Gauge theoretic signal processing. II. Zero-latency whitening for early warning pipelines
规范理论信号处理 II:面向早期预警管线的零延迟白化
Low-latency gravitational-wave search pipelines are essential for providing early warning alerts of multimessenger astrophysical transients. Current production pipelines whiten the incoming data stream using acausal, linear-phase filters, which require a look-ahead data buffer that introduces several seconds of algorithmic latency. Eliminating this latency requires transitioning to causal, minimum-phase whitening filters that use only past data. However, operating causal filters under nonstationary noise requires more than simply substituting the filter type: the drifting power spectral density must be tracked without degrading the matched-filter signal-to-noise ratio, filter updates must preserve the minimum-phase condition, and the altered phase response must be compensated to maintain sky localization accuracy. In paper I of this series, we introduced a gauge theoretic signal processing framework and showed that the minimum-phase connection on the manifold of power spectra provides a geometrically exact update rule for causal filters. In this paper, we validate that framework numerically and operationally, demonstrating that parallel transport along this connection strictly preserves the minimum-phase property while exactly conserving the matched-filter signal-to-noise ratio. We numerically certify the flatness of this connection, demonstrating that the optimal filter is a path-independent state function of the instantaneous noise. Through an injection campaign on O3 data with 15,347 binary black hole signals across the LIGO-Virgo network, we confirm that this zero-look-ahead architecture preserves the detection sensitivity and interdetector timing and phase accuracy of the standard linear-phase baseline. Implementing this framework within the production SGNL matched-filter pipeline reduces the whitening latency by 1.0 second (33%) relative to the standard linear-phase baseline at a 4-second noise estimation cadence, confirmed both in controlled local tests and on live O3 replay data at production scale. Stride reduction experiments demonstrate that up to 91% of the baseline trigger latency can be eliminated with subsecond pipeline cadence.
展开 ▾首次将微分几何式的规范理论信号处理从连续形式学落到实处:用对数功率谱流形上的平坦联络做因果最小相位滤波器的平行移动,一次解决「最小相位集非向量空间」与「非稳态噪声下动态更新」两个难题,白化延迟与 PSD 估计窗长彻底解耦(恒为 2 个 stride),把省下的数秒延迟变成可编程的早期预警预算。
低延迟搜索管线的灵敏度取决于对漂移噪声功率谱的实时估计与白化,而多信使早期预警的可用时间又被算法延迟严格卡住 Messick+ 2017;过去十余年里,从最初面向早期预警的滤波架构到 SPIIR 等在线管线,都在设法压缩这一延迟 Cannon+ 2012 Kovalam+ 2022,零延迟白化滤波器的构想也早已被提出,但因非稳态噪声下的更新难题而未能进入生产管线 Tsukada+ 2018。当前 LIGO-Virgo-KAGRA 的生产管线仍依赖非因果、线性相位的 WOLA 白化,以避免最小相位滤波器的插值失稳,代价是随 PSD 估计窗线性增长的前瞻缓冲 Sachdev+ 2019 Ewing+ 2024。本文延续并验证了 Paper I 的规范理论信号处理框架 Kennington+ 2026,证明最小相位连接是唯一既因果又在谱纤维丛上平坦、且与匹配滤波内积相容的联络,并在生产 sgnl 管线中确认其灵敏度、时间延迟与相位精度均与基线统计等价 Huang+ 2025。展望未来,随着 O4/O5 灵敏度提升带来更强的非稳态性,把滤波器更新视作谱流形上的几何输运有望成为标准做法;触发延迟与 PSD 估窗解耦后,省下的时延预算可直接前移早期预警阶段,为电磁波段的快速后随与快速定位争取关键秒数 Magee+ 2021。
预印本 2026-04-28 · 刊出 2026-11-06 · 收录 2026-11-15