Searching for gravitational waves from stellar-mass binary black holes early inspiral

TL;DR

This paper presents a novel machine learning approach combining IPCA and CNNs to detect and estimate parameters of early inspiral gravitational waves from stellar-mass binary black holes for space-based detectors like TianQin.

Contribution

It introduces a dimensionality reduction and CNN-based framework for efficient detection and parameter estimation of gravitational wave signals, overcoming the limitations of traditional matched filtering.

Findings

Achieves 95.6% variance retention with IPCA on simulated signals.

Detection model reaches 86.5% true alarm probability at 5% false alarm for SNR 50.

Estimates chirp mass with a standard deviation error of 2.49 solar masses.

Abstract

The early inspiral from stellar-mass binary black holes (sBBHs) can emit milli-Hertz gravitational wave signals, making them detectable sources for space-borne gravitational wave missions like TianQin. However, the traditional matched filtering technique poses a significant challenge for analyzing this kind of signal, as it requires an impractically high number of templates ranging from $10^{31}$ to $10^{40}$. We propose a search strategy that involves two main parts: initially, we reduce the dimensionality of the simulated signals using incremental principal component analysis (IPCA). Subsequently, we train the convolutional neural networks (CNNs) based on the compressed TianQin data obtained from IPCA, aiming to develop both a detection model and a point parameter estimation model. The compression efficiency for the trained IPCA model achieves a cumulative variance ratio of 95.6% when applied to $10^6$ simulated signals. To evaluate the performance of CNN we generate the receiver operating characteristic curve for the detection model which is applied to the test data with varying signal-to-noise ratios. At a false alarm probability of 5%, the corresponding true alarm probability for signals with a signal-to-noise ratio of 50 is 86.5%. Subsequently, we introduce the point estimation model to evaluate the value of the chirp mass of corresponding sBBH signals with an error. For signals with a signal-to-noise ratio of 50, the trained point estimation CNN model can estimate the chirp mass of most test events, with a standard deviation error of 2.49 $M_{\odot}$and a relative error precision of 0.13.