gwforge.GWForge.ifo.reproducible_noise

Reproducible coloured Gaussian noise and zero noise, in pure numpy. Probably equivalent to pycbc.noise.reproduceable (agrees to FFT round-off), but without the PyCBC dependency. The noise is reproducible in the sense that a given (seed, GPS time) always yields the same samples, so adjacent requests stitch together seamlessly. This makes it suitable for generating long observations in chunks.

Attributes

BLOCK_SAMPLES

DEFAULT_FILTER_DURATION

Functions

block(seed, sample_rate)

Return one fixed-length block of white Gaussian samples.

normal(start, end[, sample_rate, seed])

Return white Gaussian noise covering [start, end) in GPS seconds.

time_slice(samples, epoch, sample_rate, start, end)

Crop samples (starting at GPS epoch) to [start, end).

zero_noise(start_time, end_time[, sample_rate])

Return an all-zero strain series covering [start_time, end_time).

interpolate_psd(frequencies, psd, length, delta_f, ...)

Interpolate a tabulated PSD onto a uniform grid, log-log.

_resample(series, delta_f, new_delta_f[, length])

Linearly resample a uniformly sampled series to a new delta_f.

_inverse_spectrum_truncation(psd, delta_f, ...)

Coarse-grain a PSD by truncating its inverse ASD's impulse response.

colored_noise(psd, delta_f, start_time, end_time[, ...])

Return Gaussian noise coloured by psd, covering [start, end).

noise_from_interferometer(interferometer, start_time, ...)

Generate noise for a bilby Interferometer.

Module Contents

gwforge.GWForge.ifo.reproducible_noise.BLOCK_SAMPLES = 1638400[source]
gwforge.GWForge.ifo.reproducible_noise.DEFAULT_FILTER_DURATION = 128[source]
gwforge.GWForge.ifo.reproducible_noise.block(seed, sample_rate)[source]

Return one fixed-length block of white Gaussian samples.

Parameters:

  • seed (int) – Block seed. Reduced modulo 2**32 for RandomState.

  • sample_rate (float) – Sampling rate in Hz. Sets the variance to sample_rate / 2 so that the one-sided PSD of the result is unity.

Returns:

BLOCK_SAMPLES normally distributed samples.

Return type:

numpy.ndarray

gwforge.GWForge.ifo.reproducible_noise.normal(start, end, sample_rate=16384, seed=0)[source]

Return white Gaussian noise covering [start, end) in GPS seconds.

The samples depend only on (seed, absolute GPS time): adjacent requests stitch together seamlessly, which is what lets the workflow generate a long observation as independent per-chunk Condor jobs.

Returns:

The samples and the GPS epoch of the first one (which is start).

Return type:

tuple of (numpy.ndarray, float)

gwforge.GWForge.ifo.reproducible_noise.time_slice(samples, epoch, sample_rate, start, end)[source]

Crop samples (starting at GPS epoch) to [start, end).

Returns:

The cropped samples and their GPS epoch, which is start.

Return type:

tuple of (numpy.ndarray, float)

gwforge.GWForge.ifo.reproducible_noise.zero_noise(start_time, end_time, sample_rate=16384)[source]

Return an all-zero strain series covering [start_time, end_time).

Same return contract as colored_noise(), so callers can switch between the two without special-casing anything.

gwforge.GWForge.ifo.reproducible_noise.interpolate_psd(frequencies, psd, length, delta_f, low_frequency_cutoff)[source]

Interpolate a tabulated PSD onto a uniform grid, log-log.

Port of pycbc.psd.read.from_numpy_arrays. Interpolation is linear in (log f, log S), which is the right space for noise curves spanning many decades. Bins below low_frequency_cutoff are set to zero.

Parameters:

  • frequencies (array_like) – Tabulated one-sided PSD. frequencies must be ascending and positive.

  • psd (array_like) – Tabulated one-sided PSD. frequencies must be ascending and positive.

