I finally got around to watch the FFT-video. Nice work, this was quite informative!
A couple of remarks nevertheless…
An average spectrum analyzer, especially an entry level model, can be a rather doubtful reference for certain parameters, like spurious free dynamic range and distortion. You really need to know how to setup the instrument for its maximum performance and a good DSO might easily outperform an entry level SA at frequencies of just a few MHz, especially when it can provide more resolution than just 8 bits. And your tests revealed this quite clearly.
The attached screenshot shows the actual distortion of a Siglent SDG6052X at 1 MHz. The strongest harmonics are -76 dBc down. I have no doubt that any SDG2000X will be at least as good as this, and it is quite evident that both the RTB and SDS got quite a bit closer to the truth than the “reference” SA. It is quite possible that the SA could have been able to deliver more accurate results with a careful setup.
Speaking of the SA, at the end of the video you said the SA was so much easier to setup compared to the scopes. Yes, it might be the instrument that is quickest to get some result, but as much as we value quick results, we usually value any results even more if we can trust them to be reasonably accurate.
One minor additional flaw is the comparison of different window functions. The SA inevitably uses FlatTop, whereas Hanning has been used on the DSOs, which can cause a bit of additional error if a measured frequency doesn’t happen to fall at the exact center of a bin. With a binary FFT, like 128 kpts (131072) and straight decimal frequencies like 1 MHz (1000000) it is extremely unlikely to be lucky and additional errors are inevitable.
The spurs are not that straight forward. Of course, if you disconnect the signal source, then everything that remains visible has to come from the analyzer or DSO itself. Any spurs emerging when the signal is applied need not come from the signal source though. Look at the attached screenshot again: the strongest spurs are below -110 dBc, so they could never be visible on a FFT that has a noise floor around -100 dBc.
So we learn that there is a third type of spurious signals, that are generated within the analyzer only when the external signal is present. These are known as intermodulation products, coming from interactions with internal signals that might be far outside the view of just 0-3 MHz but resulting in low frequency signals that happen to show up within our analysis span.
Of course, the probability to see such intermodulation products gets higher with more bandwidth and the situation gets better with higher resolution acquisition systems. A dense carpet of spurious signals might originate in the granular noise of the ADC. A genuine 10-bit instrument is more likely to cause less distortion and in any case, it should produce less granular noise. This is where the SA shines, because the digital signal processing is at least 14 bits, and usually more than that.
I’m not convinced that we need to get the proper FFT-settings by “trial & error”. It is all very predictable.
Reply #23 in the following thread has a complete checklist for setting up the FFT for a Siglent SDS2000. Many of the hints there will apply to any FFT implementation.
https://www.eevblog.com/forum/testgear/rohde-schwarz-rtb2002-vs-siglent-sds2104x-plus/msg3239832/#msg3239832At the first part, you were unhappy because of the slow update rate on the SDS. It should be very clear that a 2 Mpts FFT at a slow timebase like 10 ms/div cannot update as fast as 128 kpts or even 64 kpts on faster timebases. Later, when you used 128 kpts on the SDS, it appeared to be the fastest in that group.
No need for speculations about the benefits of long FFT. Since the RBW is the sample rate divided by the FFT length, it is obvious that a long 2 Mpts FFT is just as desirable like narrow resolution bandwidths on a SA. While most classical SA can provide constant RBW throughout their bandwidth, the FFT bin width (hence also the RBW) in a DSO depends on the FFT-bandwidth, which in turn depends on the effective FFT-sample rate. The RBW can only be narrowed by either reducing the effective sample rate or increasing the FFT length. At 2 GSa/s and 2 Mpts FFT we get a bin width of 1 kHz, which will result in a RBW of ~3.5 kHz in case of the FlatTop window. This is not very narrow, but then again this also results in a FFT-bandwidth of 1 GHz, which the DSO cannot provide anyway. For a 500 MHz FFT-bandwidth, we can just limit the sample rate to 1 GSa/s – by selecting a slow enough timebase to get more than 2 Mpts record length.
Why is a narrower RBW desirable? Because it provides better frequency resolution and lowers the noise floor.
With 2 Mpts, you can have a bin width of 50 Hz below 50 MHz (at 100 MSa/s) and this would enable you to check the modulation of even narrowband radio signals. Try the same with just 64 kpts and ~1500 Hz bin width – just hopeless.
I’m not sure if I missed something, but I seem to remember that you left the SDS at the default value of 4 when averaging. For comparable results, all contenders should use the same number of averages of course.
Axis annotation will have a sensible resolution in the next firmware. Among other things, this means just one place after the comma for the y-axis (dB).
It has been brought up by 2N3055 already, but since you repeated it over and over again throughout the video I want to make it clear again that the sorting of the markers is done in the table and it works as it should. In fact, the markers make absolutely no sense without the table, because without table we know absolutely nothing about them. The only reason why we can switch the table off is to get it out of the way if needed, without disabling the marker function altogether.
The icons in the file manager get an additional annotation in future firmware. So no more guesswork.
LISN: The SDS has a formula editor, which, among other things, allows FFT on the sum/difference of two input channels with only one math channel. That’s probably the trick that Keysight does implicitly to get that functionality out of a single math channel.
Since FFT is a math channel, the SDS can also show two FFTs at the same time. As a consequence, SDS could show one FFT on the common mode signal and another one on the differential mode signal simultaneously.