scipy: Bug in scipy.signal.resample() when using a window

The signal.resample() function accepts an optional parameter to specify a window in frequency domain. If that’s done with a tuple with name and parameter, the window is retrieved from another function. Relevant code from signaltools.py:

if window is not None:
    if callable(window):
        # ...
    elif isinstance(window, ndarray):
        # ...
    else:
        W = fftpack.ifftshift(get_window(window, Nx))

However that get_window() function returns the window in time- and not frequency domain! Which is obviously not correct. The get_window() can be found here in the windows.py.

About this issue

Original URL
State: closed
Created 8 years ago
Comments: 17 (14 by maintainers)

Most upvoted comments

However, resample does not have any ability to minimize spectral leakage via time-domain windowing, which is a real deficiency.

I don’t see this as a deficiency so much as a reasonable design choice. You can accomplish such time-domain windowing to deal with edge artifacts or spectral leakage with resample(x * window, ...) in your own code, whereas the Fourier-domain windowing occurs in the middle of transforms (after FFT, before truncation and IFFT) so it has to be a parameter…

larsoner on Apr 15, 2016