基于卷积长短时网络的调制识别技术研究

Research on Modulation Recognition Technology Based on Convolutional Long Short-Term Networks

随着深度学习的发展,更多的研究者将调制方式识别与深度学习网络结合,充分利用了神经网络无需先验信息的特点,促进了自动调制识别(Automatic Modulation Recognition, AMR)技术的发展。但是AMR技术存在低信噪比下泛化能力有限,分类精度不高的问题,提出了一种由卷积神经网络(Convolutional Neural Network, CNN),长短时记忆(Long Short-Term Memory, LSTM)网络和深度神经网络组成的模型-卷积长短时深度神经网络(Convolutional Long Short-Term Deep Neural Networks, CLDNN)。将预处理的数据集通过CNN对信号进行空间特征提取,再通过LSTM模块对数据集进行时间特征提取,最后通过全连接层对数据集分类。实验结果表明,CLDNN模型相对于主流使用的残差网络(Residual Network, ResNet)模型、CNN模型等在性能上有显著的提高,在信噪比为30 dB时保持93.09%的高效识别,在信噪比为-10 dB时,实现54.32%的有效识别。

With the development of deep learning,more researchers have combined modulation recognition with deep learning networks,fully utilizing the characteristic of neural networks that do not require prior information,which has promoted the development of automatic modulation recognition(AMR)technology.However,AMR technology has limited generalization capability under low signal-to-noise ratio(SNR)conditions and is not highly accurate in classification.A model composed of a convolutional neural network(CNN),long short-term memory(LSTM)network,and deep neural network is proposed—convolutional long short-term deep neural networks(CLDNN).The preprocessed dataset is processed through a CNN to extract spatial features from the signals.Then the LSTM module is used to extract temporal features from the dataset,and finally the dataset is classified through a fully connected layer.Experimental results show that the CLDNN model significantly outperforms mainstream models such as the residual network(ResNet)and CNN models in performance.It maintains an efficient recognition rate of 93.09%at an SNR of 30 dB and achieves an effective recognition rate of 54.32%at an SNR of-10 dB.