Address issues with whitening defaults
Created by: alurban
@duncanmmacleod, this PR includes the following changes:
- Window the first and last half-filter-length of input data before convolving, and return a
TimeSeries
the same length as the input (this is mostly to stay more in line with the behavior of the old whitener) - Update the documentation’s whitening example to reflect this
- Change default whitening filter length to 2 seconds, which better reproduces the published GW150914 Q-spectrogram
- Refactor FIR filter design into its own function,
signal.filter_design.fir_from_transfer()
, including a unit test - Use
scipy.signal.fftconvolve()
to apply the whitening filter, anticipating #883 - Edit unit tests affected by changes to
TimeSeries.whiten()
This fixes #890 (closed).