Hello all,

I am trying to use Orekitv10.1 for determining the Orbit and Clock offset of a GPS satellite. As an input I have Rinex files from 15 (not all of them have visibility of the satellite) different stations.

So far I have succeeded in parsing the Rinex files, extracting the Range, and adding modifiers (Tropo, and Iono) to the observations. My satellite and station configuration file is set-up as follows:

- Orbit estimation (set-up as a driver)
- Satellite clock offset estimated (set-up as a driver)
- Holmes Featherstone attraction model
- Station clock offsets estimated (set-up as a driver)

I also have build a propagator based on the Dormand-Prince propagator and build my Kalman filter.

The first problem Iāve found is that I have little idea of the process noise matrix values for the clock parameters? Do you have some guesses regarding these values for the clock ?

My second question is regarding the algorithm to propagate the clock states for the Kalman filtering. How is this done ? When I am initializing the covariance matrix I am using what is provided in the test AbstractDetermination.java file and that is synthetized in the following lines:

// Orbital covariance matrix initialization // Jacobian of the orbital parameters w/r to Cartesian final double[][] dYdC = new double[6][6]; initialGuess.getJacobianWrtCartesian(propagatorBuilder.getPositionAngle(), dYdC); final RealMatrix Jac = MatrixUtils.createRealMatrix(dYdC); RealMatrix orbitalP = Jac.multiply(cartesianOrbitalP.multiply(Jac.transpose()));

Should I also compute the Jacobian with respect the parameters ?

Another question is regarding the estimated parameters. Before going to the kalman filter I parse all Rinex files and multiplex theme according to their epoch as per the example above. Then I insert the all drivers. I was wondering that when the Kalman filter is executed at every iteration is trying to estimate all the drivers or only those that are observed ? When a new station is visible how is this included into the Kalman ?

My final point is regarding the performance of the Kalman filtering. I am executing 43200 measurement. The first 10000 are fairly quick to process but afterwards it slows down and it takes more than one day to complete. Iāve seen in the forum that you had identified a bug regarding this point which you solved by adding the NewtonianAttraction model in the force models. I am using version 10.1 (with the python wrapper) which supposedly resolves this problem. Could it be that the problem is not fully resolved ?

Thanks in advance for your help.

Jordi