Broadcasting of FFT
yoshipon opened this issue · 5 comments
yoshipon commented
Thank you for developing xtensor-fftw.
I would like to request the following extension that will be improve its usability.
It will be better if there is a broadcasting in xtensor-fftw.
I'm studying on multi-channel acoustic signal processing, and often have to conduct FFTs to microphones.
In numpy, we can broadcast FFT like this way
np.fft.rfft(np.ndarray([M, T]), axis=1).shape
In my case M and T indicates the number of microphones and time frames.
This will be simplify codes of such signal processings.
Best,
egpbos commented
It might be possible to implement this very efficiently using the multiple FFT functionality of FFTW.
yoshipon commented
Thank you for the reply.
That's a good idea to use fftw_plan_many_*.
spectre-ns commented
This would be great!