Adding glitches

Real detector data carry non-Gaussian transients aka glitches that are mostly local to one detector. gwforge_glitch simulates such glitches as sine-Gaussian farts

\[h(t) = A \, e^{-(t - t_0)^2 / \tau^2} \cos\left(2\pi f_0 (t - t_0) + \phi\right), \qquad \tau = \frac{Q}{\sqrt{2}\,\pi f_0},\]

following Kat Chatziioannou et al. 2021, arXiv:2101.01200, and can add them incoherently in every detector. Currently, Claudio and I have shipped only two varieties — blips and tomtes:

[IFOS]
detectors = ['CE20', 'CE40', 'ET']

[Blip]
rate = 1
frequency-range = [50, 500]
q-range = [2, 6]
snr-range = [5, 200]
snr-index = 2.3
phases = [0, 3.141592653589793]
seed = 0

[Tomte]
rate = 3
frequency-range = [30, 65]
q-range = [4, 12]
snr-range = [5, 200]
snr-index = 2.2
phases = [3.141592653589793]
seed = 0

These glitches are Poisson distributed with mean rate per hour times the file duration (rate may be a per-detector dict, {'CE40': 1.3, 'ET1': 0.65, ...}); \(f_0\) is log-uniform in frequency-range; the optimal SNR follows the power law \(p(\rho) \propto \rho^{-\alpha}\) with \(\alpha\) = snr-index on snr-range (snr-index = 1 is log-uniform); \(Q\) is uniform in q-range; \(\phi\) is drawn from phases when given, else uniform; and \(t_0\) is uniform but kept six envelope widths inside the file so no burst is cut by a file edge. The amplitude \(A\) is then set so that the optimal SNR against the detector’s PSD equals the drawn value.

Where the numbers come from

They are ~~stolen~~ motivated by public Gravity Spy O3a Omicron tables (Zenodo 5649212, classifications with confidence above 0.9) and the glitch literature:

class	rate [per hour]	peak frequency, 5–95 % [Hz]	Omicron SNR power-law index	duration
Blip	1.3 at H1, 0.65 at L1 (about 2 in O1/O2)	93–412 at H1, 111–630 at L1	2.2–2.4	\(\mathcal{O}(10)\) ms
Tomte	0.22 at H1, 6.2 at L1	32–61	2.0–2.4	about 0.25 s

You can execute it as follows:

gwforge_glitch --config-file blip_tomte.ini --data-directory data --gps-start-time 1893024018 --gps-end-time 1893187858

The GPS range may also be written as gps-start-time and gps-end-time in [IFOS]; the command line wins when both are given. The draws are keyed on (seed, class, detector, file start), so the same seed and range reproduce the same glitches, and the per-chunk jobs of the A workflow for gwforge (pass --glitch-configuration-file) produce the same glitches as one long job. What went in is written per detector to glitches-{start}.h5 in the data directory, with the datasets class, time, frequency, q, snr, phase, amplitude.

Note

Glitches are incoherent: there is no sky position and no projection, and the detectors of a network, including the three ET interferometers, each draw their own. Run gwforge_glitch after gwforge_inject, as the workflow does; the order does not change the data, but it keeps the SNRs gwforge_inject records free of glitch power.