Sub: Rigol's DHO800 Oscilloscope (Gibbs Effect & Aliasing Misunderstanding)

[Mechatrommer]

On to perusing this thread https://www.eevblog.com/forum/testgear/rigols-new-dho800-oscilloscope-unbox-teardown/ to get as much useful info on DHO800/900 hack/mod/upgrade. putting bookmarks on most important and useful posts/infos... now all pages perused, not much to add so far... except few commentaries that got me interested since they are not well, or partially answered. so lets me add something to clear things up. what interest me is peculiar SIGLENT's DSO lines "non-causal ringing" plot just before signal "rise" and signal "down" also known as and let us termed it as "event"... ringing should be caused by the "event", ie after the "event", not before... siglent's plot show as if the "event" is caused by the ringing, not vice versa which should be the natural phenomena in nature in general, and in circuit debugging in specific...

the arguments was that RIGOL's plot are inaccurate due to "no ringing", this "artifact" has been explained by experienced members, for example by Kleinstein (inside DSO's front end artifact), but got sidetracked maybe due to nitpicking agenda? so let me add a bit about this matter. we know ringing can be caused by reflection of mismatched impedance between source and destination, and also due to resonance inside LC networks inside the scope. so different scope will have different internal respond to external signal feed to it. even with my newly acquired DHO800, at its standard setting (70MHz BW) pulses from Leo Bodnar's and UTG962 shows smooth transition with no significant ringing, but when changed to DHO924 FW (230MHz BW), "causal ringing" becomes so significant, due to change in BW limiting mechanism inside the scope. i can provide proves if required, but you can check in my posts if you are interested. anyway, i can explain more, but that is not my main point to tackle here.

the main suspicion is on the "non-causal ringing" on Siglent's DSOs plots, lets stick to the plan remember? just to avoid unnecessary confusion and nitpicking the wrong direction... attached is the area i'm talking about (circled), taken from one of the samples provided here. my take on this is that... the plots are post-processing "fabrication" inside the siglent scope, and it can trick newbies into thinking that is a real signal, but in fact actually it was never there in reality, not even in the dso's front end stage! imho. bad thing about this is it can result in wrong decision making, argumentations and debates later on among newbies vs expert alike. so if we talk about giving false data (signal) i must question again who's giving the false data? this need to be taken into consideration whoever newbies want to buy a new scope, esp the pricey one, iirc this "non-causal" effect also demonstrated by high end $4-5K range siglent scope somewhere outside this thread. when can we see this "non-causal" effect? is when we try to approximate square wave with limited parameters in numerical or fourier analysis Fourier Approximation and Gibbs Phenomenon i believe this is also has to do with "finite" amount of parameters during sinc(x) reconstruction of a signal, so i believe this is happening inside the dsp IC or fpga/asics, but that i cannot be sure,  since i have no proof. its just highly likely, where else it can be? LC circuit cannot produce "non-causal" effect, can it?

now why i dare to take this conclusion? is due to sheer amount of proofs screenshots by other instruments, including the most high end R&S DSO, posted by members and land owners Dave Jones, and HP? or Tek? sampling scope test made by Leo Bodnar himself, that this non-causal (false or lie) ringing does not exist! to add value to the existing one, also attached are plots and setup from Lecroy SDA6000 scope. among the most prominent posts i was refering to and bookmarked are listed below for reference. please dont get it wrong, this is not intended for a shame game, nor a fanboyism dick waving contest. this just to add infos and possibly further discussions, so we can gain more knowledge as to why is this, and newbies can do more informed decision when buying scopes. i stand corrected if you have more academic and technical explanations, i'm also in need of learning. that is for the night, cheers and best regards.

https://www.eevblog.com/forum/testgear/rigols-new-dho800-oscilloscope-unbox-teardown/msg5076118/#msg5076118
https://www.eevblog.com/forum/testgear/rigols-new-dho800-oscilloscope-unbox-teardown/msg5077231/#msg5077231
https://www.eevblog.com/forum/testgear/rigols-new-dho800-oscilloscope-unbox-teardown/msg5078086/#msg5078086
https://www.eevblog.com/forum/testgear/rigols-new-dho800-oscilloscope-unbox-teardown/msg5078230/#msg5078230
https://www.eevblog.com/forum/testgear/rigols-new-dho800-oscilloscope-unbox-teardown/msg5121846/#msg5121846

[nctnico]

the main suspicion is on the "non-causal ringing" on Siglent's DSOs plots,

These are typical 'Gibb's ears' caused by sin x / x signal reconstruction where the signal frequency versus sampling rate approaches the minimum samplerate / nyquist limit where sin x / x works. It is for the user to decide whether the Gibb's ears are allowable or whether the bandwidth of the oscilloscope needs to be reduced.

