Automatic Modulation Classification: A Novel Convolutional Neural Network Based Approach

Abstract

Deep learning (DL) is a new paradigm of machine learning (ML) that has shown exceptional performance in image, voice and natural language processing. However, researchers have not explored the use of DL to wireless communication to its full potential. The use of DL technology for wireless communication applications has recently gained popularity. This paper looks into the application of deep learning based approach for automatic modulation classification (AMC). Automatic modulation classification has a diverse applications ranging from civilian to military. A deep learning based convolutional neural network (CNN) architecture for AMC is proposed in this paper. We make use of Gaussian noise layer after convolution layers in our proposed architecture which has a regularization effect while training and it reduces over fitting problem. We demonstrate using experiments that the proposed architecture outperforms the existing CNN based architectures for AMC. We also demonstrate the effects of different architecture parameters on the performance of the proposed algorithm.