Anal. Chem. 1901, 63, 1441-1448
1441
Kalman Filtering for Data Reduction in Inductively Coupled Plasma Atomic Emission Spectrometry E. H. van Veen* and M. T. C. de Loos-Vollebregt Laboratory of Analytical Chemistry, Delft University of Technology, de Vries van Heystplantsoen 2, 2628 RZ Delft, The Netherlands
I n ICP-AES, the anaiyte slgnai is superimposed on a background signal. When separating these dgnak, essentially by manual or automated Wee-poW background correction, there are many tnstances in whlch the data redudon f a h e.g., wllh noisy anaiyte signals, nonlinear background signals, large and heavily structured background, and spectral interference. On the bade of scans in a fad scanning mode, the Kaiman fltter approach ykids in all cases more accurate and preclse rewtts. I t does not use a search for peak posttion or speclflc background polnts and processes peak area instead of peak height. I t attalns, up to 2 orders of magnitude, lower detection iknns in the sample solutlon, for real-worid samples. The applkation of the filter results in l-pm spectral resolution at 20pm spectral bandwidth and reduces the need for llne seilectlon, high-rerokrtlonspectrometers, and chemical separation of analyte and matrix. The Improved analytical results obtalned with the Kaiman filter are due to Its capabliity of noise averaglng, of multlple llne analysis, and of ellmlnatlon of the degradatlon of detection limit caused by llne overlap. I f line selection Is needed, H Is done quantitatively. The power of the fully automated approach Is illustrated with several applkationrr: the analyde of Mgh-purHy uranium and dudgo reference material. I n every respect, data reductlon by Kaiman filtering is superior to three-pdnt background correction.
INTRODUCTION In inductively coupled plasma atomic emission spectrometry (ICP-AES), analyte signals are superimposed on a relatively high background signal. The measured gross signals have to be corrected for this background, which determines the detection limit by its amplitude and noise characteristics. Although many samples from a series may resemble each other, the background level will vary from sample to sample. Therefore, the preparation of one blank and measurement of its signal, followed by subtraction of this value from all the sample signals, is not generally feasible. Since the spectrometer is fixed at the analyte wavelength, this method is called static or on-line background correction. Common types of background correction are based on twoor three-point, off-line background correction (1). As long as analyte signals, originating from elements at the mg/L level in aqueous solution, dominate a smooth continuum background, data reduction by this technique is straightforward. Dependent on the spectrometer used, the top of the analyte signal is found by direct peaking or peak search routines. The one or two background points are selected manually, and measurements are always made at these fixed positions. A drawback is that no information is obtained about the profile of the emission line or about the background shape in the vicinity of the line. Another possibility is the automatic selection of background points from a sample scan, by an algorithm based on the second derivative in the spectral window of the analyte (2,3).Using the derivative technique, one may 00082700/9 110383-144 1$02.50/0
select more than two points, allowing for correction for nonlinear continuum background (2). When the off-line background exhibits structured interferent emission, the search for background points still seems possible (3). At low analyte signals, when the noise becomes dominant, the results of the search for peak and background points easily are in error. With heavily structured background, the three points may not be found. In the case of overlapping emission lines, the automated algorithms fail. An additional algorithm, like Fourier deconvolution for the instrument function of the spectrometer, may be applied, but this approach is only successful if the peak separation is not less than the bandwidth of the spectrometer (4). Interelement correction with direct readers, or dynamic three-point background correction with slew-scan monochromators, is needed, leading to a degradation of detection limits by 1order of magnitude or even more (5). This is caused by optical instability and wavelength positioning errors. In an attempt to remedy the degradation of detection limits, several approaches can be applied, e.g., use of high-resolution (
