New methods and applications in signal separation

• Grant number: 12/01546-4
• Support Opportunities: Regular Research Grants
• Start date: August 01, 2012
• End date: July 31, 2014
• Field of knowledge: Engineering - Electrical Engineering
• Principal Investigator: Leonardo Tomazeli Duarte
• Grantee: Leonardo Tomazeli Duarte
• Host Institution: Faculdade de Ciências Aplicadas (FCA). Universidade Estadual de Campinas (UNICAMP). Limeira , SP, Brazil

Abstract

In Blind Source Separation (BSS), the goal is to retrieve a set of signals (sources) by only considering observations that correspond to mixed versions of the sources. BSS problems can be found in many practical applications, such as biomedical signal analysis and audio signal processing. Classically, BSS methods consider a linear mixing model and that the sources can be modeled as independent random variables. However, since these assumptions are not always verified in actual situations, there is currently a great interest in developing methods that go beyond this classical paradigm. This research project pursues this goal. More precisely, we shall investigate new solutions that are able to perform source separation in nonlinear models and scenarios in which the sources are dependent. In order to achieve this goal, our study will exploit other types of prior information such as the fact that the sources are sparse when represented by a given basis. We will also focus on alternatives formulations of the BSS problem using linear additive models. In this line of work, we intend to analyze the application of methods based on the so-called robust principal component analysis, which has been currently gaining considerable attention in the signal processing community. Finally, in order to make use of different types of prior information at the same time, we shall investigate the formulation of the BSS problem by considering a multi-objective optimization approach. Throughout the project, our research will be guided by important applications in the areas of chemical sensing, statistical process control, geophysical signal processing, and image processing. (AU)