Selecting the Smoothing Parameter for Estimation of Slowly Changing Evoked Potential Signals

Abstract

Brain evoked potential (EP) data consist of a true response ("signal") and random background activity ("noise"), which are observed over repeated stimulus presentations ("trials"). A signal that changes slowly from trial to trial can be estimated by smoothing across trials and over time within trials. We present a method for selecting the smoothing parameter by minimizing an estimate of the mean average squared error (MASE). We evaluate the performance of this method using simulated EP data, and apply the method to an example set of real flash evoked potentials.