The paper presents methods of training brain-computer interfaces (BCI) firmware to select a subset of electrodes from an aggregate, so that this subset forms the best signal/noise ratio for the subsequent spatial and temporal signal accumulation. These methods are proposed for three kinds of synchronous BCI: based on visual evoked potentials, sustainable visual evoked potentials and cognitive evoked potentials with P300 component. An algorithm offered to construct wavelet transformations of a specialized filter for evaluation of evoked potentials on the basis of analysis of the chains of local maxima and minima in the matrix of squares of the coefficients. The proposed approach allows to significantly simplify training system for BCI in case of position change of the recording electrodes, and, thereby, to increase the functionality of synchronous BCI.