AI 中文总结
研究人员开发了基于SGN框架的开源流程SGNAX，统一了snax与omicron的搜索功能，可高效分析引力波探测器数据，经测试其性能与精度符合要求，还修正了snax的振幅误差。
AI 中文摘要
我们提出了SGNAX，这是一款用于引力波探测器表征的开源流程，可从单一流式数据流图中输出匹配滤波与超额功率瞬变触发信号。SGNAX构建于Stream Graph Navigator（SGN）框架之上，将snax的多速率正弦高斯匹配滤波搜索与omicron的多分辨率Q变换搜索统一为一个Python原生包。干涉仪的辅助通道被作为共享数据读取和白化阶段的并行分支分析，随着通道数量增加，降低了单通道处理成本。对16 kHz应变的全天数据进行分析仅需14分钟，在32个通道时，Q变换处理的单通道效率是单个通道时的2.8倍。匹配滤波相关计算使用PyTorch，可在CPU或GPU上运行。数据源包括离线帧缓存、共享内存缓冲区和arrakis分发服务，相同配置支持离线和在线运行。匹配滤波的端到端触发延迟约为5秒，而Q变换的延迟为数十秒。注入实验中，在参数处于预期模板失配范围内时，可恢复99.4%的可恢复正弦高斯注入，且宽带白噪声爆发的恢复结果与已确立的事件触发生成器一致。在24小时的存档LIGO应变数据上，SGNAX在10-100 Hz频段重现了omicron触发群体，触发率和信噪比（SNR）的偏差在几个百分点以内；在生产辅助通道上，它以每个区间80%的符合率恢复了snax的显著特征群体。经注入校准的重新实现还发现，生产snax存在多波段振幅误差，该误差会将25.6 Hz以下的报告信噪比放大√2至2倍，我们已确定该误差的机制并给出修正方案。
英文摘要
We present SGNAX, an open-source pipeline for gravitational-wave detector characterization that delivers matched-filter and excess-power transient triggers from a single streaming dataflow graph. Built on the Stream Graph Navigator (sgn) framework, SGNAX unifies the multi-rate sine-Gaussian matched-filter search of snax and the multi-resolution Q-transform search of omicron in one Python-native package. Auxiliary channels from an interferometer are analyzed as parallel branches sharing data-read and whitening stages, reducing per-channel processing cost as channels are added. A full day of 16 kHz strain is analyzed in 14 minutes, and Q-transform processing is 2.8 times more efficient per channel at 32 channels than at one. Matched-filter correlations use PyTorch and run on CPU or GPU. Data sources include offline frame caches, shared-memory buffers, and the arrakis distribution service, with the same configuration supporting offline and online operation. The matched filter delivers triggers at about five seconds end-to-end latency, while the Q-transform operates at latencies of tens of seconds. Injection campaigns recover 99.4% of recoverable sine-Gaussian injections with parameters within the expected template mismatch and show broadband white-noise-burst recovery consistent with established event-trigger generators. On 24 hours of archival LIGO strain, SGNAX reproduces the omicron trigger population at 10--100 Hz, with trigger rates and SNRs agreeing to a few percent. On production auxiliary channels, it recovers the snax loud-feature population with 80% per-bin coincidence. The injection-calibrated reimplementation also reveals a multiband amplitude error in production snax that inflates reported SNRs below 25.6 Hz by factors of $\sqrt{2}$--2, for which we identify the mechanism and correction.
Comments25 pages, 15 figures, 5 tables
