Deep Learning-Based Modulation Classification Leveraging Dual-Type Image

Abstract

In this paper, we present an automatic modulation classification (AMC) algorithm for identifying overlapped signals. The proposed algorithm leverages dual-type images as deep learning input data, which is composed of spectrogram and signal images. The dual-type images have the features of both individual images, such as frequency change over time and information on amplitude and phase. We improve modulation classification performance by reflecting various features of a single signal. The simulation results show that the proposed algorithm has accurate classification performance compared to single-type input images, especially in analog modulation classification.