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

[mawyatt]

as long as "fundamentals" is not exceeding nyquist limit, there will be no aliasing.

In my opinion, this statement is a bit false.

But of course.
When the system does not produce an alias frequency from fundamental (aka 1st harmonic) then fundamental is not aliasing.

When harmonics are downconverted by ADC these down conversion products are there. And these down converted "false" signals we call a bit carelessly aliases. This is true with all harmonics, including also 1st. 1st harmonic aka fundamental.

Many seem to forget an ADC is actually like an Impulse Sampler in the time domain (Zero order hold), or close approximation of such. This is true because the ADC produces a digital output rendering of the analog input signal at precise time interval sample points.

This along with Nyquist, produces frequency translation similar to a classic heterodyne mixer from the RF world. We utilized this Nyquist "Feature" with custom ultra-high speed Sampling CCD Convolvers (Chirp Z Transform) which operated in various Nyquist "Zones" back in 1980 for a handheld RTSA we developed.

The reverse is also true as one can employ Nyquist to "Produce" output signals from a DAC in various Nyquist frequency "zones", very powerful techniques indeed!!

Best,

[Fungus]

Turning off sinc is NOT a good idea.

i give up! You win! do whatever you like!

What do you propose the 'scope should show instead?

[Mechatrommer]

As you (rf loop) also agree, this is not siglent vs rigol war.. this to increase awareness.. if you dont nitpick, you should see there i put rigol name there as well.. not for tug war, but thats 2 name we usually own, and S people usually come and degraded cheaper R based on delusion. I can as well ask you to do it on r&s, lecroy, tektronix et al, for the sake of technical/theoritical/academic discussion.. but probably hi end brands implements interpolation a bit differently.. please also do so whats on your bench and convenient to you.

[Mechatrommer]

Turning off sinc is NOT a good idea.

i give up! You win! do whatever you like!

What do you propose the 'scope should show instead?

what good proposal is? if only for you to deny? And you keep putting 2 contradicting statements in one post. Amazing!

[Mechatrommer]

here i tried to replicate 2N3055's capture of Gibbs Effect in my DS1054Z. i dont have fast rise time 16MHz that can produce 10mVpp, but since the focus is on Gibbs Effect (whether its real or not?), the closest i can get is Leo Bodnar's pulse 10MHz @ 1Vpp. and then i figured i was actually struggling to get as low sampling rate i can get similar to 2N3055, i have to turn all channels ON and have to go up to 2us/div, reduce memory depth down to minimum 3Kpts/ch, then only i get 125MSps, but at this time/div, the cycles are too densed together to give any useful info for the purpose. so i have to dial back down 20ns/div and thats at 250MSps, thats the closest i can get. we cant set sampling rate to our heart content in Rigol, apart from reducing mem depth and increase time/div... other than that its automatic, maybe to avoid nasty Gibbs? who know?

turn on Sinc, vector plot (sinc_on.jpg)
turn off Sinc, dot mode (while running/capturing) (dot_on_run.jpg)
turn off Sinc, dot mode (Stopped to better see points) (dot_on_stop.jpg)
turn off Sinc, vector plot (sinc_off.jpg)

Gibbs effect succesfully produced in (sinc_on.jpg), but when in dot mode (dot_on_run.jpg or dot_on_stop.jpg) Gibbs dissappear, if you try imaginatively (if you dont want to try it on paper) to connect straight line between point to point in dot mode, you wont see any Gibbs effect. Gibbs only appear when Sinc interpolator is turned on in scope. do you understand what i meant before posting?

so tell me Mr sharp trained eyes Fungus (and others who have sharp eyes or brain) which signal do you like? or think its true happening on circuit?.. sinc_on.jpg? sinc_off.jpg? or dot_on_stop.jpg? or all true? because when i turn on Sinc, suddenly something on front end DSO pcb clicked and activate 8th order non-causal Chebysev?

and can any S or other brand DSO to do this test? and demonstrate here? cheers..

[Mechatrommer]

In dot / vector mode, you won't get the Gibbs effect because it comes with sin x /x reconstruction. Still, the sin x/x reconstructed signal is the closest you'll get to seeing something that resembles the real signal when you are getting near the nyquist limit....

no... see above...

[gf]

Gibbs effect succesfully produced in (sinc_on.jpg), but when in dot mode (dot_on_run.jpg or dot_on_stop.jpg) Gibbs dissappear, if you try imaginatively (if you dont want to try it on paper) to connect straight line between point to point in dot mode, you wont see any Gibbs effect. Gibbs only appear when Sinc interpolator is turned on in scope. do you understand what i meant before posting?

I guess that your signal is undersampled. Then It is expected to happen. Consider it an aliasing artifact. A properly bandwidth-limited signal is reconstructed almost exactly by sinc reconstruction, but if the signal fed into the ADC violates the sampling theorem, then perfect reconstruction from the samples fails and the reconstructed signal looks different from the original signal you fed into the ADC.

[gf]

To me that looks exactly like a bandwidth limited square wave:
...
Nothing to do with sinc function.

