Rapid continuous-time identification of linear and nonlinear systems using modulation function approaches

Abstract

At the present, system identification through modulation functions has a wide range of methods. Many of them have reached maturity levels that surpass customary Kalmanfilter approaches for discrete-time identification. In this thesis, the modulation function technique is analyzed in view of its real-time capability, as well as the possible unification of the modulation function methods based on the frequency spectrum, and ability to deal with nonlinearities. Besides, to increase the rate of convergence, the optimal parameter estimation with constraints of Byrski et al. [BFN03] is applied on integrable and convolvable systems. Furthermore, the modulated white Gaussian noise influence on linear systems is examined. The proposed methods together with the Loab-Cahen modulation functions are compared in performance for linear and convolvable systems concerning three different inputs, three normalizations, identification parameters and computational cost.