necessary to manually set the threshold.Especially in non-cooperative communication,there is less a priori knowledge and the channel environment is worse,which makes it more difficult to correctly identify the modulation mode.At present,deep learning technology is developing rapidly.In the field of wireless communication,most scholars have gradually applied neural network algorithms to the recognition technology of communication signal modulation methods,and achieved good results.This article will continue to deeply study the application of neural network algorithm in communication signal modulation recognition and the performance effect it can achieve.The main research contents of this article are as follows:Aiming at the complicated problem of traditional preprocessing of communication signals,for communication complex baseband signals,an improved CLDNN(Convolutional,Long short-term memory,Fully connected Deep Neural Networks)model is constructed in this paper In the field of communication signal modulation recognition.The use of complex convolutional layers instead of real convolutional layers achieves end-to-end modulation signal identification.By using public data sets to train the improved CLDNN model,the influence of the number of complex convolutional layers,the number of LSTM layers and the number of convolution kernels on the network recognition performance is studied,so as to determine some parameters when the network performance is better.At the same time,the simulation results show that compared with the recognition performance of other network structures,the recognition accuracy of this network structure is higher,and it is easier to converge to a certain extent,which is better than the average recognition performance of modulation recognition algorithms based on higher-order cumulants,Illustrates the superiority of the improved CLDNN network.During the research,it was found that the neural network is sensitive to noise,and the modulation recognition performance is poor when the signal-to-noise ratio is low.In order to solve this problem,this paper uses Finetune transfer learning and center loss algorithm in the convolutional neural network,and applies the network model to the modulation signal recognition.Through the simulation results,it can be seen that for the modulated signals interfered with by different degrees of noise,within a certain range,the application of Finetune transfer learning algorithm and center loss algorithm improves the recognition accuracy and improves the stability of network training to a certain extent.It shows that the combined use of the two algorithms can effectively improve the recognition performance of modulated signals contaminated by noise. |
