A stabilizing controller for jump linear Gaussian systems with noisy state observations (Q2564018): Difference between revisions

The authors consider linear, discrete-time systems with Gaussian state and observation noise, for which all the system matrices depend on an underlying, finite state Markov chain with transition probability matrix \(\pi=(p_{ij})\). Such a system is called mean square stabilizable if there exists an (appropriately measurable) input sequence such that the second moment of the resulting closed loop state response is bounded. In [the authors, Proc. 32th CDC, 375 (1993)], a linear stabilizing feedback \(L^+\) was proposed for systems with fully observable state using an iid background noise with distribution \(p^+_j=\max_i p_{ij}\); here \(L^+\) is obtained via an appropriate Riccati equation. This paper contains two main results: The linear feedback \(L^+\) stabilizes the partially observed system if the state estimator (for the closed-loop system with \(L^+\)) has bounded estimate covariance and if the fully observed system is stabilizable. Furthermore, \(L^+\) is robust in the sense that it is still stabilizing if the state equation contains an uncertain term with linear growth of slope less than one. This latter result is proved for the fully observed system. Both results are proved by exploiting the slack in the Riccati equation associated with \(L^+\). Two two-dimensional examples illustrate the results. (Note that about half of the references are cited with wrong numbers).