[evlatests] Non-closing wobbles still exist!

[Rick Perley]

    I took an hour's data last week on 3C84 , and some nearby calibrator 
sources, in order to establish whether our 'PDif Compression' problem 
affects the visibility amplitudes.  What has happened since then 
illustrates perfectly my rule 'When we look for something, we find 
something else'.  Note in what follows that we are in A configuration -- 
the baselines are long, and the fringe rates fast! 

    To make this test as clean as possible, I utilized Ku-band, in 8-bit 
1 GHz mode, with the A/C IFs at 14.5 GHz, and the BD IFs at 17.5 GHz.   
There is no visible RFI in either band.  I used the standard setup -- 8 
subbands in each IF pair, each of 128 MHz width, with 2 MHz channel 
resolution. 

    The data in IFs A, B, and C are as perfect as any I have ever seen.  
An image of 3C84, made in a single subband, gives over 500,000:1 dynamic 
range -- and this with only 18 minutes of data! 

    But subband 'D' (LCP at 17 -- 18 GHz) is something quite different.  
Here I find, only many baselines, a strong, perfectly sinusoidal signal, 
covering multiple subbands, but particularly strong in the first two.  
The basic characteristics of this 'wobble' are:

    1) In LCP only, on a single (analog) IF 'D'.  The effect is much 
stronger on the lower two subbands within this IF, but is visible on all 
eight.  The largest wobbles are seen in the 20 channels at the bottom 
end of the first subband (following the bandpass correction).  This is 
true on *all* baselines which show the effect, which I claim is strong 
evidence that the interfering signal is being introduced *downstream* of 
the anti-aliasing filter located in the 1-2 GHz path of the T304. 
    2) It is equally visible in both amplitude and phase, with the same 
period and SNR.  Visualized in the complex plane, the visibility vector 
terminates on a circle, centered on the correct amplitude and phase. 
    3) The amplitude of the wobble is typically 1 -- 2 % -- but on some 
baselines, it is as high as 6%.  On phase plots, the wobble is up to +/- 
5 degrees. 
    4) The wobble period can be amazingly long (given that we're in A 
configuration, in Ku band) -- up to 2 minutes!  About ten baselines have 
periodicity of longer than 30 seconds.  Many longer baselines have 
periods of 5 seconds or less.  It is possible that all baselines have 
the effect, but that the period is too short for most to be seen with 
the 1 second averaging. 
    5) The effect does not close.  This is evident from inspecting the 
wobble patterns from any triplet of antennas.   The program CLPLT (which 
plots closure phases) shows this beautifully. 
    5) There is a tendency for shorter baselines to have longer 
periods.  The cluster of four antennas near the end of the north arm (ea 
1, 2, 23 and 28) all have high amplitude and long periods.   However, 
counter-examples can be found.  And short baselines near the center of 
the array show short periods,  or no wobbles. 
    6) Some longer baselines have clear beating patterns.    
    7)  There is no clear spatial relationships between high beat 
amplitudes and baseline/antenna length or position.  Antenna ea04 -- at 
W56 (way, way out there) has large, visible 'wobbles' on nearly all of 
its baselines.  But neither of the two antennas on each side of it (17 
and 21) show any visible wobbles at all.  I think this alone makes an 
external origin impossible. 
    8) All the monitoring data (PSum, PDif) are normal and clean.  The 
autocorrelation spectra are clean.  The bandpasses are clean.  There is 
no evidence of an external signal within the 1.024 GHz bandpass. 

    I consulted with the available gurus  (Ken and Vivek), with 
essentially no agreement, other than the effect is unlikely to be driven 
by an external, radiated signal. 

    I attach some amplitude and phase plots (encapsulated postscript), 
from subband #2 of the C and D IFs.  The blue trace is RCP (IF 'C'), and 
red is LCP (IF 'D').  The amplitude plots are first, the phases follow.  
The antenna numbers and locations are written on the plots. 

    Any ideas?  We're desperate!
 
    Rick
-------------- next part --------------
A non-text attachment was scrubbed...
Name: WOBBLE-D
Type: application/postscript
Size: 489939 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20121227/0b8fd8eb/attachment.ai>