Single trial analysis of event related potentials sensitivity of the subspace regularization based methods to the value of regularization parameter

The effect of the regularization parameter to the single-trial estimates obtained with two subspace regularization based estimation methods is studied. The first method relies on the single channel data. The second method is an extension to the multi channel case. It takes the spatial correlation of the channels into account on the estimation. Methods are compared using simulated data. Simulations are based on a combination of the real measured background EEG and simulated scalp potentials caused by two time-varying dipoles. Both methods are shown to be quite robust to non optimal selection of the regularization parameter. It is also shown that the ability of the multi channel method to preserve amplitude differences between channels is not due to selection of the regularization parameter.