search

ratt-ru / xova

dask-ms/codex-africanus MS averaging tool
Other
0 stars 2 forks source link

Averager testing #8

Open bennahugo opened 4 years ago

bennahugo commented 4 years ago

@sjperkins @sphemakh @o-smirnov I'm reasonably happy this works as advertised. I've Performed basic visibility inspection after averaging with XOVA to test the averager in production use - as a drop in replacement in my polarization calibration pipeline. From #7, after adding a switch to average "CORRECTED_DATA" I can confirm that the flags are correctly handled - the percentages are spot on. I also run through a basic pol cal to make sure that the data can be calibrated succesfully. Indeed there are no problems here:

Spectrum is spot on: image Phase is spot on: image

Prepolcal:

PKSB1934-638:

image image

3C286 (coloured per correlation - crosshands at about 10% linearly polarized):

Original dataset: image image

Average dataset: image image

Postpolcal (Ignore RFI affected bands - cannot be leakage calibrated):

PKSB1934-638

image

3C286:

image image

bennahugo commented 4 years ago

It is at least an order of magnitude faster to average this compared to oldsplit

bennahugo commented 4 years ago

ok just checked the resulting images to ensure the uv coordinates are correct. Everything seems to be in place the maximum error is about e-4 down from the peak

sjperkins commented 4 years ago

@bennahugo the UVW coordinates are lerped, I guess if that ever becomes a problem then spline fitting may be appropriate.

bennahugo commented 4 years ago

I don't think it is a problem @sjperkins everything seems to be well within smearing tolerances

SpheMakh commented 4 years ago

Great stuff!

bennahugo commented 4 years ago

@sjperkins if you give me permission to push I can push my small fixes up