Comments on: Interpret FFT, complex DFT, frequency bins & FFTShift

By: Mathuranathan

Mathuranathan — Wed, 30 Mar 2022 10:44:56 +0000

In reply to Mike F. +ve is just a mathematical abbreviation for Positive -ve is for Negative

By: Mike F

Mike F — Tue, 29 Mar 2022 04:45:22 +0000

Some of the drawings are labeled +ve and -ve. What does ve mean in this context?

By: Mathuranathan

Mathuranathan — Mon, 04 Oct 2021 04:37:55 +0000

In reply to Kadhiem Ayob. Thanks for catching the mistake. The formula stands corrected.

By: Kadhiem Ayob

Kadhiem Ayob — Sun, 03 Oct 2021 15:17:52 +0000

Thanks for the article. For the even fft case e.g. 100 the number of bins becomes indeed 100 based on your expression of [ -N/2:1:N/2-1] but for the odd case e.g 101 using your expression [-(N+1)/2:1:(N+1)/2-1] it becomes 102. Am I missing something? Thanks

By: Jonas Stein

Jonas Stein — Tue, 13 Nov 2018 16:22:11 +0000

I have been coming around FFT quite often and worked my way up to the certain level of understanding you need as an engineer whos name is not Gauss. This sums it all up and hands it to everybody very easily. I would have been really happy if iI found that years ago. Nice work. Maybe it is so easy to get into this article becauce i am pretty well introduced to thre topic already though. Nevertheless recommandable!

By: T S

T S — Tue, 06 Mar 2018 13:00:00 +0000

A very helpful article, but I just have one question about the FFT-Shift section: In the case of an odd length (N), the frequency axis f = (f_s/N) * [ -(N+1)/2 : 1 : (N+1)/2 – 1] contains N+1 points, so I guess that either the signal needs to be zero-padded by 1 at the end or the last point of the f-axis needs to stop at (N+1)/2 – 2 instead of (N+1)/2 – 1? And in order for the Nyquist Frequency (f_s/2) to match the first index, shouldn’t the multiplier be (f_s/(N+1))?

By: Mathuranathan

Mathuranathan — Wed, 21 Feb 2018 10:07:00 +0000

In reply to Terry Frangakis.

Thank you !!! Honestly, I have not used FFT for roughness analysis, so i would suggest this article for finer details (refer section 2: Discretization of the problem. Roughness ) .
https://cdn.intechopen.com/pdfs-wm/21958.pdf

By: Terry Frangakis

Terry Frangakis — Mon, 19 Feb 2018 18:19:00 +0000

I agree, this is the best explanation of FFT that I have seen, particularly with reference to scaling / plotting of the axes. I have a question in this regard, though. In the time-based examples given the sampling frequencies are in Hz, and the frequency bins are therefore also in Hz in the frequency domain. I am doing 1D surface roughness measurements using a stylus profiler and have z[n] as a function of x[n], where z[n] is the height of the profile at n discrete points, and x[n] is the horizontal displacement of the stylus. I use a 1000 Hz sampling rate to discretise the profile z. The stylus gearing ratio produces a feed rate of 3 mm / minute or 0.05 mm/s, so the dx value is 0.05 um. How do I remove time from the equations and replace it with the x displacement variable? How would this affect the equations, and what units would the FFT frequency axis have?

By: Mathuranathan

Mathuranathan — Thu, 25 Jan 2018 08:44:00 +0000

In reply to Ömer Kaşdarma. Thanks for mentioning it.

By: Mathuranathan

Mathuranathan — Thu, 25 Jan 2018 08:43:00 +0000

In reply to sndn. Thanks for catching the error. Updated the Matlab indices on Figure 5