Spectral factorization with imaginary-axis zeros (Q5961585)

A number of problems in systems and control essentially reduces to spectral factorization of a proper, rational spectral density in the frequency domain or to finding Hermitian solutions of a linear matrix inequality in state space. When the spectral density is invertible at infinity, the minimum-rank solutions of the linear matrix inequality are exactly the solutions of an algebraic Riccati equation. Moreover, in this case, the finite zeros of the spectral density are exactly the eigenvalues of an associated Hamiltonian matrix. In this paper, the authors give a simple but uniform treatment of the calculation of solutions of the linear matrix inequality. They show that the existence of zeros of the associated spectral density on the finite or infinite imaginary axis allows the construction of an explicit fixed part of all solutions of the linear matrix inequality and a consequent reduction or deflation of the problem. These results are not new, but the proof is particularly simple. They do not propose that the derivation presented in this paper is necessarily suitable for computation but it is simple and directly reveals some of the structure of the set of solutions of the linear matrix inequality. The techniques used in this paper are easily transferable to the discrete-time spectral factorization problem and related linear matrix inequality.