[daip] FITLD for different digitizers

[Leonia Kogan]

Michael,

FITLD takes care about the difference in saturation effect for DIFX and 
"old" correlator. By the way, the saturation effect at the "old" 
correlator depends on the number of polarization and this effect is 
treated at FITLD also. As a result the mean value of the auto 
correlation spectrum must be near 1 for both old and DIFX correlator.
The deviation off 1 is explained by the error at the digitizers and 
therefore can be applied to cross correlation to correct for this error.

Leonia

Leonia
Dr Michael Bietenholz wrote:
> Hi Amy, Leonia
> 
> Okay, after some discussion with Walter Brisken, I think part
> of the problem is that, aside from any 1-bit/2-bit issues,
> FITLD is applying the digital corrections for the old, hardware VLBA
> correlator to DiFX data.
> 
> For the old hardware correlator, there was a different correction for
> x-correlations and auto-correlations, since the latter saturated.  The
> correction factors were those given in VLBA Scientific Memo #12  (in the
> 1-bit case, 0.3442 for x-corrs and 0.2739 for autocorrs).
> 
> I think that for DiFX the DIGICOR=1 scaling should be the same for 
> auto-correlations and x-corrs.  The FITLD explain file suggests
> that FITLD can detect DiFX data and do something different for it.
> 
> But as far as I can tell, FITLD (MNJ of yesterday), when run
> on an FITS-IDI data set from the DiFX (CORRVERS = DIFX-2.0.1.),
> is still applying the corrections of 0.3442 for x-corrs and
> 0.2739 for auto-corrs.
> 
> Since FITLD breaking the relationship between autocorrelations
> and x-corrs, I think this means that the ACCOR corrections will not do the
> right thing for the x-corrs.
> 
>                michael b
> 
>> So reproduced your finding (ratio=(digicor=1/digicor=-1)):
>> BR-HN (both antennas recording 1 bit) ratio=0.3441
>> AR-EB (both antennas recording 2 bits) ratio=0.3441
>> BR-AR (one bit correlated with two bits) ratio=0.3441
>>
>> We suspect the error is on the DiFX side, in labeling the baselines 1 or
> 2
>> bit.
>>     I will be away for the next week, so Leonia was going to pursue
> this.
>> Thanks!
>>
>> Amy
>>
>> Dr Michael Bietenholz wrote:
>>> Hi Amy
>>>   I've got the rest of this correspondence on my HartRAO email,
>>> so I'm replying using this one (and I am currently at Hart, too). In
> answer to your questions:
>>> The data set where I stumbled upon the problem was GB072 I think Leonia
> might have been looking at a similar dataset, GB071 whose PI was
> Andreas Brunthaler.  I think GB070 (which is now public domain) may be
> similarly affected, but I haven't confirmed that).
>>> You learn something everyday - I used UVAVG to do the merging, not
> being aware of VBMERGE.  However the problem with scaling seems to
> happen before then. Basically, the visibility scaling produced
>>> by FITLD:DIGICOR=1 is the same for 1-bit and 2-bit data in
>>> a mixed data set, which I think means it must be wrong in one
>>> of the two cases - see below for a copy of an email I sent Leonia
> detailing what I'd learned so far.
>>> As far as data, each of the three pieces of GB072 has both 1-bit and
> 2-bit antennas, so you just need one piece to check.  Should you need
> it,
>>> you have my explicit permission to  get the GB072 data from the
> archive,
>>> which would be far more efficient  than me trying to ftp it to you from
> South Africa.  However, if you need I'm happy to ftp it.
>>>> Leonia asked me to look into this last week but then I immediately got
> sick
>>>> and was out for the rest of the week.  I plan to download your data
> and
>>> take
>>>> a look at it.  I asked Leonia to forward some of your discussion to
> me,
>>>> but
>>>> he seems to have lost it, although he has given me some hardcopies, so I
>>>> apologize if I ask you some of the same things that Leonia did. Since
> this experiment was correlated in 3 passes did you load them all and
>>>> then use VBMRG to merge them?  If not which of the 3 passes are you
> looking
>>> Here's what I did to test this FITLD with DIGICORR=1 -> call this
> GB072_DC.
>>> (I just used the first correlator pass for this)
>>> Then also FITLD with DIGICORR = -1 (but nothing else changed)
>>> -> GB072_XX
>>> Now I looked at the ratio of visibility amplitudes between
>>> GB072_DC and GB072_XX.  This ratio should give the "digital
>>> correction" which was applied by FITLD.
>>> For this run,  BR and HN recorded with one bit
>>> AR and EB recorded with two bits.
>>> I would expect that the "digital correction", ie. ratio of the
>>> visibility amplitudes GB072_DC/GB072_XX would be different for
>>> each of these three baselines:
>>> BR-HN (both antennas recording 1 bit)
>>> AR-EB (both antennas recording 2 bits)
>>> BR-AR (one bit correlated with two bits)
>>> What I found, however, was that for each of those three baselines
> (BR-HN, AR-EB and BR-AR), the ratio GB072_DC/GB072_XX was exactly 
> 0.3441.
>>>  (For auto-correlations, this ratio was always 0.22,
>>> although for the autocorrelations it varied by 1% or so, but
>>> not with any pattern related to whether antennas had recorded
>>> 1-bit or 2-bit data.)
>>> Now I only have a vague idea of what the "digital correction"
>>> is, so I could be wrong in thinking that it should
>>> depend on whether its one-bit or two-bit data....
>>>                   michael
>>
> 
> 
>