Identification of time-varying systems using a two-dimensional B-spline algorithm

This paper presents a new method which non-parametrically estimates a two dimensional impulse response function h_LTV(t, τ) of slowly time-varying systems. A generalized B-spline technique is used for double smoothing: once over the different excitation times τ (which refers to the system memory) and once over the actual excitation time t (referring to the system behavior). If the change of the parameters of the observed system is sufficiently slow, with respect to the system dynamics, we will be able to 1) reduce the disturbing noise by additional smoothing 2) reduce the number of model parameters that need to be stored.