Research On High-or-der Modulation Recognition Based On Machine Learning

With the development of wireless communication technology,new wireless communication devices and services are emerging,which brings great convenience to people's daily life and aggravates the complication of electromagnetic environment.Meanwhile,it brings opportunities and challenges to the development of radio monitoring and spectrum management technologies.In order to enhance the perception and analysis ability of the radio monitoring system,this thesis explores machine learning based high-order modulation signal recognition methods in complicate electromagnetic environment.The main contents of this thesis include the following two aspects.1.Modulation recognition based on multi-gene genetic programming with structural risk minimization principleIn this thesis,we propose a genetic programming-based modulation classification method(GPMC).The proposed GPMC uses cumulants as the original discriminant features and consists of two stages,the training stage and the classification stage.In the training stage,multi-gene genetic programming(MGP)-based feature engineering transforms sample estimates of cumulants into highly discriminative MGP-features iteratively,until optimal MGP-features are obtained.At the same time,the structural risk minimization principle(SRMP)is employed to evaluate the classification performance of MGP-features and train the classifier.Moreover,a self-adaptive genetic operation is designed to accelerate the feature engineering process.In the classification stage,the classification decision is made by the trained SRMP classifier using the optimal MGP-features.Through simulation results,we demonstrate that the proposed scheme outperforms other existing cumulant based methods in terms of classification performance and robustness in case of low signal-to-noise ratio and fading channels.2.Grid constellation matrix based modulation recognition using contrastive fully convolutional networkIn view of the fact that manually designed features are unable to characterize various modulation formats completely and differentially,deep learning(DL)theory is applied to modulation recognition to obtain high-dimensional representations.On this base,we propose a contrastive fully convolutional network(CFCN)based modulation classification method using grid constellation matrix(GCM).In this method,GCMs are input into the network,which are transformed from the received signals using low-complexity preprocessing.Moreover,a loss function with contrastive loss is designed to train CFCN,which boosts the discrepancies among different modulations and obtains discriminative representations.Extensive simulations demonstrate that CFCN performs superior classification performance and better robustness to model mismatches with low training time comparing with other recent DL based methods.Finally,we summarize this thesis and present the future work.