Identification of Kalman-Bucy filters from noisy measurement arrays

Abstract

The paper presents a method for jointly identifying all the parameters of the transition, input-covariance and measurement-noise eovarianee matrices from noisy measurements if no a priori parameter knowledge is available. The method is first derived for single-input—single-output processes and is further extended to single-input-multi-output processes and finally to multi-input-multi-output processes. The convergence of the resulting parameter estimates is discussed and computation results are given for the singlo-input—single-output ease, for the single-input-multi-output case (two outputs) ami for the multi-input—multi-output ease (two inputs and two outputs) to demonstrate the readability of the method.