Full text
| |
| Author(s): |
Renan Del Buono Brotto
Total Authors: 1
|
| Document type: | Master's Dissertation |
| Press: | Campinas, SP. |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Elétrica e de Computação |
| Defense date: | 2019-08-15 |
| Examining board members: |
João Marcos Travassos Romano;
Charles Casimiro Cavalcante;
Rafael Ferrari
|
| Advisor: | Kenji Nose Filho; João Marcos Travassos Romano |
| Abstract | |
This dissertation deals with the recovering of information from the perspective of signal processing concepts. We focus our studies in two classical problems of the area: unsupervised deconvolution and blind source separation. We aim to retrieve the desired information applying lp norms based-criteria, which allow us to explore sparse (p=1) and antisparse (p equal to infinity) features of the involved signals. We justify the relation between the value of p and the property of interest by estabilishing a relationship between the lp norms and the Maximum Likelihood estimator for generalized Gaussian distributions. We approach the deconvolution problem by means of the lp prediction error filter, with p different from 2, in channels with minimum and maximum phase response. The proposed method shows a superior performance when compared with the classical approach, allowing nonminimum phase response for the filters. The lp PEF, with p different from 2, are also able to provide nonlinear uncorrelated prediction error signals. We also apply the lp norms in blind source separation, proving that the antisparsity can be used as prior information for the task. The proposed separation method is composed of two steps: first, as a pre-processing step, we apply a whitenning filter on the mixtures; then we apply a rotation on the uncorrelated mixtures, completing the separation. The rotation matrix is adjusted by the l1 norm, when we are interested in sparse sources, and by the infinity norm when dealing with antisparse signals. In our simulations, we were able to recover up to 10 sources and the proposed method presented a good performance in noisy scenarios (AU) | |
| FAPESP's process: | 17/13025-2 - About the use of Lp norms in the problems of unsupervised deconvolution and blind source separation |
| Grantee: | Renan Del Buono Brotto |
| Support Opportunities: | Scholarships in Brazil - Master |