Adaptive RLS algorithms under stochastic excitation-L/sup 2/ convergence analysis

A very general class of RLS (recursive least squares) algorithms having a forgetting factor is considered. The basic assumptions are that the data generation mechanism is free of disturbances and that the observation vector is a stochastic process satisfying a phi -mixing condition. A stochastic characterization of persistent excitation is first given. Then, it is proved that the algorithm is exponentially convergent in the mean-square sense.<
>