Convolutional neural network and multi-feature fusion for automatic modulation classification

Abstract

Automatic modulation classification (AMC) lies at the core of cognitive radio and spectrum sensing. In this Letter, the authors propose a novel convolutional neural network (CNN)-based AMC method with multi-feature fusion. First, the modulation signals are transformed into two image representations of cyclic spectra (CS) and constellation diagram (CD), respectively. Then, a two-branch CNN model is developed, a gradient decent strategy is adopted and a multi-feature fusion technique is exploited to integrate the features learned from CS and CD. The proposed method is computationally efficient, benefited from its simple neural network. Experimental results show that the proposed method can achieve identical or better results with much reduced learned parameters and training time, compared with the state-of-the-art deep learning-based methods.