Consistency of a Class of Robust Estimators of Crosscorrelation.
JOHNS HOPKINS UNIV BALTIMORE MD
Pagination or Media Count:
In a previous work, a class of non-linear recursive estimators of cross-correlation was shown through simulations to be robust in identifying the parameters of linear, stationary, single-input-single-output systems whose output measurements are contaminated by noise which is not completely specified the measurement noise distribution f is given by F1-epsilon K epsilon C, where K is completely known and C belongs to the class of zero-mean, symmetric, finite variance distributions. The principal result of this report is a complement to these earlier results -- namely, a proof of consistency, with mean square convergence, for a general sub-class of the non-linear estimators of crosscorrelation defined in the earlier work. The proof is along the same lines as those followed in the Robbins-Monro stochastic approximation method.
- Statistics and Probability