BISPECTRUM AND MULTIDIMENSIONAL POWER SPECTRUM ESTIMATION ALGORITHMS BASED ON PARAMETRIC MODELS WITH APPLICATIONS TO THE ANALYSIS OF ECG DATA (SPECTRAL ANALYSIS, NONLINEAR INTERACTIONS)

Date of Completion

January 1984

Keywords

Engineering, Electronics and Electrical

Degree

Ph.D.

Abstract

A parametric frequency-wavenumber spectrum estimation technique is developed for uniform planar sensor arrays. Important theoretical results are the development of 3-d Minimum Variance Representations (MVR) and spectral estimation algorithms based on these using the Covariance Least Squares Criterion (CLS). Simulation examples that show the superior resolving abilities of this method compared to the Maximum-Likelihood Method (MLM) of Capon are given. The method also shows good spectral matching for wideband parametric random fields. The second part of the work concerns the bispectrum, an important tool in investigating non-Gaussianness, quadratic nonlinearity and phase measurements. This information is actually recovered from the third moments and existing bispectrum estimation methods are of the Fourier-type and they possess some limitations. This works develops parametric methods for bispectrum estimation based on a non-Gaussian white noise (NGWN) driven autoregressive (AR) model which approximates the third moment bispectrum estimates of higher fidelity for parametric processes than that provided by known methods, and for detecting quadratic phase coupling effects, they show higher resolving abilities. Multidimensional power spectrum estimation in the form of 3-d CLS spectrum estimation and Zero Delay Wavenumber Spectrum (ZDWS) estimation is applied to canine epicardial data. These techniques demonstrate abilities to decompose cardiac electrical activation into dominant planewaves and to obtain their bearings and speed. The ZDWS provides a compact way of representing information and differences between normal and ischemic conditions are clearly seen. ^

Link to Full Text

Share

COinS