The New Jersey Institute of Technology's
Electronic Theses & Dissertations Project

Title:	Signal processing techniques for analysis of heart sounds and electrocardiograms
Author:	Bernstein, Alan D.
View Online:	njit-etd1979-007 (xxv, 354 pages ~ 9.1 MB pdf)
Department:	Department of Electrical Engineering
Degree:	Doctor of Engineering Science
Program:	Electrical Engineering
Document Type:	Dissertation
Advisory Committee:	Zambuto, Mauro (Committee chair) Klapper, Jacob (Committee member) Reisman, Stanley S. (Committee member) Foster, Achilles E. (Committee member)
Date:	1979
Keywords:	Heart--Sounds. Heart function tests. Electrocardiography.
Availability:	Unrestricted
Abstract:	Audible heart sounds represent less than 5% of the vibrational energy associated with the cardiac cycle. In this study, experiments have been conducted to explore the feasibility of examining cardiac vibration by means of a single display encompassing the entire bandwidth of the oscillations and relating components at different frequencies. Zero-phase-shift digital filtering is shown to be required in producing such displays, which extend from a recognizable phonocardiogram at one frequency extreme to a recognizable apexcardiogram at the other. Certain features in mid-systole and early diastole, observed by means of this technique, appear not to have been previously described. Frequency modulation of an audio-frequency sinusoid by a complex signal is shown to be effective in generating sounds analogous to that signal and containing the same information, but occupying a bandwidth suitable to optimum human auditory perception. The generation of such sounds using an exponential-response voltage- controlled oscillator is found to be most appropriate for converting amplitude as well as frequency changes in the original signal into pitch changes in the new sounds, utilizing the human auditory system's more acute discrimination of pitch changes than amplitude changes. Pseudologarithmic compression of the input signal is shown to facilitate emphasis in the converted sounds upon changes at high or low amplitudes in the original signal. A noise-control circuit has been implemented for amplitude modulation of the converted signal to de- emphasize sounds arising from portions of the input signal below a chosen amplitude threshold. This method is shown to facilitate the transmission of analogs of audible and normally inaudible sounds over standard telephone channels, and to permit the "slowing down" of the converted sounds with no loss of information due to decreased frequencies. The approximation of an arbitrary waveform by a piecewise-linear (PL) function is shown to permit economical digital storage in parametric form. Fourier series and Fourier transforms may be readily calculated directly from the PL breakpoint parameters without further approximation, and the number of breakpoints needed to define the PL approximation is significantly lower than the number of uniformly-spaced samples required to satisfy the Nyquist sampling criterion; aliasing problems are shown not to arise. Thus data compression is feasible by this means without recourse to a parametric model defined for the signal (e.g., speech) being processed. Methods of automatic adaptive PL sampling and waveform reconstruction are discussed, and microcomputer algorithms implemented for this purpose are described in detail. Examples are given of the application of PL techniques to electrocardiography, phonocardiography, and the digitization of speech.