Recursive identification for Wiener-Hammerstein systems with non-Gaussian input

In this paper an identification method is discussed that deals with the Wiener-Hammerstein systems, in which ARX dynamics, non-invertible general static nonlinear function, and non-Gaussian inputs are admitted. By introducing a suitable instrumental variable a new algorithm is presented to recursively estimate the linear subsystems using stochastic approximation algorithm. Based on the kernel method the nonlinear function is estimated recursively. The proposed estimates are proved to be consistent under mild condition. A simulation example is provided justifying this method.