...
Many times when running the pipeline there will be an spw with unphysically high delays on one or more antennas, to the tune of nearly 250ns. We have been flagging the affected data, even if it was the only flagging wanted/needed. We then investigated whether such flagging was really necessary or even perhaps was a bad idea. Although the number of examples are small, it appears that in 8/12 cases, the flagging actually leads to a slightly worse rms in the final image. Even in the remaining cases where the flagging helped the image rms, the improvement is << 1% so is not worth our time to flag.
Analysis and Results
SRDP projects from January and February 2024 were collected and rerun through the imaging pipeline without the original SRDP flags. We singled out the observations where high delays were flagged, both on their own and with other flags. These high delays are specifically single spectral windows exhibiting high delays on the order of ~250 ns.
For each of these observation, the percent delta between heuristics from the SRDP-flags image and the no-flags image were determined. Those heuristics include the RMS of the full image in Carta, the RMS reported by the weblog, and the dynamic range from the weblog. The percent delta was calculated by the equation (SRDP RMS - NonSRDP RMS) / NonSRDP RMS or (SRDP DR - NonSRDP DR) / NonSRDP DR. A positive percent for RMS would suggest that having no flags improved the RMS while a negative percent for dynamic range would suggest that having no flags improved the dynamic range.
Nineteen observations had high delays flagged during these two months. Of those nineteen, eight of these had only high delays flagged. The percent with an improved RMS of these eight resulting from no flags was 66.7%, and those with an improved dynamic range was 58.3%. The average RMS delta was -0.03%, and the average dynamic range delta was -0.43%.
The difference in RMS and dynamic range between the different runs was very minimal. Whether flags harm or improve the image, the difference resulting from their presence is usually much less than 1%, and therefore likely negligible. The conclusion for this case is that an observation does not need to be recalibrated if the unphysical high delays are the only issue in the calibration.
Action items
- Toni Norton to populate this page with a description of her analysis and results