| Full text | |
| Author(s): |
Moraes, Caroline P. A.
;
Saldanha, Juliana
;
Neves, Aline
;
Fantinato, Denis G.
;
Attux, Romis
;
Duarte, Leonardo Tomazeli
;
IEEE
Total Authors: 7
|
| Document type: | Journal article |
| Source: | 2021 IEEE STATISTICAL SIGNAL PROCESSING WORKSHOP (SSP); v. N/A, p. 5-pg., 2021-01-01. |
| Abstract | |
In the Blind Source Separation problem, Post-Nonlinear models are among the few nonlinear type of mixtures that may be separated by independent component analysis methods. However, such methods usually involve higher order statistics, neural networks or even metaheuristics. In this paper, we propose a simple separation method based on the gradient-descent approach, using elements of two classical second order statistic-based methods, AMUSE and SOBI. Adaptation is performed in two stages, the linear and the nonlinear one. Necessary conditions are that the sources be temporally colored and certain constraints on the separation structure be met. First results show a good performance of the proposed algorithm. (AU) | |
| FAPESP's process: | 20/01089-9 - Unsupervised signal separation: a study on the applicability of Generative Adversarial Networks and on nonlinear models based on the Choquet Integral |
| Grantee: | Leonardo Tomazeli Duarte |
| Support Opportunities: | Regular Research Grants |