Data augmentation with conditional GAN for automatic modulation classification

Abstract

Deep learning has great potential for automatic modulation classification (AMC). However, its performance largely hinges upon the availability of sufficient high-quality labeled data. In this paper, we propose data augmentation with conditional generative adversarial network (CGAN) for convolutional neural network (CNN) based AMC, which provides an effective solution to the limited data problem. We present the design of the proposed CGAN based data augmentation method, and validate its performance with a public dataset. The experiment results show that CNN-based modulation classification can greatly benefit from the proposed data augmentation approach with greatly improved accuracy.