Title :
Regularized fast recursive least squares algorithms
Author :
Houacine, Amrane
Author_Institution :
Inst. of Electron., Univ. of Sci. & Technol. of Algiers, Algeria
Abstract :
Chandrasekhar type factorization is used to develop new fast recursive least squares (FRLS) algorithms for finite memory filtering. Statistical priors are used to get a regularized solution which presents better numerical stability properties than that of the conventional least squares one. The algorithms presented have a unified matrix formulation, and their numerical complexity is related to the factorization rank and then depends on the a priori solution covariance matrix used. Simulation results are presented to illustrate the approach
Keywords :
filtering and prediction theory; least squares approximations; matrix algebra; Chandrasekhar type factorization; covariance matrix; factorization rank; fast recursive least squares algorithms; finite memory filtering; numerical complexity; numerical stability; regularized solution; simulation results; unified matrix formulation; Covariance matrix; Equations; Filtering algorithms; Least squares approximation; Least squares methods; Numerical stability; Recursive estimation;
Conference_Titel :
Acoustics, Speech, and Signal Processing, 1990. ICASSP-90., 1990 International Conference on
Conference_Location :
Albuquerque, NM
DOI :
10.1109/ICASSP.1990.115725
