Identification of systems with fast and slow dynamics using non-uniform sampling

This paper presents a new approach for modeling linear time invariant (LTI) discrete systems with fast and slow dynamics from its input-output data. The new method selects the `best' sample spacing among the output samples to best fit the model to the input-output data. The singular value decomposition (SVD) is used to find the `best' sample spacing to reduce the size of the data matrix in a way such that we catch both slow and fast system dynamics on one hand and still improve the numerical condition of the reduced matrix in order to increase the immunity of the parameter estimation problem against data noise.