Bartlett Method Research Articles

The use of nonuniform element spacing in array processing algorithms.

Most array signal processing algorithms use uniform element spacing to estimate source bearing. This paper demonstrates the benefit of using nonuniform element spacing in Bartlett, linear prediction (LP), and minimum variance (MV) array processing algorithms. By using optimum element spacing results obtained by previous investigators for sidelobe reduction of the Bartlett method, better MV and LP performances, in terms of array output power, are obtained.

Objective quality measures applied to enhanced speech

Previously reported experiments have indicated that computer enhancement of single channel speech in wideband noise generally does not result in increased intelligibility. Our research goal has been to investigate the effect of such enhancement on speech quality. Our experimental framework employed natural speech degraded by additive white, non‐white, and aircraft cockpit noise over a wide range of signal‐to‐noise ratios (− 15 to 20 dB). An iterative enhancement technique based on a method reported by Lim and Oppenheim, was generalized to allow nonwhite additive noise. Spectral estimates of the noise were made using maximum‐likelihood method, maximum‐entropy method, Burg's method, periodogram, Bartlett's method, and Pisarenko harmonic decomposition. Quality was estimated using a variety of objective measures which have been shown to correlate well with subjective quality scores resulting from the diagnostic acceptibility measure (DAM). Over a large number of noise conditions the best objective qualities resulted when Bartlett, Burg, and maximum‐entropy spectra were used. In general, significant improvements in terms of objective measures were observed when enhancement took place. [Work supported by Lockheed Georgia Company.]

Optimal Procedures for Estimation of Factor Scores

This study identifies three different applications of factor scores which lead to three different sets of criteria for optimal estimation procedures. The regression method and the Bartlett method best satisfy the first two sets of criteria, and a generalized Anderson and Rubin procedure is optimal for the third set of criteria. The differences between the methods have been illustrated by an example.

Bivariate Linear Models in Biometry

Kuhry, B., and L. F. Marcus (Paleontological Institute, P. O. Box 558, 75122 Uppsala, Sweden, and Department of Biology, Queens College, Flushing, New York 11367) 1977. Bivariate linear models in biometry. Sys!. Zool. 26:201-209.-This paper focuses on the estimation of parameters in the bivariate linear model, especially in the context of bivariate size-shape relationships or allometry. Existing estimation procedures (regression, major axis, reduced major axis) all depend on a priori assumptions on the ratio of the residuals, usually called errors, in both variables. These assumptions are reviewed and evaluated. The Bartlett method is not independent of assumptions on the residuals as has been often claimed. A method which does not require assumptions on the ratio of residuals, providing data from a third variable are available, is given. All of the methods discussed are illustrated with data measured on planktonic Foraminifera. [Bivariate; allometry; regression; major axis; Foraminifera. ] The means of measurements on a sample of organisms are strongly dependent on size fluctuations due to the age distribution of the sample, environmental fluctuations, and -in fossils-sedimentary sorting. Shape factors are in general found to be more stable, and the majority of distinguishing criteria between taxa tend to be formulated in terms of shape. Intuitively, shape is most readily represented by ratios between measurements. However, individuals tend to change their shape during the growth process, and different sized individuals in populations are often different in shape. Hence, ratios may be seriously misleading, and a characterization of samples should be based on the dynamic relationship between size and shape. A priori considerations and supporting evidence from practical studies indicate that the allometry equation Y = f30Xfh is a reasonable basic model in the majority of size-shape problems. Growth allometry refers to the relationships of two morphometric variables in a growing organism, while size allometry refers to the relationship of such variables in samples of organisms. We are concerned with the latter in the development and examples which follow. A linear model is obtained after a log1 Present address: Rijksmuseum voor Geologie en Mineralogie, Hooglandse Kerkgracht 17, Leiden, The Netherlands. arithmic transformation of the allometry equation. Estimation of the parameters in this linear model is not the usual regression problem requiring one of the variables to be free of error. We are also not interested in the prediction of one variable from the other. It is rather a problem of finding a structural relationship between the two random variables which are both subject to error. Recourse to textbooks and review articles on methods in biometrics (Imbrie, 1956; Simpson, Roe and Lewontin, 1960; Miller and Kahn, 1962; Sokal and Rohlf, 1969; Ricker, 1973) for advice on fitting a straight line under these conditions suggests a number of different methods: major axis, reduced major axis, and the Bartlett threegroup method. Although these authors make outspoken recommendations for the most suitable method, they are rather vague, and in some cases even incorrect, on the implied assumptions which are concomitant with the specific procedures which they

A note on the bias in the Kalman-Bucy filter

In an earlier paper using the ‘regression approach’ to factor analysis the well-known Kalman-Bucy filter is derived. In this note we obtain expressions for the bias of this filter. Using Bartlett's method of estimation of the ‘factors’, we obtain a new recursive filter which is always unbiased and derive its variance-covariance matrix.

Mercury in Hair of Bobcats and Raccoons

Mercury in Hair of Bobcats and Raccoons