Attention Mechanism Based ResNeXt Network for Automatic Modulation Classification

Abstract

Automatic modulation classification (AMC) is becoming increasingly important in modern wireless communication. In this paper, we proposed a novel integrative approach for AMC based on feature and deep learning. The time-frequency spectrograms are extracted by short-time Fourier transform (STFT) on the received communication signals, which are used as the inputs of the deep learning (DL) network. The ResNeXt network is designed as the backbone, and two dual attention mechanism modules and customized classification module are incorporated. ResNeXt introduces a new dimension named Cardinality, making ResNeXt own excellent feature extraction ability. The dual attention mechanism module combines the channel attention and spatial attention modules to enhance the salient features and suppress the redundant features. Furthermore, the customized classification header improves the robustness of the classifier. Experimental results on the RadioML2016.10B dataset demonstrate its high accuracy and robust performance compared with other state-of-the-art techniques, surpassing them by 2% to 10% in terms of accuracy.