A.A. Lomov (Russia)
Identification, Identifiability, Linear Systems
1 We consider the identifiability of time-invariant parametric models used in transient signal processing on short obser vation intervals. The utilized nonlinear least-squares iden tification scheme is based on joint signals and parameters estimation of a linear model of the phenomena. This ap proach is related to errors-in-variables identification prob lem, being extended to models with block-Toeplitz system matrices. In the report, the simple finitely verifiable suffi cient conditions for parameter identifiability of stochastic and deterministic linear models are given. For ”proper” model parameterizations it is found that (a) the identifia bility conditions become sufficient and necessary, (b) local identifiability implies the global one, (c) the identifiability property is generic, in the sense that if the model is iden tifiable at a parameter point, then it is identifiable almost everywhere.
Important Links:
Go Back
