Investigating the relationship between Canonical Correlation Analysis (CCA), and Signal Fraction Analysis (SFA) in signal separation

摘要

In this paper, we study Maximum Noise Fraction (MNF) and Canonical Correlation Analysis (CCA), we generalize the concept of MNF to Signal Fraction Analysis (SFA), and we theoretically compare SFA with CCA. CCA is an approach for measuring the linear relationship between two multidimensional data sets. It finds two bases, one for each data set those which are optimal with respect to the correlation between the two variables in the new coordinates. The paper confirms that theoretically CCA and SFA are very similar methods. We run toy examples showing that CCA is capable of separating mixed signals.