A Globally Convergent Matricial Algorithm for Multivariate Spectral Estimation
Federico Ramponi, Augusto Ferrante, Michele Pavon · IEEE Transactions on Automatic Control · 2009
In this paper, we first describe amatricialNewton-type algorithm designed to solve the multivariable spectrum approximation problem. We then prove itsglobalconvergence. Finally, we apply this approximation procedure tomultivariate spectral estimation, and test its effectiveness through simulation. Simulation shows that, in the case ofshort observation records, this method may provide a valid alternative to standard multivariable identification techniques such as Matlab's PEM and Matlab's N4SID.