[evlatests] [widar-wg] Data Drops

[Bryan Butler]

well, at least we understand them now.  now on to the wobbles (again)...

	-bryan


Ken Sowinski wrote, On 8/19/09 17:32:
> On Fri, 31 Jul 2009, Rick Perley wrote:
> 
>>    New:
>>
>>    A significant fraction (perhaps 5 -- 10%) of the visibilities have
>> amplitudes low by a few tens of percent.  The phase is completely
>> unaffected.  The drops are on single records, and are spaced 10 seconds
>> apart -- always on seconds ending in 4 or 5 (like, 14, 24, 34, ...).
>> These drops are independent on all correlations, and are not antenna
>> based -- they cannot be calibrated out.  They were not present in the
>> single-polarization data taken three nights ago.  They are difficult to
>> handle, due to their varying amplitude  (i.e.  CLIP can't get them all).
> 
> This has at last been tracked down.  The story is not too long,
> but involved many people for too long a time.
> 
> While chasing the wobbles a few weeks ago it was suggested that
> we should be sure that the subband delay modules, which we had
> been ignoring, be fed something sensible just in case they were
> sulking at being left alone.  This was done by enabling the
> feature in the model generator and verifying that it generated
> delays in the middle of the subband delay range and a delay rate
> of zero.
> 
> But there were two errors that led to all this confusion.  First,
> it sometimes happened that the code feeding the delay 'models'
> to the hardware 'ran out of input' and sent garbage, changing
> the delay by many microseconds.  This happened preferentially at
> ten second intervals, ten seconds after the delay polynomials
> were last sent.  The effect of this was to send delay into
> never-never land for a fraction of a second wiping out coherency.
> The second error is that subband delays are lost on the way to
> path 1 of the station board, but not path 0.  The result of this
> is that any correlated data that came from path 0 was subject to
> this degradation, but not that from path 1.  Given the way we
> usually connect streams to baseline boards this means only subband
> two (or four or six) was affected.  Why subband zero was not affected
> is still unexplained.
> 
> We will fix this in the short term by disabling the subband delay
> option of the model generating task.  In the longer term the
> fundamental problems will be addressed.
> 
> On to mapping once again.
> 
> ________________________________________________________________________________
> There is a web page, used as a document repository, associated with this list.  
> The URL for the web page is:
> http://listmgr.cv.nrao.edu/pipermail/widar-wg/