Order estimation in the context of MOESP subspace identification methods

Subspace algorithms are used for the identification of linear, time invariant, finite dimensional, discrete time state space systems. In the identification the order of the estimated system has to be specified. This paper provides a criterion for choosing the order. The asymptotic properties of the examined procedure are derived under appropriate assumptions on the data generating process. It is shown, that the criterion leads to asymptotically consistent estimates of the order only in a generic case. In a simulation study the effects of some user choices on the performance of the order estimates is demonstrated.