Research On Adaptive Filter Denoising Method Based On Component Decomposition

Noise reduction is a classical problem in the field of signal processing.There are many kinds of signals and noises in nature,however,most algorithms only have good denoising effects on specific signals and noises.It is difficult but meaningful to find a common method.Although the adaptive filtering method needs an additional reference channel,it has strong adaptability to all kinds of noise,so it is still of great research value.As one of the adaptive time-frequency analysis methods,adaptive component decomposition methods can decompose an original signal into several components according to the nature of the signal.Through setting a threshold or some other way,we can select some components to reconstruct the useful signal.For fully preserving the adaptability of component decomposition method,and overcoming the problem of occasional slow convergence and bad effect on non-stationary signal of the LMS adaptive filtering algorithm,in this paper we mainly study the method of adaptive noise component decomposition method and its noise reduction method.With high correlation noises or noises containing multi-frequency components,the convergence speed of Least Mean Squre(LMS)type adaptive filter is slow.To solve this problem,we introduce the adaptive component decomposition method,Empirical wavelet transform(EWT),which is an adaptive frequency band decomposition method based on spectrum division.The basic segmentation method is to find the segmentation boundaries according to the maximum and minimum points in the spectrum.For complex signals,we study a scale space based no-parameters spectral segmentation method.This paper presents an adaptive filtering method based on EWT.Firstly the noise signal is decomposed into several sub bands using EWT.Then the empirical wavelet functions are retained and are used to decompose mixed signal.After that each sub-band of each noise signal and mixed signal is adaptively filtered and denoised.Finally,the filtering results of each sub-band are accumulated to get the final noise reduction signal.This paper studies the effect of gamma parameters of EWT of the segmentation.From the view of the reconstruction error for periodic deterministic signals,the normalized error is in the level of 10-4,which can improve the denoising effect if is reduced.Meanwhile,the improvement of the subband dynamic spectrum in the spectral range is also verified by experiments.In the end,the paper carries out the noise reduction experiment which shows that the proposed method can achieve better convergence performance than the direct adaptive filter and better effect on noise reduction in pink noise environment.In view of LMS algorithm has limit noise reduction effect on non-stationary signal or noise,another component decomposition method,empirical mode decomposition(EMD),is introduced.Quite different from the Fourier analysis framework,the EMD method can extract the single component signal iteratively by averaging the envelope of the signal extremum.It can better reflect the local characteristics of signals,and is more suitable for non-stationary signal analysis.In this paper,we first study two noise reduction methods based on EMD and adaptive filter proposed by other researchers.Analyzing their problems,we propose to use multivariate empirical mode decomposition(MEMD)to decomposite mixed signals and noise signals synchronously.After decomposition,the stability of each IMF is enhanced so the signal is more suitable for adaptive filter.The effectiveness of the proposed method is verified by noise reduction experiments.In this paper we propose a new idea for noise reduction which combines adaptive component decomposition and adaptive filter.From two aspects we introduce two different component decomposition methods.The noise reduction experiments verify the feasibility of this idea.