Date of Award

Spring 1971

Document Type

Dissertation

Degree Name

Doctor of Engineering Science in Electrical Engineering

Department

Electrical Engineering

First Advisor

Mauro Zambuto

Second Advisor

Marvin Kurland

Third Advisor

Anthony R. Marsh

Fourth Advisor

Benedict C. Sun

Abstract

Several types of signal processing systems in which the signal flows along parallel channels in a fashion similar to the auditory system have been investigated. The effect of excitation with signals containing both single and multiple spectral peaks (formants) was considered. In particular, the effect of nonlinear interaction between channels, referred to as centering, in the presence of noise was studied.

These systems were investigated for their value, both as information processing networks and as models of the auditory system.

The analysis indicates that parallel channel systems, in general, exhibit excellent performance in the presence of noise, and that a parallel channel system, with a limited overall bandwidth, can be made to process large amounts of information per unit time if used in conjunction with an appropriate centering network. Furthermore, these systems permit detailed frequency analysis of signals in the presence of noise without impairing their temporal discrimination capability.

Of the centering processes investigated, maximum likelihood centering provides an optimum estimate of formant frequency in the presence of noise, while lateral inhibitory centering probably represents the most practical process for implementation.

The performances of various centering processes are compared to the known characteristics of the auditory system, and the most promising of these, lateral inhibitory centering, is employed in a model of the peripheral auditory system.

The response of this model, when simulated on the digital computer, correlates closely with many of the characteristics of the peripheral auditory system. The model, however, does not adequately "explain" the spectral resolving ability displayed by the ear. An extension of the model was suggested which should not be subject to this limitation.

Share

COinS