Scholarlink

<p>With the evolution of 5G-Advanced and 6G technologies, wireless communication environments are becoming increasingly complex, and automatic modulation classification (AMC) has become a key technology to enhance spectral efficiency and guarantee communication security. Traditional methods are limited by channel model dependency and insufficient manual feature design, while existing deep learning models still have limitations in feature fusion and timing modeling. To this end, this paper proposes a multimodal feature fusion model MMF-GNN based on graph neural network, which extracts the multidomain features of the signal through time-frequency modal branching, time-sequence dynamic branching, and spatial modal branching, and realizes cross-modal feature interactions with the heterogeneous graph fusion module. Experiments on the RML2016.10a dataset show that the MMF-GNN achieves an average classification accuracy of 63.26% at all signal-to-noise ratios, which significantly outperforms comparative models such as MCLDNN and AMC-Net. The ablation experiments validate the effectiveness of the branches, with the heterogeneous graph fusion module contributing the most. MMF-GNN performs well in high-order modulation and low signal-to-noise ratio scenarios. This study provides an efficient multimodal fusion framework for modulation classification in complex electromagnetic environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automatic Modulation Classification Based on Efficient Multimodal Feature Fusion

  • 期刊论文

  • 发表日期：

    2025年8月1日

查看全文
  • Yu Lin,
  • Qinghua Chen,
  • Feng Wang,
  • Qiang Lu,
  • Zhaoxuan Zhang

摘要

With the evolution of 5G-Advanced and 6G technologies, wireless communication environments are becoming increasingly complex, and automatic modulation classification (AMC) has become a key technology to enhance spectral efficiency and guarantee communication security. Traditional methods are limited by channel model dependency and insufficient manual feature design, while existing deep learning models still have limitations in feature fusion and timing modeling. To this end, this paper proposes a multimodal feature fusion model MMF-GNN based on graph neural network, which extracts the multidomain features of the signal through time-frequency modal branching, time-sequence dynamic branching, and spatial modal branching, and realizes cross-modal feature interactions with the heterogeneous graph fusion module. Experiments on the RML2016.10a dataset show that the MMF-GNN achieves an average classification accuracy of 63.26% at all signal-to-noise ratios, which significantly outperforms comparative models such as MCLDNN and AMC-Net. The ablation experiments validate the effectiveness of the branches, with the heterogeneous graph fusion module contributing the most. MMF-GNN performs well in high-order modulation and low signal-to-noise ratio scenarios. This study provides an efficient multimodal fusion framework for modulation classification in complex electromagnetic environments.