Fine-Grained Modulation Classification Using Multi-Scale Radio Transformer With Dual-Channel Representation
Qinghe Zheng, Penghui Zhao, Hongjun Wang, Abdussalam Elhanashi, Sergio Saponara
Abstract
Automatic modulation classification (AMC) plays a critical role in both civilian and military applications. In this letter, we propose a multi-scale radio transformer (Ms-RaT) with dual-channel representation for fine-grained modulation classification (FMC). In Ms-RaT, a dual-channel representation (DcR) of radio signals is designed to help the model learn discriminative features by converging the multi-modality information, including frequency, amplitude, and phase. During the learning process, multi-scale analysis is introduced into the model to form the tighter decision boundary. Finally, extensive simulation results demonstrate that Ms-RaT achieves superior modulation classification accuracy with similar or lower computational complexity than existing state-of-the-art deep learning methods. Through ablation studies, we also validate the effectiveness of DcR and multi-scale analysis in Ms-RaT.