Matching of linear dynamic model of process based on known frequency characteristic using a non-quadratic measure of model error

The paper presents a new approach to derivation of a linear model representing an approximation of dynamic response determined for a model estimated within discrete time identification. The approximation is calculated using with new algorithm, that minimize the approximation error with application a non-quadratic performance index. The used performance index is quite close to the absolute norm of error, but has better convergence features. The recursive formula for derivation of the approximation model coefficients is derived and tested on two different examples of two dynamic processes.