How could I choose the FFT length ?

[electronic_engineer]

FFT Length

Hi DSP Guys,

I have a simple question, I have been trying to compute the FFT of data consisted of 2 million points with minimum no. of FFT point. I just wanted to get the right amplitude of the signal.

How could I choose the FFT length ? Please help me out.

Regards,
EE

 

[neodgreat]

Re: FFT Length

@ electronic_engineer

If u are working on computation using matlab.. make sure u select the next power of 2.. for example if u had 220 points.. choosing fft length of 256 would be faster.. this is what i guess.. the excess points would result in zeros which u can discard later.. hope my understanding would be correct.. but if u r working on a specific DSP processor.. i would like to see how you are working on this..

 

[electronic_engineer]

Re: FFT Length

@ electronic_engineer

If u are working on computation using matlab.. make sure u select the next power of 2.. for example if u had 220 points.. choosing fft length of 256 would be faster.. this is what i guess.. the excess points would result in zeros which u can discard later.. hope my understanding would be correct.. but if u r working on a specific DSP processor.. i would like to see how you are working on this..

Actually I am working with DSP therefore, Length of FFT matters otherwise there is no problem with DSP.

Any DSP guy can answer me please.

 

[Jack// ani]

FFT Length

Larger the number of FFT points, better is your frequency resolution.

So it got nothing to do with signal amplitude!

 

[electronic_engineer]

Re: FFT Length

Larger the number of FFT points, better is your frequency resolution.

So it got nothing to do with signal amplitude!

I know, but if I can't calculate FFT of 2 Million points data with FFT length of 2 Million. DSP memory will not allow me to do so...

Any one who has worked with FFT on DSP processors ? Please help me...

 

[doraemon]

Re: FFT Length

Hello!

Just in case, here is a simulation program (not published yet) that gives a good
idea of what FFT does.



The input is a sine wave and its frequency can vary between 0 and 0.5 fs.

The number of points can be changed in realtime.

There is also a possibility of changing the input windowing and the results of the
windowing appears immediately in the input window. The window can be chosen
among rectangle, Bartlett, Hann, Hamming and Blackmann.

At one point, I have entered 0.25 manually as the input frequency. In this case,
as it is a submultiple of the sampling frequency, the coefficient cancel each other
nicely and the result peak is extremely acute.

Note that the resolution (variable from 256 to 4096) has a great influence on the
peak shape. At 4096, it becomes very acute, close to the ideal Dirac pulse in any
case.

NB: since the spectrum is symmetric, the program displays only half of it.

Dora

 

[zorx]

Re: FFT Length

Everybody gave correct answer.
It really depends on your application. Maybe an FFT in 64 points would be enough and maybe FFT in 16384 won't be enough. If you just want to detect the existence of signals of certain level, than maybe 64 points FFT is ok. But if you wont to separate very close (in frequency) signals, then you need FFT in more points.
If you can tell little bit more about your application, then you could get more specific answer.