FFTW Accuracy vs Matlab Accuracy

Clyde wrote:

> pgw wrote:
> 
>> How it look without windowing?
> 
> They are different, and the problem still remains.
> 
>> You can try to generate one harmonic signal and do FFTW (without windowing) 
>> on it and then show us the result.
> 
> Here is the result:
> http://i45.photobucket.com/albums/f77/yesitookmypills/FFT.png
> (blue is FFTW, red is Matlab FFT)
> 
> I used fftw_plan_r2r_1d(dim, in[0], out[0], FFTW_R2HC, FFTW_ESTIMATE)
> and i compute the magnitude in this way:
> magnitude[i] = sqrt(out[i]*out[i]);

I used FFTW only one time, so I'm not familiar with it.
Try use fftw_plan_dft_r2c_1d and then compute magnitude as:
sqrt(real_of_outpute^2 + imaginary_of_outpute^2). That will have to work.

-- 
pgw

Scott Seidman :

>> Don't you need to deal with the "half complex" format that fftw is
>> asked to produce before you deal with the magnitude?

Yes, you are right.
I missed that in half complex format complex is in reversed order, 
starting from the end -2 (!)
Now it seem to work!
I'll do some more analysis with windows.

Thanks _A LOT_ you all


--

Clyde

pgw wrote:

> I used FFTW only one time, so I'm not familiar with it.
> Try use fftw_plan_dft_r2c_1d and then compute magnitude as:
> sqrt(real_of_outpute^2 + imaginary_of_outpute^2). That will have to work.

I tried that, but i think there's something about the complex order that 
i wasn't getting.
I have to admint that with r2c_1d also real numbers were different from 
matlab..


--

Clyde