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

[Fungus]

At 5ns/div what do you see 400ps before the rising edge ? Zip, zero, nada !

Yep, and that's exactly what I want when I've got a good sample rate to bandwidth ratio (ie. with a single channel enabled).

Would I like a sharper, Siglent-like bandwidth rolloff when I turn on more channels and the ratio changess? Yes, but I can't have everything.

That's the dilemma... the choice a manufacturer has to make when they design an oscilloscope front end. 

[Mechatrommer]

From Leo's website a capture from a fast Tek clearly displaying the Gibbs ears other scopes choose to suppress.

1) gibbs ears are not "after the EVENT" (rising or falling edge), gibbs ears are before the "EVENT".
2) although overshoot and undershoot after the "EVENT" are similar (symetric) appearance. thats why its easily confused, if not read carefully.
2) overshoot and undershoot are real, gibbs ears are not...
3) overshoot and undershoot are "causal" phenomena, caused by the rising and falling EVENT, gibbs ears, if it really exist in reality, are "non-causal"
4) gibbs ear are fabricated in mathematical post processing in DSP IC due to imperfect (not infinite) terms sinc interpolation.
5) you yourself demonstrated the gibbs ear by showing 2c captures, one is using X, another is sinc interpolation, but it turned out you are looking at the wrong side (the actual post-event) of "gibbs ear"

you have similar confusion as Fungus who are mistakenly identify Gibbs ear as aliasing, thats why we keep reacting round and round... here again, on the correct side (the unrealistic, the DSP fabricated) of the gibbs ear i was talking about. not the other side...



here again (attached) i should have provided more proper picture, i thought i was clear enough (my mistake). and esp to sales representative need to get this right. cheers

[Mechatrommer]

There's a simple way to know if you're seeing aliasing: Turn some channels off and see if the displayed waveform changes.
(assuming you're not STOPped...)

aliasing is false low freq representation of actual high frequency element above nyquist/2.5 leo bodnar test shows no high frequency ringing, so no, i dont see any aliasing. either you look at the wrong side, or misinterpret DSP math post-processing with aliasing. aliasing will not caused non-causality... the picture you showed, is math post-processing, unless you can give better proof, fwiw..

[Fungus]

4) gibbs ear are fabricated in mathematical post processing in DSP IC due to imperfect (not infinite) terms sinc interpolation.

Nope.

A pulse is the sum of a load of sine waves (infinite). Gibbs ears are caused by some of those sine waves being missing.

(because 'scopes have finite bandwidth)

[ebastler]

4) gibbs ear are fabricated in mathematical post processing in DSP IC due to imperfect (not infinite) terms sinc interpolation.

Nope.

Thanks. And actually it's "nope" to most of the other statements in Mechatrommer's list of claims in that post. But I think we have already beaten this to death here and this discussion is going nowhere.

Could we all agree that the DHO800 pulse response is fine, and go back to the scheduled program?

(Edit: For clarity, I consider the Siglent pulse response fine as well. But that is the controversy over the past X pages of the thread, and the part where I think the discussion with Mechatrommer is not getting anywhere. And it is off-topic in a "DHO800 unbox and teardown" thread.)

[Serg65536]

gibbs ears are before the "EVENT".

That's complete nonsense.
Linear phase filters (as should be used in oscilloscopes) produce ripple BEFORE AND AFTER fast transients in the signal.
I have been using the Fourier transform analysis almost daily for many years.

Please, read the article carefully: https://en.wikipedia.org/wiki/Gibbs_phenomenon, https://wiki.seg.org/wiki/Dictionary:Gibbs%E2%80%99_phenomenon

[Mechatrommer]

4) gibbs ear are fabricated in mathematical post processing in DSP IC due to imperfect (not infinite) terms sinc interpolation.

Nope.

A pulse is the sum of a load of sine waves (infinite). Gibbs ears are caused by some of those sine waves being missing.

(because 'scopes have finite bandwidth)

i know, and your explanation is exactly what i tried to explain about missing hi order math components. the irony is you are contradicting your own explanation by denying it this is the effect when people too obsessed with math while losing grasp of what reality is imho.. you are mixing up analog filter with digital filter. Aliasing and under damped analog filter respond..

This is what analog filters do when bw limited


This is what digital filters do when bw (parameters/terms) limited


They have opposite respond/effect/representation.

And this is 'under-damped' analog filter respond, and possibly less aggressive digital filter processing.. NOT aliasing!



Please note all differing screenshots above came from the very same pulser device, same dso, except the later has been hacked upgraded... can you tell me how the actual pulse in reality supposed to look like?

