Bounded error identification of time-varying parameters by RLS techniques

The performance of the recursive least squares algorithm with constant forgetting factor in the identification of time-varying parameters is studied in a stochastic framework. It is shown that the mean square tracking error keeps bounded if and only if the so-called covariance matrix of the algorithm is L/sup 1/-bounded. Then, a feasibility range for the forgetting factor is worked out in correspondence of which the covariance matrix (and therefore the tracking error) keeps bounded.<
>