search

ratt-ru / CubiCal

A fast radio interferometric calibration suite.
GNU General Public License v2.0
18 stars 13 forks source link

CASA gain calibration not parallelizing very well #177

Open bennahugo opened 6 years ago

bennahugo commented 6 years ago

So... revisiting this - it may be worthwhile to test whether 1GC calibration with cubical isn't faster than CASA's 1GC calibration if we assume we can preaverage data. This is fast becoming a pipeline bottleneck (in the wideband case).

The only problem is if we use cubical there is no "fluxscale" (hack) equivalent (as far as I can see). In CASA a model for the primary is plugged in to set the scale and the solutions of the secondary is normalized using a flat spectrum unitary model by default. @gigjozsa @o-smirnov It is hard to tell whether the bias of not using a proper spectrum for the secondary will be worse than perhaps only transferring phase corrections for long baselines with cubical if the system is temperature stabilized?

o-smirnov commented 6 years ago

I already have a unitary model implemented (--model-list 1), so I'm not sure what you're still needing. Doesn't setjy() just populate the model column?

bennahugo commented 6 years ago

I think the solutions are additionally scaled (per channel) assuming that the ratio in gains between the flux reference and the gain calibrator field is solely due to the incorrect model

bennahugo commented 6 years ago

https://casa.nrao.edu/docs/taskref/fluxscale-task.html