Astrophysically motivated time–frequency clustering for burst gravitational wave search: application to TAMA300 data

Abstract

We have developed a method of ‘time–frequency (TF) clustering’ to find burst gravitational waves for TAMA data analysis. The TF clustering method on the sonogram (spectrogram) shows some characteristics of short-duration signals. Burst gravitational waveforms from stellar-core collapse of supernovae that are predicted by Dimmelmeier et al [1, 2] (DFM waveforms) have short durations on the order of 10 ms and have a large spike and ringing tail in time series. On the other hand, typical detector instrumental noise transients of the same timescale have different waveforms as like as simpler spikes. Since the numerically predicted waveforms may not be reliable given conditions and model dependency, using one search algorithm is not robust to differentiate gravitational waves from instrumental noises. Our proposal for performing the separation is to compare many parameters of the cluster that represent the signal waveform. This approach will be useful for cases when the difference between gravitational waves and noise is not clear for one parameter. We employ TF clustering to represent the waveform characteristics. We calculated the parameters of each respective cluster, such as the magnitude and the Nth momentum around the center of a power distribution of the cluster. Using these parameters, we can efficiently identify some predicted gravitational waveforms and can exclude the TAMA detector's typical unstable spike-like noises due to the instruments. Our selection criteria for TF cluster shape parameters achieved an average efficiency of roughly 50% for injected DFM waveforms of (source distance of 350 pc) with false alarm rate of ∼1 Hz. In addition, the false alarm rate for larger noises, such as SNR > 100, is improved 10-fold by applying the selection criteria for TF cluster parameters.