[evlatests] Fun with 3 bits ...

[Rick Perley]

    Michael arranged a short (40 minutes) observation, early Wednesday 
morning of 3C84, a target point source a few degrees away, and a blank 
field, using the 3/8 combination mode.  (3bit on the AC side (A1C1 and 
A2C2) and 8bit on the BD side (B0D0)).  The A1C1 and A2C2 frequency 
setups were the same (spanning 12.976 through 14.960 GHz).  The B0D0 was 
set to the center of the above band. 

    The primary goal was to demonstrate that phase calibration works as 
well in 3-bit as 8-bit. 

    Some Results:

    First, the boring part:  The data quality was, overall, 
outstanding.  Amplitude and phase stability were excellent.  The phase 
connectivity between calibrator and target is *exactly* the same in 
3-bit as in 8-bits.  No surprise there. 

    Now, the more interesting bits. 

    1) Two antennas (18 and 21) had just been moved the prior day.  
Although both fringed, they were clearly not in an operational state.  
ea18 was very weak, and ea21 had a delay error well outside the range 
that can be solved for.  Both had to be flagged out. 

    2) The six baselines involving the four antennas ea01, ea02, ea20, 
and ea23, have a notable amplitude closure error of about +1.2% 
(meaning, they are too high).  There is no closure error in phase.   The 
same closure error was on all three IF paths.  And the error was the 
same in RCP as in LCP.   The error is constant over the 40 minute 
duration of the observations.  Thinking (and hoping) that this was 
caused by the 'double-D' mechanism,  I did a polarization calibration, 
using 3C84.  The results show these antennas have very small D-terms 
(less than 1% w.r.t. the reference antenna, ea16) -- as do nearly all 
others.  So this mechanism is not likely. 
    If the closure is due to bandpass effects, these have to be on a 
scale much less than the channel resolution.  (2 MHz).  (note that each 
observation of 3C84 had its own bandpass solution applied, and its own 
self-calibrated amplitude and gain applied). 

    (NB)  The AIPS  task BLCHN, which solves for the closure error 
channel-by-channel, is an amazingly useful tool here.  It outputs a BN 
table which can be plotted with POSSM. 

    All other baselines have fabulously small closure errors -- the 
*maximum* was about 0.1%.  Most are closer to .05% or less -- limited by 
the noise.  Phase closure errors are similarly minuscule. 

    3) Bandpass stability was very good overall -- much better than 1%.  
However, the 40 minutes is not a long period to make a strong 
statement.  The 8-bit stability looks the same as 3-bit.  However, the 
3-bit path spectra are 'lumpier' than 8-bit. 

    4) The PDif compression problem is easily visible in the 3-bit 
paths, as expected.  However, it is also visible in *some* of the 8-bit 
data, albeit at a much lower level.  In the 8-bit path, the maximum 
compression is about 1% on 3C84, which adds about 10% to the system 
noise.  By comparison, the *typical* compression on the 3-bit side is 
about 6% for the same situation. 

    5) I ran the histogram program 'UVHGM' using the data from the noisy 
field, and all baselines.  The gaussian widths were 57.9, 66.0, and 69.2 
mJy for B0D0, A1C1, and A2C2, respectively.  Hence, a global drop of 14 
and 19%, respectively.  Rather more than I had expected, especially on 
A2C2.  I haven't yet done the antenna-dependent exercise.  I'll do this 
tomorrow.