“…Designed in this manner, the linear filter h is known as the Hotelling observer, or pre-whitening matched filter 11,48 . Ideally one would desire an infinite amount of data in order to calculate the exact covariance matrix K. Its estimation using a limited amount of data is a classical signal processing problem, conventionally solved with help of information-theoretic dimension reduction criteria like AIC, MDL or EDC 49,50 . In our tests, the EDC2 criterion 50 was the most robust (data not shown), and we therefore used it to estimate the reduced-dimension covariance matrix, and then to invert the matrix.…”