Research On Digital Modulation Recognition Technology Based On Machine Learning

Automatic Modulation Recognition is the technology that the receiver determines the modulation style of the received signal when the modulation information is unknown in the non cooperative communication system.As a research hotspot in the field of wireless communication,modulation recognition technology plays a key role in signal monitoring,spectrum management,battlefield reconnaissance and electronic countermeasures.With the development of artificial intelligence technology,the application of machine learning algorithm to modulation recognition has become an irreversible trend.Compared with the traditional modulation recognition algorithm,the biggest advantage of machine learning algorithm is its excellent feature extraction ability and classification ability,which greatly reduces the workload and achieves better classification performance.Therefore,in this thesis,the modulation recognition technology of digital signal based on machine learning is studied in depth.The main results are as follows:1.An Evolutionary Neural Network(ENN)with multi-dimensional features as input is proposed.Based on Genetic Algorithm(GA),this method automatically designs the network structure and super parameters,and evolves the classification network with the best modulation recognition performance.The results show that the modulation recognition accuracy of the proposed method is improved compared with that of the traditional method.2.From the perspective of sample enhancement,a sample enhancement method based on Fast Gradient Sign Method(FGSM)is proposed.This method can expand and enhance the sample set,and train the convolutional neural network with more robustness and mobility.The results show that the network fine-tuning using the counter samples generated by this method can effectively alleviate the over fitting problem and improve the accuracy of modulation recognition under low SNR.3.From the perspective of signal preprocessing,a signal preprocessing method of Connected Convolutional Denoising Auto-Encoder(CCDAE)is proposed.This method can reduce the noise of the original signal sample and improve the accuracy of modulation recognition.4.From the perspective of network structure,a Deep Evolutionary Convolutional Neural Network(DECNN)based on Genetic Algorithm is proposed.This method takes the original signal data as input,optimizes the network structure and super parameters based on genetic algorithm,and can automatically design the modulation recognition model with the highest accuracy.5.A recognition method of unknown modulation pattern signal based on deep embedding feature is proposed,which uses cross entropy and triple loss.Based on Pearson correlation coefficient,the feature similarity of unknown signal and known signal is measured,which can recognize unknown modulation type signal on the premise of ensuring the recognition accuracy of known modulation type signal.All of the above work has been verified by experimental simulation,and a set of modulation recognition flow scheme based on machine learning can be constructed to solve the problems such as sample distortion,over fitting,low recognition accuracy and heavy human workload in modulation recognition.