Until you can read up some preliminary literatures on these subjects, we have not much to add. Its amazing to see how people are less and less interested in reading and understanding technical literatures, more and more interested in giving personal thought/belief or hearsay... it fits nicely with the prophecy's prediction

I remember when i was very young going back home on plane. next to me is a factory operator, we were discussing how a plane fly and gravity. . he told me gravity is what makes a plane floats. But I told him gravity is what makes us fall.. he insisted on his point, so i stop the discussion and quiet, let him win.. cheers


 

[Mechatrommer]

gibbs ears are before the "EVENT".

That's complete nonsense.
Linear phase filters (as should be used in oscilloscopes) produce ripple BEFORE AND AFTER fast transients in the signal.
I have been using the Fourier transform analysis almost daily for many years.

Please, read the article carefully: https://en.wikipedia.org/wiki/Gibbs_phenomenon, https://wiki.seg.org/wiki/Dictionary:Gibbs%E2%80%99_phenomenon

there is no mentioning anywhere in the article about actual physical filter that produces giibs effect, the article is talking about mathematical property, did you read?

The Gibbs phenomenon is a behavior of the Fourier series

do you know what is fourier series? and your linear phase filter is also the digital one https://en.wikipedia.org/wiki/Linear_phase https://en.wikipedia.org/wiki/Finite_impulse_response you have many thank you there, so congratulation if played well enough with FFT. if you do brickwall at f (cutoff freq), setting FFT elements past that f to zeros, and do inverse FFT, you'll have gibbs effect. unfortunately i tried to relate it (digital filter imperfection whether FIR or FFT or whatever implemented) to commentaries and confusion that its fabrication on screen as if its the actual signal and most accurate representation as it happened on circuit, which is not and a total misinformation to readers. you (Fungus and Tautech and possibly bunches of people thanking you) really missed the fact that gibbs ear (the pre-ringing one) effect is not in existence when plotted with R&S, Lecroy and Tek sampling scope used by Leo Bodnar. are you saying those 10GSps scopes not using proper "linear phase" filter?

[iMo]

You cannot get those ears before the rising edge starts to rise in a real hw system.
You see it on the LCD display because the sinc interpolation has been used to interpolate between couple of sampled points as I wrote above. Thus it is a graphical tool only. No FFT or any miracles are involved.
If you looked on raw data (off the memory after the ADC sampling) you would not see the ears before the start of the rising edge of the signal (for example).

PS: a low pass or any other filter (hw or sw) starts to ring "after" something happened at its input (the "causality principle" Mechatrommer mentioned above). The real physical system cannot react (knowing what to do) before comes to an input disturbance/change (like the rising/falling edge).

For example the Thermodynamic/Causal Arrow of Time..
https://en.wikipedia.org/wiki/Arrow_of_time

[Mechatrommer]

You cannot get those ears before the rising edge starts to rise in a real hw system.
You see it on the LCD display because the sinc interpolation has been used to interpolate between couple of sampled points as I wrote above. No FFT or any miracles are involved.
If you looked on raw data (off the memory after the ADC sampling) you would not see the ears before the start of the rising edge of the signal (for example).

yes this is exactly my point. i think when we present educational materials in public we should be careful by removing other factors that may mess up the subject we are discussing. the issue was "accurate representation" of the actual signal... imho we should turn off the questionable digital filter (sinc) feature in scope, use vector or dot plot, turn on peak detect etc and then claim it to be the most accurate. like in this capture...



i would say, this is "acceptable", Lecroy also presented a little bit of pre-ringing... but not as extreme as presented by 2N3055 in https://www.eevblog.com/forum/testgear/rigols-new-dho800-oscilloscope-unbox-teardown/msg5121846/#msg5121846 on subject of "not properly designed DSO due to aliasing"



this is trying to prove thing at it worse as we can see Sinc is messing the otherwise perfectly square wave.. and OT on that subject, he should feed the siglent while at 100MSps with > 50MHz sine or square signal instead of 16MHz arduino clock, aliasing guaranteed! whether Sinc is turned on or off. but then it seems even "aliasing" can be confusing to some people discussing it. cheers.


https://en.wikipedia.org/wiki/Aliasing

[tautech]

imho we should turn off the questionable digital filter (sinc) feature in scope, use vector or dot plot, turn on peak detect etc and then claim it to be the most accurate. like in this capture...

You have been busy digging out my previously posted screenshots from SDS6204A. 

[iMo]

That is the same as this example :
I give you 0 apples on Sunday, 1 apple on Monday, 2 apples on Tuesday, 3 apples on Wednesday.
On Thursday I calculate the average (for example), like (0+1+2+3)/4 = 1.5 and I write on your DHO804 LCD "You got from me 1.5 apples per day since Sunday".
Where the 1.5 apple on Sunday came from ??
 

