refactor: Write smoothed states in GX2F #3584
Conversation
|
The GX2F states are a hard prediction of a propagation with the fitted parameters. There was a similar discussion in the beginning of the GX2F-development. The GX2F doesn't do any filtering or smoothing, therefore those states don't exist. Of course the An idea on how to have a common set of resulting parameters would be to rework the naming completely, e.g. having for the best fit a set called |
|
I think
This is what we call smoothed states in the current terminology. |
|
As far as I remember, in signal processing terms smoothing is used to reduce random fluctuations of a signal, basically all noise, that cannot be caught by frequency based filters. In the best interpretation, the GX2F does not do any of this. In the worst, it does the exact opposite, be introducing additional scattering angles that make the final trajectory less smooth. |
|
In today's meeting, we converged to the following resolution: We rename to smoothed, because the resulting states should be the same as the resulting KF-states. We need to write into the documentation, that smoothed means finalised for some fitters and should be used for all trackstates, that come back nicely, even if the fitter is not doing any explicit smoothing. |
|
I don't think there is a good definition of what smoothing is but in the KF formalism it means reversing the filtering process to inform "past" states with "future" measurements. Currently we keep 3 track states in our EDM
Given these three as a choice for the GX2F only the |
|
The global chi2 fitter should write smoothed states like the other fitters since the states will contain information about previous and following measurements.
Apart from that I explicitly bound I renamed the free function
updateCovariancePredictedtoupdateGx2fCovariancebecause it is too generic for the publicActsnamespace and bound the implementation explicitly to the namespace to avoid linker errors while renaming.