Strange nois on FFT when 1kHz it's ok, byt 12kHz spectrum wrong - Rigol DS1054Z

[Neurosurg]

Hi Everything,
I have a problem with my Rigol DS1054Z
When I give him the square signal 1Khz, 10, 100 etc, FFT looks normal, but for f=12kHz, 15kHz, 8hHz etc something goes wrong with the spectrum.
Please look at the screen shots. Co-procesor for match is failed ? Help my.

Thanks.

[TUMEMBER]

You got the departments wrong. This one is about metrology, not repair. You should put the post in the test equipment section. It's not an "electroda.pl" - they did not deign to answer there? Because I saw an identical post there.

https://www.eevblog.com/forum/testgear/new-rigol-ds1054z-oscilloscope/

[RoGeorge]

There is nothing wrong with the oscilloscope, that is how FFT works.  Similar FFT leakage will appear for the settings used in the first picture, if the input signal would be 1.2kHz instead of 1kHz.

https://www.gaussianwaves.com/2011/01/fft-and-spectral-leakage-2/

[Scottie]

I have a DS1054Z and can confirm the FFT function is pretty lousy. As far as I know the FFT operates on the display sample buffer which is likely 1024 points.

The scope downsamples the acquisition buffer to display buffer by simply dropping samples. So, many of the square wave harmonics end up aliasing and folding back down to the first Nyquist band.

If you FFT a sine-wave input you should have better luck.

You can also try adding an FFT window.

[RoGeorge]

The scope downsamples the acquisition buffer to display buffer by simply dropping samples.

DS1054z has a setting to avoid that.  In the FFT settings, on the second page of settings, the "Mode" button can switch between "Trace" (FFT is done using only the samples that are seen on the oscilloscope's display), and "Memory*" (all the captured samples are used to calculate the FFT, not only the samples that are displayed on the screen).

In practice, "Memory*" mode will usually give a better frequency resolution of the FFT, but FFT leakage will still happen.  Unless the sampling time is an exact multiple of the input signal length, then windowing comes into effect, and FFT will show frequency leaks no matter what.

[Scottie]

DS1054z has a setting to avoid that.  In the FFT settings, on the second page of settings, the "Mode" button can switch between "Trace" (FFT is done using only the samples that are seen on the oscilloscope's display), and "Memory*" (all the captured samples are used to calculate the FFT, not only the samples that are displayed on the screen).

That is good to know, though I don't ever use the FFT function on the scope, I always run FFTs in Matlab.

OP: I see about the same for a 12 kHz square wave. I also included a 12 kHz sine wave (just had to push the sine button on my function generator).