2) Probes. The 2104X+ comes with 200MHz probes. This will impact measuring even a 50MHz square wave, to a degree.
Even the most expensive passive high-impedance probe with the widest bandwidth rating will not have a perfectly flat frequency response, thus impact measurements to a degree indeed.
Apart from that, how do you know this? Have you ever measured the system bandwidth with the supplied probes?
3) matching/termination. Also called out elsewhere. If you are doing bandwidth tests, 50Ohm termination with proper matched coax and signal source. Also note that 50ohm termination can impact the signal level depending on your signal source. IIRC Dave did a video on this subject.
No. It
can not impact the signal level, it
will impact the signal level. To be more precise, it will halve the signal level when compared to the High Z termination. This is so because the source impedance will be 50 ohms (if it's not, then we don't have a proper matched signal source anymore).
4) Sampling rate. This is a big thing on this scope. With a single channel enabled, you get 2Gs/s. At 500Mhz, that' only 4 samples per cycle. Trying to look at anything other than a sine wave, or something very close to it, is going to give you a poor representation of the signal. Signal interpolation only does so good.
Big thing? This scope? Only good for sine waves? Now fasten your seat belts: any scope (analog or digital) will only be able to properly display and analyze sine waves as soon as you exceed half its bandwidth. Surprise, surprise!
So it all comes down to the (obvious) fact, that for non-sinusoidal signals, much more bandwidth is needed than just covering the fundamental frequency. There have been rules of thumb for almost forever, how for example you should have at least 5 times the bandwidth for a square wave – 7 or even 9 times is better of course. It all comes down what you want to see and measure, i.e. the exact shape of a waveform is often not important, we just want to know the amplitude and where the transitions (i.e. zero crossings) are.
Only 4 samples per cycle? And what does the sampling theorem request?
Do you have test results that indicate that 4 samples per cycle aren't sufficient to reproduce a sine wave?
In theory, we need just 2 samples per cycle. Theory exists only in textbooks, as we all know, so practice is different indeed. Some decades ago, early realtime sampling DSOs like Tek TDS220 could only have 100 MHz bandwidth despite a 1 GSa/s sample rate, just because they only had linear interpolation instead of a proper sin(x)/x reconstruction. Digital signal processing just was not very common back then. As a result, a sample rate ten times higher than the maximum signal frequency was indeed a requirement.
In order to come close to the theory, we need a few things; one of them a proper reconstruction (not interpolation!) at the output of the sampling system. Since this system is not only sampling, but also digital, we have the opportunity to have our reconstruction filter in the digital domain as part of the digital signal processing chain. This means, we can have a FIR filter with near ideal brickwall characteristics and constant group delay. Or should I better say: we could have it, if the number of data points would be infinite. Instead of this, we just have a couple samples per screen width at short timebase settings. This means we cannot go straight to Nyquist, e.g. 500 MHz bandwidth at 1 GSa/s samplerate, it is not the factor of 2.0 from the textbook, but rather 2.5. This means that in today’s DSOs, 1 GSa/s is good for up to 400 Mhz.
If you turn on both channels on the same ADC (1+2/3+4), the same rate drops to 1Gsa/s. That's not enough for 500MHz, full stop. Calling it good for 350MHz is a stretch. 250Mhz gets you the same 4x sampling that you had with a single channel at 500MHz.
Once again I read wild claims – do you have any evidence that 350 MHz is “a stretch”?
The cheap Siglent SDS2000X+ gives you the
option to use it up to 500 MHz in half-channel mode. Only in this mode, the input bandwidth is 500 MHz; as soon as the sample rate drops to 1 GSa/s, the input bandwidth is limited to 350 MHz. All this is clearly indicated in the input channel tabs.
Of course the input bandwidth limit is not very effective - it is only 1st order. We cannot have anything more effective in the analog domain, because more than anything else we need a constant group delay in a scope frontend. Consequently, an effective AA filter is not going to happen, but in practice we can live with that quite well.
For waveforms, the 2504x+ is, IMHO, a 200MHz (1 channel), 100MHz (2 channel) scope. Above that, you need to know what you are doing and how to interpret what you are seeing as sample rates start to limit representation. Where the 500Mhz bandwidth will help to a degree, is with representing high frequency components of slower wave forms -- the aforementioned 50MHz square wave. Even that is dependent on your probing setup.
There is a clear definition: The bandwidth of any scope is where the amplitude of a sine wave drops by 3 dB.
Other than that you are still caught in the (early) nineties of last century, where a factor of ten for sample rate to bandwidth was indeed mandatory – even though with ridiculously short memories like the 2.5k in a TDS220, the sample rate (hence also the bandwidth) drops like a stone for time bases any slower than ~200 ns/div…
It would be much more appropriate to worry about aliasing, because the input bandwidth limit isn’t very effective (it just cannot be) and so we only have a weak AA filtering. This means you can provoke aliasing artefacts if you throw signals with a much wider bandwidth than 500 MHz at the scope in full channel mode.
With fast signals that have a wide spectrum, vastly exceeding 500 MHz, you can indeed provoke aliasing artefacts in full-channel mode, i.e. at only 1 GSa/s. But rest assured that you won’t be able to do the same in half-channel mode at 2 GSa/s, even though the true input bandwidth is 570 MHz then. Anything that goes beyond Nyquist, i.e. > 1GHz is already sufficiently attenuated so aliasing won’t be a problem.
Finally I agree: it’s always desirable if you know what you are doing. Above but also below 100 or 200 MHz.