Noise in various interferometer systems can sometimes couple non-linearly to create excess noise in the gravitational wave (GW) strain data. Third-order statistics, such as bicoherence and biphase, can identify these couplings and help discriminate those occurrences from astrophysical GW signals. However, the conventional analysis can yield large bicoherence values even when no phase-coupling is present, thereby, resulting in false identifications. Introducing artificial phase randomization in computing the bicoherence reduces such occurrences with negligible impact on its effectiveness for detecting true phase-coupled disturbances. We demonstrate this property with simulated disturbances—focusing only on short-duration ones (lasting up to a few seconds) and employing mainly the auto-bicoherence in this work. Statistical hypothesis testing is used for distinguishing phase-coupled disturbances from non-phase coupled ones when employing the phase-randomized bicoherence. We also obtain an expression for the bicoherence value that minimizes the sum of the probabilities of false positives and false negatives. This can be chosen as a threshold for shortlisting bicoherence triggers for further scrutiny for the presence of non-linear coupling. Finally, the utility of the phase-randomized bicoherence analysis in GW time-series data is demonstrated for the following three scenarios: (1) Finding third-order statistical similarities within categories of noise transients, such as blips and koi fish. If these non-Gaussian noise transients, or glitches, have a common source, their bicoherence maps can have similarities arising from common bifrequencies related to that source. (2) Differentiating linear or non-linear phase-coupled glitches from compact binary coalescence signals through their bicoherence maps. This is explained with a simulated signal. (3) Identifying repeated bifrequencies in the second and third observation runs (i.e. O2 and O3) of LIGO and Virgo.

中文翻译：

---

识别非线性耦合引起的引力波数据中的噪声瞬变


各种干涉仪系统中的噪声有时会非线性耦合，从而在引力波 （GW） 应变数据中产生过多的噪声。三阶统计量，如双相干和双相，可以识别这些耦合，并帮助区分这些耦合与天体物理GW信号。然而，即使不存在相位耦合，常规分析也可以产生较大的双相干值，从而导致错误的识别。在计算双相干性时引入人工相位随机化可以减少此类事件的发生，而对其检测真实相位耦合干扰的有效性的影响可以忽略不计。我们通过模拟干扰来演示这一特性——只关注持续时间短的干扰（持续几秒钟），并在这项工作中主要采用自双相干性。当采用相位随机双相干时，统计假设检验用于区分相位耦合干扰和非相位耦合干扰。我们还获得了一个双相干值的表达式，该值使假阳性和假阴性的概率之和最小。这可以选择作为筛选双相干触发器的阈值，以便进一步审查是否存在非线性耦合。最后，在以下三种情况下证明了相位随机双相干分析在 GW 时间序列数据中的效用：（1） 在噪声瞬态类别中查找三阶统计相似性，例如光点和锦鲤。如果这些非高斯噪声瞬变或毛刺具有共同的源，则它们的双相干图可能具有与该源相关的常见双频产生的相似性。 （2） 通过双相干图区分线性或非线性相位耦合毛刺与紧凑的二进制合并信号。这可以通过模拟信号来解释。（3） 在 LIGO 和 Virgo 的第二次和第三次观测运行（即 O2 和 O3）中识别重复的双频。