Guaranteed accuracy of linear nonstationary systems in the presence of perturbing processes (Q1921553)

The paper deals with the problem of calculating the guaranteed accuracy of a linear discrete-time system \[ X(n+1) = LX(n) + BU(n) + \Sigma\xi(n+1),\;\;X(n_0) = X_0, \] \[ V(n) = E X(n), \] where \(X\) is the state vector, \(\xi\) is the vector standard discrete white noise, \(U\) is the perturbation process and \(V\) is the scalar output. The vector perturbing process \(U\) is not completely known but has the property: the components \(U_j(n)\) are mutually noncorrelated and are not correlated also with the system noise \(\Sigma \xi (n)\) and the initial vector \(X_0\). Given numbers \(a_j\), the following constraints \[ MU_j^2(n) \geq a_j^2, \;\;j=1, \dots , r, \] are imposed on the unknown variances. The problem of maximizing the variance \(D=MV^2(n)\) of the system output \(V\) is studied. The solution is given for each of three cases concerning the structure of perturbation \(U\) which may be a nonstationary process, a stationary process or a quasistationary one. An example of calculation of the minimax filter accuracy for a second-order system is given.