Description of glitches PTmon is searching for

    As a veto monitor, PTmon is searching for glitches on environmental and auxiliary channels that could be propagated to the gravity wave channel, AS_Q.  After downsampling the data to a frequency above the Nyquist of the sampling frequency, it sets up a time series of Peak-Trough or peak-to-peak values.  These are compared with two thresholds, currently set at 4 and 3.25 sigma on all channels.  For a glitch to set a trigger, it must either (1) cross the higher threshold at least once, and the lower threshold at least twice more within a specified  time window, or (2) cross the lower threshold at least four times within the same time window.  In practice, this forces a signal to go through the mean beyond the noise band before being accepted as a physical signal that could be propagated to AS_Q.  Graphically,
/\     is accepted as a glitch, but /\ is rejected.
   \/
    When simulating test data using DMTGen, this means that the parameter Q for Damped Sine or Sine Gaussian waveform must be set high enough to allow more than a single one-sided peak, and the parameter sigma must be set high enough for the Gaussian Busrts.

Other DMTGen paramters

    The other parameters that need to be set are the frequency or frequency band, specified by F, and the amplitude or power A, which needs to be large enough to produce events that will have snr > 4.  The rate should also be adjusted (if possible) not to produce glitches that are so close together that they will be recorded as a single glitch for veto purposes.  With the rate 'r' set at 0.2, some of the injections were separated by < 0.06 s.  While it would be possible to separate these, that would not be appropriate for a veto monitor, and some fancy footwork would then be needed to avoid missing the peak amplitude of slow-building bursts.

Results of test carried out so far

    Eight minutes worth of fames were generated with white noise background for each of the following signals.  Each set of triggers produced includes a big startup glitch that corresponds to lock acquisition.  Two injections, one  sine Gaussian and one damped sine are currently unaccounted for.  There are no false alarms above six sigma.

    Checking a high frequency filter 700-850 Hz, the Gaussian burst similations were re-used.  For some reason, only the first twelve frames were processed, but these were sufficient to test the filter.  There were 39 burst injections in the frames processed.  All of these were found,  and again there were no false alarms above six sigma.

Gaussian Bursts

    104 triggers were set, including one at startup, and 95 signals were injected.

    Using a 70-300 Hz bandpass filter,all of the injections were found when looking at all triggers with snr >= 4, but three had snr < 5, and so would not  normally be reported.    Two triggers were false alarms - both had snr < 6.  Some of the large amplitude and raggedy bursts had two triggers set in quick succession, and these would result in one veto.  An example is shown in this frame.  The first injection at a nominal time of 730000272.06 s had a single trigger set, but the next injection at a reported time of 730000273.44 had a pair of triggers set.

    The same set of frames was rerun to check a 700-850 Hz filter.

Sine Gaussian

    96 triggers were set for 88 injections.
Comparing these sets, there were 56 matches, seventeen cases where a pair of triggers were set for a single injection, six close pairs of injections each produced a single trigger, and one triplet of injections produced a single trigger.  Three injections and triggers were separated by ~0.5 s.  These are differences between start time (PTmon) and peak time (DMTGen).  Initially three injections were unaccounted for, but  a change in the code that shouldn't have mattered - putting braces around a single target statement for an 'if' - found two of them.

Damped Sine

    110 triggers were produced including one at startup.  These were compared with 110 signals injected.
104 were matched with time difference < 0.3 s, and 91 with time difference < 0.02 seconds.  In fact, 86 triggers were late by the same amount, and this could be corrected.

    Four of the glitches not found occurred within 0.06 s of another glitch, so the pair set one trigger for veto purposes.  One pair that set only one trigger were separated by ~ 0.5 s, and should probably have been separated, although this could mean that a low-frequency burst (< 100 Hz) might have its amplitude under-reported, using the present technique for finding this.  One injection that certainly should have been found was missed.

    There were no false alarms above 6 sigma.  There were four false alarms between 5 and 6 sigma, and the startup trigger.