I may come up with more comments, but here are a couple right off of
the bat:
The truncation effect is almost surely related to MIRIAD's default
storage of visibilities as 16-bit integers paired with a scale factor
... which can be overridden, at a cost of disk space. The discussion of
precision in section 3 seems a bit inaccurate to me.
Figure 1 is not fully correct. MIRIAD also supports the "uvfits" data
ordering ... and in fact this is preferred, since then you can perform
on-the-fly Stokes resampling when reading a file. But, for instance,
early ATA output files were in the format showed on the left side of
the figure, until I rewrote the data emitter to be better.
Peter
On Wed, 2016-03-09 at 12:00 -0500, Jonathan Pober wrote:
> Hi everyone,
>
> I attach a memo by Adam Lanman for posting to the HERA memo series.
> What started as a simple task of reading PAPER data out of both
> MIRIAD and uvfits formats revealed small, but non-trivial differences
> between the two files. We've mainly been discussing this on MWA
> telecons, since they have been the main users of the MIRIAD -> uvfits
> conversion, but I think the final findings do reveal interesting
> things about our MIRIAD data analysis. We can lead a short
> discussion of the findings during the HERA data telecon in ~30
> minutes.
>
> Cheers,
>
> Jonathan
