Extraction of binary neutron star gravitational wave waveforms from Einstein Telescope using deep learning
基于深度学习的爱因斯坦望远镜双中子星引力波波形提取
In the future, the third-generation gravitational wave (GW) detectors, exemplified by the Einstein Telescope (ET), will be operational. Many binary neutron star (BNS) coalescence events are expected to be observed. There are many machine learning works on binary black hole data analysis published, but works on BNSs are fewer. This paper explores the extraction of BNS waveforms from ET. Drawing inspiration from summed parallel infinite impulse response's matched filtering approach, we introduce a novel framework based on leveraging deep learning for BNS waveform extraction. By integrating denoised outputs of time-delayed strain, we can reconstruct the embedded BNS waveform. We have established three distinct BNS GW denoising models, each tailored to address the early inspiral, later inspiral, and merger phases of BNS GWs, respectively. To further regulate the waveform shape, we propose an amplitude regularity model which takes the denoised output as input and output the regulated waveform. The experiments conducted on test data demonstrate the efficacy of the denoising models and the amplitude regularity model, as well as the overall waveform construction method. To the best of our knowledge, this marks the first instance of deep learning being utilized for the task of BNS waveform extraction. We also introduce a method for detecting BNS GWs that relies on waveform extraction. Our detection results demonstrate that this proposed method can accurately identify all GW signals with a SNR exceeding 20, without any false triggers. Furthermore, we believe that this GW extraction method also holds great promise for various other applications in BNS GW data processing, including localization and parameter estimation.
展开 ▾受SPIIR匹配滤波通道划分启发,提出分段降噪结合幅度规律模型的新框架,首次解决BNS长信号、高频率并合的降噪难题;验证Transformer与幅度模型在低信噪比下显著提升波形重合度。
传统引力波搜索依赖模板匹配滤波,但面对第三代探测器如爱因斯坦望远镜的巨大计算压力,深度学习已被广泛用于探测与参数估计,其中针对双黑洞信号的降噪已取得进展[Wei & Huerta 2020; Chatterjee+ 2021]。本课题组此前提出基于波形包络的搜索模型[Ma+ 2023]和利用降噪预测匹配信噪比的框架[Ma+ 2024],但双中子星波形的降噪与重建仍然空白。本文借鉴SPIIR方法将数据划分为多通道,分别处理早期旋进、晚期旋进和并合阶段,并借助Transformer捕获长程依赖,辅以幅度规律模型,首次实现了BNS完整波形的提取。实验在信噪比20以上时重合度超90%,为实时预警和定位奠定了基础。未来可结合更精确的波形模型与低延迟管线,进一步提升低信噪比信号的复原能力,直接助力多信使天文学的快速跟进[Baltus+ 2021]。
预印本 2024-06-22 · 刊出 2025-06-30 · 收录 2026-10-13