Analysis of the CNV waveform in the time and frequency domains.

The Contingent Negative Variation (CNV) response of the Electroencephalogram (EEC) is often obscured by the background EEC and/or ocular artefacts (OAs). These necessitate the application of a variety of signal processing techniques to reduce the influence of such phenomena on the measured signal. The methods discussed in this thesis are: ocular artefact removal (OAR) in the time domain and the use of data tapering and Fourier Transforms to give frequency domain amplitude and phase spectra.Two OAR methods were compared: a non-recursive method using ordinary least squares parameter estimation developed by Nichols and a recursive least squares technique developed by Ifeachor. Recursive OAR was observed to distort the CNV response. This was discovered to be due to omission of the response from the algorithm being used, an omission ocurring in both techniques. To correct this the inclusion of a model of the response in the algorithm was proposed. A number of data sets were devised in order to investigatethis effect and the successfulness of including the response model. It was shown that response modelling gave more efficient OAR and reduced response distortion. Similarinvestigations were performed on recorded response data which showed that modelling was essential for recursive OAR but that non-recursive OAR was relatively insensitive to the inclusion or omission of response modelling.The use of data tapering was included in order to help improve the spectral analysis of short epochs of EEC data.A comparison of statistical properties of the harmonics of the amplitude and phase spectra of the CNVs of normal,patient and at risk subjects was made using Predictive Statistical Diagnosis (PrSD) and Discriminant Analysis (DA). The investigation compared results between PrSD and DA and between 1 and 4 second inter-stimulus interval CNVs and suggested that the 1 second data gave better discrimination.