problem with FFT IP core

[habbas33]

I have generated FFT ouput from FFT IPcore 7.1 for a signal and i wanted to match it with the FFT result from matlab. as shown the result of both outputs are normalised to compare butl i found a little missmatch with the result. can any one indicate what is the reason for this mismatch.

There are 512 samples of the signal therefore the transform length of IP core is set to 512 with unscaled output.

for matlab i used
fft_maltab = real(fft(x,512))

 

[TrickyDicky]

are the differences related to real and fixed point representations?

 

[habbas33]

i dont think so because this real fxn is used to get the real component from the complex FFT result. the fixed point result from the IP core is converted into integer and read into matlab for plotting

 

[TrickyDicky]

This is the problem - the matlab version is floating point, the IP core is fixed point. Hence difference in implementation (and lots of scope for differences). You need to make the matlab model completly fixed.

 

[habbas33]

This is the problem - the matlab version is floating point, the IP core is fixed point. Hence difference in implementation (and lots of scope for differences). You need to make the matlab model completly fixed.

I have converted the system on matlab into fixed point one but still having the same problem