Identification of multi-input multi-output systems by observability range space extraction

The observability range space extraction system identification technique, a time-domain technique for state-space model identification of linear multi-input multi-output systems, is discussed. It extracts the base vectors of the observability range space of a linear system from a semipositive definite data matrix and obtains a state-space model from these vectors. Input excitation conditions which are required by this technique to obtain a transfer function equivalent model are discussed. Identification errors due to measurement noises are analyzed, and the conditions for transfer-function equivalent identification in the presence of measurement noises are derived. Simulation results show that this technique is able to produce relatively accurate models in a colored noise environment and under low signal-to-noise-ratio conditions