  • length (int) – Number of samples in the output (DC to Nyquist inclusive).

  • delta_f (float) – Frequency resolution of the output in Hz.

  • low_frequency_cutoff (float) – Bins below this are zeroed.

Returns:

The PSD on numpy.arange(length) * delta_f.

Return type:

numpy.ndarray

gwforge.GWForge.ifo.reproducible_noise._resample(series, delta_f, new_delta_f, length=None)[source]

Linearly resample a uniformly sampled series to a new delta_f.

Port of pycbc.psd.interpolate.

gwforge.GWForge.ifo.reproducible_noise._inverse_spectrum_truncation(psd, delta_f, max_filter_length, low_frequency_cutoff)[source]

Coarse-grain a PSD by truncating its inverse ASD’s impulse response.

Port of pycbc.psd.estimate.inverse_spectrum_truncation restricted to the two options GWForge uses: which_spectrum='invasd' and trunc_method='hann'. See arXiv:gr-qc/0509116 for the method.

The goal is to remove the sharp spectral features that would otherwise give the colouring filter an impulse response longer than the data segment.

gwforge.GWForge.ifo.reproducible_noise.colored_noise(psd, delta_f, start_time, end_time, seed=0, sample_rate=16384, low_frequency_cutoff=1.0, filter_duration=DEFAULT_FILTER_DURATION, scale=1.0)[source]

Return Gaussian noise coloured by psd, covering [start, end).

Numerically equivalent to pycbc.noise.reproduceable.colored_noise (agrees to FFT round-off), including the seed/GPS reproducibility contract described in the module docstring.

Parameters:

  • psd (array_like) – One-sided PSD sampled uniformly from DC at spacing delta_f.

  • delta_f (float) – Frequency resolution of psd in Hz.

  • start_time (float) – GPS bounds of the requested data. The returned series covers [start_time, end_time).

  • end_time (float) – GPS bounds of the requested data. The returned series covers [start_time, end_time).

  • seed (int) – Noise seed. A given (seed, GPS time) always yields the same samples.

  • sample_rate (float) – Output sampling rate in Hz. Must be held constant to keep continuity between disjoint spans.

  • low_frequency_cutoff (float) – The PSD is zeroed below this, so the output carries no power there.

  • filter_duration (float) – Length in seconds of the colouring filter.

  • scale (float) – Extra multiplicative factor on the amplitude spectrum.

Returns:

The coloured samples and their GPS epoch (which is start_time).

Return type:

tuple of (numpy.ndarray, float)

gwforge.GWForge.ifo.reproducible_noise.noise_from_interferometer(interferometer, start_time, end_time, seed=0, sample_rate=16384, noise_type='gaussian', low_frequency_cutoff=None, filter_duration=DEFAULT_FILTER_DURATION, psd_delta_f=1.0 / 16)[source]

Generate noise for a bilby Interferometer.

The PSD is taken from interferometer.power_spectral_density as arrays (frequency_array / psd_array) rather than by re-reading the file. Those arrays are populated no matter how the PSD was supplied – PSD file, ASD file, or raw arrays – so this works for ET (which ships a PSD file) and for every user-supplied psd-dict entry, both of which the old file-sniffing path could not handle.

Parameters:

  • interferometer (bilby.gw.detector.Interferometer) – Detector whose PSD and minimum_frequency are used.

  • start_time (float) – GPS bounds of the requested data.

  • end_time (float) – GPS bounds of the requested data.

  • seed (int) – Noise seed; ignored for zero noise.

  • sample_rate (float) – Output sampling rate in Hz.

  • noise_type (str) – 'gaussian' or 'zero'.

  • low_frequency_cutoff (float or None) – Defaults to the interferometer’s minimum_frequency.

  • filter_duration (float) – Length in seconds of the colouring filter.

  • psd_delta_f (float) – Resolution the tabulated PSD is interpolated onto before colouring.

Returns:

The strain samples and their GPS epoch.

Return type:

tuple of (numpy.ndarray, float)