The element-wise weighted total least squares problem

Download

PDF Download (318Kb)

Description/Abstract

A new technique for parameter estimation is considered in a linear measurement error model AX approx B, A = A0 + tilde A, B = B0 + tilde B, with row-wise independent and non-identically distributed measurement errors tilde A, tilde B. The total least squares method yields an inconsistent estimate of the parameter X in this case. We formulate a modified total least squares problem, called element-wise weighted total least squares, which provides a consistent estimator, and propose two iterative algorithms for its solution. A local convergence and the rate of convergence of the algorithms is discussed. As a computationally cheap initial approximation we use the generalized total least squares estimate.