Combinations of adaptive filters.

• Ano de defesa: 2015
• Autor(a) principal: Chamon, Luiz Fernando de Oliveira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Adaptive filtering; Algoritmos; Coefficients feedback; Combinação de filtros adaptativos; Combination of adaptive filters; Filtragem adaptativa; Incremental topology; Processamento de sinais; Processamento de sinais adaptativos; Realimentação de coeficientes; Topologia incremental
• Link de acesso: http://www.teses.usp.br/teses/disponiveis/3/3142/tde-14072016-143633/

Adaptive filtering has grown to become a fundamental topic in signal processing, increasingly attracting attention from the community. Important factors in this popularization were their low computational complexity and model-free nature, adapting even to nonstationary characteristics of the systems and/or signals under study. Nevertheless, many adaptive algorithms introduce trade-offs, for instance, between convergence rate, nonstationary signals tracking, and steady-state error, which can hinder their use in practical applications. Furthermore, some adaptive filters can become unstable when word length is reduced and/or the input data are highly correlated. Recently, combination of adaptive filters was put forward as a solution for such issues. This approach consists in combining a pool of filters by means of a supervisor that attempts to make the overall system at least as good (usually in the mean-square sense) as the best filter in the set. Examples of these structures have been shown to successfully solve this problem, although well-known limitations remain to be addressed. Moreover, due to the relative novelty of this topic, developments in combination of adaptive filters are difficult to accommodate into a common theoretical framework. This work studies combination of adaptive filters and addresses the aforementioned issue by (i) classifying the existing combinations and proposing a taxonomy that exposes the similarities and differences in their forms; (ii) proposing new combinations; (iii) devising a general framework for studying combinations of adaptive filters and using such framework in performance analyses.