EVALUATION OF 3 ZERO-AREA DIGITAL-FILTERS FOR PEAK RECOGNITION AND INTERFERENCE DETECTION IN AUTOMATED SPECTRAL DATA-ANALYSIS

An intermediate step in the automatic evaluation of complex spectra is the determination of positions and number of individual lines. In practice, peak analysis is often hampered by statistical noise, presence of a significant background, and influence of substantial interference. One possibility to avoid some of these peak detection problems consists in transforming the accumulated (spectral) data into another spectrum, by means of convolution. Transformed signals have a very characteristic shape, allowing recognition of peaks in an easier and more explicity way. Furthermore, by choosing the proper filter and its parameters, a resolution enhancement can be achieved. In the present work, the performance of square-wave, Gaussian, and triangular zero-area digital filters is compared when applied to single Gaussian lines, multiplets of Gaussian profiles, and Voigt profiles. In order to obtain a general overview of the properties of the filters just mentioned, analytical expressions for the convolution of a Gaussian profile with each of the filters are derived. It is proved in general that a linear background component of a signal is completely filtered out by an even zero-area digital filter as opposed to higher order components. The parameters that are looked at in the convolution signal are full-width at half-maximum, intensity at the position of the maximum, signal-to-noise ratio, second-order background contribution, and resolution enhancement as a function of some typical filter parameters. Besides its use for peak detection, the convolution signal can also provide some information about the presence of interferences. When combined with Zimmermann's method, digital filtering is able to detect most spectral interferences or can at least give a warning about irregularities in the original spectrum.