
300
S. Hosur et al.
If the input vector xn is transformed to un = ETxn, where E is the unitary eigenvector
matrix of ttx, then the output process zn would be de-correlated. However, this implies
that we need to perform an eigen decomposition of the autocorrelation matrix or a singular
value decomposition of the data matrix at every adaptation, implying a computational
computational complexity of O(N 3) for every adaptation. One could replace E with V,
where V is any unitary matrix which block diagonalizes Rx, separating the signal subspaces
from the noise subspace, but this still takes O(N ...