Automatic Modulation Classification Based on wavelet Image and Convolutional Neural Network

Abstract

Automatic modulation recognition is an important fundamental technology in the field of wireless communication, and effective modulation identification of target signals is a necessary prerequisite for applications such as software radio and spectrum management. The complexity of modulation patterns and the increasing congestion of spectrum occupation bring great challenges to traditional classification methods. The authors propose an end-to-end signal modulation classification method based on wavelet image and convolutional neural network, which characterizes the intra-pulse signal based on continuous wavelet transform(CWT), and design a convolutional neural network to extract and classify the characteristics of signal wavelet time-frequency images. At the same time, a small sample dataset of 8 signal modulation types is generated, and the effectiveness of the proposed method is verified by combining data enhancement in 10 signal-to-noise ratio environments, and the classification accuracy is about 87% when the signal-to-noise ratio is 8dB.