The output of laser interferometric gravitational wave detectors consists principally of noise, of which the primary sources are fundamental: photon shot noise, thermal noise and the Heisenberg uncertainty in the mirror positions. Experience suggests, however, that there are many other noise sources present as well, not all of which will be clearly identifiable. The goal of detector characterization is to analyze noise from all sources, using known properties whenever a source can be identified, and obtain a statistical description in a way which will best aid detection while minimizing a false rejection rate. With this aim, my current research interests can be grouped into the following categories:
A wide range of techniques for gravitational wave detection are currently under investigation, the most common being matched filtering and time-frequency methods. However, many signal detection methods assume that the underlying noise has a Gaussian distribution. This is generally not the case: it is known that there are many non-Gaussian noise components present in interferometer data. These include coherent spectral lines at harmonics of the mains frequency, resonances of the mirror suspension wires, and other as-yet poorly understood sources. Characterizing such noise sources is essential to the successful detection of gravitational waves.
Characterizing noise well enough to remove it should increase the residual signal-to-noise ratio, making detection of gravity wave events more likely. However, such ``cleaning up'' of the data may degrade it's statistical properties to the point where no effective enhancement of detection is gained. We are presently analyzing the statistical properties of data and it's frequency distribution before and after such procedures, thereby allowing a quantitative evaluation of the procedures being considered.
The steps outlined above provide detailed information about the distributions of frequencies over the long term. When these distributions are known, likelihood ratio comparison can be used to identify whether a section of data is ``typical'', or else exhibits some transient phenomena. This may be due to some instrumental property, in which case the tests can be used as a diagnostic tool, or it may indicate the presence of a signal which can then be investigated more thoroughly.