脉冲噪声下基于CNN-FRFT的线性调频信号参数估计方法  

LFM Signal Parameter Estimation Method Based on CNN-FRFT Under Impulsive Noise

作　　者：卢景琳 郭勇 杨立东[1] LU Jinglin;GUO Yong;YANG Lidong(School of Information Engineering,Inner Mongolia University of Science and Technology,Baotou 014010,China;School of Science,Inner Mongolia University of Science and Technology,Baotou 014010,China)

机构地区：[1]内蒙古科技大学信息工程学院,内蒙古包头014010 [2]内蒙古科技大学理学院,内蒙古包头014010

出　　处：《探测与控制学报》2024年第1期96-104,113,共10页Journal of Detection & Control

基　　金：国家自然科学基金项目(62161040,62201298);内蒙古科技大学创新基金项目(2019QDL-B39)。

摘　　要：由于脉冲噪声破坏了线性调频(LFM)信号的分数谱特征,使得基于分数谱特征的参数估计方法无法有效估计参数。针对这个问题,提出一种脉冲噪声环境下基于CNN-FRFT的LFM信号参数估计方法。首先,利用α稳定分布拟合随机脉冲噪声,构建加性含噪信号,输入卷积神经网络(CNN)进行训练和测试;其次,利用训练好的CNN模型对信号进行去噪,并验证模型的去噪能力和泛化能力;最后,利用分数阶傅里叶变换(FRFT)建立去噪信号的分数谱,通过峰值点位置来估计LFM信号的参数。实验结果表明,相比于传统的基于非线性函数的方法,该方法在强脉冲噪声环境下具有更好的精度和噪声鲁棒性,CNN的应用使其具有更强的泛化能力,在实测脉冲噪声下仍可以准确估计参数。The impulse noise destroys the fractional spectrum characteristics of linear frequency modulation(LFM)signal,so the parameter estimation method based on the fractional spectrum characteristics cannot estimate the parameters effectively.To solve this issue,a LFM signal parameter estimation method under impulsive noise environment was proposed based on CNN-FRFT.Firstly,the alpha-stable distribution was used to fit the random impulse noise,and the additive noisy signal was constructed and input into CNN for training and testing.Secondly,the trained CNN model was used to denoise the signal,and the ability of denoising and generalization of CNN model was verified.Finally,the fractional spectrum of the denoised signal was established using FRFT,and the parameters of LFM signal were estimated by the position of the peak point.Experimental results showed that the proposed method had better accuracy and noise robustness in the strong impulsive noise environment in comparison with the traditional method based on nonlinear function.The application of CNN made the proposed method a stronger generalization ability with the measured impulse noise.

关 键 词：脉冲噪声 线性调频信号 参数估计 卷积神经网络 分数阶傅里叶变换 

分 类 号：TN957.51[电子电信—信号与信息处理]