The γ-attenuation problem for systems with state dependent noise (Q4248570)

The authors consider the general \(\gamma\)-attenuation problem for systems with state dependent noise described by \[ dx= (A_0x+ B_1u_1+ B_2u_2) dt+ \sum^r_{j=1} A_jx dw_j,\tag{1} \] \[ y_1 = C_1x+ D_{11}u_1+ D_{12}u_2,\quad y_2= C_2x+ D_{21}u_1, \] where \(x\), \(u_1\), \(u_2\), \(y_1\) and \(y_2\) are \(n\), \(m_1\), \(m_2\), \(p_1\) and \(p_2\) dimensional vectors, respectively; \(A_j\), \(0\leq j\leq r\), \(B_i\), \(C_i\), \(D_{ik}\), \(i,k= 1,2\) are constant matrices of appropriate dimensions; \(w_j\), \(1\leq j\leq r\) are standard mutually independent Wiener processes and \(\gamma> 0\). The \(\gamma\)-attenuation problem consists in finding a controller with the following imposed structure: \[ dx_c(t)= [A_cx_c(t)+ B_cu_c(t)]+ \sum^r_{j= 1} A_jx_c dw_j,\tag{2} \] \[ y_c= C_cx_c(t)+ D_cu_c(t). \] It is required that the system obtained by the substitution \(u_2= y_c\), \(u_c= y_2\) and \(u_1=0\) is exponentially stable and the input-output operator from \(u_1\) to \(y_1\) has the norm less than \(\gamma\). The authors show that if a controller exists which stabilizes system (1) then certain linear matrix inequalities (LMI's) must have positive-definite solutions satisfying a complementary rank condition. If such solutions exist for the considered LMI's the controller is obtained also by solving an LMI.