Cognitive identification of systems using nonlinear dynamics

In this paper, we address the problem of semi-blind parameter estimation of linear systems driven by chaotic signal. We propose an on-line expectation maximization (EM) based extended fixed-lag Kalman smoothing (EM-EFLKS) estimator to simultaneously estimate parameters of system along with the chaotic signal. The performance of the proposed estimator is compared to its offline counterpart, EM-EKS estimator, which achieves ML estimation performance but requires an entire sequence of observation to be processed backwards. Both theoretical analysis and numerical evaluations show that the EM-EFLKS is superior to EM-EKS and achieves the performance of non-blind estimator asymptotically.