Vetoing techniques (time domain outlier test)

In an ideal world, the output of an interferometer would be a stationary signal described by Gaussian statistics (with very rare superposed binary inspiral chirps and other gravitational-wave signals). This is unfortunately not the case, as can be quickly determined by simply listening to the raw (whitened) interferometer output. Typically the output is a stationary-sounding hiss, interrupted every few minutes by an obvious irregularity in the data stream. These are typically ``pops", ``bumps", ``clicks", ``howlers", ``scrapers" and other recognizable categories of noises. In at least some cases, there are ``suspects" for these events. For example the pops and bumps might be problems in any of the hundreds of BNC cable connectors used in the instrument.

It is an unfortunate fact that the output of an optimal filter strongly reflects these events. As you have seen in the previous section, a delta-function-like impulse signal in the IFO output can cause a large signal in the optimal filter. And in practice, this happens all of the time - the outputs of optimal chirp filters are frequently triggered by identifiable events in the IFO data stream that are clearly not binary inspiral chirps. Distinguishing these events from real inspiral chirps is called vetoing. We have found that two vetoing techniques work particularly well.

The first technique operates in the time domain, and is documented in the routine is_gaussian(). The idea is straightforward: if a chirp detector (optimal filter) is triggered, then we look in the data stream for an impulse event that might be responsible. Such events can be found by looking at the statistical distribution of the points in the time domain. If this distribution is significantly non-Gaussian then it indicates that some large transient event caused the filter to trigger, and the event is rejected. In Figures and we show a typical stretch of time-domain raw interferometer output, that does not contain any outlier points. This stretch of raw data ``passes" the time-domain outlier test. Figures and show a typical stretch of time-domain raw interferometer output, that does contain outlier points, and ``fails" the time-domain outlier test.

Figure: A short stretch of raw IFO data in the time domain, which passes the outlier test.
Figure: A histogram of this data shows that it has no outlier points.

Figure: A short stretch of raw IFO data in the time domain, which fails the outlier test.
Figure: A histogram of this data shows that it has a number of outlier points - which is why it fails to outlier test.