Scientific Newsletter - April 29, 2002


log in
The Signal Candidate Scoring System
Eric Person, Jeff Cobb, Steve Fulton, Eric Korpela, Dan Werthimer

A signal candidate is a group of two or more events detected at different times in the same one-beamwidth area of the sky. Events are identified by SETI@home's pattern detection algorithms as Gaussians, spikes, pulses, and triplets. A fundamental goal of SETI@home is to identify these signal candidates and determine which ones are the most likely to represent extraterrestrial communication.

Prioritizing these candidates takes place on several levels: the event level, the candidate level, and the multiple-algorithm level.

  1. On the event level, scores are assigned to individual events that are algorithm-specific. Gaussians, pulses, spikes, and triplets all have different scoring systems unrelated to one another. Thus, we can prioritize Gaussians relative to other Gaussians, for example, but at this level we can't directly compare a Gaussian to a spike.

  2. On the candidate level, we calculate the probability that a set of persistent events could occur by noise alone. Thus, candidate scores are probabilities where values close to zero are more interesting than values approaching 1. The events comprising each candidate all come from the same detection algorithm. (For example, as of April 21, 2002, the candidates on our best signal candidates list are each composed of Gaussian events alone.) The candidate scoring formula (and a more specific explanation) is given in the box below.

    Score = (Ne Nd * FrequencyFactor * PositionFactor) / (Nd)!
    1. Ne = number of events across all frequencies
    2. Nd = number of persistent events (events occurring within the frequency window fwin)
    3. FrequencyFactor = [(fwin/ftot)Nd-1]*[(RMS(f)+Freq_uncert)/Freq_uncert]
      • fwin = size of the allowed frequency window: 125 Hz
      • ftot = total searched frequency band: 2.5 MHz
      • Freq_uncert = the barycentric uncertainty across the telescope beam: 50 Hz
    4. PositionFactor = [(Pwin/Ptot)Nd-1]*[( SQRT(RMS(ra)2+RMS(dec)2)+Pos_uncert)/Pos_uncert]
      • Pwin = size of the allowed position window: 6*6 arc minutes
      • Ptot = total searched region: 360*40*3600 arc minutes
      • Pos_uncert = pointing uncertainty of telescope beam: 3 arc minutes

    Formula Summary:
    Score is a Poisson statistic comparing the number of events from a given candidate (Nd) to the number of events expected by chance within that candidate's frequency window (fwin). In this case we use frequency and not position because, unlike position, the distribution of hits across frequency is approximately even (a requirement of Poisson statistics). A candidate's quality is also proportional to the closeness in frequency ("FrequencyFactor") and sky position ("PositionFactor") among its members.

  3. On the multiple-algorithm level, we compare spikes, Gaussians, triplets, and pulses together to identify important "hot spots" in space. For example, some detections might be identified as part of both a triplet and a Gaussian. Directly comparing these different events is difficult, since the algorithms that identify them are not orthogonal (statistically independent) to one another. Nonetheless, making such comparisons can help "highlight" certain regions of space for further study and observation.

With even the highest-scoring signal candidates, there is always a high likelihood they were created by RFI, satellites, or natural astronomical phenomena (like supernovas). Much re-assessment, re-observation, and independent corroboration is required to conclusively determine evidence of extraterrestrial intelligence. (See "identifying final candidates" for more details.) Still, as SETI@home users process more data and discover stronger candidates, our chances of answering a fundamental existential question become better and better: Are we alone?

Return to Science Newsletter main menu

Copyright © 2014 University of California