Research On Efficient Deep Neural Network For Digital Modulation Recognition

Modulation recognition is a non-cooperative communication recognition technology between signal detection and demodulation,the main task is to realize intelligent reception,processing and classification of the modulated signals.The accuracy of digital signal recognition is related to many aspects of the national economy and people's livelihood.Modulation recognition can achieve radio spectrum management and intelligent control in the civil field,it also can monitor enemy intelligence and protect national information security in the military field.After years of research,many achievements have been obtained.With the increasing complexity of modulation mode and channel environment,how to accurately identify various modulation modes in low SNR has become an important issue.Therefore,in the face of the challenges to be met in the future,combining with new technology and new methods to improve and optimize the modulation recognition,has always been a research hotspot at home and abroad.In recent years,deep learning has developed rapidly,which has superior big data processing ability and good classification ability.It has been widely used in various fields,such as computer vision,image processing,language recognition,and many breakthrough progresses has been made.For the modulation recognition in the field of wireless communication,the deep learning algorithm also has a preliminary application.In view of the shortcomings of neural network,some optimization algorithms are needed to improve it.This paper mainly studies the efficient deep neural network algorithm for the field of digital modulation recognition,which aims to provide a more effective new method for the field of modulation recognition and make communications more intelligent.Different from the traditional algorithm,this paper uses the deep neural network architecture combined with the feature extraction of modulation signals,and use the feature matrix to train the neural network to build a more intelligent model of digital modulation recognition for wireless communication.This paper mainly includes the following aspects:(1)Data preprocessing.In this paper,12-dimensional feature parameters are extracted from the noisy modulated signals,which not only reduces the dimension of the original data,but also improves the calculation speed,and helps the neural network gets a better training to achieve higher accuracy.(2)Construction of digital modulation recognition model.The deep neural network has the problems of over fitting,easy to fall into local extremum and the number of hidden layer nodes are not fixed.In this paper,particle swarm optimization algorithm and another method are used to optimize the deep neural network.Simulation experiments show that the two optimized deep neural network classification algorithms both obtain the good results when identifying six kinds of commonly used modulation signals.The overall recognition accuracy of the deep neural network optimized by particle swarm optimization is higher than that of the traditional algorithm.The proposed method achieves the accuracy above95%,which is 8% and 8.8% higher than the DNN algorithm and the SVM algorithm respectively(SNR ? 1d B);Compared with traditional methods,the accuracy of the deep neural network optimized by Dropout is also improved,and the time of calculation is reduced.(3)Analysis of experimental parameters.This paper also analyzes the influence of different parameters on the experimental results and gets the suitable parameters for this experiment by simulations.The experimental results effectively prove that the proposed method has better performance in the environment of additive Gaussian white noise channel environment,and provide a more robust and efficient method for the field of digital modulation recognition,which has certain guiding significance and application value.