[evlatests] [Fwd: 3bit digitizer / sampler clock jitter]

[Rick Perley]

    Inspired by Vivek's report, I looked at these data in a different 
way -- using the rms noise histograms. 
    Although Vivek gives a convincing demonstration that 22R shows the 
effect of increasing noise with subband number, presumably due to clock 
jitter, I think it more important to demonstrate the effect is present 
in those correlators which aren't obviously broken, but which show the 
undesired extra noise. 

    The 'bottom line' here is that there is no evidence for increased 
noise in the LCP side correlations, as a function of subband number, 
which is over and above the known dependence of sensitivity with 
frequency. 

    Details:

    The data taken Thursday are very nice, about 22 minutes on 3C286 and 
an adjacent blank field.  Calibration/editing followed the standard 
methods. 

    The rms noise matrix, for any one subband, shows all the 
now-familiar characteristics.  Important for this study is that all six 
LCP correlations have about the same noise -- from 0.154 to 0.169 mJy in 
a subband free from RFI. 

    I then generated the rms noise statistics as a function of subband 
number, combing all six baselines to improve the accuracy.  As I 
understand it, the phase jitter effect is expected to increase the noise 
as we progress downwards from the highest numbere subband at C-band.  
Here are the results (remember that 0.14 is the 'correct' value). 

Subband #    1       2       3      4        5       6        7      
8       9      10     11     12     13     14    15    16
Frequency 5520 5648 5776 5904 6032 6160 6288 6416 6544 6672 6800 6928 
7056 7184 7312 7440
-------------------------------------------------------------------------------------------------------------------------------------
rms noise  .206  .186  .194  .203  RFI    .181  RFI   .173   .175  RFI   
.160  .162 .168  .162  .177  .222

    Although it is clear that the low-numbered subbands have higher 
noise than the higher numbered ones, this difference is close to that 
expected by the well-known better performance of our C-band systems 
above 6 GHz than below.  It will take a bit more work to separate this 
effect from the phase jitter effect. 
    Note that subbands 1 and 16 are anomalously high -- can somebody 
explain this? 

   

-------- Original Message --------
Subject: 	[evlatests] 3bit digitizer / sampler clock jitter
Date: 	Fri, 10 Sep 2010 15:42:59 -0600
From: 	Vivek Dhawan <vdhawan at nrao.edu>
Organization: 	NRAO
To: 	EVLA Tests <evlatests at nrao.edu>



Fallout from yesterday's meeting: I had looked at the phase noise
by subband for the 3bit data, did not see a trend, and dismissed it.

Prompted by Barry's comments, I looked harder in the right place, on
ea22 RCP, and see evidence of clock jitter: the phase noise goes up at 
frequencies away from the baseband edge, dominating the thermal noise
a few 100 MHz into the BB. This is on 2GHz wide data, from yesterday,
before the cable swaps.

This effect is very clear on 22 R, (now back in the lab), but it may
be present at lower level on the remaining 3-bit paths.

----------------

Here are some plots that purport to clinch this case, unfortunately
they require wordy captions:

Both plots show phase +_15 degrees, vs. time ~3 minutes.

plot 1. Blue is for a good baseline, 12-15, subband 2, phase difference
between RR and LL, so L301, L302, atmosphere are common-mode and cancel.
i.e. the noise shows how much worse RR phase is, compared to LL.

Red trace is the baseline 15-22, much worse!

For the setup at hand, C band, 2GHz 3bit, subband 2 is farthest from BB
edge, subband 14 is closest, verified to be clean of RFI. And, as expected
for clock jitter, subb 3,4,8,9,14 are progressively better, but still not
as good as the corresp ones from the good baseline. (not shown in plot)

plot 2. shows 2 adjacent subbands, 2&3, from the bad baseline, 15-22,
the extra noise is highly correlated between subbands, so it is not
front end noise, it is there on every subband and scales with subband
offset as above. (red/blue overlap makes green in this plot)

Vivek.