较低信噪比下信号去噪方法研究

发布时间：2018-01-30 16:23

本文关键词： 经验模态解小波变换随机内插自相关 ICA 低信噪比子空间　出处：《电子科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：在通信系统中,信号从产生到传输,再到处理整个过程中都会受到噪声的干扰,这些噪声污染了真实信号,对后续的分析,判断,识别工作产生一定的干扰,因此,信号去噪一直是该领域中的重要研究课题,当前的信号去噪方法中,大多假设信噪比较高,但在实际通信中,常是噪声强于信号甚至会出现信号完全淹没在噪声中,此时,基于先前假设的去噪方法性能下降甚至无法进行。本文研究了常用的几种去噪算法,并对它们在较低信噪比下的去噪可行性进行了深入研究。本文首先介绍几种常用的信号去噪方法,接着重点探讨了这些算法的去噪原理及性能,并将之应用于较低信噪比下,分析对比两种情况下的去噪性能。在经验模态分解(EMD)中,研究了低阶系数舍弃、连续均方误差准则去噪方法,分别仿真分析了这两种算法在较高和较低信噪比下的去噪性能。信噪比较高条件下,均取得较好的去噪效果,但在信噪比较低时,经由EMD分解得到的IMF分量中,一些分量之间能量的差异并不大,此时采用低阶系数舍弃法由于K值的选取不当,会引起混叠效应;基于连续均方误差准则的EMD去噪为界值K的选取提供了理论依据,但在信噪比较低的情况下,即使存在某个分量的能量是局部最小值,但不一定是噪声与信号主导区间的分界点,某些极端情况下,甚至找不到全局最小值,这时该算法完全无效。在小波变换中,分析对比软、硬阈值去噪性能,在较高信噪比下,性能较优。但在较低信噪比下由于阈值较大,此时,会将信号部分的小波系数视为噪声,误剔除,影响去噪性能,故在较低信噪比下不宜采用传统的小波去噪方法。通过分析、对比噪声自相关函数和信号自相关函数的数学特征,提出一种基于自相关EMD与自适应小波阈值联合去噪方法,通过计算IMF分量的自相关函数,选取噪声信号分界点,对界点之前的分量进行门限处理,重构后的信号作为下一级小波分析的输入。信号的不连续会造成吉布斯效应,影响去噪性能,针对此提出一种基于随机邻域内插的信号去噪算法,该算法通过随机的邻域插值,改变信号特征和小波基函数之间的匹配条件,达到去噪目的。基于ICA的信号去噪算法,在较高输入信噪比下,可以很好的完成去噪工作,但在较低信噪比下,信号的自相关矩阵可能会出现零特征值,即此时,自相关阵为奇异矩阵,无法完成白化过程,此时基于ICA的去噪方法失效,研究了一种基于子空间投影的ICA分析。
[Abstract]:In a communication system, signal transmission from generation to will be disturbed by the noise in the whole process, the noise pollution of the true signal, judgment of the subsequent analyses, certain interference, so the identification work, signal denoising has been the important research topic in the field, the current signal denoising methods, assuming high SNR, but in the actual communication, is often stronger than the noise signal or even signal submerged in noise, at this time, the previously assumed denoising method based on performance degradation or even unable to. This paper studies several common denoising algorithms, and their in low SNR denoising feasibility is studied. This paper introduces several common signal denoising method, and then discusses the algorithm's principle and performance, and applied in low SNR analysis Than the denoising performance under two kinds of situations. In the empirical mode decomposition (EMD), the low order coefficients of renunciation, denoising method of continuous mean square error criterion, respectively analyzed the two algorithms of denoising performance at high and low SNR at high SNR conditions. Both have good denoising effect, but when the SNR is low, can be obtained through the EMD decomposition of the IMF component, some components of energy between the difference is not large, the low order coefficient method because the selected K value up properly, will cause aliasing effect; denoising provides a theoretical basis for the selection of K as the boundary value for mean square error criterion based on the EMD, but in the low SNR case, even if there is a component of the energy is a local minimum, but not necessarily is the dividing point of the noise and signal leading interval, some extreme cases, even can not find the global minimum, then the algorithm. All invalid. In wavelet transform, comparative analysis of soft and hard threshold denoising performance at high SNR, the performance was better. But when the SNR is low because the threshold is larger, at this time, will be regarded as part of the wavelet coefficients of signal noise, false rejection, affect the denoising performance, so in the low SNR should not adopt the traditional wavelet denoising methods. Through the analysis, comparison of noise autocorrelation function and signal mathematical characteristics of autocorrelation function, we propose a self correlation EMD and adaptive wavelet threshold denoising method based on joint, by calculating the autocorrelation function of the IMF component, select the noise signal cut-off point. The point before the component threshold. The reconstructed signal analysis as the next level of input signal. Wavelet discontinuity will cause the Gibbs effect, effect of denoising performance, put forward a random signal de-noising algorithm based on neighborhood interpolation, the algorithm By random interpolation, changing the matching condition between the signal characteristics and the wavelet function, achieve denoising. ICA denoising algorithm based on a signal at high SNR, can complete a very good denoising, but in low SNR, the correlation matrix of the signals may be zero eigenvalue, i.e., autocorrelation matrix is singular, unable to complete the whitening process, based on ICA denoising method of failure, a subspace projection based on ICA analysis.

【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN911.4

【相似文献】