Improved estimation performance using known linear constraints (Q705143)

The paper deals with an optimization problem that is often referred to as a nonlinear-least-squares problem in signal processing and system identification. This type of problem appears when the underlying dynamics is described by a system of differential and difference equations. The unknown boundary (or initial) conditions appear as nuisance parameters. For the class of problems that is addressed here, the boundary conditions are a priori known to satisfy a few additional linear constraints. The goal is to develop a procedure to use these additional constraints and thereby achieve better accuracy of the parameter estimates. Asymptotic accuracy expressions are derived and the approach, which exploits the constants, yields better estimates. This approach is applied in the identification of parameters of viscoelastic materials and in wave propagation experiments. Some results from numerical simulations are presented.