search

ratt-ru / pfb-imaging

Preconditioned forward/backward clean algorithm
MIT License
7 stars 5 forks source link

Recompute uvw during averaging #112

Open landmanbester opened 4 months ago

landmanbester commented 4 months ago

The init app has options to do simple averaging in frequency and time based BDA using the averaging functionality in codex-africanus. If I understand correctly, these do not recompute the uvw coordinates from the averaged time and frequency centroids but approximates them with a Euclidean average (is this correct or is it a weighted average @sjperkins @bennahugo?). We should probably rather recompute them as is done in xova. Since chunks are processed in separate processes and we don't have to write back to the MS it should be possible to do this in parallel

sjperkins commented 4 months ago

Yes it is a weighted average and technically the UVW coordinates should be recomputed.

sjperkins commented 4 months ago

In general this is something that I'd like to implement in arcae, but that will have to wait until the data processing churn has quietened down.

landmanbester commented 4 months ago

From my tests it doesn't seem to make a big difference if you are doing a moderate amount of averaging. At least for the L-band MeerKAT data I have been working with setting -max-fov 2.2 and using a bda-dorr 0.98 I don't see any difference in the 2 degree fov I am reconstructing and yet I get a factor of about 5 in compression. Big win so thanks. Low priority but thought I would document it for posterity