-
Daiana Caroline dos Santos Gomes
Universidade Federal do Maranhão, Av. dos Portugueses, 1966, Bacanga, São Luís, Maranhão
-
Ginalber Luiz de Oliveira Serra
Instituto Federal de Educação, Ciência e Tecnologia do Maranhão, Av. Getúlio Vargas, 04, Monte Castelo, São Luís, Maranhão
Keywords:
Intelligent systems, Fuzzy inference system, Adaptive Filters, Normalized Least Mean Square Algorithm, Hypercomplex domain
Abstract
In this paper, a fuzzy descent method applied to adaptive filtering algorithms in the hypercomplex domain is proposed. The adopted methodology consists of Mamdani fuzzy inference system design, in the hypercomplex domain, applied to the computation of adaptation step size to be adopted for the convergence of the adaptive filtering algorithm. The partitioning of the universe of discourse regarding to the components of input and output hypercomplex variables of fuzzy inference system, to define the rule base, is performed by the expert’s knowledge about the application. For validation purposes, the proposed methodology was applied to a Normalized Least Mean Square (NLMS) filter variation, which is based on the minimization of cost function of the Recursive Least Square (RLS) algorithm. Computational results and comparative analysis illustrates the efficiency and applicability of the proposed methodology, providing lower computational cost and lower final misfit for estimating the optimal weights of the adaptive filter.
