A practical algorithm for least-squares spline approximation of data containing noise

Abstract

A simple yet robust algorithm is presented to approximate data containing noise by smooth spline functions. The method is easy to implement and fast. Experience has shown that in the great majority of cases no user decisions are required to obtain adequate approximants, which is a useful property for routine processing of experimental data. The statistical basis of the algorithm is a generalized version of the Durbin–Watson statistic, which performs well even in the presence of correlated noise. Examples are given to illustrate the power of the algorithm. © 1998 American Institute of Physics.