SG滤波;Savitzky-Golay
570
Anal. Chem. 1990, 62, 570-573 Obrusnik I.; Eckschhger, K. J. Radioanal. Nucl. Chem., Articles l987, 112, 233-242. Eckschlager. K.;&p6nek, V . cdlect. Czech. Chem. Commn. 1985., so. .., 1359-1367. . . . . . Eckschlager, K.; Fusek, J. Collect. Czech. Chem. Commun. 1988, 53. 3021-3028. Lindstrom, R. M. Nuclaer Ana&tical Methods in Standards certlffcation; IAEA-TECDOC435; IAEA: Vienna, 1987. Becker, D. Analytical Design in Activation Analysis: The R d e of Accuracy and Precislon. In Accwacy in Trace Analysis: Sampling, Sample HendUng. and Analysis; NBS Special Publication 422; NBS: Washington, DC, 1976; pp 1143-1 155. De Bruin, M. GammaRay Spectroscopy in Neutron Activation Analysis. In Quality Assuance in BbmedkW Neuiron Acthatkn Analysis; IAEA-TECDOC-323 IAEA: Vienna. 1984; pp 163-177. Greenberg, R. R. J. Radioanal. Nucl. Chem., Articles 1987, 113, 233-247. H e n, R. E. Control of Sample Conf&mtion as an AM to Accuracy in INAA; NBS Special Publication 422; NBS: Geithersbura. MD. 1976; pp 1275-128i. Ozmutlu, C.; Ortaovaii, A. 2. Nucl. Instrum. Methods 1978, 133, 149- 155.
Moreover, we cannot derive the bias level 6 from theory. In such cases we can estimate u and especially 6 empirically by the analysis of reference materials with certified concentrations of an analyte n,. Then
6 (9%)= x 100 nc
If,
-&I
(6)
where f, is an average result of the analysis of an RM (see details in ref 7). We can expect that information theory can be successfully used in the optimization and comparison of various analytical methods and procedures, in the evaluation of the results obtained in collaborative tests for the certification of new RMs, and in many other applications.
LITERATURE CITED (1) Eckschlager, K.; S t b p h k, V . Anal. Chem. 1982, 54, 1115A-1127A. (2) Eckschlager, K.;$t&phk, V. Information Theory as Applied to Chemical Analyst; Wiley: New’fork, 1979. (3) Eckschlager, K.; S& tp n i&, V . A n a w l Meesvement and Information; Research Studies Press: Letchworth. Great Britain, 1985. (4) Eckschlager, K.; Stbplnek. V. Chemom. Intel/. b b . Syst.~ 8 7 1,, 273-204.
(14) Eckschfager, K. Anal. Chem. 1977. 49, 1265-1267. (15) Fleming, R. F.; Lindstrom, R. M. J. Radioanal. Nucl. Chem., Articles 1987. 113, 35-42.
RECEIVED for review A u P t 7,1989. Accepted November 27, 1989.
General Least-Squares Smoothing and Differentiation by the Convolution (Savitzky-Golay) Method Peter A. Gorry Department of Chemistry, University of Manchester, Manchester, England M13 9PL
S m and d#ferentlatkn of large data sets by plecewlse least-squares polynomlal fffllng are now wklely used technlques. The calculation speed Is very greatly enhanced H a convolution formalism is used to perform the calcuiatlons. Prevlously tables of convolution weights for the center-pdnt least-squares evaluatlon of 2m 1 points have been presented. A major drawback of the technique Is that the end polnts of the data sets are kot (2m pohts for a 2m 1 point fllter). Convdutlon weig
hts have also been presented In the speclal case of Inltlai-point values. I n this paper a sknple general procedure for calculatlng the convolution weights at all podtlons, for all polynomial orders, all fitter lengths, and any derlvatlve Is presented. The method, based on the recursive properties of Gram polynomials, enables the convolutlon technique to be extended to cover all points In the spectrum.
+
+
INTRODUCTION In 1964 Savitzky and Golay (1) provided a simplified method for calculating smoothing and differentiation of data by a least-squares technique. Since then the Savitzky-Golay approach has been widely used because it produces a significant improvement in computational speed-replacing the traditional lengthy least-squares calculation by a simple, but equivalent, convolution. In this approach the least-squares value of a given point is calculated as a weighted combination of itself and m points on either side of it. This corresponds to performing a moving (2m+ 1point) least-squares fit across the data. The purpose of the original work was to provide a methodology for calculating the required weights and to
provide tables of commonly used values. The original tables of convolution weights contained several errors however and were subsequently corrected by Steinier et al. (2) who recast the calculation in a matrix form. The Savitzky-Golay approach suffers from one major drawback it truncates the data by m points at each end. This occurs because the convolution requires m points to the left and right of a point in order to calculate the required leastsquares value. The problem of data truncation is compounded if repeated smoothing/differentiation is applied since each application removes a further 2m points. For large spectra with zero values at the ends this is not important, and indeed zero values can easily be reinserted. However for more limited data sets without base-line values at both ends, the Savitzky-Golay algorithm is not applicable. In principle the above limitation is easy to overcome. The first 2m+ 1 point least-squares fit can simply be used to evaluate least-squares values for the first m points as well as the m+ 1. Similarly the last 2m+ 1 point fit can be used to calculate the last m leasbsquares values. Indeed if a normal least-squares calculation has been performed, yielding the least-squares coefficients, evaluating the fit of any of the m values, rather than the middle, is a simple task. This approach has been adopted by Khan (3) in a matrix formulation of the full least-squares method. Unfortunatelywe now lose the main advantage of the convolution method-its greatly enhanced speed. Extending the convolution approach to accommodate“end points” requires calculating a different set of weights for each position. The special case of initial-point smoothing and slope evaluation has been considered by Leach, Carter, and Harris ( 4 ) where the Savitzky-Golay approach was modified to provide quadratic convolutio
n weights for smoothing and the
0003-2700/90/0362-0570$02.50/0 0 1990 American Chemical Society

F08 – Least-squares and Eigenvalue Problems (LAPACK) F08UEF (SSBGST/DSBGST) – NAG Fortran Library Routine Document Note. Before using this routine, ...
On Least-Squares Bias in the AR(p) Models: Bias Correction Using the ...In this paper, in more general formulation which can be applied to all ...
On Least-Squares Bias in the AR(p) Models: Bias Correction Using the ...In this paper, in more general formulation which can be applied to all ...
热门文档