Effects of correlation matrix estimation on correlated source cancellation in beamformers

We examine the effects of using different methods to estimate the spatial correlation matrix of multi-epoch MEG time series data on the correlated source cancellation phenomenon in linearly constrained minimum variance (LCMV) beamforming. The method used to estimate the correlation matrix impacts the level of effective correlation between sources that is seen by the beamformer and the consequent performance. We present an analysis of correlated source cancellation based on a general framework for estimating the spatial correlation matrix. Our analysis is exact at all signal to noise ratios and is used to compare the performance obtained using different correlation matrix estimates. Thus, the analysis provides guidance on the selection of a correlation matrix estimation strategy. For example, over certain ranges of signal to noise ratio, we show that averaging across epochs before estimating the correlation matrix can create significant correlated source cancellation relative to estimating the correlation matrix from the single trial data. Simulations confirm the conclusions of the analysis.