Model order identification using an auxiliary system

This paper introduces an alternative approach to model-order testing for single-input/single-output system identification from input-output data. The essence of the technique is to adjoin a known system to the normal parameter estimation equations and assess the suitability of the model order of the unknown system through the behaviour of the estimated auxiliary system parameters. In this way the identification of system order is embedded in an equivalent parameter estimation problem. The main advantage of this procedure vis-a-vis many existing order testing algorithms, is that it does not require statistical information concerning the quality of parameter estimates. An important consequence of this is that the technique can be used with practically any parameter estimation or fitting algorithm. In particular, it allows the elegance of least-square parameter fitting to be brought to bear on the order-determination problem.