Estimation of Evoked Fields Using a Time-Sequenced Adaptive Filter with the Modified P-Vector Algorithm.

摘要

This theses describes a time sequenced adaptive filter developed to estimate visually evoked fields (EF) using visually evoked potentials (EP). These non-stationary signals are buried in strong background noise. The two types of noise are magnetoencephalogram (MEG) and electronencephalogram (EEG). The filter implementation is based on the Ferrara Time Sequenced Adaptive (TSAF) using the Least-Mean-Square (LMS) algorithm and the Williams modified P-vector algorithm (mPa). This essentially results in two filters. A two stage filter structure is used in which the first stage removes the time-varying mean of the input signals. This allows the second stage to process zero-mean signals which increases the convergence speed of the filter. The theory for the two filters is overviewed with the input signals to the filters modelled as the sum of three uncorrelated components: average signal response, signal jitter, and noise. The signal model is verified based on a statistical analysis of simulated EP data files. The software implementation is then shown to be error free. (KR)