Extended Kalman Filter Jacobian

Hello!

The non linearity of my EKF implementation evolves mainly from coordinate conversions (ECEF, LLA, and more). As far as I have checked, the numeric derivation of these conversions (for the sake of yielding the Jacobian) is sufficiently fast and accurate. Is there a known limitation for this numeric solution relative to the analytic one? Is it customary to use numeric derivation for EKF or is the analytic one preffered?

Thank you!

Answer 1

[John Petersen]

I would suggest you do the coordinate conversions outside of the EKF equations. Can you stay in a local coordinate frame within the EKF equations?