[evlatests] Strong on-linearity at 335 MHz (P-band)

[Rick Perley]

    I have reduced the data taken in the September run on the major 
low-frequency calibrators.  These data were taken during the 
reconfiguration time, and total 12 hours, over a 14-hour period. 

    I'll discuss the 'PDif' compression issues at another time.  Here I 
report on the flux transfer results -- these are independent of the PDif 
compression problems.

    First -- the good news:

    The ratios amongst the major interferometric calibrators:  3C196, 
3C286, 3C295, and 3C123, are  the same (to within ~1%) with the new 
system as with the old (P-band) system.  Below I give the current ratios 
and the mean ratios determined during the many (about a dozen) 'flux 
density' runs.

    Source Pair         Old              New
-------------------------------------------------------------
    286/295             0.430            0.430
    196/295             0.776            0.769
    196/123             0.322            0.318
    196/286             1.806            1.787
    123/286             5.62              5.61
    123/295             2.41              2.42
-------------------------------------------------------------

    Using the Scaife-Heald flux density of 3C196 as the reference (45.86 
Jy), I get the following results for the other calibrators (not 
including the FourBigSources, discussed below).    All these are for 335 
MHz. 

    3C48      43.1
    3C123   144
    3C138   16.64
    3C147   52.76
    3C196   45.9
    3C286   25.7
    3C295   59.7
    3C380   40.4

    B)  The Bad News

    Sadly, the good results listed above do not extend to the four 
'absolute' calibrators:  3C144 (Taurus A), 3C274 (Virgo A), 3C405 
(Cygnus A), and 3C461 (Cassiopeia A). 
    Using the values listed above for the (weak) interferometric 
calibrators, I then determined the flux densities of these four sources. 
   
    Below is a table, showing the 'expected' and 'derived' values. 

    Source        Baars            VLA         Ratio
---------------------------------------------------------
    3C144         ~1235           1080         0.87
    3C274           730               630         0.86
    3C405          5938            4200         0.71
    3C461          6305            3910         0.62
----------------------------------------------------------

    We can ignore 3C461 -- Cas A is known to have a secular decrease, so 
the very low ratio for this source is definitely too low. 
    But we can't ignore the others.  A deficit of 5 to 10% might be 
easily explained.  But not a 30% deficit (in the case of Cygnus A).  
There is no way the pre-1980s work could be wrong by this factor!   The 
VLA is clearly seriously saturating -- somewhere -- when we observe 
these stronger objects.

    This is definitely a global problem -- all antennas are doing it.  
As noted above, for 3C144 and 3C274, the various antennas agree quite 
well in their gains -- so they have all dropped by about the same 
fraction, typically 10 -- 15%. 
    For Cygnus A, although there is great scatter in the calibrated 
visibilities,  no antenna pair (baseline) gives a visibility even close 
to what is correct. 

    Fortunately, a clear explanation of this is not hard to find. 

    In comparing the gains needed to put the fluxes of those four 
sources to the right level to the actual power levels seen in the 
station board (i.e., correlating the CALIB gains against the PSum 
values), it's dead-easy to see what the problem is:

    *** We are running the power levels far too high ***

    Cas and Cyg typically triple to quadruple the total system power.    
The recorded PSum values during the experiment, while on either of these 
sources, were typically 60 counts.  Due to the troubles in the 
'set-and-remember' procedures that I've reported on earlier, we actually 
had digital power levels varying between 10 and 120 counts.  (!).  In 
comparing the gain calibration values against the PSum power, I find 
that the only instance in which the gains, on Cyg A, were correct, was 
when the power level was near 10 counts (one antenna, in one 
polarization, for a single 2-hour block).  (This was ea03, in 'R' 
polarization).  Even when we were near the desired 14 counts, it appears 
that a significant gain correction (more than unity) was required.

    I believe it is critically important to follow up on this was an 
on-off Cygnus A experiment, where we set the on-Cygnus A power to 14 
counts -- or less -- then compare this to a nearby standard calibrator 
(3C380 will do well here).