Abstract
The procedure of averaging generally applied for noise reduction of evoked potentials can probably be improved by weighting the sweeps in dependence on their noise energy, which fluctuates considerably. This article demonstrates that weighting factors must be free of the signal part in the sweeps if underestimation and distortion of the reconstructed evoked potential are to be avoided. This can be achieved by pairing the sweeps and weighting the sum of a pair by the reciprocal sample variance from its difference. In order to enhance precision in variance determination, pre-whitening the data has been suggested (Wernicke, 1992). In a feasibility study, this technique is compared with several other algorithms known from the literature.