Forums

Overlapped FFT optimization

Started by abhijitk October 15, 2012
Hi Guys,

I am doing a 256 point FFT in FPGA on High speed ADC data with a custom FFT
core...Now i need to do a 4:1 overlap(64 points for 256 FFT) on ADC
data...that is first 256 point FFT will be done on sample no 1 to 256, next
FFT on sample no 65 to 320 and so on...One solution is to use four such FFT
cores to perform the operations..I want to know whether there is any other
method/optimization so that i can use lesser number of FFT cores owing to
the fact that i am using the same (256 -64) points for the next FFT?

Regards
Abhijit Kulkarni
Bangalore India
abhijitk <51485@dsprelated> wrote:

> I am doing a 256 point FFT in FPGA on High speed ADC data with a custom FFT > core...Now i need to do a 4:1 overlap(64 points for 256 FFT) on ADC > data...that is first 256 point FFT will be done on sample no 1 to 256, next > FFT on sample no 65 to 320 and so on...One solution is to use four such FFT > cores to perform the operations..I want to know whether there is any other > method/optimization so that i can use lesser number of FFT cores owing to > the fact that i am using the same (256 -64) points for the next FFT?
The whole idea behind the FFT is that you can generate a 2N point transform by combining two N point transforms in a simple way. I presume you are using a systolic array. That is, at least, my favorite way of doing it in an FPGA. Anyway, I believe that there should be a convenient way to combine the transforms. -- glen
On Sun, 14 Oct 2012 23:05:43 -0500, "abhijitk" <51485@dsprelated>
wrote:

>I want to know whether there is any other >method/optimization so that i can use lesser number of FFT cores owing to >the fact that i am using the same (256 -64) points for the next FFT?
http://groups.google.com/group/comp.dsp/msg/06800ec6bc25deb8?hl=en, in particular the final paragraph. Greg