[evlatests] Slow Wobbles -- back with a vengeance!

[Rick Perley]

    Brent:

    The amplification of the effect at the bottom edge of subband #1 is 
likely from the bandpass correction.  The  attenuation is caused by an 
anti-aliasing filter within the T304.  This suggests that the effect is 
added in after this filter.  But then this would seem to contradict the 
indication that the effect is proportional to the correlated power, 
which (I think) would cause the wobble amplitudes to decrease after that 
filter.  On the other hand, if the effect is related to some complicated 
aliasing from some non-linear process, then the filter effect would be 
avoided. 

    We do see, very faintly, wobbles at other IFs.  However, in nearly 
all cases, these are only seen in the two or three channels on each side 
of the subband filter 'notches'. 

    I hope these clues help ...

    Rick

Brent Carlson wrote:
> Hi Rick,
>
> It is strange that even though it shows up in sub-band 2, it is not 
> uniform across all channels in that sub-band.
>
> A possibility is that in A config, with a few Hz of earth-rotation phase 
> rate, that the corr chip dump/blanking time is beating with the net 
> phase rate in the correlator...an effect that is known to cause issues.  
> I don't believe that corr chip integration times are set to mitigate 
> this effect as pointed out in NRC-EVLA Memo #032.  I also don't know if 
> 3C84 for these observations is located such that very low delay rate 
> effects, if not taken out by post-correlation "discrete-step delay 
> correction", can become evident as amplitude and phase 
> oscillations--although you'd think this would show up the same in all 
> sub-bands and be increasingly worse moving away from the center of a 
> sub-band, and normally an acceleration signature is seen as well.
>
> I also note that it seems worst at the lowest edge of sub-band 1, which 
> turns out to be the highest I/F frequency region feeding into the 
> sampler (i.e. since it is Nyquist zone 2, with resulting frequency sense 
> flipping, unless the plots are already corrected for this), and is 
> normally caused by sampler clock cross-talk into the sampler.  Don't 
> know what this means exactly, if anything.
>
> --Brent
>
> On 17/01/2013 10:57 AM, Rick Perley wrote:
>   
>>    I reported, a couple weeks ago, that we had baseline-dependent slow 
>> (periods of seconds to minutes) oscillations in amplitude and phase on 
>> some baselines.  These showed up in Ku-band data on 3C84 (a strong, 
>> unresolved object) in A-configuration.  The magnitude of the effect 
>> was modest -- a few percent and a few degrees.  The effect was seen 
>> only on  IF 'C', and it was clear that it was a non-closing effect 
>> (i.e., cannot be removed by antenna-based calibration).    A 
>> subsequent test, done at the same band, with the same source, with the 
>> same setup, failed to find these 'wobbles'.
>>    But ... They're back!      About ten days ago, we observed 3C84 at 
>> Ku and K bands (alternating), while still in A configuration.  The 
>> duration of this test was 6 hours, and the goal is to find our 
>> 'ultimate' dynamic range -- and to uncover any remaining subtle 
>> effects produced by our system.    Overall, the data quality is simply 
>> amazing.  Closure levels (except for what I'm about to report on 
>> below) are effectively unmeasureable -- likely less than 0.01%.
>>    But in reviewing the data carefully, the 'slow wobble' problem 
>> found earlier is present throughout the entire new dataset.  But since 
>> we have two frequencies in this long dataset, some new and hopefully 
>> useful information has been found.
>>    1) The strong slow periodicity is seen in both K and Ku bands for 
>> each baseline on which the effect can be seen.  The periodicity seems 
>> unrelated to the baseline coordinates.  (This has yet to be proven).  
>> Some antennas are more likely to show wobbles than others.  There is 
>> no spatial relationship between antenna pairs showing the effect.   
>> There is no apparent relationship with antenna number or f-shift.      
>> 2) For both, the wobbles are only seen in IF 'C'.  One can make a case 
>> for them being very faintly visible in the others IFs, but the effect 
>> is at least an order of magnitude lower.
>>    3) The wobbles are seen identically in amplitude and phase, with 
>> the same period.  Viewed in the complex plane, the 'wobble vector' 
>> circles about the (stationary) visibility vector.
>>    4) The effect is not visible on a secondary source that was 
>> observed.  This source has 1/15 of the flux of 3C84 -- from this we 
>> conclude that the wobble effect is multiplicative, not additive, and 
>> is proportional to the correlated power -- not the total system power.
>>    5) The effect is strongest by far in subband #2 (#1 in 
>> CASA-speak).  It is also easily visible in subband #1, and faintly 
>> visible in subbands 5 and 6.  The wobbles are contiguous between 
>> subbands 1 and 2 -- which Ken says is evidence that the baseline 
>> boards are not to blame.  (See the attached figures).
>>    6) The period of the wobbles is shorter at K band than Ku band, 
>> roughly at the ratio of the sky frequencies.  For the baseline 2 x 23 
>> (see attached figures), the wobble period at K-band is 20 seconds.  At 
>> Ku-band, the period is 30 seconds.   The periodicity of the wobbles is 
>> different on all baselines, varying from a few seconds to a few tens 
>> of seconds.
>>    7) The amplitude of the wobbles is much higher in K-band than in 
>> Ku-band.  At K-band, the effect is about 5% of the source flux, and 
>> 3.5 degrees in phase.  (pk-pk).  At Ku-band, the effect is about half 
>> of this.
>>    The latter two characteristics make us wonder what happens at Ka 
>> and Q bands.  A short test this evening -- if there is time available 
>> -- might help here.
>>    I have attached four plots to show the effect -- amplitude and 
>> phase for Ku and K bands.  All plots are of the baseline which shows 
>> the effect most strongly:  2 x 23.
>>    About the only thing I'm willing to conclude from this is that the 
>> effect is not caused by an external agent, such as RFI.
>>    Speculations as to origin, or suggestions for further testing, are 
>> welcome! 
>>     
>
>