摘要
针对自动调制识别(Automatic Modulation Recognition,AMR)技术在复杂电磁环境下部分调制信号易混淆、识别准确率较低的问题,提出一种基于多注意力残差网络和门控循环单元(Gated Recurrent Unit,GRU)的AMR模型。通过数据预处理增强信号的相位特征信息,利用自注意力机制使模型有效提取信号的相位偏移特征;设计了由坐标注意力机制、多尺度卷积和通道注意力机制组成的融合注意力残差模块(Fusion Attention Residual Block,FARB),增强对信号空间特征的关注度,有效提取信号的空间特征;使用GRU提取信号的时序特征,通过结合信号的时空特征,提高调制识别精度;通过全连接层进行调制信号分类。仿真结果表明,在RadioML2016.10b数据集上,提出的模型识别准确率有较大提升,且模型参数量少于大多现有模型。此外,对于其他模型易混淆的16-QAM和64-QAM两种信号,所提模型具有较好的识别能力。
Considering the problems of Automatic Modulation Recognition(AMR)technology in complex electromagnetic environments where some modulated signals are easily confused and the recognition accuracy is low,an AMR model based on multi-attention residual networks and Gated Recurrent Unit(GRU)is proposed.Firstly,signal phase characteristics are enhanced through data preprocessing,enabling the model to effectively extract signal phase offset features using self-attention mechanisms.Secondly,a Fusion Attention Residual Block(FARB)is devised,incorporating coordinate attention mechanisms,multi-scale convolution,and channel attention mechanisms to improve focus on signal spatial features and facilitate their extraction.Subsequently,GRU is used to extract the temporal features of the signal,improving modulation recognition accuracy by combining the signal's spatiotemporal features.Finally,modulation signal classification is performed using fully connected layers.Simulation results show that the proposed model has a significant improvement in recognition accuracy on RadioML2016.10b dataset,and the number of model parameters is smaller than most existing models.Moreover,the proposed model has good recognition ability for 16-QAM and 64-QAM signals that are easily confused by other models.
作者
李鸣皓
解志斌
颜培玉
李思
宋科宁
LI Minghao;XIE Zhibin;YAN Peiyu;LI Si;SONG Kening(Ocean College,Jiangsu University of Science and Technology,Zhenjiang 212003,China;Unit 95829,PLA,Xiaogan 432100,China)
出处
《无线电工程》
2025年第1期36-44,共9页
Radio Engineering
基金
高端外国专家引进计划(G2023014110)
国家自然科学基金(62276117)。
