MIMO system identification using common denominator and numerators with known degrees

摘要

In system identification, prior knowledge about the model structure may be available. However, imposing this structure on the identified model may be nontrivial. A new discrete-time linear time-invariant identification method is presented in the article that imposes prior knowledge of the degree of the common denominator of the system's transfer function matrix and the degrees of the numerators. First, a method is outlined for the solution in case of exact data. Then, this method is extended for noisy data in the output error setting. An initial estimate obtained by a subspace method is improved by a structured low-rank approximation method. The performance of the method imposing the structure is compared on simulated data with the performance of classical identification methods that do not impose the structure.