Add xtalk error calculation #56
Conversation
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
| @@ -7,7 +7,7 @@ | |||
| "source": [ | |||
There was a problem hiding this comment.
Line #12. if band_slice is None:
It is weird to me that you can have an inconsistent band_ind and band_slice. I'd suggest picking one or the other to specify.
Reply via ReviewNB
| @@ -7,7 +7,7 @@ | |||
| "source": [ | |||
There was a problem hiding this comment.
Line #24. xtalk_frop = get_frop_wrapper(pol=pol, stream_ind=stream_ind, band_ind=spw,
For baselines and bands with little to no overlap in the xtalk notch and the main beam, do we want to just skip all this?
Reply via ReviewNB
There was a problem hiding this comment.
OK, do we want to automate the check with some criterion or make a global switch?
There was a problem hiding this comment.
OK I added a global setting called CORR_MATRIX_NOTCH_CUTOFF defaulted to 30m which means by default it will skip the notch filter in the error calculation in the EW projection is longer than 30m. This skips the notch filter part by a little less than 75% of the single-baseline runs I believe.
This adds the cross talk filter to the noise covariance calculation (and corresponding correction factors). It turned out to involve a sizable matrix multiply that made the calculation take 15 minutes without any adjustments.
I tried different speedup schemes to ameliorate this including generating one filtration operator per spectral window, only using one of the streams, and decimating in frequency. I found that decimating in frequency by a factor of 10 improves the speed by an order of magnitude (just under two minutes to complete) with an error compared to the full calculation of one part in 10^4 at worst on the couple baselines that I checked.
This option is adjusted with the CORR_MATRIX_FREQ_DECIMATION global, defaulted to 10.