[evlatests] Spectral window power variations

[Rick Perley]

         I wrote last week about the very wide range of power seen 
within the spectral windows at P-band.  Typical ratios in PSum between 
the highest frequency spectral window  and SPW 6 to 8 (typically the 
highest powers, outside of RFI) are greater than ten for most antennas 
-- and as high as 50 for some.

     Recent testing at L, S, C and X bands, using the 8-bit samplers, 
show that a similar wide disparity occurs at all of these bands as well.

     The cause at S, C, and X bands is the anti-aliasing filter within 
the T304.   This causes a strong attenuation in spectral power at the 
high frequency end at X-band, and at the low frequency end at C and S 
bands.  Typical powers in the edge spectral window is 6 to 8 dB lower 
than central spectral windows.  For the worst cases, it's as high as 12 
dB.  The situation is worst at X-band, where this attenuation is added 
to that of the receivers themselves at the high frequency end.  The SPW 
adjacent to the end (second highest at X-band, second lowest at S or C 
bands) is also affected, although not nearly as strongly. (Note that 
this rolloff is not present when using the 3-bit samplers in C or X bands).

     At L-band, due to a different tuning setup, the anti-aliasing 
rolloff is not present.  However, it's clear that gain variations within 
the receivers cause a drop in SPW power, usually at the high frequency 
end, which is 10 or more dB weaker compared to the higher-power central 
SPWs.

     The cost of this is a loss in sensitivity.  A 12 dB lower power is 
two bits, reducing the 4-bit requantizer to a 2-bit requantizer (roughly 
speaking).  Although this is not disastrous, it is avoidable, simply by 
having the requantizers adjust to the optimum levels for each subband -- 
an operation already standard in the 3-bit correlator configurations.  
Requantizer settings are now very fast -- about two seconds -- so the 
loss in observing time (if set up properly) is trivial.

     Further testing (using Cygnus A, 3C286 and 3C295, with full 
requantizer settings at P, L, S, C, and X bands) is being conducted this 
evening.