Gravitational waves (GWs), a solution to Einstein’s theory of general relativity, are ripples in the fabric of spacetime that carry energy across space and emanate from astrophysical sources involving extreme masses. The energy given off by these masses is high enough to perturb the spacetime. Such perturbations are observed and are recorded by LIGO-Virgointerferometers like LIGO and Virgo that generate signals corresponding to the same disturbances in spacetime. LIGO has arm length as long as 2.5 miles (~4 km) and has two US stations. In contrast, the Virgo interferometer, LIGO’s Italian counterpart, has arm lengths of 1.8 miles (~2.9 km). The upcoming space-based interferometer LISA (the Laser Interferometer Space Antenna) has an arm length of over 3,000,000 miles (~5,000,000 km)!
These signals usually are mixed with the noise that affects the short-time low-amplitude change in the gravitational wave signal through which any astrophysical event could be detected. In this project, we used a filtering method for end-to-end time-series signal processing. This method is employed for classification and regression, specifically for detecting GW signals in noisy time-series data streams collected by the LIGO-Virgo interferometers. The results are fed in an artificial neural network (ANN) model and trained to predict any astrophysical events and their corresponding features. The data used for training the model include O2 4 KHz (detectors placed at Hanford, Livingston, and Louisiana) datasets, out of which 20% of the testing datasets were used for testing the model. The model showed an accuracy of 92% with validation accuracy of 91.6% on experimenting with various model parameters.
The workflow of data generation scheme used
Accuracy and Loss plot after 100 epochs
The research paper on this study is under review process, a preprint for the same will be updated soon.