A method for non-parametric waveform estimation based on filter banks

A non-parametric estimation method based on a series expansion of a signal by a set of basis functions is presented. The method is used to estimate the waveform of event-related signals which have a transient behaviour with unknown shapes and time delays which vary randomly after a known event time. The quality of the estimate depends on the choice of the set of basis functions. Two different sets of basis functions are designed from a template, which contains a priori information, by using one uniform and one non-uniform bandpass filter bank, respectively. Each of the two sets is enlarged by a number of delayed copies of each basis function by employing FIR-filters, the coefficients of which are determined in terms of squared error. The method is evaluated and compared with the Karhunen-Loeve expansion