The situation When I calculate the fft of a signal, I want to plot it almost every time. However, there is a number of operations that I want to do each time, as the output from fft and fftfreq is honestly terrible. So what is so bad about them? Several things. Problems with fft and fftfreq First …

I much prefer the fft function from FFTW.jl to be consistent with the definition of the FFT. Any subsequent data massaging is better left to the users. If you are calculating the one sided spectra of real signals, then I’d suggest using rfft instead of fft.

I usually use DSP.jl for that kind of things, makes things much easier : https://docs.juliadsp.org/stable/periodograms/#DSP.Periodograms.periodogram-Union{Tuple{AbstractVector{T}},%20Tuple{T}}%20where%20T<:Number

Agreed, consistency is a must. I am arguing that the data massaging function should ship with the fft function, with the package. And the dosctring should refer normal users to the assisted version.

That is a great point, and something i just discovered myself. But the AbstractFFTs doctrings should then refer users to use periodogram rather than FFTW, right? It just seems like it takes too much fiddling/looking to make/find such good solutions, and that they should be advertised from the start.…

A periodogram is not an FFT (which computes a discrete Fourier transform / DFT). You’re making pretty strong assumptions of what a “normal” FFT user is if you are assuming that most people want a periodogram rather than a DFT. e.g. some of us want to solve PDEs, or compute Chebyshev polynomials, …

Fair point. I have been quite biased in my basis for a “normal user”. My personal context is as an engineering student, of which there are a few. My best guess is that all engineers learn about FFT’s at some point. A key part is hands-on calculation and visualisation. So can we agree that at minimu…

[image] mbaz: If you are calculating the one sided spectra of real signals, then I’d suggest using rfft instead of fft. This is exactly what I need, except for the scaling by length. Thanks!

I realize that I am talking to one of the creators of FFTW, and I want to be clear: I am grateful for the great work put into making an amazing library, faster than anything else. I am also sure that there are good algorithmic reasons why the return-values are ordered as they are. I am however sim…

[image] TheLateKronos: … I should simply “vote with my feet” and make such a “more julian” (by my estimate) library myself, building on … For what it’s worth, I had a go at some point. But instead of “values and frequencies as a named tuple” they live together in a fancy array: using AxisKey…

[image] TheLateKronos: dispatching on the element type to use rfft automatically rfft is a different transformation than fft. Think of fft(x) as F*x where F is a matrix. rfft is a different matrix (a submatrix of F). Because of that, I don’t think it would be appropriate to silently chang…

Why are the return-values of fft and fftfreq so terrible?

Specific Domains Visualization

TheLateKronos November 12, 2022, 10:48pm 17

I finally made the package announcement here. Check if out if you are interested

1 Like

Topic		Replies	Views
[ANN] Announcing EasyFFTs.jl Package Announcements	1	1035	November 13, 2022
How do frequencies shift after a FFT? Signal and Image Processing fftw , dsp , fft	11	2330	October 9, 2022
Comparison of different fft-implementations Signal and Image Processing fftw , signal-processing , fft	11	2286	October 7, 2022
Fft() function does not return the correct analytic result Signal and Image Processing fftw , dsp	5	2143	November 26, 2020
Replacing FFT from Numerical Recipes with FFTW Numerics question , fftw	8	1711	August 11, 2024

Why are the return-values of fft and fftfreq so terrible?

Related topics