An Optimal Weighting Function for the Savitzky-Golay Filter

TL;DR

This paper demonstrates that using a quadratic residual weighting function in the Savitzky-Golay filter optimizes the smoothness of the output, outperforming the traditional equal-weight approach.

Contribution

It introduces a quadratic weighting function that enhances the filter's smoothness, providing a novel optimal weighting strategy for the Savitzky-Golay filter.

Findings

Quadratic weighting improves filter smoothness significantly.

Optimal weighting enhances frequency response in the stopband.

Compared to equal weights, the quadratic function yields better results.

Abstract

The Savitzky-Golay FIR digital filter is based on a least-squares polynomial fit to a hypothetical sample of equally spaced data. This gives the filter the ability to preserve moments of features like peaks in the input. Descriptions of the filter typically consider the case where equal weights are implicitly applied to the residuals of the fit. In a largely overlooked paper Turton showed that weighting the residuals with a triangular function significantly improves the frequency response of the filter in the stopband. The Savitzky-Golay filter is commonly referred to as a smoothing filter. This paper uses a particular measure of smoothness to show that a quadratic residual weighting function optimizes the smoothness of the filter output for a given sample size and degree of the fitting polynomial. This weighting function can provide substantially better smoothness than that with a constant weighting function.