Rigol DHO804 Test and Compare Thread

[2N3055]

I´ve tried it a dozen of times* but yes, for you I´ll try it one more time...

Thanks so much -- I did not mean to annoy you! But I am afraid I might still not have gotten my point across.

If you look at e.g. RigolDS15.png in reply #219, you have apparently captured the signal shorty after the minimum of the sine wave -- the curves are almost horizontal on the left of the screen and then begin to climb gradually. Overall, their slope is significantly lower than on the Siglent screenshots, where the signal goes up by a total of > 200 mV over the capture time of 1 ms, or even 350 mV over 0.65 ms in the SDS1104X-E screenshot.

Larger slope -> larger DAC step size -> more pronounced steps in the unfiltered signal. Hence my suggestion to try and make sure that you capture the signal at the same phase of the sine oscillation on both scopes. (Or use a sawtooth instead if that's available on the demo board.)

Instead of thinking on terms of time, it is better to make sure that trigger level is same. Maybe trigger level of 1.5V and 0 delay ? That will insure look at the same position in a slope.

[Performa01]

Ah, right, that scope can switch in internal 50 Ohm termination. Luxury! 

It's not just an "internal termination", but a proper 50 ohms channel that maintains a VSWR of <1.5 over the full bandwidth of 500 MHz (and beyond).

Any 50 ohm termination of a 1meg scope input, be it internal or external via through terminator, yields acceptable results up to 100 MHz at best.

[ebastler]

It's not just an "internal termination", but a proper 50 ohms channel that maintains a VSWR of <1.5 over the full bandwidth of 500 MHz (and beyond).

Any 50 ohm termination of a 1meg scope input, be it internal or external via through terminator, yields acceptable results up to 100 MHz at best.

So how do passive probes with bandwidths up to 500 MHz manage to push the signal over a much longer 1 MOhm connection?

I don't mean to quarrel, but am genuinely curious -- if the short internal high-impedance connection from BNC jack to front end is already considered problematic at > 100 MHz, what is the impact of a passive probe's cable?

[Martin72]

We had this topic before, regarding the probes, I just don't know in which thread it was.

Yes, that's what I tried to say in #222 -- different trigger levels were apparently used on the Siglent vs. Rigol captures.

Then look again at the pictures to see if the difference can really be decisive.
But I will do it again and "normalize" and keep me at the rigol from the values and then set the siglent so.
Is easier, because the rigol is a bit bitchy with the trigger.

[TurboTom]

The problem isn't so much the distance from through-terminator / BNC to the scope's internal circuitry but rather the few tens of picofarads of input capacitance. Rigol's specified 15pf input capacitance result in a reactance of 106Ohms at 100MHz. An there goes the 1Mohm input impedance by a large margin...

Passive high bandwidth probes are always 10:1 (or another scaling factor that is) and use a capacitive/resistive divider approach along with a dissipative cable. The approximately divide the input capacitance by their scaling factor, as their ohmic resistance increases likewise. Designing good high bandwidth probes is difficult, that's why they aren't cheap.

The channel input capacitance is required to achieve the required linearity of the high-impedance attenuator. A low-impedance attenuator may be designed differently and with lower capacitance.

[Performa01]

So how do passive probes with bandwidths up to 500 MHz manage to push the signal over a much longer 1 MOhm connection?

I don't mean to quarrel, but am genuinely curious -- if the short internal high-impedance connection from BNC jack to front end is already considered problematic at > 100 MHz, what is the impact of a passive probe's cable?

A passive high impedance probe is a rather complex construction. The key is in the attenuation (20 dB for x10), which leaves enough headroom to compensate for the mismatch on both ends of the probe cable. Here I've "designed" some simple passive x10 high impedance probe, that performed well up to at least 300 MHz in the simulation (reply #521):

https://www.eevblog.com/forum/testgear/siglent-sds1204x-e-released-for-domestic-markets-in-china/msg1434317/#msg1434317


It is no secret that passive high impedance probes with their high input capacitances have limited use, as the above posting illustrates. Even the not too high source impedance of 1 kohm limits the bandwidth to about 10 MHz. Consider how many wannabe experts might have gotten just house numbers as results because of ignoring this.

