dsp.FFT does not work with HDL coder

Hi,

I have a code and a testbench code that I'm trying to convert to VHDL.

The code uses FFT. When I use canned FFT, I get "Function 'fft' not supported for fixed point conversion" at "Verify" step of Fixed Point Conversion".

When I instead use dsp.FFT it instead complains about ' "dsp.FFT" cannot be displayed ' at the Analyze step of fixed point converions.

I cannot use dsphdl.FFT since we don't have the license for that.

Can this be solved or do we need to buy dsphdl.FFT?

Incidentally, is there a real-to-complex FFT, taking N real points and returning N/2+1 complex ones like FFTW or np.fft.rfft in python? Using complex FFT on real data wastes resource by a factor of 2, but all the examples seem to be using that.

Thanks,

anze

Answer 1

[Kiran Kintali]

dsphdl.FFT ships with DSP HDL Toolbox.

You can also find some examples of how to write custom FFT in MATLAB suitable for HDL Code Generation in the attached examples.

[Anze Slosar]

Thanks for this, this is also quite useful to illustrate some other concepts.

But surely, I should not need to write custom FFTs in 21st century. Presumably I can count on dsphdl.FFT to produce an optimal result for radix-2 problem and also correctly optimize for real input?

[Kiran Kintali]

Yes, those requirements are supported out of the box. Check the Performance section of dsphdl.FFT for additional details related to optimizations. It covers topics related to real inputs and output ordering.