| With the integrated design and application of radar,communication and electronic reconnaissance,automatic modulation classification is becoming increasingly significant in various applications,such as spectrum monitoring and cognitive radio,and has entered a stage of remarkable development.Recently,automatic modulation classification technology is confronted with difficulties and challenges such as complex and diverse signal types,low signal-to-noise ratio and poor algorithm robustness,and so forth.Specifically,the main contributions of the thesis are as follows:1.To remedy the flaws that the poor robustness of traditional decision tree algorithm under low SNR,this thesis presents a modulation classification scheme based on signal statistical characteristics and machine learning.Based on the statistical characteristics of signals,an auto-encoder and an XGboost is utilized for feature dimension reduction and modulation classification,respectively.To extract statistical features with high stability in low SNR,this thesis presents a feature selection algorithm based on PSO-XGboost,which greatly shortens the training time of the algorithm.2.To remedy the flaws that the low stability of statistical characteristics in low SNR,this thesis presents a modulation classification scheme based on deep learning.In this algorithm,a Res Ne Xt network is used for feature extraction and modulation classification,which improves the classification performance of convolutional neural network effectively.To improve the accuracy of modulation classification,this thesis presents a modulation classification algorithm combining Res Ne Xt and GRU,which improves the generalization performance of the network.3.To remedy the flaws that the complementarity between different transform domain features,this thesis presents a modulation classification scheme based on multi modal fusion.First,a deep migration learning algorithm is utilized to extract the feature of the signal,and the DCA algorithm is applied to fuse the statistical features and features extracted by Res Ne Xt and GRU.In the latter,a XGboost model is used for modulation classification,and the classification of the signal is realized through the fusion of the decision-making level,finally.It is worth noting that the simulations results reveal the superior performance of the proposed approach compared to the existing algorithms. |
