Dimension of measurements

The ObservedMeasurement, EstimatedMeasurementBase and EstimatedMeasurement interfaces support measurements with dimension greater than one. In practice, most measurements are mono-dimensional, the only multi-dimensional measurements are angular, position and position-velocity. CCSDS TDM files do not allow multi-dimensional measurements, this has led to problems when using TDM with Orekit angular measurements: we have to make sure pairs of measurements are in the tracking message to load them as a single two-dimensional measurement.

I wonder if we should change our design and reduce measurements to mono-dimensional only, splitting angular, position and position-velocity as sets of several measurements. This of course could be done only at a major release.

What do you think?

From my perspective, limiting the capability of OREKIT to align with a particular presentation format (TDM) is not the preferred approach. Notwithstanding this, you may have other reasons to slim down the OREKIT measurement model.

In terms of impact on my system, removing the 2D angular observations would be a barrier to adopting the next major release, but ultimately all (correlated and coincident) observations are wrapped into MultiplexedMeasurement - so provided you keep that, impact would not (prima facie) be a blocker. I hope you are not considering removal of MultiplexedMeasurement.

Incidentally, because we need the precision of an observation, we do need to correlate angular observations in our calling code anyway, because often the angle_1 precision (for example right ascension) is unknown without angle_2 (e.g. declination) to provide the measure correction (sec(angle_2)).

I would guess that if we start taking position and PV with your proposal we would have to unpack them to mon-directional observations with a scale and direction (e.g. X in GCRF), which doesn’t feel like an improvement.

You made very good points, and yes I totally forgot about multiplex measurements (which is a shame since I introduced them).

So scratch my proposal, it was stupid, we must keep multi-dimensional measurements.

Sorry for the noise.

No problem!
And thank you for introducing multiplexed measurement, it makes outlier rejection nice and clean :slight_smile: