Page 1 of 1

Amplitude self-calibration

PostPosted: Sun Aug 31, 2014 6:03 pm
by thomas
I have encountered a problem with amplitude self-calibration. I have been applying this to individual pointings at 1.3-1.8 GHz. I am finding that amplitude selfcal is affecting the fluxes of faint (SNR < 50 or S < 1 mJy) sources – see plot below which compares the flux after amplitude selfcal with the flux before amplitude selfcal (in both cases I applied phase selfcal). I used all CLEAN components above 5 sigma as input model for amplitude selfcal. Could this occur if there is a significant amount of flux which is missing in the model? I did not set options=noscale, i.e. I am using the default which is to scale the gains so that average solution amplitudes is 1.0. Am I right in thinking that scaling the gains in this way is supposed to avoid this problem?

Amplitude selfcal does a good job at removing artefacts around bright sources but for some pointings also introduces stripes across the image because of anomalously high amplitude solutions. I have found that a ‘solution’ to this problem is to flag the data where the difference between the visibilities and model is greater than 5 sigma. I did this using uvmodel, setting options=flag and sigma=5. This can flag up to 0.5% of the data and seems to be quite effective at removing bad selfcal solutions. I am mentioning this in case it might be related to the above problem.

Re: Amplitude self-calibration

PostPosted: Sun Aug 31, 2014 8:19 pm
by ste616
Hi Thomas,

Have you tried using gpboot to put the rescaled amplitudes back onto the proper flux scale?

Re: Amplitude self-calibration

PostPosted: Mon Sep 01, 2014 10:08 am
by Mark.Wieringa
Hi Tom,

Since the strong sources show up with the correct flux, the scaling is correct. I think you are detecting the bias that selfcal can put into the gains when you have few antennas. Any sources not fully contained in the model tend to get absorbed into the gains and are then reduced in amplitude in the image after selfcal. The effect is of the order of 2/N, where N is the number of antennas, so it could be 30-40% for ATCA in the worst case of 1 solution per integration for continuum data. You can reduce the effect somewhat by using longer solution intervals and wider frequency bins - but since the data tends to be correlated in time and freq, the improvement is not as large as you might hope.
For your case of mosaic data, there is often not a lot you can do, since the data is sparse - doing fewer solutions (smaller nfbin) across the bandwidth when mfs imaging should reduce the effect. The same effect will reduce the noise in the calibrated data by absorbing the visibility noise into the gains. I looked at this stuff during my PhD... (



Re: Amplitude self-calibration

PostPosted: Sat Sep 27, 2014 4:11 pm
by thomas
Hi Mark,

Thanks, that’s very helpful to know. I’ve discussed this problem with other members of the ATLAS team and given that this is a fundamental problem due to the low number of antennas we’ve decided not to apply amplitude selfcal. We are extracting source counts from the data so it’s important to get accurate fluxes for faint sources. Most of the improvement in the dynamic range comes from applying phase selfcal and we can live with the artefacts that remain in the image due to not applying amplitude selfcal.