Title :
Recursive local orthogonality filtering
Author :
Bodenschatz, John S. ; Nikias, Chrysostomos L.
Author_Institution :
Dept. of Electr. Eng., Univ. of Southern California, Los Angeles, CA, USA
fDate :
9/1/1997 12:00:00 AM
Abstract :
Recursive local orthogonality (RLO) is a stochastic Newton algorithm to achieve a condition we term local orthogonality, which only requires unimodal symmetric densities with continuous nonzero second derivatives near the origin. For Gaussian systems, RLO reduces to recursive least squares. Local orthogonality is both a subset of median orthogonality and a form of constrained maximum-likelihood optimization. Fast multichannel time and multichannel frequency domain implementations are given. Simulations show the utility for system identification and inverse modeling
Keywords :
Gaussian processes; Newton method; frequency-domain analysis; identification; inverse problems; least squares approximations; maximum likelihood estimation; optimisation; recursive filters; time-domain analysis; Gaussian systems; RLO; constrained maximum-likelihood optimization; continuous nonzero second derivatives; inverse modeling; median orthogonality; multichannel frequency domain implementation; recursive least squares; recursive local orthogonality filtering; stochastic Newton algorithm; system identification; time domain implementation; unimodal symmetric densities; Equations; Filtering; Finite impulse response filter; Frequency domain analysis; Inverse problems; Least squares methods; Stochastic processes; System identification; Transversal filters; Working environment noise;
Journal_Title :
Signal Processing, IEEE Transactions on
