Baseline correction of overlapping event-related responses using a linear deconvolution technique

Abstract

The time it takes an event-related response (ERR) to subside is often longer than the interval between successive events so that the response to a new event interferes with a baseline formed by responses to preceding events. Thus, without proper baseline correction, the interpretation of an event-triggered average (ETA) of recorded data can be problematic. As the spectral compositions of ERR and baseline typically overlap, filtering the ETA is not always an adequate solution. The approach introduced here exploits that ETA and ERR are linearly related. Unless the series of events is exactly periodic, the ERR can be derived from the ETA by linear deconvolution. The performance of the method is illustrated with simulated examples as well as data from an auditory evoked field (AEF) study. It is also outlined how to handle experiments with two or more different events. Intriguing applications beyond the scope of baseline correction arise from the fact that the ERR is invariably estimated for a time window longer than the mean interval between successive events. The method may help, for example, to better understand the relationship between transient and steady-state responses or to delineate the component structure of a specific ERR.