I am currently running the Kalman Filter on satellite in low earth orbit and using the Python Wrapper. For this satellite, it would greatly simplify things for me to have the covariance and process noise matrices represented in a local RIC (or TNW) frame. From what I understand, since my initial state and filter output is in the J2000 frame (and kalman propagator is inheriting these in cartesian), both of the aforementioned matrices are represented in this frame.
My current idea for this is to have my getProcessNoiseMatrix function (instance of CovarianceMatrixProvider) take the input J2000 covariance, transform it to the RIC frame, perform the evolution similar to here on p. 5 CovarianceReference, then transform it back to J2000. This method of 2 transforms seems a bit clunky, and also makes any intermediate covariance propagation between measurements difficult as each step will have to do the same. Plus I would guess that this will significantly increase the filter processing time.