[evlatests] Some Good News!
Michael Rupen
mrupen at nrao.edu
Mon Jun 15 09:43:02 EDT 2009
Hi Rick --
whew! Nice to have some really _good_ news on Monday morning! A few
comments/questions --
>
> Differences from former tests are (so far as I know)...
>
> 1) X-band, rather than C-band.
> 2) 2 IFs, (sub-bands) rather than 4.
> 3) Point source was offset from phase center by a small amount
> (about 0.3 arcseconds, I think).
Should have been much more than this! 45 arcseconds east of the center.
Another difference is the use of much longer scans (10-15 minutes,
vs. the previous 40 seconds).
> There is a world of difference between this database, and all the
> others taken this month ...
>
> 1) There are *NO* 180 degree phase jumps. All data are completely
> seamless in amplitude and phase.
I think this is due to the long scans -- phase jumps occur at scan
boundaries, and also seem to occur more often with shorter scans.
I wanted to check the "lumpy-bumpy" problem, not the phase jumps, this
time around. Scan boundaries seem to be associated with most of the bad data
we've been seeing, and I didn't see any point making people spend all
their time flagging.
So, don't get too excited about not seeing phase jumps!
> 2) Except for the initial couple of minutes, there are no bad data.
> I cannot recall ever seeing such lovely data ...
See above.
> 3) Histograms of the single-channel distributions (both real and
> imaginary) show absolutely perfect gaussians, with 1-sigma widths of
> 0.655 Jy -- a bit better than what I expected.
> 3) Initial images showed our 'lumpy-bumpy' problem. But read on ...
>
> To uncover the origin of these lumps/bumps, I SPLITed the data into
> a single 961-channel average (to increase SNR) and plotted the amplitude
> and phases for all baselines and times. The amplitudes looked just
> fine, but the phases showed a single baseline with an oscillatory
> pattern which persisted throughout the entire hour of observation. The
> pattern frequency decreased throughout, reaching a zero at about 14:50
> IAT, at a U-spacing of -2.1 Klambda -- i.e., this problem has nothing to
> do with a steady external stationary signal (which must maximize when
> the source fringe rate goes to zero -- at U = 0).
>
> The evil baseline is 2 x 18. The amplitudes look fine, but the
> phases are oscillating by +/- 2.5 degrees. Both IFs (sub-bands) behave
> identically (except that IF#2's phase oscillations are inverted w.r.t.
> #1).
Ah! that is very interesting. IF 1 = subband 1 ; IF 2= subband 2.
We should take some data with more subbands to track this down.
Could you send around a plot of the phase effect? I do not see any similar
problems with 2-18 in the 8jun09 C band run. 2-19 however *does* show
fast oscillations in phase; I'll track this down and see what comes up.
2 and 18 are in no special place in the correlator (thank the gods) --
Rack 1, Slot 2, and Rack 3, Slot 0.
> I then flagged the evil baseline, and re-calibrated. BLCAL was run
> (single solution for entire hour). The baseline corrections are
> *extraordinarily* small -- typically 1 part in 10^5, consistent with
> zero, I believe.
>
> The single-IF (#1) map which resulted from this is almost perfect.
> The rms (it doesn't matter where it is measured, the background is
> completely flat) is 55 microJy -- this is *exactly* what it should be
> based on the single-channel histogram width reduced by the square root
> of the number of points (150602) and channels (916). The resulting
> formal DR is 78200.
>
> There is a barely detectable waviness seen beneath the noise. As
> this is hardly attenuated at all throughout the image, it appears these
> could be a very few unflagged points which escaped by (almost
> non-existent) flagging.
>
> So! Why is this X-band test so good, while the preceding C-band
> ones so bad?
I also took some X band old-correlator data yesterday; it will be interesting
to see whether that is also better than C band (which had some nasty
problems).
> And -- we still have a single very bad baseline. What is wrong with
> it?
>
> Finally -- other than that bad baseline, the corrrelator looks to me
> to be behaving precisely right.
Do you see any sign of increased rms noise on baselines between
{18, 19, 25}? A simple VPLOT of amplitude vs. time on your SPLIT (pre-BLCAL
and pre-CALIB) data set would suffice to show this.
Another question -- did you ever run FRING on this data set? If not,
that's another difference between this and previous reductions.
Michael
More information about the evlatests
mailing list