[evlatests] Curious Problem with Polarization Spectral Line Mode

[Rick Perley]

       As I reported earlier, use of polarization spectral line mode 
('PA') clearly demonstrated that the residuals seen in Q and U images of 
an unpolarized source observed in the continuum modes are due to 
'closure errors' (baseline-dependent offsets) due to the mis-matched 
bandpasses. 

    In preparing a proper report on this, I made images of the source 
OQ208, at X-band, with data taken in mode 'PA'.   The data were filled 
as corrrelation coefficients, and calibrated using the known flux 
densities of OQ208 and 3C286.  (Both sources are in the database).  The 
data used for the images were a sum of 11 channels, each of 780 kHz.  
All images appear noise-limited -- but the noise in the I and V images 
was exactly twice that in Q and U!  As the I and V visibilities are 
computed from the sum and difference (respectively) of the RR and LL 
correlations, while the Q and U visibilities arise from similar 
combinations of the RL and LR correlations, I checked to see if the 
noise in these share the same characteristics.  They do.  Histograms of 
the real and imaginary parts of the four fundamental correlations showed 
the spread in RR and LL to be twice that of RL and LR.  Using the basic 
relations to compute the SEFD from these noises, it's clear that the RL 
and LR correlations have the correct noise, while those in RR and LL are 
twice the correct value. 

    This problem, however, does not occur in the continuum data.  The 
observations I took were equally split between 50 MHz continuum, and 
12.5 MHz spectral line.  The histograms of the four correlations in 
continuum mode are all the same, and all show the correct distribution:  
sigma = 20 mJy -- corresponding to an SEFD of 280 Jy for a BW of 46 MHz, 
and the time averaging of 3.3 seconds.  Exactly correct, and equal to 
the values we determine through the 'stress test' procedures. 

    The problem noted above is not due to the summation of the 
channels.  I copied out the single, central channel, and determined the 
noises with this one alone.  The result is the same as for the sum over 
11 channels:  -- the noise histograms are twice as wide in RR and LL 
than in RL and LR.  The widths displayed are correct for the cross-hand, 
two times too wide for the parallel hand. 

    This is not a calibration issue, so far as I can tell.  The PCAL 
program, used to determine the cross-polarization, finds the correct 
polarized flux -- about 570 mJy for all cases (50 MHz continuum, 780 kHz 
single line, 11*780 kHz = 8.6 MHz combined spectral line).  All 
procedures utilized were identical. 

    Any ideas of what has caused this?