Next: Bibliography Up: Automated Editing of Radio Previous: Examples

Conclusions

Pieflag can efficiently search interferometry data for man-made RFI. The implemented algorithms can deal with a variety of observational circumstances, are robust and find virtually all RFI-affected data which would be found by visual inspection. Usage of Pieflag makes the flagging procedure much faster, relieves observers from an annoying part of the data calibration, and can easily be set up to run in data calibration scripts without any human interaction.

Figure 2: Top: Visibility amplitudes of one frequency channel with RFI on one baseline during an observation at 1.4GHz. Dots indicate unflagged data, crosses indicate data which have been flagged by Pieflag. 20 pointings were observed for one minute each, only one of these has a flux density which is significantly higher than the receiver noise, of around 0.2. After 10:00h, the data in this channel are dominated by terrestrial RFI, whereas the reference channel is RFI-free. Note the few data at 11:50h with amplitudes below 0.2, which have been flagged by the second postprocessing step because they are surrounded by bad data. Bottom: The visibility amplitudes of the reference channel used to detect the bad data in other channels. There are no obvious outliers or noisy sections.

Figure 3: Visibility amplitudes on the same baseline as in Figure 2, but showing another frequency channel with a different kind of RFI, and only 4.5h of data. As the outliers have large amplitudes, the amplitude range of 0 to 1 has been magnified in the lower panel for clarity. This plot illustrates the effect of the two postprocessing algorithms. Where only few outliers have been detected, such as between 4:30h and 5:30h, only small margins around the bad data were flagged during postprocessing. However, between 7:45h and 8:30h, the fraction of bad data within a 30min window exceeded the threshold of 0.15, and extended margins were flagged. The gaps in the data are due to calibrator observations.

Figure 4: Top: Visibility amplitudes of one channel on a baseline observing a source with significant structure. The amplitudes are therefore not noise-dominated but source-dominated, and the amplitude-based flagging failed, except for the data taken after 16:30h, when the amplitude exceeded

. Two periods of RFI have been flagged by rms-based flagging, one between 6:30h and 7:40h, and one between 15:00h and 15:30h. Margins have been flagged around both of these periods in postprocessing. Bottom: The reference channel on the same baseline. The interference at the beginning of the experiment is also present, but the usage of the median instead of the mean when computing comparison values tolerates small amounts of bad data in the reference channel. Data courtesy of M. Dahlem.

Figure 5: Images made from a 1.4GHz ATCA detection experiment with a net integration time of 25min. The effective bandwidth is 208MHz, and the data have been calibrated in Miriad according to the Miriad User Guide, then phase self-calibrated, and imaged. The same colour transfer function has been used for the two images. Top: No flagging was carried out, and the noise in the image is 0.5mJy, approximately four times the thermal noise limit of 0.13mJy (as reported by the Miriad task ``invert''). Bottom: The same data, but 10.6% of the data have been flagged using Pieflag. The image sensitivity is now 0.16mJy, and many more sources than without flagging are clearly visible.

Next: Bibliography Up: Automated Editing of Radio Previous: Examples