<i>H</i>_∞ estimation for discrete‐time linear uncertain systems (Q3987690)

Given a prescribed level of noise attenuation \(\gamma>0\), the authors show sufficient conditions for designing an estimator \(\{x_ e,z_ e\}\) of the system with uncertainties \[ \Delta: x_{k+1}=(A+\Delta_ 1(k))x_ k+Bw_ k, y_ k=(C+\Delta_ 2(k))x_ k+Dw_ k, z_ k=Lx_ k, \] \[ [\Delta_ 1,\Delta_ 2]^ T=[H_ 1,H_ 2]^ T F(k), F^ T(k)F(k)=I, \] such that the augmented of the system and the estimator is quadratically stable and \(\|{\mathcal G}\|<\gamma\) under zero initial conditions for \(x(k)\) and \(x_ e(k)\), where \(\|{\mathcal G}\|\) is the induced operator norm of the mapping \(\mathcal G\) from the noise, \(w\), to the estimation error, \(e=\{z(k)-z_ e(k)\}\), and the inequality holds for all the mentioned \(F(k)\). In the case when \(A\) is stable and invertible, \((C,A)\) is detectable and \(DD^ T>0\), a solution to such a robust \(H_ \infty\) estimation problem has been given in terms of two algebraic Riccati equations. It is similar to the authors' results for continuous-time case [\(H_ \infty\) estimation for continuous-time linear uncertain systems, Proc. IFAC symp. on design methods of control systems (1991, to appear)]. The Riccati equation approach allows them to solve a \(H_ \infty\) estimation problem associated with a linear certain system with the parameters \(\gamma/\varepsilon\cdot H_ 1\), \(\gamma/\varepsilon\cdot H_ 2\), \(\varepsilon>0\), and then the main result is obtained by converting the robust \(H_ \infty\) estimation problem to solving a discrete-time co-spectral factorization with a suitable state-space realization and the standard \(H_ \infty\) estimation problem.