LPCXpresso IDEのトピックRe: LPC4088 N point FFT Problem

LPC4088 N point FFT Problem

lpcware — Thu, 16 Jun 2016 02:03:23 GMT

Content originally posted in LPCWare by maheshg on Sun Jun 15 22:32:34 MST 2014
Hi

I am working on a project ,in which i need to use FFT on 100000 point of input data .I am using CMSIS_DSPLIB_Latest library but it supports only a maximum of 4096 point FFT .So is there any way to find the N point FFT (Where N is Variable length ).
An early response will be appriciated.

Thanks in advance.

Re: LPC4088 N point FFT Problem

lpcware — Thu, 16 Jun 2016 02:03:24 GMT

Content originally posted in LPCWare by nerd herd on Fri Mar 06 08:26:00 MST 2015
Hi maheshg,

If you are asking if NXP has any sort of library or sample code that can do the FFT you are require, then I'm sorry but we do not. Your best bet is to use Google to see if someone has made such a function.

Re: LPC4088 N point FFT Problem

lpcware — Thu, 16 Jun 2016 02:03:25 GMT

Content originally posted in LPCWare by emh203 on Fri Mar 06 08:47:37 MST 2015

I think you need to rethink your problem. You cannot do an FFT on a non power of 2 size input buffer. If you absolutely need to do a non power of 2 buffer size, you must use a DFT (which will be very slow). The FFT algorithm is a special case of the DFT for power of 2 size inputs.

Also, the reason the FFT size is limited to 4096 is that you can always use a decimation in time algorithm to break your input buffer into smaller FFT sizes.

Re: LPC4088 N point FFT Problem

lpcware — Thu, 16 Jun 2016 02:03:25 GMT

Content originally posted in LPCWare by starblue on Mon Mar 09 02:40:31 MST 2015

Quote: emh203
You cannot do an FFT on a non power of 2 size input buffer.

That's not quite true, you can use a modified FFT if your input size has only small factors. It is more complicated (e.g. for a factor of five you'll need a 5-fold analogue of the butterfly), and efficiency will get worse the larger the factors become. So it is better to avoid it.

I agree that rethinking the problem in order to avoid an overly large and possibly nonstandard FFT computation is much better.