Research On Automatic Modulation Classification Based On Deep Learning

Automatic Modulation Classification(AMC)is a key step in signal demodulation and is critical for subsequent demodulation work.Due to the variety of modulation methods and the complexity of the communication environment,the modulation mode of receives communication signals is often unknown in a non-cooperative communication system.In order to solve this problem,scholars have proposed a number of AMC algorithms.How to identify the modulation of signals effectively has become an important research topic.In the feature-based automatic modulation recognition algorithm,most of them are realized by artificially designing features and then extracting feature and classifying.Therefore,a large amount of calculation is needed in signal preprocessing,and there is not good robustness.Due to the outstanding performance of deep learning in pattern recognition,two AMC algorithms based on deep learning has been proposed.Aiming at 11 common digital modulation signals,a residual network model is designed to realize the classification of digital modulated signals under Gaussian white noise.The difference between this model and the Convolutional Neural Network(CNN)model is that it can effectively alleviate the problem of gradient disappearance,to accelerate training and convergence to a certain extent.The simulation results show that the residual network with adaptive learning rate can provide better recognition results.When the signal-to-noise ratio is greater than 0 dB,the recognition rate reaches 96%.Compared with other deep learning models,the residual network model has faster convergence speed and higher recognition rate,which proves the effectiveness of the algorithm.For the Orthogonal Frequency Division Multiplexing(OFDM)signals of six different modulation schemes,a CNN model is designed to realize the classification of OFDM signals under multipath fading channels.Firstly,the received OFDM signal is preprocessed,and the complex signal is converted into the pixel matrix of the constellation.The difference of the constellation diagrams of different modulation modes is used as the classification feature,and the classification of the modulation mode is completed by the CNN model.The simulation results show that the proposed convolutional neural network has a recognition rate of 90% when the signal-to-noise ratio is greater than 10 dB,and the recognition rate in different channel environments is above 87%,which proves the robustness of the algorithm.In summary,applying deep learning to AMC is a viable method with better recognition performance and greater applicability.