Abstract
A stochastic subspace identification method has been developed based on a “block LQ decomposition” [1] with the help of stochastic realization theories. The algorithm, however, does not guarantee that an estimate of the forward innovation representation is stable and of minimum phase. In order to overcome this difficulty, this paper discusses a method of estimating the spectral density function of the output process, and present a prototype of a subspace identification algorithm guaranteeing minimum phase via a spectral factorization technique.
