Deep Learning Search for Gravitational Waves from Compact Binary Coalescence

Author(s)

Mobilia, Lorenzo, Dal Canton, Tito, Guidi, Gianluca Maria

Abstract

Gravitational wave searches rely on a combination of methods, including matched filtering, coherent analyses, and more recent machine learning based pipelines. For compact binary coalescences, where signals originate from the relativistic dynamics of compact objects, matched filtering remains a central element, but its computational cost will increase substantially with the data volumes and parameter-space coverage required by next-generation interferometers such as the Einstein Telescope. Developing complementary strategies that reduce computational load while preserving detection performance is therefore essential. We investigate a hybrid approach that combines matched-filtering concepts with Convolutional Neural Networks, enabling efficient signal searches without relying on the usual $χ^2$ rejection test. Using simulated data sets that include injected signals in Gaussian noise, transient noise, and physical effects not represented in template bank, such as eccentricity, precession and higher-order modes, we show that the method achieves a detection efficiency comparable to a standard matched-filtering search while offering a more resource efficient pipeline. These results indicate that deep learning assisted searches can support sustainable gravitational-wave data analysis in future detector eras.

Figures

Caption

Examples of the TT-SNR-Map obtained by piling up the $\rho(t)$ after the normalization procedure. The left picture corresponds to Gaussian Noise solely, on the center to an injected signal and on the right the TT-SNR Map resulting from an injected signal with a glitch superimposed.

Caption

EasyResNet architecture. A convolutional stem (Conv2D-BN-ReLU) processes the input and feeds three residual blocks (ResBlock1-3). The final feature maps are reduced by global average pooling (GAP) to a feature vector, regularized with dropout, and passed to a fully connected (FC) layer for classification. Arrows indicate data flow; numbers below the blocks denote channel counts (and, where shown, spatial dimensions)

Caption

Residual block. The lower arrow represents the input passing through a sequence of convolutions and normalizations. The upper one carries the identity. The green \(\oplus\) denotes element-wise addition. Arrow directions indicate data flow through the layers (Conv2D-BN-ReLU-Conv2D-BN-Conv2D in this example).

Caption

: Distribution of injected signals for $\rho_\mathrm{opt}$ - distance Mpc - CNN output

Caption

: CDFs for Noise - Signal population

Caption

: Simulation 2: Distribution of injected signals for $\rho_\mathrm{opt}$ - chirp distance $d_\mathrm{chirp}$ [Mpc] - CNN output

Caption

: Simulation 3: Distribution of injected signals for $\rho_\mathrm{opt}$ - chirp distance $d_\mathrm{chirp}$ [Mpc] - CNN output

Caption

: BNS non-spinning ($1.4 - 1.4\,M_\odot$) superimposed signals. The TT-SNR MAP clearly encodes the presence of two signals that are closely separated in time and nearly identical in parameter space. An interference-like pattern emerges from the superposition of the signal representations produced by the template-based SNR time series, highlighting the TT-SNR MAP’s ability to capture overlapping events.

Caption

: BNS non-spinning ($1.4 - 1.4\,M_\odot$) superimposed signals, zoomed view around the merging time.

Caption

Example of a BNS ($1.4 - 1.4 M_\odot$) waveform including higher-order modes, injected with inclination $\iota = \frac{\pi}{2}$. The TT-SNR MAP shows a clear modification of the central region, where the sharp rectangular features typically present in the standard representation are smoothed out by the contribution of higher-order modes. A similar structure is also visible at lower amplitudes, repeating regularly at later times.

Caption

TT-SNR MAP of precessing NSBH system ($1.4 - 25M_\odot$) injected with inclination $\iota = \frac{\pi}{2}$. The effect of spin-induced precession manifests as a modulation of the TT-SNR MAP structure, producing beam-like features radiating from the center of the map. These patterns reflect the time-dependent orbital-plane precession imprinted on the signal morphology.

Caption

Representative example illustrating the effect of eccentricity in the TT-SNR MAP. Here it is shown a $1.4 - 6M_\odot$ non spinning system with $e = 0.3$, chosen outside the injection families for illustrative purposes. The imprint of eccentricity is clearly visible as a repeating structure beneath the main TT-SNR MAP pattern, appearing twice and introducing distinctive features associated with the non-circular orbital dynamics.

Caption

Results of the ROC curves for the 5 families injected signals campaign encoding physics not included in the template bank used to construct the TT-SNR Map. The signals are injected in Gaussian noise polluted with sine-Gaussian glitches.

Caption

: Spectrogram of systems with $M_\mathrm{c}^1 = 0.95\,M_{\odot}$ merging time at $t_1 = 38\,\mathrm{s}$ (red line), $M_\mathrm{c}^\mathrm{ref}$ time at $t_0 = 40\,\mathrm{s}$ (green line), and $M_\mathrm{c}^1 = 0.95\,M_{\odot}$ merging time at $t_2 = 42\,\mathrm{s}$ (blue line).

Caption

: Spectrogram of systems with $M_\mathrm{c}^2 = 4.35\,M_{\odot}$ merging time at $t_1 = 38\,\mathrm{s}$ (red line), $M_\mathrm{c}^\mathrm{ref}$ time at $t_0 = 40\,\mathrm{s}$ (green line), and $M_\mathrm{c}^2 = 4.35\,M_{\odot}$ merging time at $t_2 = 42\,\mathrm{s}$ (blue line).