APP下载

基于矩阵秩最小化和统计修正的信号降噪方法研究

2016-01-15李文峰,许爱强,戴豪民等

振动与冲击 2015年15期

基于矩阵秩最小化和统计修正的信号降噪方法研究

李文峰1,2,许爱强1,戴豪民1,王丰3

(1.海军航空工程学院科研部,山东烟台264001;2.92635部队山东青岛266041; 3.海军航空工程学院91206部队,山东青岛266108)

摘要:针对奇异值分解在信号降噪时有效秩的选择问题,提出一种基于矩阵秩最小化和统计修正的信号降噪方法。首先,利用矩阵秩最小化理论将奇异值有效秩选择问题转化为秩的约束优化问题;然后,通过凸优化求解,得到干净信号的Hankel矩阵,实现一次降噪;最后,根据奇异值子集标准差对干净信号Hankel矩阵进行统计修正,进一步优化降噪效果。模拟信号和真实信号的实验结果表明:该方法可以有效的消除脉冲干扰和高斯噪声,能够最大限度的降低信号均方误差,提高信噪比,增强了奇异值分解在信号降噪中的通用性。

关键词:奇异值分解;奇异值扰动理论;矩阵秩最小化;奇异值子集标准差

中图分类号:TN911.7文献标志码:A

Signal denoising method based on matrix rank minimization and statistical modification

LIWen-feng1,2,XUAi-qiang1,DAIHao-min1,WANGFeng3(1. Research Department, Naval Aeronautical and Astronautical University, Yantai 264001, China;2. No. 92635 Unit of PLA,Qingdao 266041, China;3. No. 91206 Unit of PLA, Naval Aeronautical and Astronautical University, Qingdao 266108, China)

Abstract:Aiming at the selection problem of effective ranks in singular value decomposition for signal denoising, a signal denoising method based on matrix rank minimization and statistical modification was proposed. Firstly, the effective rank selection problem of singular value decomposition was transformed into a constrained optimization problem of rank by using the matrix rank minimization theory. Secondly, Hankel matrix of a clean signal was obtained with a convex optimization to realize the first noise reduction. At last, the statistical correction was performed for the clean signal’s Hankel matrix with subset standard deviation of a singular value to further optimize the noise reduction effect. The simulated signal and real signal test results showed that the method can effectively eliminate pulse noise and Gaussian noise; at the same time, the method can reduce the maximum signal mean square error and improve the signal-to-noise ratio; so the method can enhance the universality of singular value decomposition in signal denoising.

Key words:singular value decomposition (SVD); matrix perturbation theory; matrix rank minimization; subset standard deviation of singular value

信号在产生、采集和传递时往往伴有噪声的干扰,因此信号预处理中的信号降噪是保证后续信号分析的重要前提。传统的降噪方法如小波降噪和基于谱分析的滤波方法,其原理是利用信号与噪声的频谱不同达到降噪的效果。但是,当有用信号和噪声频带混叠严重时,传统的降噪方法则具有较大的局限性。近年来,奇异值分解(Singular Value Decomposition,SVD)在信号降噪问题上已经得到了广泛应用,其中奇异值有效秩选择问题一直是当前研究的热点和难点。Donoho等[1]提出了奇异值优化硬阈值法,给出了硬阈值的优化选择标准;……

登录APP查看全文