Precision averaging for real-time analysis

Authors:
Mark J. Cloutier;Matthew J. Friedman
Affiliations:
Logic Associates and Clark Franklin Kingston Press, White River Junction, VT;Veterans Administration Hospital, White River Junction, VT, and Dartmouth Medical School, Dartmouth, NH
Venue:
Communications of the ACM
Year:
1983

Citing 5
Cited 0

Computing standard deviations: accuracy

Communications of the ACM
Letters to the editor: Dealing with Neely's algorithms

Communications of the ACM
Error-free methods for statistical computations

Communications of the ACM
Comparison of several algorithms for computation of means, standard deviations and correlation coefficients

Communications of the ACM
Updating formulae and a pairwise algorithm for computing sample variances

Updating formulae and a pairwise algorithm for computing sample variances

Quantified Score

Hi-index	48.22

Visualization

Abstract

An analysis of the computation of the arithmetic mean using only single-precision fixed-point arithmetic is presented. This is done to ease the timing constraints common to many on-line applications. Others have introduced several averaging algorithms in floating-point arithmetic for use in inferential statistics. In this paper, these algorithms are evaluated with respect to their feasibility as fixed-point methods in the context of real-time analysis. Modifications of these algorithms are presented, and previously unpublished ones are introduced in the interest of avoiding overflow (necessary) and minimizing truncation errors (highly desirable). All algorithms presented are tested for speed and accuracy on several sets of data, including their own “worst case.” The applicability of each algorithm is discussed with respect to some of the basic functions that real-time programs must perform.