Dual partitioning and subspace decomposition in neuro physiological signal processing

The authors propose a methodology to extract the evoked potential from noise-corrupted data. This is achieved by projecting the observation vector onto the signal subspace of a symmetric cross-correlation matrix derived using dual partitioning of the observation vector. It is shown that this technique is asymptotically independent of the signal-to-noise ratio. Compared to the existing subspace approaches, this technique does not require the assumption that the variances of the noise in different realizations are equal.<>