Automatic Modulation Classification Using Hybrid Convolutional Neural Network

Abstract

Automatic modulation classification (AMC) plays an essential role in signal demodulation and interference identifica-tion. In this paper, we propose a novel AMC method using the Hybrid Convolutional Neural Network (HCNN), which combines with two different convolutional neural networks (CNNs) jointly using various signal features. In the former CNN, spectral correlation features (SCFs) are generated as network input, to classify FSK and BPSK. In the latter CNN, the Attention-based Densely Convolutional Neural Network (AD-CNN), which is trained using regular constellation images (RCs), is proposed to identify the modulation formats that are hardly recognized by the former CNN, such as QPSK, 16-QAM and 64-QAM. The simulation results demonstrate that HCNN displays superior classification performance than existing AMC methods with lower computational complexity.