Next: About this document ... Up: Usage Previous: Execution
Tips
- data sets dominated by noise
Pieflag should work ok without much tuning of command line parameters, because its defaults are chosen for detection experiments, in which the signal on a baseline is dominated by receiver noise.
- data sets dominated by source structure
Amplitude-based flagging may not give good results, because the measure of what is "good" data is contaminated by real fluctuations of the visibility amplitudes. Inspect the flags in Pieflag's display, looking at a single source at a time only. This will plot the thresholds on the right of the plot and give you an idea above which level data have been deemed bad. Large outliers can still be captured by amplitude-based flagging, but you shouldn't lower the threshold using the "-cutoff" switch, or you may flag good data. Try to tune the rms-based flagging instead.
The best thing for data sets dominated by source structure is probably to subtract a source model from the visibility data and to run Pieflag on that data set. The ``-filename'' option can be used to run Pieflag on the model-subtracted data set but to prepare the flagging commands for the original, unmodified data set.
- non-interactive automated flagging
If you think you understand what Pieflag is doing, and you have a lot more similar data to flag, consider using the "-noplot" command line switch to skip plotting and interactive flagging.
- too much data flagged in postprocessing (ie, lots of red points without yellow crosses in the display)
You may want to reduce the width of the window in which the running sum is computed using the -long option.
- edge channels flagged despite signal is apparently RFI-free
In the case of strong sources with large spectral indices, Pieflag's assumption that signal levels in channels are comparable breaks down, and amplitude-based flagging may be inappropriate. Rms-based flagging is not affected by this, because the noise levels are the same in all channels (to first order).
Next: About this document ... Up: Usage Previous: Execution
Enno Middelberg 2006-03-21