首页|脉冲噪声下基于DCNN的LFM信号去噪方法

脉冲噪声下基于DCNN的LFM信号去噪方法

扫码查看
由于脉冲噪声具有明显的尖峰脉冲特性,使得基于高斯假设的传统去噪方法无法有效滤除脉冲噪声.针对这个问题,文中提出了一种脉冲噪声下基于深度卷积神经网络(DCNN)的线性调频(LFM)信号去噪方法.首先,生成LFM信号和随机脉冲噪声,构建不同广义信噪比下的数据集,输入DCNN进行训练和测试.进而,从时域波形图、分数谱、时频分布三个方面验证模型的去噪能力.最后,对去噪LFM信号进行分数阶傅里叶变换,通过搜寻分数谱中的峰值点来估计LFM信号的参数.仿真实验结果表明,文中方法不仅能够有效去除含噪信号中的随机脉冲噪声,而且还可以保持LFM信号的时域特征、分数谱特征和时频特征基本不变,进而提高了参数估计的噪声鲁棒性.与传统的基于非线性变换的方法相比,本文方法在低信噪比下仍能有效保持信号的分数谱特征和时频特征,具有更好的去噪性能和泛化能力.
LFM Signal Denoising Method Based on DCNN Under Impulsive Noise
Since impulse noise has obvious pulse characteristic,the traditional denoising methods based on Gaussian hypothesis can't effectively filter impulse noise.To solve this problem,a LFM signal denoising method based on DCNN under impulsive noise environment is proposed.Firstly,the LFM signal and random impulse noise are generated,and the data sets with different general-ized SNR are constructed and input into DCNN network for training and testing.Furthermore,the denoising ability of the model is verified from three aspects:time-domain waveform,FRFT spectrum and time-frequency distribution.Finally,the denoised LFM signal is subjected to fractional Fourier transform,and the parameters of LFM signal are estimated by searching the peak points in the fractional spectrum.The simulation results show that this method can not only effectively remove the random impulse noise in the noisy signal,but also keep the time domain characteristics,fractional spectrum characteristics and time-frequency characteris-tics of LFM signals,and thus improves the noise robustness of the parameter estimation method.Compared with the traditional methods based on nonlinear transform,this method can still effectively maintain the fractional spectral characteristics and time-fre-quency characteristics of the signal at low GSNR,and has better denoising performance and generalization ability.

impulsive noisedeep convolution neural networklinear frequency modulation signalfractional Fourier transform

卢景琳、郭勇、杨立东

展开 >

内蒙古科技大学 信息工程学院,内蒙古 包头 014010

内蒙古科技大学 理学院,内蒙古 包头 014010

脉冲噪声 深度卷积神经网络 线性调频信号 分数阶傅里叶变换

国家自然科学基金资助项目国家自然科学基金资助项目内蒙古自然科学基金资助项目内蒙古科技大学创新基金资助项目内蒙古科技大学基本科研业务费专项资金资助项目

62201298621610402024LHMS060032019QDL-B39

2024

现代雷达
南京电子技术研究所

现代雷达

CSTPCD北大核心
影响因子:0.568
ISSN:1004-7859
年,卷(期):2024.46(10)