[evlatests] WIDAR RFI Tolerance test

[Rick Perley]

    The 'Golden 3C286 Dataset', taken on 28 July, has provided (as 
reported earlier) some excellent images of the field of 3C286, with 
dynamic ranges up ~200,000:1.  The observations were taken in RCP only, 
with 11 antennas at L-band,with 4 subbands of 128 MHz width each, and 
were about 6.5 hours long. 

    The imaging done earlier were utilized only a single sub-band -- the 
one most free of RFI (indeed, almost entirely free).   I report here on 
an imaging test done on sub-band #3 -- the one most heavily clobbered by 
RFI. 

    The AIPS program SPFLG is very useful in perusing for RFI (except 
that for this purpose, and ability to utilize a logarithmic transfer 
function would be very useful!!!).  In displaying portions of subband 3 
(which extends from 1556 to 1684 MHz), it was noted that (following 
basic calibration) that the RFI which occupies nearly the entire 
frequency range from the lower frequency edge through 1628 MHz has an 
intensity of up to 5000 Jy, and is typically 1000 Jy.  I noted that 
there is a narrow slot, of width 3 MHz, centered at 1613 MHz, which 
showed no RFI at all -- this lies between the Glonass and Iridium 
transmissions -- each of which are thousands of Janskys at times, and 
are located about 5 MHz (ten channels) away on each side. 

    The test was then to determine the imaging performance within this 
apparently narrow empty slot. 

    I extracted, with SPLIT, the 6 'empty' channels, applying Hanning 
smoothing to damp down the Gibbs ringing from the adjacent RFI.  I did 
not apply any bandpass solution.  This formed a single channel of width 
3 MHz, centered at 1613 MHz.  Perusal of these amplitudes showed a 
nearly complete absence of any instabilities in amplitude or phase which 
might be the signature of nearby RFI causing 'ringing' or 'echoes' in 
the adjacent channels.  Near the middle of the 6.5 hour run is about a 
half of of ratty data -- there is nothing special about the behavior of 
the RFI signals at that time, so I attribute this to actual emission at 
the frequency of the narrow slot.    I removed these on the basis of the 
calibration solutions (via EDITA), and by CLIP on the final data 
following calibration. 

    No 'heroic' measures were incorporated in the calibration -- I 
simply started with a point source model, and iterated twice as the 
background sources appeared.  (No symmetrization occurred).   As noted 
above, EDITA was employed to remove antennas with deviant solutions -- 
nearly all of these were in the short period of time noted above. 

    The final image is quite outstanding:  rms noise of 0.27 mJy, so a 
DR of about 52,000:1.  The background 'noise' looks normal for the 
amount of data actually gridded -- we are likely noise-limited.  This 
noise is a factor of about 20 million below the peak RFI noted only 6 
MHz away on each side (in frequency). 

    This little test strongly indicates the correlator is linear to a 
very high degree -- but certainly does not constitute any proof of such.