A weighted total least squares estimator for multivariable systems with nearly maximum likelihood properties

The aim of the present paper is to develop a parametric estimator for linear time-invariant multivariable systems with nearly maximum likelihood properties. The estimator is based on the total least squares (TLS) method. It can be seen as an “optimally” weighted iterative generalized total least squares (GTLS) estimator, combining the nice asymptotic properties of the maximum likelihood (ML) method with the global minimization property of the GTLS estimator. The main difference with, for instance, the IQML method and the method of Sanathanan and Koerner is that it generates consistent estimates in each iteration step