Single epoch adaptive estimation of evoked potentials using the modified p-vector algorithm

Abstract

A new adaptive filtering algorithm (the modified P-vector algorithm) and special multistage filter structure was developed to resolve epoch-by-epoch variations in single epoch evoked responses. The evoked responses were first modeled as the sum of three signal components; a constant ensemble average (M) across all epochs, noise (N), and an epoch-by-epoch stochastic signal variation (Q). A two stage time sequenced adaptive filter structure decouples the M and Q components of each new signal vector. The result is improved convergence performance. The modified P-vector algorithm (mPa) was developed to eliminate the need for a separate desired signal. As a result, the filter input can also be used as the desired or training signal. The mPa adaptive filter was tested using simulated and human data sets. The mPa filter was able to resolve signal variations on an epoch-by-epoch basis.<>