Abstract
In this paper a new system identification method is proposed. This method is based on representing an unknown system by a model with complex coefficients. Much variety of methods have been known which can estimate the parameters of a linear discrete-time system from the input-output records. In their methods usually the real coefficient models are assumed for the unknown system.
In the proposed method, we transform the real input-output records into the analytic signals, which are the complex signals that have no negative frequency components. And then we estimate the complex coefficient model by using the analytic signals. The model with complex coefficients have the meaning only for the positive frequency characteristic of the unknown system and the model order is reduced by a factor of two.
When the complex modeling method is applied to the instrumental variable approaches, which have been often used for the system identification, the obtained complex instrumental variable method requires slightly more computational burden than the conventional one, but considerably improves the estimation accuracy. This fact is verified by several computer simulations.