It all gets worse if you use your probes with the supplied ground lead, as the following posting demonstrates (reply #62):

https://www.eevblog.com/forum/testgear/siglent-sds1104x-e-in-depth-review/msg1435196/#msg1435196

[ebastler]

Then look again at the pictures to see if the difference can really be decisive.

Yes, I think so. Half the DAC step size in mV, a vertically compressed display due to the "letterbox" format, and fatter traces due to more noise -- all three taken together seem to explain why the steps are less obvious on the DHO800. I don't think one has to assume the presence of any filtering.

[Martin72]

Something about the topic..
I have now measured again, both scopes have been set exactly the same, at each step.
And this is the result, I can not change...

[ebastler]

Something about the topic..
I have now measured again, both scopes have been set exactly the same, at each step.
And this is the result, I can not change...

I don't get it. If you compare e.g. the "Rigol_50mV" vs. "Siglent_50mV" screenshots: Why does the signal rise by 2*50mV on the Rigol, but by 5*50mV on the Siglent, during the same capture period of 10*50µS? 

Am I reading the displays wrong or have some other kind of mental block? Or are the two scopes still showing different segments of the sine function, with different slopes and hence different DAC step sizes?

[2N3055]

OK,
I'm ashamed it took me so long....
Not enough coffee I guess..

It is obvious why steps on Siglent are higher.
Because screen display is taller. 
Rigol has weird panoramic ratio, waveform plot area is some 20% "squished" vertically compared to Siglent.

EDIT.

Yeah something not right anyways..


Look at the graticule.

Aside from that, I see slight noise in DHO800 trace and less details on top.. That is partially because 1/3 of sampling rate too.. (625MS/s compared to 2GS/s)
So it is not really as good as better 12 bit scopes.

But for the price it looks good... Now they should finish the product 

[ebastler]

It is obvious why steps on Siglent are higher.
Because screen display is taller. 
Rigol has weird panoramic ratio, waveform plot area is some 20% "squished" vertically compared to Siglent.

That's what I meant in #236 when talking about the vertically compressed display due to the "letterbox" format.

But as mentioned I don't think that is the only effect. The displayed step size in mV also seems to differ, unless I am looking at things totally wrong. Please see my post #238 and Martin's most recent screenshots.

[Martin72]

Now something completely different, before I leave for the next hours...Just for fun.

https://youtu.be/UOFe9wmdIKk

[Performa01]

Just a little demonstration of internal/external termination vs. a proper 50 ohms signal path. This seems to be required, as I’ve often seen claims like “who needs a scope with 50 ohms inputs, we can always use a BNC-T with a terminator”.

Initially I should define the goal: a return loss of 14 dB (VSWR 1.5:1) is generally considered acceptable in the industry. This means, a proper scope shouldn’t drop below 14 dB RL (or above 1.5 VSWR) within its bandwidth, in order to be usable, i.e. show reasonable pulse fidelity and measurement accuracy for all the ones who use their oscilloscopes for serious tasks.

First a proper through terminator, a venerable HP10100C. This performs beautifully when just connected at the end of a 50 ohms coax cable, but gets pretty ugly as soon as it’s connected to the scope input:

SDS2354X+_RL_200mV_HP10100C

The limit of 14 dB RL is reached at 64 MHz already. Now the alternative, even cheaper solution, a BNC-T with a terminator connected to one leg and the input signal to the other:

SDS2354X+_RL_200mV_BNC-T

As can be seen, the result is pretty much the same. Of course – at frequencies that low a few centimeters aren’t a big deal. This time I’ve measured the minimum RL within the 500 (actually 600) MHz bandwidth, and it’s only 6.94 dB. Utterly useless.

Finally the real thing – the scope switched to 50 ohms, which alters the internal signal path and eliminates the input capacitance that’s always present in high impedance mode:

SDS2354X+_RL_200mV

We get a return loss of more than 17 dB throughout the entire measured range (up to 1 GHz). At 200 MHz, we still have 30 dB RL, which is equivalent to 1.07:1 VSWR.

