cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Nitin_Kumar
Contributor
Contributor
306 Views
Registered: ‎07-28-2020

a set of incorrect values from IFFT

Hello,

     I am doing FFT on 1024 Single floating point Complex values packed as 0:31 real & 32:63 Imag part, the answer is correct in comparison to MATLAB. Then i do some division on these values by first converting them from float to integer, perform division and then again store them as float. When I give these values as input to 1024 point IFFT, the output results match for first and last 450 values and does not match for central 124 values. What could be the possible source of such error?

I used Automatic as well as Radix 4 burst I/O architecture, with scaling 0x2AB, the forward /inverse for both blocks.

Please help me with this.

Regards

Nitin Kumar

P.S. I know IFFT does not scale by 1/N in vivado.

0 Kudos
2 Replies
vkanchan
Xilinx Employee
Xilinx Employee
185 Views
Registered: ‎09-18-2018

Hi @Nitin_Kumar ,

Is the FFT configured for Floating point Input and output?  Scaling schedule is not applied when the FFT is configured for Floating point I/O.  

Can you check the FFT - > IFFT process and compare the result ? What is the output ordering used in the FFT IP ? If it is bit reversed, then the input to the IFFT has to be re-arranged to be in normal order.

Nitin_Kumar
Contributor
Contributor
86 Views
Registered: ‎07-28-2020

Hi @vkanchan ,

Thanks for your reply, yes the FFT is configured for floating point and now I understand that scaling schedule is not applied for this case. I figured out the source of my problem, it was the sample values given as input to the FFT.

My sample set contains several zeros in the middle of the packet with only one positive value placed at the center, So it is like integers followed by zeros, one positive value at center again followed by zeros and then some integers. I figured out if i remove zeros and place some very very small values close to zeros, I get proper results from FFT or IFFT.

Thanks 

Nitin Kumar

0 Kudos