Kalman Filter not converging

I am trying to understand how unscented Kalman filter works using simulated multiplexed tdoa measurements. I am propagating with cartesian state vectors.

  1. Does the KF extract the measurement covariance values from the sigma values in the measurement class instance?
    note: Based on the decorate method in KalmanEstimatorUtil class, the measurement covariance is set to identity matrix for all measurement types except Position and PV. Is there a way to override this ?
  2. Is the Kalman Observer applicable to UKF as well?
  3. My KF physical covariance value seems to be exponentially diverging. But the innovation covariance is an identity matrix for all timestamps. I am confused as to why this is happening? (The measurement sigma values are around 10ns.)
    I am using 1km sigma for initial position and 1m/s sigma for initial velocity uncertainties. Also the process noise uncertainty for position values are set at 1km and velocity is set at 10m/s.
    Thank you.

If the tdoa sigma = 10ns, then the measurement covariance would be in 10^(-16) levels. Does this affect inversion of innovation covariance and is that why the value of measurement covariance is set to identity matrix ? ( That explains why innovation covariance is always = I ). Appreciate any clarifications.