空间注意力机制优化的通信调制样式深度识别方法  

Optimized Communication Signal Modulation Recognition Method with Spatial Attention Mechanism

作　　者：袁中群 陈卫 王成东 梁栋 YUAN Zhong-qun;CHEN Wei;WANG Cheng-dong;LIANG Dong(School of Internet,Anhui University,He'fei 230601,China)

机构地区：[1]安徽大学互联网学院,安徽合肥230601

出　　处：《中国电子科学研究院学报》2023年第12期1086-1093,1118,共9页Journal of China Academy of Electronics and Information Technology

基　　金：天基综合信息系统重点实验室开放基金。

摘　　要：为了提高无线通信系统对调制信号的识别性能,减小信号调制样式识别时间,提出一种基于空间注意力机制优化的通信调制深度识别方法,以空间注意力模块对CNN2网络进行优化,并与CNN1、CNN2、ResNet、CLDNN四种典型调制识别网络作对比。在RML2016.10b公共数据集上,所提方法在0 dB~20 dB信噪比范围内取得了92%的平均识别准确率,相较于CNN2的90%有一定提升。同时,所提方法的训练时间减少了20%,并提升了14%的推理速度,展现出更高的计算效率。实验结果表明,所提方法能进一步提高调制识别准确率,提高训练和识别效率。To enhance the recognition performance of wireless communication systems towards modulation signals and reduce the time required for signal modulation pattern recognition,a communication modulation deep recognition method optimized with a spatial attention mechanism is proposed.This method leverages the spatial attention module to optimize the CNN2 network and is compared with four typical modulation recognition networks:CNN1,CNN2,ResNet,and CLDNN.On the RML2016.10b public dataset,the proposed method achieves an average recognition accuracy of 92%in the signal-to-noise ratio range of 0 dB to 20 dB,showing a slight improvement compared to CNN2's 90%.Additionally,the proposed method reduces training time by 20%and improves inference speed by 14%,demonstrating higher computational efficiency.Experimental results indicate that the proposed method can further enhance modulation recognition accuracy while improving training and inference efficiency.

关 键 词：自动调制识别 空间注意力机制 卷积神经网络 深度学习 

分 类 号：TN911.3[电子电信—通信与信息系统]