Abstract : We propose a theoretical performance analysis for a class of reconstruction problems, formulated as coupled canonical polyadic decompositions of two low-resolution tensor observations. We study a particular case when all the modes of the tensors are coupled. Unlike the case of a single coupling constraint, a fully-coupled model requires nonlinear constraints in some estimation scenarios. Thus we introduce two probabilistic scenarios. For each scenario, we derive the constrained Cramér-Rao bounds for the parameters and for the mean-squared error of the reconstructed tensor. We show that with a carefully chosen initialization, the maximum likelihood estimators reach the bounds, even in challenging cases (low signal-to-noise ratio or large tensor rank).
https://hal.archives-ouvertes.fr/hal-03651874 Contributor : David BrieConnect in order to contact the contributor Submitted on : Tuesday, April 26, 2022 - 10:17:37 AM Last modification on : Monday, May 2, 2022 - 4:14:44 PM
File
Restricted access
To satisfy the distribution rights of the publisher, the document is embargoed
until : 2022-10-12
Clémence Prévost, Konstantin Usevich, Martin Haardt, Pierre Comon, David Brie. Constrained Cramér–Rao bounds for reconstruction problems formulated as coupled canonical polyadic decompositions. Signal Processing, Elsevier, 2022, 198, pp.108573. ⟨10.1016/j.sigpro.2022.108573⟩. ⟨hal-03651874⟩