The relation is the inverse Fourier transform of the frequency response of a linear phase brickwall filter. It is a sinc function in the time domain. IOW, applying such a brickwall lowpass filter in the frequency domain is equivalent to a convolution with a sinc function in the time domain.

[Fungus]

turn off Sinc...
[/quote]

You can turn off sinc on a DS1054Z when there's 4 channels enabled but it replaces it with a different filter. You're not seeing raw data.

[2N3055]

here i tried to replicate 2N3055's capture of Gibbs Effect in my DS1054Z. i dont have fast rise time 16MHz that can produce 10mVpp, but since the focus is on Gibbs Effect (whether its real or not?), the closest i can get is Leo Bodnar's pulse 10MHz @ 1Vpp. and then i figured i was actually struggling to get as low sampling rate i can get similar to 2N3055, i have to turn all channels ON and have to go up to 2us/div, reduce memory depth down to minimum 3Kpts/ch, then only i get 125MSps, but at this time/div, the cycles are too densed together to give any useful info for the purpose. so i have to dial back down 20ns/div and thats at 250MSps, thats the closest i can get. we cant set sampling rate to our heart content in Rigol, apart from reducing mem depth and increase time/div... other than that its automatic, maybe to avoid nasty Gibbs? who know?

turn on Sinc, vector plot (sinc_on.jpg)
turn off Sinc, dot mode (while running/capturing) (dot_on_run.jpg)
turn off Sinc, dot mode (Stopped to better see points) (dot_on_stop.jpg)
turn off Sinc, vector plot (sinc_off.jpg)

Gibbs effect succesfully produced in (sinc_on.jpg), but when in dot mode (dot_on_run.jpg or dot_on_stop.jpg) Gibbs dissappear, if you try imaginatively (if you dont want to try it on paper) to connect straight line between point to point in dot mode, you wont see any Gibbs effect. Gibbs only appear when Sinc interpolator is turned on in scope. do you understand what i meant before posting?

so tell me Mr sharp trained eyes Fungus (and others who have sharp eyes or brain) which signal do you like? or think its true happening on circuit?.. sinc_on.jpg? sinc_off.jpg? or dot_on_stop.jpg? or all true? because when i turn on Sinc, suddenly something on front end DSO pcb clicked and activate 8th order non-causal Chebysev?

and can any S or other brand DSO to do this test? and demonstrate here? cheers..

I repeatedly politely asked you to stop using my images and FALSELY claim what I said they mean and ascribe to them completely wrong conclusions you are spewing.

I explained what they are, how they were made and what they mean.

Your weird explanations are just that. Weird.

You have some grasp about math theory behind and some grasp of how sampling and DSO internals works. Just enough to arrogantly think you know it all. Better than all of the rest of us, including, among others, PhD who designed ADC for living, some people that designed scopes, and other very good engineers.....

Stop doing it and listen to explanation why are you wrong. You are behaving like flat earth people. There is no conspiracy, not everybody except you are stupid. It is simple. You are wrong.

Gf very nicely, eloquently and very concisely said what is going on.

That being said.
You seem to be confused by dot mode.
You are correct that dot mode can (in some cases) better show signal shape when signal is partially undersampled.

Do you understand why?

As an example I can show 10 MHz square wave signal, sampled at very low 25MS/s in dot mode. Signal is band limited with 200 MHz filter, so there are no frequencies (odd harmonics up to 19th in this case) above that. Simple edge trigger. Scope is my Siglent SDS6000.
Reference image is same signal, just sampled at 5GS/s.

Signal sampled with 25MS/s is perfectly reconstructed.
Can you explain why?

Let's start with that. Then we will proceed from there. Explain that first.

[Fungus]

so tell me Mr sharp trained eyes Fungus (and others who have sharp eyes or brain) which signal do you like? or think its true happening on circuit?

I still have no idea what you're trying to prove, but:

When I see a signal like that my first thought is "square wave". My second thought is to turn off some other channels to get a better picture.

At no point do I think that's what's happening on the circuit - a circuit cannot predict the future as that signal implies.

Nor do I think "sinc is evil, I wish I could turn it off".

Sinc artifacts are useful because "pre-ringing" is a visible warning that your sample rate is too low. They also give you a rough idea of how much "post-ringing" can be ignored.


My DHO800 even has a measurement called "preshoot" to measure the value.    IIRC your DS1054Z has it too.


This is the same signal with only CH1 enabled. Preshoot is gone so now I can start to believe what I see on screen.

[Mechatrommer]

Nor do I think "sinc is evil, I wish I could turn it off". It's actually helpful to see the "pre-ringing" because it gives you a rough idea of how much "post-ringing" can be ignored.

predicting actual event from false event? ok good for you, thats new. thats not how we work... but then i believe you've stated you want to avoid such artifact as much as you can by lowering scope's BW or in need of more sampling rate?... you can stick with 100MHz BW and be safe, cheers.

[Mechatrommer]

Gibbs effect successfully produced in (sinc_on.jpg), but when in dot mode (dot_on_run.jpg or dot_on_stop.jpg) Gibbs disappear, if you try imaginatively (if you dont want to try it on paper) to connect straight line between point to point in dot mode, you wont see any Gibbs effect. Gibbs only appear when Sinc interpolator is turned on in scope. do you understand what i meant before posting?

