FFT Error on FPGA with FPGA in the Loop workflow

Question

Jan Gleinser 2019-8-8

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/475360-fft-error-on-fpga-with-fpga-in-the-loop-workflow

回答： zhang zheng 2020-4-18

I recently started to work with FPGA's and I am currently trying to implement an FFT on the FPGA of my Zedboard.

For this purpose I started with this simple example:

https://de.mathworks.com/help/dsp/examples/generate-hdl-code-for-fft-hdl-optimized-block.html

My main aim would be to use IP Core Generation Workflow, however, therefore no complex values are supported yet (if somebody knows how to solve this problem tell me pls!).

So, instead of IP Core Generation I am trying to use the "FPGA in the Loop" Workflow. Unfortunetaly the results when I try to do the FFT on the FPGA diverges massively from when I just do it with the HDL Coder Block in Simulink (see in the Pictures, on the left the FFT on the FPGA).

Obviously it is indeed the same signal but somehow it looks like a quantisation Error or something to me.

The FIL-model is also attached to the question.

So my question is, if somebody got an idea through which causes this kind of error can arise.

Thanks in advance!

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Navya Seelam 2019-9-26

Why did you use Complex to Real-Imag block when the Out1 is always real?

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

zhang zheng 2020-4-18

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/475360-fft-error-on-fpga-with-fpga-in-the-loop-workflow#answer_426803

i've tested the simulink model you attached. i think maybe you attached the wrong model?

in the attached model,

1. you combine the real and imag part togeter as an uint32 as the input to the fft. i think no need to do this. you can just input the data as complex data, and fixdt(1,16,13).

2. the dimension of the data source is 8, i think this may lead 8 fft processors in parallel. you can try scalar input first.