Deep learning classification of the continuous gravitational-wave signal candidates from the time-domain F-statistic search. (arXiv:1907.06917v1 [astro-ph.IM])
<a href="http://arxiv.org/find/astro-ph/1/au:+Morawski_F/0/1/0/all/0/1">Filip Morawski</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Bejger_M/0/1/0/all/0/1">Michał Bejger</a>, <a href="http://arxiv.org/find/astro-ph/1/au:+Ciecielag_P/0/1/0/all/0/1">Paweł Ciecieląg</a>
Many potential sources of gravitational waves still await for detection.
Among them, particular attention is given to a non-axisymmetric neutron star.
The emitted, almost monochromatic signal, is expected to be detected in the
near future by LIGO and Virgo detectors. Although the gravitational waves
waveform is well known, its small amplitude makes it extremely hard to detect.
The accepted approach in searching for continuous gravitational waves is a
matched filter technique, known as the F-statistic method. The method consists
in cross correlation of the collected data stream with signal templates in the
frequency domain. Thus, for an all-sky search in which the parameters of the
sources are not known, large number of templates have to be checked and
therefore a large number of candidate gravitational-wave signals is produced
and further analyzed. In this work, we propose deep learning as a fast method
of classification for various types of candidates. We consider three types of
signals: the Gaussian noise, the continuous gravitational wave, and the
stationary line mimicking local artifacts in the detector. We demonstrate one
and two-dimensional implementations of a convolutional neural network
classifier. We present the limitations of our model with respect to the various
signal-to-noise ratios and frequencies of the signal. The following work
presents deep learning as a supporting method for the matched filtering
detection pipeline.
Many potential sources of gravitational waves still await for detection.
Among them, particular attention is given to a non-axisymmetric neutron star.
The emitted, almost monochromatic signal, is expected to be detected in the
near future by LIGO and Virgo detectors. Although the gravitational waves
waveform is well known, its small amplitude makes it extremely hard to detect.
The accepted approach in searching for continuous gravitational waves is a
matched filter technique, known as the F-statistic method. The method consists
in cross correlation of the collected data stream with signal templates in the
frequency domain. Thus, for an all-sky search in which the parameters of the
sources are not known, large number of templates have to be checked and
therefore a large number of candidate gravitational-wave signals is produced
and further analyzed. In this work, we propose deep learning as a fast method
of classification for various types of candidates. We consider three types of
signals: the Gaussian noise, the continuous gravitational wave, and the
stationary line mimicking local artifacts in the detector. We demonstrate one
and two-dimensional implementations of a convolutional neural network
classifier. We present the limitations of our model with respect to the various
signal-to-noise ratios and frequencies of the signal. The following work
presents deep learning as a supporting method for the matched filtering
detection pipeline.
http://arxiv.org/icons/sfx.gif
