Abstract
The paper discusses a Kalman-filter divergence problem, caused by the partial decoupling of the process-noise covariance. A simple method for determining the optimum observation matrix is presented. The effectiveness of the procedure is illustrated by a computer simulation.