Research On Automatic Classification Algorithm Of Communication Modulation Signals

Automatic modulation classification(AMC)which detects the modulation type of communication modulation signals,not only can be used in military scenarios to improve interference,reconnaissance and monitoring,but also has more important applications in civilian scenes with the rapid development of intelligent communication system.In the cognitive radio and software radio systems,automatic modulation classification and coding is an integral part of each.The communication system can choose the modulation mode according channel conditions and system specifications through the application of the two complementary system units so that spectrum utilization and reliability was improved.In this research,we proposed some improvements of the conventional methods and more intelligent automatic AMC methods after obtaining the in-depth analysis of various AMC methods and the summary of the advantages and disadvantages of existing methods under the background of the rapid development of artificial intelligence.The main research and contribution as follows:First of all,we developed some improvements of the traditional method.By using combination of the proposed new features and the selected features and applying the support vector machine in machine learning to classify modulation signals,classification accuracy was improved and the real-time classification is quite good.Secondly,for the traditional methods can not extract the feature automatically,we extracted automatically the feature from the transform domain of signals by the stacked sparse auto encoder and avoided the time-consuming and toilsome design and selection of the feature.And then the softmax classifier was used to recognize modulation type.The proposed method was shown to have a great improvement compared with the traditional algorithm under the condition of low signal noise ratio and the accuracy was higher than that of the traditional method under the condition of high signal noise ratio.At the same time,the robustness and generalization ability of the method are comparatively high.Finally,this former method can only be used to extract the feature from the trans-form domain of signals and thus be restricted by the transform domain so that the method can only obtain modulation types of inter-class.Therefore we proposed a method which used the densely connected convolutional network to directly extract features from the tof modulation signals,and then applied the softmax classifier to classify the ime domain modulation signals.it can get modulation types of intra-class and inter-class.The method has higher generalization ability and robustness and the accuracy of classification is significantly better than the traditional methods and the existing other excellent machine learning methods.