Re: FFT magnitud normalized


"Salvadorr" <salpiano@xxxxxxxxxxx> wrote in message
news:YY-dndyuyfUxZz_ZnZ2dnUVZ_rOdnZ2d@xxxxxxxxxxxxxxx
hello for everybody¡¡¡

I would thanks if somebody could help me.

I´m working with FFT and FHT, i wrote my code enC++ for raddix 2 in both
cases. When i run my programs everything is OK, the armonics are detected
very well in the correct frecuency in both cases, but my problem begin
here, the magnitud of my armonics are diferent. The signal that i analyse
is:

1 + 1 * sin(2*pi*f*n + P) + 1 * sin(2*pi*3*f*n) + 1 * sin(2*pi*5*f*n)
1 * sin(2*pi*7*f*n) + 1 * sin(2*pi*9*f*n) + 1 * sin(2*pi*11*f*n)

The dc component is almost 2 times more than fundamental armonic the
others armonic are a little smaller o a little bigger. I dont know if i
need to do some extra step inside of the algorithm or outside.

thanks for your help¡¡¡


The constant term transforms into the term at dc - and into a single sample.
The sinusoids transform into two terms; one at +f and one at -f.
The energy is what it is.
So, the sinusoids are split between frequency samples and the dc term isn't.
Thus the factor of 2.

The harmonic peak values depend on the number of samples you use.
If the time samples are spaced by T=1/11*k*f where k is an integer, then the
values should be consistent and all the other values should be zero.
Otherwise you will have a fractional number of cycles and there will be
spectral spreading that's evident.

Fred





.

Quantcast