[evlatests] WIDAR bronze dataset - C band dual pol.

[Frazer Owen]

Vivek Dhawan wrote:
>            Re-weighting the data offers no improvement.
>            --------------------------------------------
>
> I have not treated data weights with any respect so far. Now, given:
>
> 1. I am stuck at 2 (or 5) times the theoretical rms, for 5 MHz BW
>    (or 105 MHz), and
>
> 2. The image artifacts resemble the dirty beam,
>
> I thought it might be somehow related to data weights being imperfect
> over many baselines i.e., distributed over the UV coverage.  This is
> examined here.
>
> The short answer is that re-adjusting weights does not improve the
> final images. The problem is intrinsic, or has been locked in at an
> earlier stage of processing.
>
> -------
>
> Procedure: Start with previous best stage: data flagged + delay fit
> + self-cal 1sec solint A&P (5.5Jy clean-comp model model, plenty SNR)
> + BPASS every 3min using same model + averaged to 21 chan of 5MHz each.
>
> The weights in these data look realistic, poor antennas have low weight.
> This is a result of default behaviour of CALIB etc. working on data with
> intrinsic spread in SNR. This (I think) is better than using the  weights
> coming in from CASA, all 1 - no Tsys or corr-coeff normalization has been
> imposed.
>
> There are 2 ways to re-weight:
> 1. WTMOD - reset all weights to same - make a new dataset.
> 2. FIXWT - recompute weights from amp rms over a solint (3 minutes),
>            make another new dataset.
>   
    In my experience neither of these methods is useful. The only thing 
which works in practice is the T_sys based, calibrated weights.
> And 2 ways to image:
> 1. Image the new datasets directly.
> 2. Rerun CALIB on the new datasets, using the best previous model,
>    then image.
>
> I tried the 4 combinations above, averaging over all channels in the
> images but keeping the RR, LL, IF1, IF2 separate. As expected, the
> option 1,1 (equal weights and no calib) is worst, increasing rms
> by upto 2 on IF2 LL, which had the poorest data quality. None of the
> options improved the off-source rms - at best they were harmless.
> The various options did stir around the close-in sidelobes but I
> could not distill any insight from that.
>
> Onward - other ideas??
>
>
>
> |
> | The Super-Short Summary:
> |
> | __ NO dropouts. No symmetric artifacts.
> |
> | __ LCP raw data has wider range of uncalibrated amplitudes across
> |      antennas, but steady in time and calibrates out OK, finally
> |      only 10-20% worse rms than RCP. (Ant 3,IF2,LCP notably bad)
> |
> | __ Using 5 MHz channels, the rms of the central few channels is a
> |      factor 2 worse than theoretical=0.2mJy. Edge channels worsen
> |      gradually by another factor ~2. This is true of R & L.
> |
> | __ V image cubes (5 MHz) are a bit better, compared to theoretical
> |      they are a factor 1.5 worse in middle, factor 2 at edge chans.
> |
> | __ Averaging up to 105 MHz, the image rms should improve by a factor
> |      4.6. It improves by less than 2. The artifacts look more-or-less
> |      like the dirty beam.
> |
>
>
> _______________________________________________
> evlatests mailing list
> evlatests at listmgr.cv.nrao.edu
> http://listmgr.cv.nrao.edu/mailman/listinfo/evlatests
>