Filter design methods of multiple model system

In this paper, two filter design methods of multiple model system are proposed. One is the identification of ARMA model, and the other is /spl chi//sup 2/ test. The identification of ARMA model means the steady state gain matrix of Kalman filter can be identified online via recursive extended least squares method, by comparison of steady-state Kalman filter gain with the Kalman filter gain obtained from possible model, the true gain matrix can be determined by the principle of minimal error norm. The /spl chi//sup 2/ test method means the true model can be determined by detection of the whiteness of innovations process. The two methods are applied to homing guidance system. The simulation results prove that both methods are effective.<>