[ebastler]

Two things are important here, in my view:

(a) It's not the case that the ringing is actually present on the Bodnar pulser's output and the DHO800 somehow "glosses it over" while Siglent shows the truth. Tautech had claimed that early in the discussion a few weeks ago, but dropped that claim later.

(b) It is also not the case that the Siglent scopes are "wrong" and produce unacceptable artefacts. They do apparently use lowpass filters with a relatively sharp cutoff, causing the Gibbs reconstruction phenomenon as described. Rigol (and R&S, which Dave showed for comparison too) seem to use filters with a smoother roll-off, resulting in less overshoot in the reconstruction. A matter of philosophy or personal preference, I'd say.

[Mechatrommer]

They do apparently use lowpass filters with a relatively sharp cutoff, causing the Gibbs reconstruction phenomenon as described.

i believe the more accurate term is "digital sharp cutoff". i had a play with FFT and did brickwall filter, do inverse FFT and got this Gibbs effect, in the end... its NOT the real signal anywhere on the pcb... thats my main point.

[iMo]

OT comment - Bodnar's <40ps edges require special cabling, connectors and matching, imho, the BNC connectors are not the best technology here..

[ebastler]

its NOT the real signal anywhere on the pcb... thats my main point.

It is the real signal, as represented by all its harmonics that fall within the scope's Nyquist limit. Which, one could argue, is what you expect a DSO to show you.

Having said that -- personally I do prefer the smoother roll-off used by Rigol and R&S, which dampens or avoids those apparent overshoots. Otherwise it is too tempting to mis-interpret the overshoots as being actually present (and a potential problem) in the input signal.

[Mechatrommer]

OT comment - Bodnar's <40ps edges require special cabling, connectors and matching, imho, the BNC connectors are not the best technology here..

i specifically bought the BNC version to test low end DSO which always have BNC inputs. for high end reputable brand name DSO, there is not much point in testing with Leo Bodnar pulse. so there more is one more adapter seen in my attachment that should not be there, hence expect a little bit more ringing... its ok if i have to add adapter on one test rather than have to do many times of it (less reliable mismatch) on low end DSOs tests. there is no LPA (BMA connector of Lecroy's input) version of Leo's Pulser anyway. ymmv.

its NOT the real signal anywhere on the pcb... thats my main point.

It is the real signal, as represented by all its harmonics that fall within the scope's Nyquist limit. Which, one could argue, is what you expect a DSO to show you.

i believe not, unless you can provide proof that "physical" filter can give this "non-causal" effect. are you telling us that... R&S, HP, Lecroy, Rigol are telling lies by hiding this reality of "Gibbs Ear" on actual "physical" cirrcuit? it is only a crippled version of "mathematical" or "theoritical" representation of the real signal, minus the higher terms that the dsp/dso designer decided to omit due to limited resources. not the actual one on circuit.

[2N3055]

OT comment - Bodnar's <40ps edges require special cabling, connectors and matching, imho, the BNC connectors are not the best technology here..

You can connect BNC  version directly to scope input without cables...

[ebastler]

i believe not, unless you can provide proof that "physical" filter can give this "non-causal" effect. are you telling us that... R&S, HP, Lecroy, Rigol are telling lies by hiding this reality of "Gibbs Ear" on actual "physical" cirrcuit?

Yes, a steep (higher order) analog, "physical" filter will give the same effect. Edit: Hang on -- specifically referring to the "pre-ringing", analog filters will not exhibit that. They will show the Gibbs phenomenon (apparent overshoot and ringing) after the edge though, which is also an effect of the limited frequency components used to reconstruct the signal. The DHO800 does not show that either, due to its smoother filter roll-off.

