Banca de DEFESA: ALEXANDRE MICCHELETI LUCENA

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
DISCENTE : ALEXANDRE MICCHELETI LUCENA
Data: 27/05/2026
HORA: 09:00
LOCAL: Remoto (https://conferenciaweb.rnp.br/sala/ricardo-33)
TÍTULO:

On the Nonlinear Blind Source Separation: Exploring Nonlinear Mixtures as Time-Varying Linear Systems

 

PÁGINAS: 88
RESUMO:

Nonlinear blind source separation is a challenging area of signal processing, extending beyond the well-established linear paradigm. Current methodologies include specialized nonlinear models and, more recently, local linear approximation techniques that exploit temporal information. Initial investigations into these approximation methods suggested a potential research direction: the reinterpretation of certain nonlinear mixtures as time-varying linear systems. This possibility motivates further exploration into whether such a perspective could simplify separation and relax existing constraints, particularly for structured nonlinear models like Linear Quadratic mixtures. This work investigates the hypothesis that specific nonlinear, time-invariant mixture models can be effectively treated as equivalent linear time-varying systems. Building upon prior research on local approximations, we explore an alternative pathway that circumvents the dependency on derivative-based computations. The investigation specifically examines this reinterpretation for the Linear Quadratic model, assessing its transformation into a time-varying framework and the resulting implications for separation algorithms. Methodology proceeds in two phases: first, a nonlinear model investigation evaluates the approach using sinusoidal and Autoregressive sources under different nonlinear mappings; second, the LQ model is explicitly reformulated as a time-varying system. Performance is quantified using Signal-to-Interference Ratio across comprehensive synthetic scenarios, establishing conditions for viable source recovery and evaluating algorithmic compatibility. Analysis confirms separation feasibility when leveraging the time-varying interpretation. For the initial nonlinear models tested, the smoothness criterion was successfully relaxed to accommodate AR(1) sources while maintaining performance. The reformulation demonstrated consistent compatibility with adaptive ICA algorithms, providing a viable alternative to direct nonlinear separation approaches. This reinterpretation offers theoretical insights for connecting BSS methodologies and presents practical advantages for specific nonlinear scenarios. It suggests promising directions for developing alternative separation strategies for challenging nonlinear mixing models.

MEMBROS DA BANCA:
Presidente - Interno ao Programa - 1761107 - RICARDO SUYAMA
Membro Titular - Examinador(a) Interno ao Programa - 1544392 - ALINE DE OLIVEIRA NEVES PANAZIO
Membro Titular - Examinador(a) Interno ao Programa - 1761105 - MURILO BELLEZONI LOIOLA
Membro Titular - Examinador(a) Externo à Instituição - LEONARDO TOMAZELI DUARTE - UNICAMP
Membro Titular - Examinador(a) Externo à Instituição - DIEGO BARRETO HADDAD - CEFET/RJ
Membro Suplente - Examinador(a) Interno ao Programa - 1946319 - DIOGO COUTINHO SORIANO
Membro Suplente - Examinador(a) Externo à Instituição - MARCIO EISENCRAFT - USP