Model order, convergence rates and information content in noisy partial realizations

It is shown that convergence rates of recursive algorithms for parameter estimation from noisy partial realizations depend on the structure of the chosen model. The model is analyzed by considering information unique to the parameterization-the sensitivity and interconnectedness of the model parameters. The convergence analysis is independent of model order. Useful information about relative convergence rates can be inferred even when the model order does not match that of the identified system