I don't know which design of bandwidth limiting filter Siglent or each of the other scope manufacturers you mention are using. In order to implement a digital cutoff filter you need to oversample the input signal, and especially near the scope's bandwidth there is often not that much headroom in the sampling rate -- so it may well be analog-only filters in most cases?

Also, I don't know whether your implication is true that none of the other brands shows a similar Gibbs effect. Do you have access to pulse response data for all of them? For example, Lecroy -- with their focus on "analytical" scopes -- might take a similar approach as Siglent? And it would certainly be plausbile that Siglent takes some guidance from LeCroy as part of their collaboration.

[2N3055]

i believe not, unless you can provide proof that "physical" filter can give this "non-causal" effect. are you telling us that... R&S, HP, Lecroy, Rigol are telling lies by hiding this reality of "Gibbs Ear" on actual "physical" cirrcuit?

Yes, a steep (higher order) analog, "physical" filter will give the same effect.

I don't know which design of bandwidth limiting filter Siglent or each of the other scope manufacturers you mention are using. In order to implement a digital cutoff filter you need to oversample the input signal, and especially near the scope's bandwidth there is often not that much headroom in the sampling rate -- so it may well be analog-only filters in most cases?

Also, I don't know whether your implication is true that none of the other brands shows a similar Gibbs effect. Do you have access to pulse response data for all of them? For example, Lecroy -- with their focus on "analytical" scopes -- might take a similar approach as Siglent? And it would certainly be plausbile that Siglent takes some guidance from LeCroy as part of their collaboration.

Most modern scopes, especially higher BW ones will have that overshoot as a consequence of brickwall AA filter used.

DHO800 does not show it because it's 100MHz filter has gentle slope and combined with 1.25GS/s on one channel will give good response.
Enable more than one channel and check then again.
Even more will be visible on DHO900 at 200MHz BW...

If I go to both my 500MHz and 1GHz scope and enable 200 MHz BW limit, it shows perfect pulse response.

Keysight explanation, in attachment.

[Mechatrommer]

Yes, a steep (higher order) analog, "physical" filter will give the same effect.

i noticed gibbs effect is worst in siglent at much lower sampling rate, but not at its max smapling rate. from here https://www.eevblog.com/forum/testgear/rigols-new-dho800-oscilloscope-unbox-teardown/msg5121846/#msg5121846

1GSps:


100MSps:


lowering sample rate should lower the cut off frequency, not creating more high frequency elements. even sampling below nyquist and straight plot point to point (no sinc interpolation) will not "fabricate" this artifact, if its a real physical filter that caused it, why implement a physical filter that causes artifact that is not seen in high end dso?

Most modern scopes, especially higher BW ones will have that overshoot as a consequence of brickwall AA filter used.

see? i think my point will be side tracked again, i was talking about "non-causal" or "precursor" ringing... not overshoot not undershoot that we usually saw do you care to read my long post above? or at least look carefully at 1st picture attachment, that whats i'm talking about. here again...

[ebastler]

i think my point will be side tracked again, i was talking about "non-causal" or "precursor" ringing... not overshoot do you care to read my long post above? or at least look carefully at 1st picture attachment, that whats i'm talking about.

Got your point now -- please see the "Edit" in my most recent post.
Maybe your original post was a bit too long for me, with too much text. Talk about a "brickwall"... 

Good night!


Edit: I Still stand by my earlier assessment,

It is the real signal, as represented by all its harmonics that fall within the scope's Nyquist limit. Which, one could argue, is what you expect a DSO to show you.

But getting that "ideal" set of harmonics, without relative phase shifts, can't be done with an analog filter. Hence there must be digital filtering happening in the Siglent.

[Martin72]

You have made me curious...
You don't necessarily have to open it to find out what the quality is like.
Because in the end it's what comes out that matters....
I'm just thinking about a USB-C adapter on which you can make measurements.

I have now ordered USB-C breakout boards, once to look at the signal quality of the power supply and as a nice side effect, to check the accuracy of my USB-C powermeter.
If someone should have opened the liteon power supply, how many wires come from the connection line ?

