Data Domain Method
Recall the optimum linear filter solves Cw=s
The straightforward approach to solving for w is to invert the NxN covariance matrix C. This is computationally expensive. An “orthogonal triangular decomposition” reduces the cost.
The “QR” decomposition of the sample data matrix X (MxN) provides X = QR, where Q is an NxN unitary matrix (QHQ=I) and R is an upper triangular matrix. Thus:
Now it’s easier (fewer ops) to solve for w: