Subspace algorithms for system identification and stochastic realization

The subspace approach for linear realization and identiication problems is a promising alternative for the 'classical' identiication methods. It has advantages with respect to structure determination and parametrization of linear models, is computationally simple and numerically robust. A summary is given of existing techniques based upon the singular value and qr decomposition. Algebraic, geometric, statistical and numerical points are emphasized. A new idea is outlined for the joint stochastic realization-deterministic identiication problem. Several examples are given.