FFT problem

Hi,

I have implemented a 256 pts FFT at 10.24 MHz sampling frequency so I
have a frequency resolution of 40 kHz...

I compute the FFT output to have the squared magnitude : Re(X(m))^2 +
Im(X(m))^2

When I send a pure 500kHz sine to the ADC input (12 bits) I see on the
FFT output one peak at bin frequency 500 kHz and 9.74 MHz

Then when I send 510kHz sine to the ADC I read on the FFT output : one
bin at 510 kHz one bin at 9.73 MHz and one bin at 8.23 MHz and at 7.24
MHz

Why I have got these two bins 8.23 and 7.24 ? what's happen, it's like
an intermodulation on my signals ??

With rectangular or flat top windows I have these two strange bins...

thanks for any help

I have another problem...

When I send a pure 500 kHz sine, when I don't have the stranger
frequency at 8.23 and 7.24

I can see that the magnitude of the FFT varies in amplitude over the
time, it's like the "picket fence" effect that we have when the input
frequency send to the FFT is increasing over the time...

So it's like an amplitude modulation on my FFT magnitude.... because
the frequency sent to the ADC is asynchronous to the frequency
sampling used by my ADC and my FFT

how to don't have this problem ?

thanks

On 13 Feb 2007 03:16:56 -0800, patrick.melet@dmradiocom.fr wrote:

>Hi,
>
>I have implemented a 256 pts FFT at 10.24 MHz sampling frequency so I
>have a frequency resolution of 40 kHz...
>
>I compute the FFT output to have the squared magnitude : Re(X(m))^2 +
>Im(X(m))^2
>
>When I send a pure 500kHz sine to the ADC input (12 bits) I see on the
>FFT output one peak at bin frequency 500 kHz and 9.74 MHz
>
>Then when I send 510kHz sine to the ADC I read on the FFT output : one
>bin at 510 kHz one bin at 9.73 MHz and one bin at 8.23 MHz and at 7.24
>MHz
>
>Why I have got these two bins 8.23 and 7.24 ? what's happen, it's like
>an intermodulation on my signals ??
>
>With rectangular or flat top windows I have these two strange bins...
>
>thanks for any help

Hello Patrick,
  wow, what a strange problem!  When you're FFT 
input is 500 kHz, your FFT magnitude results are 
correct.  (Now if you run that 500 kHz test ten times 
your magnitude results should always be the same, 
but your spectral phase will probably be different 
for each test.)

If your FFT software is working properly then your 510 
kHz test should also work.  But what puzzles me is that 
during your 510 kHz test, your spectral magnitudes 
are *NOT* symmetrically centered at 5.12 MHz (Fs/2)!
That implies that the input sequence values to your 
FFT are complex-valued time samples rather than  
real-only time samples.

(Can you trust your FFT software?
What makes you think that software is correct?) 

Soooo, ... either your FFT software is not working 
properly (but I'll bet it is), or you are not 
really applying the time samples to your FFT 
that you think you are.  I suggest that you,
somehow, very carefully examine the time samples 
that you are applying to the FFT during your 
510 kHz test.

Then again, maybe your FFT software is working, 
and your FFT input samples are correct, but you're 
making a mistake in evaluating the FFT output 
during your 510 kHz test.

Good Luck,
[-Rick-]
 

Hi,

In fact I don't used a software FFT.

I use an Altera FGPA FFT IP... I post on dsp forum because this is a
signal processing problem...

For my picket fence problem, I read that I must send to the FFT, my
256 samples and "sometimes" less samples and others at zero

But how many times in time I don't send all my samples or my samples +
zeros, and how many zeros must I add ?

for my 510 kHz problem, it's a reset problem, if I reset a long time I
d'ont have my stranger 2 frequencies bins... I think the IP core need
a synchronous reset with a minimum time ....

thanks