A priori covariancegreenspun.com : LUSENET : CU Statistical Orbit Determination : One Thread
You mentioned that you fixed the mu, J2, and Cd in the a priori covariance (i.e., setting their P0 = 1e-20), which forced the model to absorb the errors in the last three terms (wr, wi, and wc). If the errors are being absorbed only in the last three terms, does that mean you've fixed some or all of the position and velocity covariances as well?
I ask this because when I include the modified state and the new phi matrix in my code, my RMS and RSS residuals jump to approximately 83m (RMS) and 175m (RSS), as would be expected. However, when I add the process noise, my RMS drops to around 11m, but RSS jumps to about 37876m. I've checked and double checked my Q matrix equations and they match those on the website. I've tried toying with tau and a priori covariance for the acceleration terms, but nothing seems to reduce the RSS to below 16000m. Can you think of anything that might be producing this significant decrease in RMS, but substantial increase in RSS, other than my initial assumptions for tau?
-- Eric Rhoden (firstname.lastname@example.org), February 18, 2000
No, I didn't fix position or velocity with a priori covariance. I did lower the a priori covariance to 10 m position and 1 m/s velocity. They still need to be adjusted, we are using the compensative accelerations instead of adjusting for drag and MU/J2. Did you try leaving the a priori covariance for MU, J2 and CD high?
As far as reasons your problems...my residual RMS drops significantly, to under 10 meters, but the RSS of my position error dropped as well. You are using xyzDMC right? Try lowering you tau and sigmas to small numbers (1e-10). The effect of this will be as if you aren't using any process noise. Use this to check to see if you have things coded ok. If that doesn't show you something, email me all of your code and I'll take a look.
-- David Goldstein (email@example.com), February 18, 2000.