[evlatests] Flux Density Accuracy via Switched Power Alone
Bryan Butler
bbutler at nrao.edu
Wed Jul 10 17:08:39 EDT 2013
i'm very much against maintaining a "peculiar gain" table.
the right answer is probably to tell people to observe a standard
calibrator, unless they don't care to 20% or so. remember that, as you
stated, this was done at a moderate elevation in excellent dry weather,
so the numbers below are best case.
as an aside, you could use a robust estimator of the rms to not be
affected by outliers in the distribution (MAD, for instance, is easy and
pretty good for that).
-bryan
Rick Perley wrote, On 7/10/13 14:27 :
> Knowing both the antenna efficiencies and the values of Tcal, we
> should -- in principle -- be able to pre-calibrate our visibility data
> with an accuracy set by the knowledge of the above quantities (and of
> the effects of atmospheric extinction). To do this, the raw
> cross-correlations need to be multiplied by a factor of
>
> sqrt(Tc1.Tc2/(e1.e2))
> where e is the antenna efficiency, and Tc the value of the cal (in K).
> (There are also some constant factors, which we think we know).
> We know that this system actually works 'pretty well' -- Vivek and I
> have both stated that it should be good to 5%. But until now, nobody
> has tried to quantify this statement. Using the 'flux density data',
> taken May 2, I have used the observations of 3C286, (taken in excellent
> dry weather) at an elevation of 55 degrees, and the associated switched
> power calibration data, to see how close we get to the right answer.
>
> I note here that getting this system to work correctly would often
> remove the need for users to observe a 'standard' radio source to set
> their flux density scale.
> The bottom-line result is: Not bad, but the 5% claim is 'a little
> optimistic' at some bands.
> Below I give a table which summarizes the essential results.
> Attached are the data from which the table comes, and histograms of the
> gain factors.
> The method I used was to used CALIB to generate antenna-based
> voltage gains, after the visibilities were modified by the switched
> power. Corrections due to atmospheric opacity were applied, but no
> elevation gain correction was applied. (Hence, I used only the 55
> degree elevation data). The gain solutions represent the antenna-based
> error: A value less than 1.0 indicates the antenna has been
> over-corrected, a value less than this means the antenna has been
> under-corrected.
> A gain value less than 1.0 means that the actual antenna efficiency
> is higher than that applied, and/or the true Tcal is lower than that
> value applied.
>
> A gain value greater 1.0 means the actual antenna efficiency is
> lower than that applied, and/or the true Tcal value is higher than that
> applied.
> In the analysis, I squared the derived gain factors, so they are now
> proportional to power. The error factors are then directly applicable
> to the efficiency or Tcal. I computed, for each of the 16 frequencies,
> the mean gain factor, and the rms of the distribution about this mean.
> Nearly all mean factors are with 10% of unity, however note the
> distressingly high deviations -- particularly at S, Ka, and Q bands.
> Band Frequency RCP LCP
> MHz mean rms mean rms
> ----------------------------------------------------------------------
> L 1465 0.98 0.11 0.99 0.17
> L 1865 1.02 0.40 0.93 0.20
> S 2565 1.09 0.39 1.11 0.44
> S 3565 0.99 0.37 0.98 0.43
> C 4885 0.98 0.16 0.98 0.11
> C 6885 0.89 0.10 0.90 0.13
> X 8435 0.87 0.13 0.90 0.13
> X 11062 0.87 0.13 0.83 0.10
> Ku 14965 0.95 0.17 0.97 0.17
> Ku 17422 1.01 0.12 1.06 0.25
> K 22450 0.90 0.12 0.90 0.18
> K 25836 0.89 0.15 0.85 0.19
> Ka 28450 1.04 0.20 0.99 0.20
> Ka 36435 0.933 0.27 0.90 0.22
> Q 43340 1.03 0.28 1.01 0.30
> Q 48425 1.39 0.60 1.29 0.56
> ------------------------------------------------------------------------
>
> More interesting and useful information is shown in the histograms
> of the gain power distribution amongst antennas. Attached are eight
> plots showing these for the two frequency pairs and two polarization for
> each band.
> I haven't bothered to put the antenna numbers into the histograms
> (this is not easy to do with the software I use). For those interested
> in the actual gain values for each antenna/IF, a table is attached. Note
> the entries here are the *amplitude* antenna solutions.
>
> Comments:
>
> 1) About half of the average power gains are within the 5% limit
> quoted by Vivek and me.
> 2) The histograms illustrate that most antennas are within about 10%
> of each other, and that at most bands, there are a few very discrepant
> antennas responsible for the very high dispersions seen in the table.
> The most discrepant antennas are generally not shown in the histograms
> (for reasons of maintaining a decent horizontal scale).
> 3) There are curious and real mean offsets by band: A good example
> is X-band, where the corrected amplitude scale is 10 to 15% too high
> (gains values less than 0.9). This implies that the efficiency we are
> using is too low by that same factor, or that the Tcal values -- on
> average -- are too high by that factor. (Or, a combination of both).
> The K-band discrepancies are of the same value. But note that Ku band,
> and the lower half of Ka band, are very close to the correct value.
> 4) The higher end of Q-band values are far too low (power gain
> values typically 1.3) -- but this is easy to explain as an efficiency
> considerably lower than that being applied. This is as expected -- AIPS
> only allows a single efficiency per band (at this time) -- we know the
> efficiency sharply decreases in Q-band towards higher frequency. Also,
> the very large dispersion in the Q-band values likely reflects a larger
> variation in efficiency between antennas (unsurprising!) at this band.
> (Or, it reflects pointing errors due to (say) collimation errors).
> 5) The very high dispersion in the values, particularly at S band,
> and in the upper half of L-band, is a worry. This is very unlikely to
> be due to efficiency errors, so likely reflects errors in the Tcals.
> These are large errors -- tens of percent! In general (except at
> Q-band), I'd have expected the dispersion in these values to be at the
> 5% level. The best we have at present is about 10%, and 20% is more
> typical. This is far too large for reliable gain calibration.
> 6) So -- the question for everybody is: What should we do about
> this? We could 'fudge' either the efficiencies or the Tcal values to
> make things come out right. We could define a 'peculiar gain' factor
> which is updated as needed, and is applied when the switched power is
> applied. Individual in-situ measurement of the receiver temperatures
> and antenna efficiencies is out of the question (we don't have the
> people or the time).
>
> I propose we discuss some of this at the 'test' meeting tomorrow.
>
>
>
>
>
> _______________________________________________
> evlatests mailing list
> evlatests at listmgr.cv.nrao.edu
> http://listmgr.cv.nrao.edu/mailman/listinfo/evlatests
>
More information about the evlatests
mailing list