160
N. S. Philip
Fig. 6 The LIGO gravitational wave detector is a technological marvel that works like a Michelson
Interferometer that has arm lengths of about 4 km each. There are two detectors of similar type,
LIGO Hanford in southeastern Washington State and LIGO Livingston that is 3002 km away [13]
that will appear as changes in the length of the arms of an interferometer kept in its
path. The changes in arm length result in changes in the fringe patterns that can be
deconvolved to reconstruct the nature of the source of the gravitational wave (Fig. 6).
The Laser Interferometer Gravity Wave Observatory (LIGO) operated by Caltech
and MIT consists of two similar interferometers separated by a distance of 3002
km. Each arm of the interferometer is about 4Km long and is sensitive enough to
detect the vibrations produced by trucks moving miles away from its location. Such
arm lengths are required to produce constructive or destructive interference by the
oscillations produced by a passing gravitational wave. Two interferometers are used
since a genuine gravitational wave should produce same patterns in both, delayed
by the speed of light to reach from one to the other. This way it is possible to
isolate fringe changes due to local disturbances such as moving trucks, foot steps of
people or animals, earthquakes, etc. To improve sensitivity and to reduce the effect of
the surrounding disturbances, the arms of the interferometers are evacuated and the
mirrors are made to virtually float in the tube. In spite of all these precautions, several
kinds of noise get into the detector and are in general called glitches. Glitches and
actual signals are transients, meaning that they appear and disappear after a short time,
see Fig. 7. Though it might appear from Fig. 7 that detection and isolation of glitches
might be a straightforward problem, removal of glitches from real signal is extremely
difficult due to the poor signal-to-noise ratio (SNR) of the detector. It happens that
the glitches are concealed in the inherent noise due to various other sources making
it invisible even for the expert eye, see Fig. 8. The LIGO detector usually have SNR
ranging from 10 to 100 and detection at low SNR is really challenging. However,
since signal detection and isolation is the sole purpose of the detector, transient
classification is an indispensable part of the gravitational wave detector pipeline.
Two postdoctoral students of IUCAA, Nikhil and Sheelu [9] observed that though
the transients hidden in noise are apparently invisible, they become visible if the
wavelet transform of the signal is taken. Since wavelet transform has good time
resolution, it is possible to identify the exact location of the transient, see Fig. 9.
They also noted that the wavelet energy for each transient is distinct that the nature
of the transient can be accurately determined from the detailed coefficient wavelet
energy patterns, see Fig. 10. The next goal was to develop a machine learning tool
N. S. Philip
Fig. 6 The LIGO gravitational wave detector is a technological marvel that works like a Michelson
Interferometer that has arm lengths of about 4 km each. There are two detectors of similar type,
LIGO Hanford in southeastern Washington State and LIGO Livingston that is 3002 km away [13]
that will appear as changes in the length of the arms of an interferometer kept in its
path. The changes in arm length result in changes in the fringe patterns that can be
deconvolved to reconstruct the nature of the source of the gravitational wave (Fig. 6).
The Laser Interferometer Gravity Wave Observatory (LIGO) operated by Caltech
and MIT consists of two similar interferometers separated by a distance of 3002
km. Each arm of the interferometer is about 4Km long and is sensitive enough to
detect the vibrations produced by trucks moving miles away from its location. Such
arm lengths are required to produce constructive or destructive interference by the
oscillations produced by a passing gravitational wave. Two interferometers are used
since a genuine gravitational wave should produce same patterns in both, delayed
by the speed of light to reach from one to the other. This way it is possible to
isolate fringe changes due to local disturbances such as moving trucks, foot steps of
people or animals, earthquakes, etc. To improve sensitivity and to reduce the effect of
the surrounding disturbances, the arms of the interferometers are evacuated and the
mirrors are made to virtually float in the tube. In spite of all these precautions, several
kinds of noise get into the detector and are in general called glitches. Glitches and
actual signals are transients, meaning that they appear and disappear after a short time,
see Fig. 7. Though it might appear from Fig. 7 that detection and isolation of glitches
might be a straightforward problem, removal of glitches from real signal is extremely
difficult due to the poor signal-to-noise ratio (SNR) of the detector. It happens that
the glitches are concealed in the inherent noise due to various other sources making
it invisible even for the expert eye, see Fig. 8. The LIGO detector usually have SNR
ranging from 10 to 100 and detection at low SNR is really challenging. However,
since signal detection and isolation is the sole purpose of the detector, transient
classification is an indispensable part of the gravitational wave detector pipeline.
Two postdoctoral students of IUCAA, Nikhil and Sheelu [9] observed that though
the transients hidden in noise are apparently invisible, they become visible if the
wavelet transform of the signal is taken. Since wavelet transform has good time
resolution, it is possible to identify the exact location of the transient, see Fig. 9.
They also noted that the wavelet energy for each transient is distinct that the nature
of the transient can be accurately determined from the detailed coefficient wavelet
energy patterns, see Fig. 10. The next goal was to develop a machine learning tool