Signal times and multiplet scores

The biggest issue recently involves timing: specifically, how the times of the signals in a multiplet should affect the multiplet's score.

Dan W. observed that many of the high-scoring multiplets in recent scoring runs consist of two signals separated by a second or less. These could be ET, and we don't want to discard them completely. But SETI@home is designed to find faint "beacons": transmitters that are constantly on, and for which we'd expect to find signals spanning years (assuming that the sky location is one that we've observed repeatedly).

One of our scientific goals is to quantify the sensitivity of the SETI@home "instrument" by seeing whether we detect birdies with various powers and other parameters. By "detect" we mean that a multiplet with signals from that birdie is in the top-scoring 1000 (or so) multiplets. If the multiplet list is clogged with these 2-signal multiplets, it will interfere with our sensitivity estimates.

Our first attempt at dealing with this issue was to add a "time factor" to multiplet scores, penalizing multiplets whose signals were clumped together in time. I won't describe this in detail because I never fully understood it (Eric came up with the formula). In any case, we eventually decided that it wasn't doing what we wanted.

Then we went to something simpler: penalize multiplets whose time range (i.e. the time from first signal to last signal) is less than 60 seconds. That got rid of the 1-second multiplets, but now there were a lot of high-scoring 65-second multiplets.

As a stop-gap measure, I got rid of the time factor in scoring, and changed the web interface so you can specify limits on time range; e.g. you can look at the high-scoring multiplets with time range at least a day, or a year.

I think that to address this issue we'll need to have a time factor that takes into account observation time. A 1-second multiplet shouldn't be penalized if we observed that sky position for only 1 second. But if we observed it for hours, over a period of years, then it should not have a high score.

Drifting RFI and birdies

I noticed that some birdies were having most of their signals flagged as RFI. Some amount of removal is normal - e.g., if a birdie's frequency range happens to overlap an RFI "zone". But I was seeing a lot of removal by the drifting algorithm. In some cases the algorithm was being too aggressive. Specifically, the signals in one of the triangular "bins" were clumped at one or two times, rather than being spread out over the bin's time range, as we'd expect for drifting RFI. Here's an example:



To address this, I refined the drifting algorithm so that it computes the time entropy of the signals in each bin, and flags bins only if this entropy is above a threshold. This seems to work in some but not all cases. More work is needed.


Web interface additions

I added some things to the web interface:

• "Mega waterfall plots" that show the spikes in a birdie (or in a multiplet, or in both) over their entire time and frequency range. This makes it easy to check whether a birdie multiplet contains as many signals as it should. These plots can be accessed via "plot" links on birdie and multiplet pages.
  
• The option to display waterfall plots with period (for pulses and triplets) and delay (for autocorrs) rather than frequency. This is useful because a group of pulses can be from the same source only if their periods are similar.
  
• The ability to bookmark birdies.
  
• Links from multiplet pages to birdie pages.
  

David:

I hope your changes would not reject an old style CW Morse code like signal which would essentially be a narrow band switched beacon, the switching carrying the data. Or perhaps a civilization using a planetary radar as we do where the signal is narrow band short pulses that the galactic medium will transform. That I know will be a tough one as we have to reject our own radar signals. Depending on the receiver you might have some luck looking at the rise and fall times. Our signals should be much quicker rise and fall were as a signal that has traveled some light years will have a much slower rise and fall but otherwise be identical.

Gary

What the hell is a Birdie ?

---

What the hell is a Birdie ?

Birdies. ;-)

Cheers.

What the hell is a Birdie ?

Birdies. ;-).

Or more properly http://www.naic.edu/~phil/rfi/rfilist.html

1) How to join to Nebula?

2) Perhaps it is better to use a neural network for very fast reliable RFI removal?

---

Nebula is not a project that we can join in the same we we join SETI@Home, for more information I suggest you read the various papers and posts by Dr Anderson:
https://setiathome.berkeley.edu/nebula_web/

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?