摘要
针对传统信号调制识别方法对信号复杂调制方式难以识别的问题,提出一种基于注意力机制的双向LSTM卷积神经网络(BLACN),利用卷积神经网络优秀的特征提取能力,实现对复杂调制方式识别特征的提取。并通过注意力机制和双向LSTM关注信号的关键特征与时序信息来提高信号的识别准确率。仿真实验表明,BLACN在信噪比(SNR)为2 dB时可以达到90%的识别准确率,相比于传统CNN神经网络准确率提高7%,证明提出的方法是有效可行的。
Considering the problem that traditional signal modulation recognition methods are difficult to recognize complex signal modulation methods,a bidirectional LSTM convolutional neural network based on attention mechanism(BLACN)is proposed.Depending on the excellent feature extraction capabilities of the convolution neural network,the extraction of the recognition features of complex modulation methods is realized.The accuracy of signal recognition improves because the attention mechanism and bidirectional LSTM focus on the key features and timing information of signal to improve the accuracy of signal recognition.Simulation experiments show that BLACN can achieve a recognition accuracy of 90%when the SNR is 2 dB,which is 7%higher than traditional CNN neural network,which proves that the method proposed is effective and feasible.
作者
杜志毅
张澄安
徐强
李保国
Du Zhiyi;Zhang Cheng'an;Xu Qiang;Li Baoguo(State Key Laboratory of Complex Electromagnetic Environment Effects on Electronics and Information System,NUDT,Changsha 410073,Hunan,China)
出处
《航天电子对抗》
2021年第5期44-48,共5页
Aerospace Electronic Warfare
基金
湖南省自然科学基金创新群体项目
非合作智能信号处理技术(2019JJ10004)。
关键词
注意力机制
双向LSTM
深度学习
调制识别
attention mechanism
bidirectional LSTM
deep learning
modulation recognition
