On-line model order determination and parameter estimation

Abstract

In this paper an algorithm is proposed to compute recursively the model order testing criteria and parameter estimates simultaneously for K nested models which have similar structures but with increasing complexity where K is the highest model order being considered. The algorithms exploit certain properties of the LDLT factorization of the augmented covariance matrix corresponding to the parameters, where D is a diagonal matrix and L is a lower triangular matrix with unit diagonal elements. It will be shown that the parameters corresponding to the different model orders are contained in the L factor while several of the well established model order testing criteria for different models can readily be computed from the D factor.