Blind Source Separation: Alternative approaches to Independent Component Analysis

• Grant number: 08/00002-5
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Effective date (Start): April 01, 2008
• Effective date (End): February 28, 2010
• Field of knowledge: Engineering - Electrical Engineering - Telecommunications
• Principal Investigator: João Marcos Travassos Romano
• Grantee: Ricardo Suyama
• Host Institution: Faculdade de Engenharia Elétrica e de Computação (FEEC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil

Abstract

The problem of Blind Source separation, introduced by Hérault and Jutten as a non-supervised learning strategy for neural networks, has attracted attention since the beginning of the 90's, and can be considered a mature research topic in the modern theory of signal processing.The great interest on blind source separation is due to the fact that several distinct applications can benefit from its framework, like applications in voice and speech signal processing, telecommunications, biomedical signal processing, among others.There is a significant number of solutions to the problem when linear mixing models are considered, in particular the case in which the number of sensors is greater than the number of sources. In such case, the independent component analysis can be used to obtain the separating system. However, there still only a few techniques to deal with the other cases. Moreover, the influence of noise is still not well understood, and in general it is assumed that there are no noise in the measurements.Therefore, the main goal of this project is to continue the work developed in [Suyama2007], investigating solutions to the problem of blind source separation in scenarios that could not be handled, in general, by the independent component analysis. In particular, we´ll be looking for original solutions for convolutive mixing models, extending the framework proposed in [Suyama2007b], and also to problem of underdetermined mixtures, in which there are more sources than sensors. For this second case, additional information about the sources will be taken into account, in order to overcome the fact that the mixture is no longer linearly invertible.