Subspace Identification of Pure Stochastic Systems

In this paper we treat the subspace identification of pure stochastic systems with no external input. The stochastic identification problem consists of computing the stochastic system matrices from given output data only. We show how this can be done using geometric operations as orthogonal projections. Implementation of the subspace identification algorithm for stochastic systems has been discussed in terms of the numerically stable and efficient singular value decomposition. The proposed algorithm is tested on a simple example simulated with a Monte Carlo experiment