The idea of multibeam RFI removal is to find groups of signals that are similar (e.g. in time and freq) and that appear in non-overlapping beams. Terrestrial RFI often appears in different beams; celestial signals do not.

As of a few weeks ago, this was removing way too many (like 90%) pulses and triplets. My last change - which tightened up the notion of "similarity" in time and frequency - reduced this to about 15%, but it was still removing too much. The issue is that pulses and triplets have an attribute, namely period, that other signal types don't. Pulses or triplets that are close in time and frequency, but not close in period, should not be viewed as similar.

So, working with Eric, I added a period check to the similarity test for pulses and triplets. For pulses, we now require that the periods agree within 1%; for triplets, they must agree within half the time resolution for that FFT length.

This seems to have worked. The multibeam removal fraction is down to about 3% for both signal types, and I think we're still removing actual RFI. I reran the pipeline with this change and results are online. Please look at waterfall plots; if you find any RFI anomalies, please bookmark them and let me know.

Thanks for your continued effort. Could you explain how often are newly completed SETI@home workunits fed into Nebula and how often is "Sky coverage" updated? Are data coming from Breakthrough Listen processed by Nebula?[/url]