Frequency domain identification with fixed zeros: first generation seismic-AMD benchmark

The frequency domain linear multi-variable system identification has been extensively studied recently. Many of the existing algorithms assume no prior knowledge of the underlying system structure, e.g. zero and pole locations that are pre-determined (or fixed). However such information, if available, helps to reduce the error in the interesting frequency range and the number of the model parameters to be estimated. In this paper, an algorithm that explicitly includes the fixed zeros is developed using the polynomial matrix curve-fitting technique. A simple and effective method is introduced to constrain the identified model to be stable. And an application study is presented to demonstrate the effectiveness of the identification algorithm.