Transfer Learning Method for Convolutional Neural Network in Automatic Modulation Classification

Abstract

Automatic modulation classification (AMC) plays an important role in many fields to identify the modulation type of signals, in which the deep learning methods have shown attractive potential development. In our research, we introduce convolutional neural network (CNN) to recognize the modulation of the input signal. We used real signal data generated by instruments as dataset for training and testing. Based on analysis of the unstable training problem of CNN for weak signals recognition with low SNR, a transfer learning method is proposed. Experiments results show that the proposed transfer learning method can locate better initial values for CNN training and converge to a good result. According to the recognition accuracy performance analysis, The CNN with the proposed transfer learning method has higher average classification accuracy and is more compatible for unstable training problem.