Modulation Classification Based On Machine Learning

In recent years,modulation classification has attracted more and more attention.The classification of the signal modulation method mainly refers to the determination of the modulation method in the case of unknown modulation information.The recognition of modulation methods is of great significance not only in the electronic warfare in the military field,but also in the civilian field.However,how to accurately identify the modulation mode of the signal in a complex and variable channel environment has always been a problem in the communication industry.At present,the successful application of machine learning in many fields makes the recognition of the modulation method based on machine learning widely studied.Especially,the deep learning has a significant advantage in the classification problem,making the application of deep learning in modulation recognition a hot research topic.Firstly,this paper introduces the basic knowledge of the basic structure of modulation recognition and feature extraction.Then,we analyze the application of traditional machine learning algorithms in modulation classification,including the Decision Tree(DT),K Nearest Neighbor(KNN),Support Vector Machine(SVM),and Neural Network(NN),and compares them.In the case of extracting the same features,the performance of the algorithm,experimental simulation and analysis were performed.After that,AlexNet neural network algorithm based on grayscale constellation is further proposed to achieve better performance of the modulation classification.Different from feature-based machine learning algorithms,the way of neural network recognition based on constellation does not require the operation of feature selection,which not only makes the algorithm more efficient,but also can avoid the information omission caused by feature selection and preserve the original signal information.In addition,in order to overcome the problem of partial information loss caused by the coincidence of constellation points on traditional constellation maps,we propose a grayscale constellation diagram,divide the constellation into multiple cells,and count the constellation points that fall into each cell.The number of points represents the gray level,and then this paper analyzes the effect of choosing different numbers of grids on the accuracy.Grayscale constellation can effectively preserve the original signal information to improve the accuracy of modulation recognition.Finally,a large number of signal samples are obtained after signal modeling,channel modeling,and signal sampling.Then,the collected signal samples are preprocessed,and the grayscale constellations are plotted according to the density to obtain experimental samples.Finally,an AlexNet neural network classifier was established and AlexNet network training was conducted using experimental samples.Experimental results show that our proposed AlexNet neural network algorithm based on grayscale constellation map achieves good accuracy and robustness.