Iterative methods for parameter estimation

Abstract

Starting with a least squares formulation of the parameter estimation problem, both fixed data and data-adaptive iterative algorithms are developed. We apply two new techniques, namely diagonal perturbation and multiple partitioning, to existing finite impulse response (FIR) and infinite impulse response (IIR) fixed data matrix splitting algorithms, resulting in improved performance. Also, we extend the fixed data algorithms to the data-adaptive case, and contrast them with FIR and IIR recursive least squares (RLS) algorithms. Computer simulations are used to evaluate the computational effectiveness of the new algorithms. We show the general rate of convergence for the algorithms, evaluate their ability to correctly represent the spectral components of simulated system frequency response in noise, and present system performance, when the order of the model is chosen to be larger than the known system order (over-modeling).