[Mechatrommer]

imho we should turn off the questionable digital filter (sinc) feature in scope, use vector or dot plot, turn on peak detect etc and then claim it to be the most accurate. like in this capture...

You have been busy digging out my previously posted screenshots from SDS6204A. 

On to perusing this thread https://www.eevblog.com/forum/testgear/split-from-rigols-new-dho800-oscilloscope-unbox-amp-teardown/msg5136138/#msg5136138 then i came across some "hilarious" posts imho thats difficult for me to let go not particularly you, but whoever did that even now i have to face one of rigol fanboyism will that made me change side? no! i'm just here, on my own side cheers.

[2N3055]

this is trying to prove thing at it worse as we can see Sinc is messing the otherwise perfectly square wave.. and OT on that subject, he should feed the siglent while at 100MSps with > 50MHz sine or square signal instead of 16MHz arduino clock, aliasing guaranteed! whether Sinc is turned on or off. but then it seems even "aliasing" can be confusing to some people discussing it. cheers.

Could you please stop twisting what people said and misrepresenting data? Please?

You are taking images I made for specific purpose with very specific set of parameters to illustrate something.
Then you use that image and say it is something else and babble about evil Sinc reconstruction filter..
It has been explained in 1948. Long ago..

What I have shown is what happens when you undersample square wave signal, how it looks.

I used 16 MHz deliberately to illustrate point of very common signal (Arduino clock), being square wave, will have much larger frequency content of 16 MHz. This is common mistake with beginners, they buy 100 MHz BW scope to look at 50 MHz square wave because it is only 50 MHz, right.

With square wave, how much frequency BW is there in signal is visible in rise/fall time of the signal.
You can have square wave signal that has repetition frequency of 1 kHz, that will have 4GHz BW of spectrum..

16 MHz square wave (with 30ps edges) will have harmonics into gigahertz range. Any of these harmonics above nyquist will alias. And then at reconstruction will be mixed back into waveform as lower frequencies, distorting pulse shape.

Undersampling will also result in not enough timing accuracy for digital triggering. That will result in trigger jitter.
So combine that with changed pulse shape and you get that shape.

Scope cannot violate math and physics. But it is not ideal representation of infinite functions.

In attachment:
10 MHz square wave on:

Keysight 3104T : 1GHz full BW 1ns/div

[nctnico]

You cannot get those ears before the rising edge starts to rise in a real hw system.
You see it on the LCD display because the sinc interpolation has been used to interpolate between couple of sampled points as I wrote above. No FFT or any miracles are involved.
If you looked on raw data (off the memory after the ADC sampling) you would not see the ears before the start of the rising edge of the signal (for example).

yes this is exactly my point. i think when we present educational materials in public we should be careful by removing other factors that may mess up the subject we are discussing. the issue was "accurate representation" of the actual signal... imho we should turn off the questionable digital filter (sinc) feature in scope, use vector or dot plot, turn on peak detect etc and then claim it to be the most accurate. like in this capture...

No. Dot and vector mode will show a signal which is even further from what the signal looks like in reality. As I wrote before: if you want to prevent Gibb's ears then make sure the signal has a limited bandwidth.

[tautech]

10 MHz square wave on:

Keysight 3104T : 1GHz full BW 1ns/div

Fat fuzzy traces. 

[2N3055]

10 MHz square wave on:

Keysight 3104T : 1GHz full BW 1ns/div

Fat fuzzy traces. 

Yep. But exactly same response as SDS6104.

[Mechatrommer]

Legends Never Die - Rigol DHO924S (ermmm... 804) looking at 100MHz RF signal (EraSynth Micro) with amplitude -70dBm to 10dBm

[mawyatt]

We must remember a DSO is NOT a scope in the sense of the older analog scopes like the ones Tek provided, these portray signals in a more "realistic" sense because they are "almost" real time, "almost" in the perspective of just slightly delayed in time to the signals occurring at the probe tip which is shown as a electron beam sweeping across a screen (not considering triggering and retracing). Note the signal of interest is delayed, scaled/amplified and filtered but NOT sampled as in a DSO, and some non-linear effects may be introduced along the way to the screen for viewing. Whereas a DSO is actually a data acquisition system, disguised as a scope which captures data with Analog to Digital Conversion with sampling techniques (both of which are non-linear functions). The sampled waveform data stored (just digital bits now) in memory and displays a rendering of the data (digital bits) captured in memory with manipulations on a computer type screen (pixels) to render what the DSO "thinks" was the original waveform in a from the user prefers. Totally different than the phosphor based screen with sweeping electron beams used with analog scopes. Here, the DSO "thinks" involves all sorts of digital data manipulation, some user requested, some not, and varies highly with DSO brand, settings, and development date.

