Rigol fft weirdness

[francesco_97]

Hi! I was trying to perform a simple FFT analysis of a 440 Hz sine wave tuned by my PC using the poor-man scope Rigol DS1102 Z-E. I can't understand if I'm too tired or if there is something effectively strange in the software. As you can see from the pictures below if I set the sampling frequency to 10 kSa/s that should be equal to 10kHz (please give me some feedback, I'm doubting my memory) everything seems to be ok: only a delta on the 440 Hz and nothing more. Nyquist is happy because the sampling frequency is more than double of the signal frequency. But but BUT BUT-BUT. If I set the sampling frequency to 2kHz I get this: . Horrible spikes everywhere. They should be ghosts but I can't understand why because 2kHz is still more than double of the signal frequency so they can't be aliasing. What kind of stuff is this? Maybe I'm too drunk and some knob is still set with the wrong value? Maybe  What do you think about it? (Obviously the solution will be something stupid but I can't see it so feel free to ask for every kind of data).

Thanks in advance.

[MasterT]

I noticed similar issue with my DS1202.  Check acquisition mode, if it in high resolution, try normal.

[francesco_97]

Nope, yet tried after checking another post

[wraper]

Nyquist is happy because the sampling frequency is more than double of the signal frequency.

He is not happy since that only works with no issues with input low pass filter with ideal frequency cutoff and with proper sin(x)/x interpolation. In the real world you often need to sample way faster.

2kHz is still more than double of the signal frequency so they can't be aliasing.

You wish.
Capture a waveform with parameters as you set, then zoom in and see if there is aliasing.

[francesco_97]

Ok so a good rule of thumb to choose the minimum frequency multiplier? 5 times the signal frequency could be ok?

[wraper]

Ok so a good rule of thumb to choose the minimum frequency multiplier? 5 times the signal frequency could be ok?

More like borderline OKish, 10X usually is enough for reasonably accurate waveform.

[GumpyGus]

Yeah Nyquist is satisfied, but you're still sampling at barely 22 times the base frequency, so there are going to be lots of sampling products.  If the ratios are commensurate the products are going to add up and be stationary.  If they're relatively primish, they will just show up as a lower level of quasi-random noise.

[Doctorandus_P]

When doing FFT there are several limits. and al sorts of noise shows up in the final result. In general, calculating a FFT over more samples gives a more accurate result. I've got an older Rigol DS1052E, and it is capable of showing a TFT graph, but it's resolution is too low to be useful in any way.

The (newer?) Siglent scopes have a much better FFT function then the Rigols. By calculating results over millions of samples they apparently get a dynamic range of some 80dB out of an 8 bit ADC, but to be fair, I do not even understand how that is possible. (You normally only have about 6dB for each bit (Each bit doubles the resolution)).

[francesco_97]

Dark witchcraft of marketing. Maybe some kind of interpolation? Maybe the subject for the next Dave's video? Anyway, thank you all for your replies.

[Doctorandus_P]

Have you seen this one:
It's weird that both the Rohde & Schwartz and the Siglent scopes appear to show a better FFT picture then the Rigol DSA815, witch does not show the 3rd harmonic above the noise floor.



I'm also a bit disappointed the Siglent is not in that comparison.

[francesco_97]

Yes, I saw it. Obviously, it's a scam but Rohde & Schwarz is also known to make good products. If I can remember well in that video there was a sort of issue by Dave in setting the Rigol because he didn't use all the available points or something similar but for my purposes, it is enough also with Dave's settings. I'm only concerned about the max capability of the Rigol because using the "10x rule" said before I have an effective FFT bandwidth of 100MHz (dividing by 10 the 1GHz sampling frequency). It seems to be a bit too much for a Rigol product assuming that a real dynamic signal analyzer has a 9kHz to 1.5GHz span. Rigol, are you lying to me? Yes of course.