[TimFox]

A lot of posts on this thread, include mention of Lenovo (supplied with my DHO914S) and LiteOn supplies.
Has anyone made a direct comparison between DHO800 or DHO900 performance with different power supplies, or, even better, between a supplied unit and a very clean non-switchmode lab supply?

[Martin72]

No, but it´s a nice idea, will do it in the next days.

DHO800 or DHO900 performance

What performance? Noise ?

[tautech]

Especially for Mech some screenshots better comparing apples.....at similar timebase settings....max on 1000X HD.
Bodnar powered from rear USB port. 
X and SinC interpolation.... captured running and Stop.

[Fungus]

i believe this is happening inside the dsp IC or fpga/asics, but that i cannot be sure,  since i have no proof. its just highly likely, where else it can be? LC circuit cannot produce "non-causal" effect, can it?

I think it's this: https://en.wikipedia.org/wiki/Gibbs_phenomenon

It's counterintuitive that you see something before the pulse but it's mathematically correct. That's what you get when you apply a low pass filter to a discontinuity (step) in a signal.

[ebastler]

i believe this is happening inside the dsp IC or fpga/asics, but that i cannot be sure,  since i have no proof. its just highly likely, where else it can be? LC circuit cannot produce "non-causal" effect, can it?

I think it's this: https://en.wikipedia.org/wiki/Gibbs_phenomenon

It's sometimes advisable to read the next couple of posts before one fires off a reply. 

[Fungus]

It's sometimes advisable to read the next couple of posts before one fires off a reply. 

Yeah, I just noticed.

Still, nobody else posted a link to Wikipedia or said "if you apply a low pass filter to a step..." 

[Mechatrommer]

i believe this is happening inside the dsp IC or fpga/asics, but that i cannot be sure,  since i have no proof. its just highly likely, where else it can be? LC circuit cannot produce "non-causal" effect, can it?

I think it's this: https://en.wikipedia.org/wiki/Gibbs_phenomenon
It's counterintuitive that you see something before the pulse but that's sin(x)/x for you.

yes i provided the link in my post. sin(x)/x is a highly theoritical and mathematical. iirc i read long ago that you cannot reconstruct back the sampled real signal back on monitor the way its appeared on electrical circuit if you dont do infinite terms reconstruction, at least thats what i understood. with limited terms, what you interpolated on screen will not match the real signal. even lesser terms, the displayed signal will be deviated even more from the actual one.

@tautech: refering to picture below. a little bit of "precursor" ringing could exist in real world i think to some specific semiconductors or logic fabrics? maybe something triggered inside the logic gate much earlier that caused ringing before actual logic state switch (rising and falling), but excessive precursor ringing is a bit unnatural, unnaturally "symmetrical" . i wonder if such extreme such as image provided by 2N3055 just above could exist to physical/analog filter at dso front end. anyway, proofs that Loe's pulse doesnt have this excessive non-causality are already presented by other brands TE. fwiw.

[Mechatrommer]

X and SinC interpolation.... captured running and Stop.

i just noticed and look carefully that you also proved the non-causal ringing appeared when using Sin(x) and does not exist when doing straight line (X) graph plotting.

[2N3055]

Yes, a steep (higher order) analog, "physical" filter will give the same effect.

i noticed gibbs effect is worst in siglent at much lower sampling rate, but not at its max smapling rate. from here https://www.eevblog.com/forum/testgear/rigols-new-dho800-oscilloscope-unbox-

lowering sample rate should lower the cut off frequency, not creating more high frequency elements. even sampling below nyquist and straight plot point to point (no sinc interpolation) will not "fabricate" this artifact, if its a real physical filter that caused it, why implement a physical filter that causes artifact that is not seen in high end dso?

Most modern scopes, especially higher BW ones will have that overshoot as a consequence of brickwall AA filter used.

see? i think my point will be side tracked again, i was talking about "non-causal" or "precursor" ringing... not overshoot not undershoot that we usually saw do you care to read my long post above? or at least look carefully at 1st picture attachment, that whats i'm talking about. here again...

