Abstract :
Continuous-time tracking filters based on the two-state exponentially correlated velocity (ECV) model are discussed, in which a correlation parameter is used to give the velocity an exponential correlation. It is shown that the correlation parameter in the model causes additional deterministic steady-state filter errors. For a given bandwidth, the performance declines with the correlation parameter even in the Kalman algorithm covariance sense. The covariance obtained from measurement noise only is also given and compared with the Kalman algorithm covariance. The paper illustrates, in a simple case, that a reasonable model extension from the Kalman filtering point of view could give a worse filter performance. The results are obtained by using filter transfer functions.
