Studies on Adaptive Filtering of Analytical Instrument Signal Based on Wavelet Theory
Yanshi Dong,Yiyu Cheng
DOI: https://doi.org/10.3321/j.issn:0251-0790.1999.03.008
1999-01-01
Abstract:In this paper, a new type of adaptive filtering algorithm, which can adaptively remove all kinds of noises from signals of analytical instruments under a variety of complex conditions, is proposed. At present, the popular filtering algorithms which are widely applied to the data processing equipment for analytical instrument are low-pass filter or band-pass filter. The fundamental of those filters depends on the fact that the frequency characteristics of real signals are different from those of noises. These filtering algorithms based on the different frequency distribution characteristics between signals and noises have an obvious defect, that is, users have to preset properly initial filter factors according to the width of peaks, which greatly influences the objectivity and veracity of computational results in analytical procedures. In the light of the wavelet transform modulus maximum theory proposed by Mallat, the characteristics of wavelet transform modulus maxima of real signals are distinctively different from those of noises in the practical signals of analytical instruments, such as chromatography. It is easy to identify them. Taking advantage of the different characteristics between real signals and noises on different scales in wavelet transformation domain, noises can be removed from the practical signals or analytical instruments while avoiding to distort the real signals. The adaptive filtering algorithm designed by this principle breaches the popular patterns of current filtering algorithms, and radically improves the filtering effects. A lot of tests using chromatography data prove that this algorithm has a serial of virtues, such as no requirement on artificially presetting filter factors, excellent separation of signals and noises, holding the position and height of peaks, and so on. Its performance in the robustness, adaptability and fidelity of peak completely satisfy the needs of signal processing for analytical instruments.