[evlatests] L-Band RFI Filter Test results

[Rick Perley]

    We ran a test today designed to reveal whether the sensitivity 
issues we see at L-band are due to aliasing of out-of-band signals 
within the IF system.  To do this, narrow-band RF filters were inserted 
at the output of the receiver of one antenna, so that the subsequent IF 
electronics only 'see' the signal from the band which is destined for 
the correlator.  The sensitivity was measured (via AIPS weights) before 
and after the filters were inserted.
    EVLA antennas 13, 14, and 16 worked.  Antenna 18 was out at the time. 

    Details:

    Antenna 14 was selected for this experiment. 

    Two filters were utilized:  1325/50 MHz was inserted in the RCP 
side, 1675/50 in the LCP side. 

    The VLA electronics were set for these frequencies, using 12.5 MHz BW. 

    The correlator was set into mode '4', providing correlations for the 
four parallel-hand combinations:  AA, BB, CC, and DD.   The A and C IFs 
were set to 1325 MHz, the B and D IFs to 1675 MHz. 

    Because the filters block the 1325 MHz signal in IF C, and the 1675 
MHz signal in IF B, the frequency assignments were reversed at the 
midpint of the experiment, so the AC side was tuned to 1675, the BD side 
to 1325 MHz.  In this case, the 'A' and 'D' channels were blocked by the 
filters. 

    The unfiltered state was observed both before and after. 

    The data were calibrated in the usual way, using well-known values 
of the flux density of 3C147.  Spectral line mode was used to avoid 
corrections for passband effects.  Only the central (198 kHz wide) 
channel was used for calibration, and calculation of the AIPS weight. 

    Results:

    1325 MHz, Channel 'A' (RCP). 

    In the unfiltered state, the AIPS weight was 124 units -- fourth 
highest amongst all antennas.  When the filter was inserted, the AIPS 
weight *dropped* to 108 -- a ~7% loss in sensitivity.  However, the 
reason for this is almost certainly in the low power level delivered to 
the sampler.  With the full bandwidth, the rms sampler level was about 8 
levels.  With the filter in place, it dropped to only ~1.8 levels, as 
expected due to the narrow bandwidth.   Ken adjusted the T304 attenuator 
to raise the sampler rms to 5 levels -- the AIPS weight promptly rose to 
120 units, effectively unchanged from the full bandwidth level.  We then 
further decreased the attenuation until the rms level rose back to 8 
levels.  The sensitivity promptly dropped to 7 units -- effectively no 
sensitivity at all.  It is suggested here that the reason is saturation 
of the VLA's system temperature measuring system.  Although the power 
delivered to the sampler is about the same as the full bandwidth, the 
spectral power density within the 50 MHz seen at the backend has risen 
by a factor of about 20 -- causing something back there to saturate.  
(The switched power dropped to nearly zero). 

    1675 MHz, Channel 'D' (LCP). 

    In the unfiltered state, the AIPS weight was a rather low 90 units 
-- worse than any VLA antenna by a significant factor.  (Indeed, only 
EVLA antennas 13 and 16 were worse!)  With the filter inserted, the aips 
weight dropped to 80 units.  However, this drop is likely due to the 
same low rms at the sampler, as the ratio is about the same seen in IF 
'A'.  (We adjusted the power level in this IF only to get to 8 levels -- 
and saw the same sensitivity collapse as seen in IF A -- no adjustment 
to 5 levels was made). 

    1325 MHz, Channel 'B'. 

    The unfiltered sensitivity was 122 units, the filtered sensitivity 
was 105 units -- again we explain the drop from the sampler statistics.  
No adjustment of power level was made in this IF. 

    1675 MHz, Channel 'C'. 

    The unfiltered sensitivity was a totally pathetic 30 units -- a 
factor of two lower in real sensitivity (G/T) than what this IF provided 
at 1325 MHz, and by far the worst operating antenna in the array.   With 
the filter in place, the sensitivity *rose* to 81, which we should bump 
up to about 100 to account for the digitizer issue noted earlier.  At 
this level, the antenna compares favorably with the VLA antennas.   So 
here we have clear evidence of a significant benefit provided by the 
filter.  However, the performance of this antenna in the normal 
(unfiltered) state was so bad that this must be a unique situation, 
generated by something else seriously out of range which was 'fixed' by 
narrowing the signal passband. 

    The other IF/frequency combinations (1325 in C, 1675 in B, 1675 in 
A, and 1325 in D) could not be tested in the filtered state.  To do so 
would have required reversing the filters in the antenna -- there was 
not time to do this.  In the unfiltered state, these IF/frequencies gave 
the usual results -- good at 1325 MHz, poor at 1675 MHz.