2. Calculating and plotting a SpectralVariance histogram

The most common visualisation of the spectral content of a data series is via the power or amplitude spectral density calculations (PSD or ASD). However, these typically involve averaging of the data over a period, which can wash out transient noise (as is often desired).

The SpectralVariance histogram provide by gwpy.frequencyseries allows us to look at the spectral sensitivity in a different manner, displaying which frequencies sit at which amplitude _most_ of the time, but also highlighting excursions from normal behaviour.

To demonstate this, we can load some data from the LIGO Livingston intereferometer around the time of the GW151226 gravitational wave detection:

from gwpy.timeseries import TimeSeries
llo = TimeSeries.fetch_open_data('L1', 1135136228, 1135140324, verbose=True)

We can then call the spectral_variance() method of the llo TimeSeries by calculating an ASD every 5 seconds and counting the amount of time each frequency bin spends at each ASD value:

variance = llo.spectral_variance(5, fftlength=2, overlap=1, log=True,
                                 low=1e-24, high=1e-19, nbins=100)

We can then plot() the SpectralVariance

plot = variance.plot(yscale='log', norm='log', vmin=.5, cmap='plasma')
ax = plot.gca()
ax.grid()
ax.set_xlim(20, 1500)
ax.set_ylim(1e-24, 1e-20)
ax.set_xlabel('Frequency [Hz]')
ax.set_ylabel(r'[strain/$\sqrt{\mathrm{Hz}}$]')
ax.set_title('LIGO-Livingston sensitivity variance')
plot.show()

(png)

From this we see that in general the sensitivity varies a few parts in 10 ^-23, while many of the lines (narrow-band peaks in the spectrum) are much more stationary.