Advanced noise filtering of EC signals through wavelet transformations

In this paper, various filtering methods are considered for the purpose of reducing the level of noise in the EC signals, and comparisons are made. The most widely used method for digital signal analysis is Fourier analysis. Unfortunately, this method is best put to use on stationary signals where the loss of time information isn't critical. EC signal's characteristics vary significantly over time in the stochastic sense. Important events (indications) are represented as transient, highly localized changes in the signal and therefore the information is not easily extractable from the spectral domain. Wavelet analysis offers a distinct approach to signal analysis because low frequency information can be analyzed on larger scales while short intervals are used for high frequency content. In this way, time information is not lost in the transformed domain. While Fourier analysis is used to decompose the signal into sinusoids with varying frequencies, wavelet transformation decomposes the signal into shifted and scaled copies of the base (mother) wavelet function. Wavelet families like Haar and Daubechies wavelets are compared. Methods for determining the optimal decomposition tree as well as several post-decomposition thresholding techniques are discussed, including automatic threshold selection. Application of wavelet de-noising algorithms implemented in the INETEC Eddy One EC Data Analysis software is presented on real-world signals collected from WWER steam generator tubes.(author)