一种基于SVD降噪和CNN分类的无线信号调制识别方法

A Wireless Signal Modulation Recognition Method Based on SVD Noise Reduction and CNN Classification

针对低信噪比下噪声干扰导致的调制信号识别精度不足的问题,提出了一种基于奇异值分解(Singular Value Decomposition,SVD)降噪和卷积神经网络(Convolutional Neural Network,CNN)分类的无线信号调制识别方法SVD-CNN。该方法提出了基于SVD的信号降噪模块来对输入信号进行降噪,设计了一维符号级CNN架构来直接识别信号特征并分类。针对高斯、瑞利信道下的调制仿真数据集,将提出的方法与典型调制识别方法如CNN识别方法、瞬时特征-全连接神经网络(Instantaneous Characteristic-Fully Connected Neural Network,IC-FCNN)识别方法进行了对比实验。实验结果表明,所提方法在低信噪比下具有更高的识别精度,在信噪比为0 dB时平均识别准确率提升近38%~49%。

In order to solve the problem of low signal recognition accuracy caused by noise interference in low signal to noise ratio,a wireless signal modulation recognition method based on Singular Value Decomposition(SVD)noise reduction and Convolutional Neural Network(CNN)classification SVD-CNN is proposed.The signal denoising module based on SVD is proposed to denoise the input signal.The one-dimensional symbolic level CNN network architecture is designed to identify and classify signal features directly.For the modulation simulation data sets in Gaussian and Rayleigh channels,the proposed method is compared with typical modulation recognition methods such as CNN recognition method and Instantaneous Characteristic-Fully Connected Neural Network(IC-FCNN)recognition method.The experimental results show that the proposed method has higher recognition accuracy at low SNR,and the average recognition accuracy is improved by 38%~49% when the SNR is 0 dB.