Research For Modulation Recognition Of Communication Signals Based On Deep Machine Learning

Automatic modulation recognition is a basic technology in the field of wireless communication,and it is also a very critical technology.Modulation recognition has been widely used in the military and civilian fields.With the rapid development and wide application of wireless communication technologies,the environment of wireless channels becomes more and more complex,also the modulation methods of communication signal become more and more complicated and diversified,which brings many new problems and challenges for modulation recognition technology.The existing modulation recognition technology mainly adopts the pattern recognition method.The recognition result of this method is too dependent on the feature extraction by manual method,and the recognition results is poor when the signal to noise ratio is low.Through the investigation about the existing modulation recognition method,this paper studies modulation recognition methods based on deep learning.In this paper,first the modulation signal is preprocessed to project the one-dimensional signal into twodimensional space,then neural network is designed as classifier to perform the recognition of modulation format.This method weakens the dependence of recognition algorithm on feature extraction,and avoids complex feature engineering.This paper designs two types of neural networks: multilayer feed-forward neural network and convolutional neural network,and proposes three recognition method to realize modulation format recognition of typical wireless communication signals: 2ASK,4ASK,2FSK,4FSK,BPSK,QPSK,16 QAM,and 64 QAM.(1)The Hilbert Transform is used to obtain complex analytical expression of the modulation signal,and a variety of statistics of the modulation signal are obtained.A multilayer feed forward neural network is designed to perform the modulation recognition.This method is the simplest,most efficient,and easy to implement.(2)The time-frequency diagram of the modulated signal is obtained by the Short-Time Fourier Transform,and the convolutional neural network is designed to recognize the timefrequency diagram of the modulation signal so as to achieve the effect of recognizing modulation signal.This method requires complex time-frequency analysis,but shows higher accuracy rate of recognition.(3)The asynchronous delay-tap sampling structure is used to sample the modulation signal to obtain a two-dimensional histogram.Then this paper designs convolutional neural network to complete the modulation recognition.The signal preprocessing of this method is very simple,and the recognition rate is high.In conclusion,simulation results show that the modulation recognition method based on deep learning weakened the dependence of recognition result on feature extraction,and compared with the traditional method this method has a higher recognition rate,at very low SNR conditions also has a very good recognition results.