Approximation of generalized frequency response functions via vector fitting

This paper describes a method for data-driven approximation of nonlinear systems, using the notion of generalized frequency response functions as provided by the Volterra series theory. Using rational approximation, frequency-domain input-output data are fitted to a bilinear model structure. The proposed algorithm performs optimization of model coefficients through a greedy approach that relies on linear least-squares solves only, thus ensuring speed and scalability.