[evlatests] Correlator Offsets, (cont.)
Rick Perley
rperley at nrao.edu
Fri Jun 22 16:09:48 EDT 2007
Because Ken wondered about the reliability of my claim of there
being no offsets in the 12.5 MHz, mode '4' test done last May 30, I
reviewed the evidence. There is indeed no sign of an offset, or 'choppy
noise' at the center of the field.
However -- in more carefully reviewing the images, I noted that the
rms noise level was about 5 times what it should have been. Using
various tests, it was found that this excess noise is on every baseline,
on every channel, on every source. The scatter in both amplitude and
phase -- both as a function of time and frequency -- is 5 times too high!
I checked the gain coefficients -- they are normal. I checked the
reported Tsys -- they are normal. I then looked at the spectral line
tests done last week (Modcomp-free, in this case), during which the same
mode and BW combination was used. All data in that test had normal
sensitivities. I thus have no explanation or even a suggestion as to
what was wrong on 30 May.
Give the sensitivity in that 30 May test, I'll have to revise my
conclusion, and state that the test was not sufficiently sensitive to
detect any offsets.
Ken Sowinski wrote:
> On Wed, 20 Jun 2007, Rick Perley wrote:
>>
>> A test was run this morning, to see if the infamous offsets remain,
>> following some work yesterday by Ken and Ray.
>
> More work was done on Thursday and we think we have banished
> the infamous correlator offsets. Unfortunately there was no
> clear sign of the mechanism. It was made to go away by
> rearranging the timing of the readback of selftest results
> from the correlator so that the entire process is spread out
> over more time. There is no hint from this work that the
> problem would have been restricted to 25 MHz; I would not
> be surprised if narrower line BWs were also affected in some
> way. We are confident that continuum and 50 MHz line was not
> affected.
>
> Still, I encourage anyone with long integrations in line mode
> on empty fields to examine the data carefully and report anything
> that looks incorrect as soon as possible.
>
> Thanks to Ray and Bruce who put in the thought, effort and time
> to fix this.
More information about the evlatests
mailing list