Please don't get angry again. I read your post. But your post is very complicated and you mixed few different things together.
And you did not understand my images I posted..

So to start from end: Overshoots you see on that Siglent are result of analog Brickwall filter when you feed it with very fast signal, much faster than it's transition band is.  I have exactly same artefacts on my Keysight MSOX3104T. That same effect is very common in higher ends DSOs, that happen to have very sharp Brickwall response filters and sample at 2.5x the BW.
Only way to avoid them is to have more oversampling. DHO800 on the image from Fungus had 100MHz BW and sampling at 1.25GS/s. Factor of 12.5x more.
Images I posted were not answer to that question but to one about insufficient sampling

Lovering sample rate does not change cut off frequency whatsoever. Where did you get that information?
Cutoff frequency is solely function of input channel's BW.
Lowering sample rate simply means you take sample of same signal (with same BW) but with samples more apart in time.

What I have shown you is EXACTLY what happens when you sample signal with 200MHz + frequency content with insufficient sampling rate.
Full bandwidth of signal reaches ADC that cannot reconstruct signal correctly anymore because sample points start to skip parts of signal curve. Since input signal is not synchronous (it has no correlation) with sampling clock of the scope, on every consecutive trigger you get another slightly randomized wrongly reconstructed shape. With those happening thousands of times per second, screen shows the cloud of funny shapes.
And apparent higher frequency content is reconstruction algorithm tryin to curve fit curve through nonsensical data.

It is not artefact of Siglent scopes, but quite the opposite: since I have full manual control of sampling rate and BW on those scopes I could produce such experiment. Normally with scope in Auto memory mode scope you won't be able to make this effect.

And experiment shows you EXACTLY what happens if you have scope with 200MHz BW, insufficient sampling rate and quite common 16MHz square wave signal with 1ns edges. I didn't even shoot Bodnar's pulser in it. I made signal quite simple and common so anybody can reproduce it.

It will happen on any scope that you can achieve those conditions. On DHO800 with 100MHz BW this should not happen that easy or at all.
But if you hack it to 200MHz, enable all 4ch (sample rate drops to 312.5 MS/s) you should see same or similar thing...

That is why I kepp repeating that DHO800 samples properly and DHO900 does not unles in very limited usage scenario.
DHO800 with 100 MHz BW should be largely immune to this. Unless you manually reduce sample rate or memory length at long timebases..

I hope I explained this well. If you don't understand what I mean please ask and I will try to explain it further/better..

Since DHO800 does not seem to have manual sample rate control you can try seeing same effect by reducing memory depth to 1kpts or 10kpts  to force smaller sample rate. Play with it a bit to understand where limits are...

[tautech]

@tautech: refering to picture below. a little bit of "precursor" ringing could exist in real world i think to some specific semiconductors or logic fabrics? maybe something triggered inside the logic gate much earlier that caused ringing before actual logic state switch (rising and falling), but excessive precursor ringing is a bit unnatural, unnaturally "symmetrical" . i wonder if such extreme such as image provided by 2N3055 just above could exist to physical/analog filter at dso front end. anyway, proofs that Loe's pulse doesnt have this excessive non-causality are already presented by other brands TE. fwiw.

Yet if we look at the real values they are small vs the full signal. From previous readings I always understood these to be Gibbs related.
From Step response adjustments I have done, the step speed (RT) is best to match the capabilities of the front end and if using anything too fast for the scope the response can be impacted so other waveforms are affected.
IMO a tradeoff is required.
YMMV

[2N3055]

X and SinC interpolation.... captured running and Stop.

i just noticed and look carefully that you also proved the non-causal ringing appeared when using Sin(x) and does not exist when doing straight line (X) graph plotting.

Because overshoot is artefact of SinC interpolation..

Again, my images has nothing to do with this overshoot behaviour.
It is simulation of insufficient sampling frequency by setting my scope pathologically wrong, manually, on purpose.

[tautech]

X and SinC interpolation.... captured running and Stop.

i just noticed and look carefully that you also proved the non-causal ringing appeared when using Sin(x) and does not exist when doing straight line (X) graph plotting.

Stop and SinX was ugly as one expects. Didn't even bother to capture it.