Sidestepping the positive real condition in RELS via multiple RLS identifiers

Abstract

A recursive ARMA identifier based on multi-step prediction error minimization is considered so as to sidestep the positive real condition of recursive extended least squares (RELS). For a single step prediction horizon the algorithm reduces to RELS while, as is shown by local convergence analysis, it tends to be a local approximation of the recursive maximum likelihood (RML) algorithm as the prediction horizon increases.