Optimal time-frequency resolution for minute-long gravitational-wave transients

Abstract

Spectrograms, which are time-frequency maps illustrating the amplitude evolution, are utilized for Gravitational Waves (GW) detection and analysis. Spectrograms have a particular resolution in time and in frequency. As the presented work will show later, spectrogram resolution is important for unmodeled signal recovery. Existing methods mainly rely on trying empirically to find the best resolution, namely the pixel size in both time and frequency, which could enhance signal detection and retrieval. The objective of our research is to determine the most effective time and frequency resolutions for a bank of signal types likely to be encountered, with the aim of maximizing signal recuperation. It could potentially enhance future GW analyses, offering valuable insights into the maximization of scientific information derived from minute-long unmodeled signals. Additionally, GW detection involves processing huge amounts of data, which is computationally expensive. Future detectors (Einstein, telescope, Laser Interferometer Space Antenna) will present enhanced sensitivity and alternative designs, allowing for the detection of new events, hence also requiring more data processing as well. Machine learning (ML) algorithms are increasingly utilized for challenging tasks, given the data structures’ recognization efficiency displayed by ML methods. ML algorithms being the future of the field, a performance increase by improving signal recognizability becomes relevant. Determining the optimal spectrogram resolution, notably, is a solution to the mentioned problem as it directly impacts algorithms’ signal recovery efficiency.