[evlatests] More on noise from spectral line

[Rick Perley]

    Gustaaf ran a test file yesterday, to investigate the noise 
discrepancy that we have recently been seeing. 

    The observations were at L-band, in both 50 MHz continuum, and 
various line combinations:  1A, 1C, 2AC, 4, (all of these with 12.5 MHz 
BW),  PA with BW = 0.781, 12.5, and 50 MHz, and PB with BW = 12.5 MHz. 

    For all correlator setups, there were two observations, one of a 
strong calibrator, the other of nearby empty field.  Unfortunately, the 
nearby field contains about 400 mJy of background sources, but this was 
not a serious issue for most of the setups (where the channelwidths are 
narrow enough that noise dominates). 

    The data were loaded as correlator coefficients, and the calibration 
was done on the central channels, using an approximately correct flux 
density.  (Any error in the flux will affect all results equally). 
    Histograms of the central channels, and maps made with the central 
channel, were made for all modes, and polarizations. 

    The results are quite odd, and worrisome. 

    A)  The only modes which gave rms noises close to the expected were 
50 MHz continuum, and mode PA, when the BW = 791 kHz.  All other gave 
noises a factor of 2 to 3 higher than expected.  A possible exception 
was mode PA with BW = 50 MHz.  For this setup, the Gibbs' ringing is so 
severe that calibration is very uncertain.  The cross-hand histograms 
for this setup on the blank field had widths close to normal. 

    B)  The originally reported problem (higher noise in RR and LL than 
in RL and LR) is repeated, but only for mode PA, BW = 12.5 MHz. 

    C) Mode PB, with 12.5 MHz BW gave equal noise in all correlations, 
but a factor of 2 to 3 too high. 

    D) All the other modes (which didn't include X-hand correlations) 
gave noise (both in histograms and in images, for both the calibrator 
and the blank field) a factor of 2 to 3 too high. 

    I recommend re-doing the test, but at X-band (where the background 
is less a problem, and the X-pol of the antennas much lower), and not 
including PA at 50 MHz.  Use 25 MHz instead.