New Performance Bounds in Signal Parameter Estimation under Mismatch

In this paper we develop a new upper bound for the mean square estimation error of a parameter that takes values on a bounded interval. The bound is based on the discretization of the region into a finite number of points, and the determination of the estimate by a maximum likelihood procedure. It is assumed that inaccurate versions of the true spectra are utilized in the implementation of the maximum likelihood estimator. As a special case, we develop new upper bounds to the performance of time delay estimation schemes