[daip] UVLIN & UVLSF questions
Bryan Butler
bbutler at aoc.nrao.edu
Mon Jun 18 20:00:07 EDT 2001
hi eric,
>> FLUX.......Max. residual flux allowed for unity weight. This
>> should be 6-8 times the rms noise for a single
>> channel in a 10 second integration.
>>
>> does it *really* mean in a 10 second integration, or is it
>
> No it means weight 1 which used to be the weight assigned a
>10-sec integration.
>
>> in the integration time that is in the data (i.e., the rms
>> per channel visibility)? if it *is* 10 seconds, why? i
>> can find nothing in the data that scales the residual or
>> the flux cutoff from the true integration time to 10 seconds,
>> so i suspect that it's the true integration time, but if so,
>> why does it so explicitly say "in a 10 second integration"
>> in the help file?
>
> The cutoff occurs when
> abs(Resid) > FLUX * sqrt(wt)
>where the the data weight is assumed = 1/sigma**2.
with the new weighting scheme, it might be nice to simply be able
to specify a cutoff such that the flagging occurs when:
abs(Resid) > FLUX
i.e., ignoring the weights.
>>
>> now, the UVSLF explain file says:
>>
>> FLUX.......If the residual flux in any channel used to fit the
>> baseline exceeds FLUX, then the spectrum for that time, IF,
>> and polarization is fully flagged. <= 0 => 1.0E20
>>
>> which is more straightforward in a sense, but gives no guideline
>> (similar to the 6-8 times rms in UVLIN) to the user. but then
>
> which is no longer really right
>
>> further down in the explain file for UVLSF it says:
>>
>> UVLSF differs from UVLIN in a number of ways. UVLIN does its
>> flagging only on the peak residual in the channels selected for
>> fitting and scales the peak by the square root of the
>> integration time before deciding on flagging. This is correct
>> for peak residuals due solely to thermal noise but is an added
>> confusion to the user. In any case, you should set the FLUX
>> cutoff at a conservative level (6 - 8 times the expected sigma
>> in the usual integration time).
>>
>> OK - here is the same recommended factor of 6-8 (why isn't it
>> somewhere above, where folks will actually see it?). but, is the
>> "usual" integration time the one that is present in the data, or
>
> usual here means the usual one in the current data set,
>assuming that there will be some short integrations at the end of
>scans.
OK...
>This advice is placed where users needing advice look. The
>fact that FLUX cuts off in a simple way is stated where lazy users who
>think they know what they are doing read.
fine...
>> and one final question - are both UVLIN and UVLSF currently
>> supported?
>
>Any task in AIPS is supported if it is distributed.
OK...
>BTW - I do not see you volunteering to write Explain files for tasks.
>The explain files should not be written by the programmers since they
>use the same words and points of view in the help and explain area.
>But the last time a scientist contributed an explain file was < 1990.
i helped write the explain file for ELINT, which, IIRC, was around
1995 or 1996. i wrote the original explain file for SETFC, which,
if i'm not mistaken, was incorporated into the "official" one.
this was summer of 1999, if memory serves. i will be happy to
work on any Explain file that you think would benefit from my input.
basically, i haven't been asked. nor, i would guess, have any of
the other scientific staff members. i may be mistaken there, and
if i am, apologize. but in my own case, i know it to be true.
-bryan
More information about the Daip
mailing list