Introduction - If you have any usage issues, please Google them yourself
ack sobi_cp3
Second Order Blind Identification via Candecomp/Parafac order-3 (CP3)
to solve linear instantaneous Blind Source Separation (BSS).
sobi_cp3 allows to handle the exactly-determined, the over-determined and
several under-determined problems.
sobi_cp3 computes a set of K covariance matrices of the observed signals.
Then, the mixing matrix is estimated by joint-approximate-diagonalization (JAD)
of this set of matrices.
The JAD problem is solved by computing the CP3 decomposition of the third-order
tensor (3-way array) built by stacking the covariance matrices along the third
dimension.
It exploits the fact that JAD is a particular case of CP3 where the slices of
the tensor are symmetric, i.e., the mode-1 and mode-2 loading matrices
of the decomposition are identical.
Here the cp3 model is fitted by ignoring this symmetry, via an alternating
least squares (ALS) coupled with exact line search.
Thanks to powerful uniqueness properties of CP3, estimation of t