New results on stationary stochastic feedback processes

Considers stationary stochastic discrete-time vector processes made up of two component processes y and u, such that the joint (y,u)-process has a rational spectral density phi /sub yu/(z). Such processes can be represented by a white noise driven matrix transfer function model, and (in most cases) by a closed-loop model. A number of new results are presented on the connections between these two representations, and on their properties: stability, invertibility, identifiability, uniqueness of the spectral factorization, detection of feedback, and continuity of spectral factors.Anglai