Principal component identification of variable single-trial evoked brain potentials

Current single-trial evoked potential estimators assume deterministic signal waveforms embedded in the background electroencephalographic brain activity. Identification of morphological changes of the evoked responses has been suggested, requiring however a skilled operator to predetermine the location of the variable components. Here, the authors propose an alternative approach for the identification of variable single-trial evoked potentials, based on reconstruction of the single responses from the principal components of the data correlation matrix and thus eliminating the requirement of a-priori knowledge of the variable component locations. The reconstruction performance is demonstrated via simulations and application to experimental evoked potential data.