And now, what’s the consequence?

Look at the next screenshot:

SDS2504X HD_PR_50

This is with the scope switched to its 50 ohms input path. It shows a nice 10 ns wide pulse with 1 ns rise and fall times – exactly as I’ve programmed it on my pulse generator. Yes, there is a little overshoot – and we certainly want to know about that.

Now let’s try the HP10100C through terminator and use the 1 megohm input channel instead:

SDS2504X HD_PR_BNC-T

Some “experts” will say: “oh, this is much nicer now, without the ugly overshoot and ringing!”

The reality is, that we’ve lost all the fine detail due to linear distortion (frequency dependent amplitude error), so we don’t see the overshoot anymore and we also cannot measure the rise and fall times accurately.

But yes, you can say it’s not really a problem for low bandwidth oscilloscopes, because the 200 MHz bandwidth limit appears to have a far worse effect on that particular pulse measurement:

SDS2504X HD_PR_50_BWL200MHz

[Fungus]

The Rigol seems to run a filter, the staircase signal does not look so pronounced, compared to another 12-bit scope.

That seem unlikely.

What's the bandwidth and/or input capacitance of the two? I bet that's the difference.

[csuhi17]

If I have seen and understood correctly, the steps at Rigol have two heights.
A large step and a small step add up to a whole step compared to Siglent.

[TurboTom]

@Martin72: How many times/steps did you zoom in on the Rigol scope to get the "Stepped" sine screenshot from the original still? The one with the 1v/div setting cannot be the original since the trigger setting is different. If the original has also been recorded at 1V/div, zooming in to 50mV/div (4 steps) should be okay. But if it's been more than five increments, the shown effect is exactly what I'ld expect due to the "funny" behavior of Rigol's digital vertical zoom engine that I tried to analyze here: https://www.eevblog.com/forum/testgear/rigol-dho804-test-and-compare-thread/msg5108949/#msg5108949

Edit: Apparently I got your approach wrong: The screenshot with the "stepped sine" isn't a magnification of the other one with the full sine. I tried to replicate your findings (though I haven't got a Siglent scope for comparison) and also no stepped sine waveform, but I "assembled" a stepped ramp of 3Vpp with 1000 3mV steps. Obviously, zooming into the still of the "whole" ramp doesn't show any steps at all, but the DHO914S resolves the magnified signal exactly as one would expect. The screenshots are in sequence: original ramp, magnified still of original ramp (vertical position slightly adjusted), triggered magnified signal.

I cannot find anything unexpected here. 

[2N3055]

It is obvious why steps on Siglent are higher.
Because screen display is taller. 
Rigol has weird panoramic ratio, waveform plot area is some 20% "squished" vertically compared to Siglent.

That's what I meant in #236 when talking about the vertically compressed display due to the "letterbox" format.

But as mentioned I don't think that is the only effect. The displayed step size in mV also seems to differ, unless I am looking at things totally wrong. Please see my post #238 and Martin's most recent screenshots.

So I made a coffee and went again, this time at least semi conscious ..
When in doubt measure and verify. Rinse and repeat until data starts to make sense..

First, we should measure in AC input mode. This way you keep offset 0, trigger level 0 etc.... This way we exclude DC offset accuracy and trigger level accuracy. Those can (and should) be measured separately.
Scope will center signal around DC mean, waveform being symmetric and all..

That been done, let's look at the signal.
Steps are not all exactly the same. There are some up to 13mV in vicinity.
Waviness on top of steps is there.

Rigol is the one not showing correctly.

Verified with SDS6000H12 Pro, Keysight MSOX3104T and king of the hill 16bit Pico 4262

Questions?

[2N3055]

And this is signal, up close and personal with full 1GHz BW.

