Signal Modulation Classification Technology Based On Deep Learning And Distributed Framework

Automatic Modulation Classification(AMC)is a technology used to identify the modulation mode of the signal in non-cooperative communication scenarios.Traditional AMC methods consist of likelihood-based(LB)methods and feature-based(FB)methods.Recently,benefiting from the outstanding classification performance of deep learning(DL),various deep neural networks(DNNs)have been introduced into AMC methods.Most current AMC methods are based on local framework(Local AMC)and centralized framework(Cent AMC).In Local AMC,there is only one local device which is used to train model.In Cent AMC,there is one central server and multiple local devices.Each local device uploads its data to the central server for neural network model training.Local AMC may not get ideal results due to insufficient data and finite computational power of single local device.Cent AMC requires a large amount of communication cost and carries a significant risk of privacy leakage during the transmission of data from the local devices to the central server.First of all,this paper chose high-order cumulants as signal features,and Support Vector Machine(SVM)as classifier for the research on ML-based AMC.And then this paper chose Convolutional Neural Network(CNN),Modulation Classification Network(MCNet)and Residual Network(Res Net)as classifers for the research on DL-based AMC.Simulation results show that the classification performance of MCNet,CNN and Res Net is superior to that of SVM under the condition of sufficient training data,which means DL-based AMC performs better than ML-based AMC.Secondly,this paper used CNN,MCNet and ResNet for the research on AMC under local framework and centralized framework.Simulation results show that:1.Under the same framework,the classification performance of Res Net is better than that of CNN and MCNet.2.Under the same network,the classification performance of the centralized framework is much better than that of the local framework.In addition,from the perspective of training cost and data security,decentralized learning is applied in AMC.This paper used CNN,MCNet and Res Net for the research on AMC under decentralized framework.Simulation results show that the performance of ResNet based on decentralized framework is better than that of CNN and MCNet.Hence,through the above theoretical analysis and simulation results,this paper proposed an AMC method based on Res Net under decentralized framework(Res Net-based Decent AMC),which can guarantee classification performance,reduce training cost and protect data privacy.