If one thinks for minute an ideal squarewave has infinite harmonics, which the analog scope low pass filters and displays as the pulse step response on each edge, thus some of the harmonics are attenuated, and higher one towards obscurity. We often think of this "response" as the Step response of the filter, but can also be viewed as a consequence of affecting the harmonics in the original waveform step. However because these harmonics are filtered in real time by the analog filters, and displayed by the pseudo-real time analog screen, the resultant display possess no magic non-causal effects (Gibbs ears).

A DSO behaves quite differently as mentioned being a sampled data system, and displaying results long after the fact and highly manipulated. Especially true wrt to the actual display, and how it's rendered, which involves "waveform reconstruction" in the sense of the discrete time digital data is arranged/manipulated to display an apparent continuous waveform to the user by means display "filters". One must remember the DSO screen is NOT continuous in time nor amplitude, but yet another discrete X and Y system, and the sampled signal digital data which is discrete in time (X) and amplitude (Y) must be "fit" into the display area for the user to interpret!!

Regarding the Gibbs ears, this is a manifestation of the digital rendering of the captured analog signal by the non-linear processes of Analog to Digital Conversion and Discrete Time Sampling, with the effects of both analog and digital filtering involved, and manipulation for displaying on the DSO screen to add further confusion.

So it should be no surprise there are significant differences between the two methods of time domain waveform displaying. One is entirely continuous in time and amplitude (analog), the other is highly discontinuous in time and amplitude (DSO).

Best,

[mawyatt]

I used 16 MHz deliberately to illustrate point of very common signal (Arduino clock), being square wave, will have much larger frequency content of 16 MHz. This is common mistake with beginners, they buy 100 MHz BW scope to look at 50 MHz square wave because it is only 50 MHz, right.

With square wave, how much frequency BW is there in signal is visible in rise/fall time of the signal.
You can have square wave signal that has repetition frequency of 1 kHz, that will have 4GHz BW of spectrum..

16 MHz square wave (with 30ps edges) will have harmonics into gigahertz range. Any of these harmonics above nyquist will alias. And then at reconstruction will be mixed back into waveform as lower frequencies, distorting pulse shape.

Agree, very common for folks to "think" a squarewave is just a fundamental waveform with a few "Harmonics" at 3X and 5X, whereas "seasoned" folks know the fast edge speed can yield results well into the Multi-GHz regions!!

A good question to ask, "what's the frequency content of a Step function, or an Impulse function!!!"

Best,

[nctnico]

16 MHz square wave (with 30ps edges) will have harmonics into gigahertz range. Any of these harmonics above nyquist will alias.

In theory yes, but given the spacing between harmonics and their levels, the chance of harmonics aliasing back as a visual artifact are not very high as you have to consider the anti-aliasing filter in a DSO. Especially when dealing with a square wave which has mostly odd harmonics which are spaced at twice the fundamental frequency.

[TimFox]

Also, the odd harmonics in a symmetric square wave fall off as 1/n in amplitude (voltage), so they eventually fall below the noise in the system.

[2N3055]

16 MHz square wave (with 30ps edges) will have harmonics into gigahertz range. Any of these harmonics above nyquist will alias.

In theory yes, but given the spacing between harmonics and their levels, the chance of harmonics aliasing back as a visual artifact are not very high as you have to consider the anti-aliasing filter in a DSO. Especially when dealing with a square wave which has mostly odd harmonics which are spaced at twice the fundamental frequency.

Problem is that they will all get compressed in same energy "blob" and will add up.
As they get folded down in frequency domain. Equivalent in time domain is, if you can imagine a small differential triangle formed by fast edge and edge of risetime of scope. Area of that triangle gets folded on top of the rising edge with a overshoot pulse that has period of 1/risetime....
If energy enters the ADC it will show somewhere..

[mawyatt]

One might be temped to create the "Gibbs ears" by means of a fast edge squarewave, then low pass filtering the waveform to reduce/eliminate higher order harmonics to create the "ears" as shown in earlier computer generated visualizations of adding harmonics to create a squarewave approximation.

If we can agree the analog scope does not "self-inflict" the Gibbs effect, then view on the analog scope and DSO at various settings.

This exercise is left to the reader!!!

Best,

[TimFox]

Of course, the output waveform of a "fast" square wave (e.g., the output of a good binary flip-flop) passed through an elementary low-pass filter (e.g., a simple one-pole R-C circuit) should not exhibit the Gibbs phenomenon on a high-bandwidth analog oscilloscope or a fast-enough sampling digital oscilloscope.