Under-Determined Tensor Diagonalization for Decomposition of Difficult Tensors

0483430 - ÚTIA 2018 RIV US eng C - Conference Paper (international conference)
Tichavský, Petr - Phan, A. H. - Cichocki, A.
Under-Determined Tensor Diagonalization for Decomposition of Difficult Tensors.
IEEE 7th International Workshop on Computational Advances in Multi-Sensor Adaptive Processing (CAMSAP) 2017. Piscataway: IEEE, 2017, s. 263-266. ISBN 978-1-5386-1250-7.
[CAMSAP 2017 - 7th IEEE International Workshop on Computational Advances in Multi-Sensor Adaptive Processing. Curacao (NL), 10.12.2017-13.12.2017]
R&D Projects: GA ČR GA17-00902S
Institutional support: RVO:67985556
Keywords : canonical polyadic decomposition * tensor decomposition * matrix multiplication
OECD category: Statistics and probability
http://library.utia.cas.cz/separaty/2017/SI/tichavsky-0483430.pdf

This paper deals with the Cramer-Rao Lower Bound (CRLB) for a novel blind source separation method called Independent Component Extraction (ICE). Compared to Independent Component Analysis (ICA), ICE aims to extract only one independent signal from a linear mixture. The target signal is assumed to be non-Gaussian, while the other signals, which are not separated, are modeled as a Gaussian mixture. A CRLBinduced Bound (CRIB) for Interference-to-Signal Ratio (ISR)
is derived. Numerical simulations compare the CRIB with the performance of an ICA and an ICE algorithm. The results show good agreement between the theory and the empirical results.
Permanent Link: http://hdl.handle.net/11104/0278758