[2N3055]

@Martin72: How many times/steps did you zoom in on the Rigol scope to get the "Stepped" sine screenshot from the original still? The one with the 1v/div setting cannot be the original since the trigger setting is different. If the original has also been recorded at 1V/div, zooming in to 50mV/div (4 steps) should be okay. But if it's been more than five increments, the shown effect is exactly what I'ld expect due to the "funny" behavior of Rigol's digital vertical zoom engine that I tried to analyze here:

Edit: Apparently I got your approach wrong: The screenshot with the "stepped sine" isn't a magnification of the other one with the full sine. I tried to replicate your findings (though I haven't got a Siglent scope for comparison) and also no stepped sine waveform, but I "assembled" a stepped ramp of 3Vpp with 1000 3mV steps. Obviously, zooming into the still of the "whole" ramp doesn't show any steps at all, but the DHO914S resolves the magnified signal exactly as one would expect. The screenshots are in sequence: original ramp, magnified still of original ramp (vertical position slightly adjusted) and magnified, triggered signal.

I cannot find anything unexpected here. 

Batronix demo board makes simple sine with rather low res so steps are visible. Also steps are not clean, there are undulations on top of steps.
That makes it interesting signal with actual three levels of waveforms: full sine, steps and detail on top of steps.. 3V-10mV- sub mV level...

As for capture, what I understood, is that it was not "magnified after the fact" but simply captured at V/div as shown...

We need to check more but I see nonlinearities in ADC....  More specifically differential nonlinearities.. In some places it even seems non monotonic..
But we need to check more in detail before passing judgment.

P.S  this is one situation where we can "overdrive" input channels slightly. Since signal is slow moving, most of the overdrive recovery happens in few first pixels of screen on the left...

[ebastler]

Heck, guys -- I don't know why I am having such a hard time to get my point across. I am not concerned about varying step sizes within one of the captures or such. I am pointing out that the images which have been compared were showing two different signals. Specifically, two different segments of the test board-generated sine waveform, with different slopes and hence step sizes (differing between the scopes). WHich made the steps look less pronounced on the Rigol.

Please have a look at the attached, which is an annotated combination of the screenshots "Rigol_50mv" and "Siglent_50mV" from Martin's post #237. Maybe that explains it better than another 1000 words?

[TurboTom]

Tested some more since one detail slipped though my alertness (or rather lack thereof considering the time I went to sleep last night...  ):
@Martin72 applied a lot of vertical shift to the signal. And apparently, that's what explains the problem. Rigol's input stays linear up to a vertical shift of +-50 vertical divisions (10mV/div). Beyond this, ugly things appear to happen, a little bit more severe at positive shifts than negative ones, see screenshots...

Strange enough, the amplification factor appears to increase, but then there's also some kind of "integration" of the signal taking place. I didn't test further, it's important to keep in mind that (at least in the 10mV/div range), you're safe within a vertical shifting range of +-50 divisions. As yet I didn't test if the "Vertical Bias" setting may help...

[TurboTom]

It's worse, fellows, it's worse...

I just used 15mV steps with the otherwise identical signal as in my last post in order to "see" the steps better in lower sensitivity ranges. Finding: you can actually see the non-linearity at a vertical shift of 500mv (ten divisions of the 50mV/div range). That's practically a no-go. I'm very curious if this is a problem inherent to Rigol's new front-end ASIC and if this effect can be observed on their DHO1000 and 4000 instruments as well. IMO, this renders a 12 bit scope useless for anything else than most basic jobs. Btw, the "Bias" setting just works the same way as the "Offset".

[Fungus]

Questions?

Where's the screenshot of the Rigol?

[2N3055]

It is obvious why steps on Siglent are higher.
Because screen display is taller. 

No, it's not that.

Look at the maximum/minimum voltages in each one. They're completely different ranges.

Turn on some cursors if it will help you see it.

Thank you, but you're late... I already corrected it.
We went on measuring and Thomas already found some very bad stuff.. That means that both 800 and 900 are suspect. We need someone to check with 1000 and 4000, they might not have the problem, different board..
Still in progress...

Best,

[2N3055]

Questions?

Where's the screenshot of the Rigol?

Martin gave it.
I performed the important task of verifying test signal shape. 4 different scopes from 3 different manufacturers show exactly the same.