On exponential convergence and robustness of least-squares identification (Q1088967)

This paper considers the exponentially-weighted least-squares (EWLS) estimation of the parameters in linear regression models. Its main results are: a) In the noise-free case the exponential convergence of the EWLS estimator is shown under an extended persistency-of-excitation condition; b) The exponential convergence in the noise-free case does not necessarily imply convergence in (bounded) noise; the basic assumption under which this non-robustness result is shown is that the regression vector is not persistently exciting (in the standard sense).