Computational Aspects of Nonlinear Biological Dielectric Spectroscopy

Abstract

This thesis describes an investigation into suitable software and hardware for the measurement of nonlinear dielectric effects in biological cultures, and the use of such measurements in the formation and application of models of properties of the cultures. The main contributions of this thesis are the application of multivariate methods to the measured nonlinearities, and the extension of such methods to provide variable selection, providing faster acquisition of measurements and more robust models. Nonlinear effects are measured by “probing” the culture using a sinusoidal waveform of controlled frequency and amplitude, and recording the distortion placed on the signal by the nonlinearity. A powerful data acquisition system has been designed to allow generation of waveforms at specified frequencies and amplitudes and sampling of the response signal. The system calculates a power spectrum, and can average a number of such spectra to increase the signal to noise ratio. Through use of a digital signal processor, these processes take place concurrently, such that for most excitation frequencies of interest, the data acquisition time is merely that required for excitation. Modelling and prediction are performed using a separate software system. This uses methods such as partial least squares regression to form robust models of biological variables. The advantages of such models over more traditional wet chemical methods are those of acquisition time and the f lexibility of the instrument. A single instrument may be used to measure a number of biological factors simultaneously. Moreover the measurement time can be of the order of seconds, and measurements can take place continuously. A novel method for variable selection, based on the generated models, is used to improve their performance. Examination of the reduced variable sets provides a valuable insight into the most relevant excitation parameters, allowing the appropriate range of excitation frequencies and amplitudes to be targeted.

Date of Award	18 Dec 1995
Original language	English
Awarding Institution	Aberystwyth University
Supervisor	Jeremy Rowland (Supervisor) & Douglas B. Kell (Supervisor)