I guess that your signal is undersampled. Then It is expected to happen. Consider it an aliasing artifact. A properly bandwidth-limited signal is reconstructed almost exactly by sinc reconstruction, but if the signal fed into the ADC violates the sampling theorem, then perfect reconstruction from the samples fails and the reconstructed signal looks different from the original signal you fed into the ADC.

yes! so far so good!... i hope people like N03* and tau* get it as you do.... i believe some sharp people already understand this... now on to why i raised this issue? because i saw arguments that can be confusing to other people such as Fung* that lead to what i believe a poor decision or reasoning....

argument 1) DHO900 is not properly designed because at its rated BW (230MHz) can alias when all channels turned on (312.5MSps)
my answer:

1) its true, but why need to push DHO900 to see at its rated BW when you want 4 channels? you know its only 312.5MSps, lets just do < 125MHz circuit. if you want to include circuit integrity or rule out possible aliasing in digital circuit, lets go down further to 30-12Mbps digital system/clock. thats what the limit is (need proper knowledge though, newbies can be trapped, who cant?) why expect more than this? why want to push DHO900's metric to be on par with other more expensive scope? i say this as DHO804 owner that upgraded to DHO924... from cheapest properly designed, to save money from hack but improperly designed scope?

2) 230MHz BW limitation has its use instead of cutting it off in circuit further with analog LP filter to the "proper" nyquist BW of 125MHz will cripple some of our ability to debug circuit. we can use 1 or 2 channels "properly" on that BW... oscillations or unexpected circuit behavior will be hidden from users before the behavior even reach the ADC, if BW crippled in first place. in DHO900, we can see fuzzy or nasty display (aliasing) indicating something went wrong. in properly cutoff DSO, we are happy with sharp and thin traces thinking the circuit works, in fact its already clamped down by analog front end LP filter. when tested with hi end scope, or suddenly circuit behave erratically during production, we will scratching head as to why?

3) and this is relating to this topic... even if properly designed BW scope, there is no stopping it from experiencing aliasing, and worse, interpolated Gibbs effect. esp when you are probing much slower "square digital" frequency at lower sampling, maybe by accident lowering mem depth or time/div large enough. as my "properly designed" DS1054Z and N03*'s scope demonstrated. maybe the argument is only at DSO's max sampling rate, as answered in (1) why push it? and expect more than $400 scope? members in forum already give behavioral testing in tear down and comparison thread, and hence readers can make decision to buy. no need bashing to prove your point, we have some use for it (230MHz). but if argument is rigol didnt give proper explanation in spec sheet of DHO900 (about LA, sample rate and stuffs), then i agree.

argument 2) Rigol scope representation is not accurate, it hides details of overshoots and undershoots here and there
my answer:
its been lengthy posts from me try to prove that here, that even with your "shiny properly designed" scope can show false details that you think "accurate", and if Gibbs Effect is the sign of improperly designed scope as N03* tried to emulate and prove, then your scope is also not properly designed due to not cutting off frequencies above nyquist limit. how we know? because what you showed and claimed as accurate is in fact a Gibbs Effect that was never there in real circuit. and why Gibbs Effect is an indication of improperly designed scope? because N03* says so, because he demonstrated hard enough so.... here again, we go back to where we began.. cheers.

[tautech]

Let me again present 2 GHz DSO 10GSa/s in DOT mode.



BW and sampling rates are everything.

[Fungus]

Nor do I think "sinc is evil, I wish I could turn it off". It's actually helpful to see the "pre-ringing" because it gives you a rough idea of how much "post-ringing" can be ignored.

predicting actual event from false event? ok good for you, thats new. thats not how we work...

I'd say it's pretty darn close in your example image.

you can stick with 100MHz BW and be safe, cheers.

I've thought about it.

Right now I'm at 200Mhz because I figure 1.25GSa/sec. is plenty for two channels at that bandwidth.

Dropping down to two channels from three/four when there's any doubts about what's on screen doesn't seem too painful.

[Fungus]

argument 1) DHO900 is not properly designed because at its rated BW (230MHz) can alias when all channels turned on (312.5MSps)
my answer:

1) its true, but why need to push DHO900 to see at its rated BW when you want 4 channels? you know its only 312.5MSps, lets just do < 125MHz circuit.

I agree that you shouldn't push it but limiting the circuit isn't practical when you have pulses.

The practical solution is to monitor the preshoot and turn off some channels when you have a lot of it and you need a better view of the circuit.

[Mechatrommer]

The practical solution is to monitor the preshoot and turn off some channels when you have a lot of it and you need a better view of the circuit.

if i got preshoot, i'll turn off Sinc, if there is significant difference, i know its artifact, but if not, then i can guess the signal is real, or either undersampled, here we need to think carefully and if needed rig up experimental setup or simply, go find 10GSps scope. and turning off channels in case of suspicious is what exactly we do.. make sure there's written 1GSps too on screen. the keyword is, Know Thy Tools... 

Let me again present 2 GHz DSO 10GSa/s in DOT mode.
[img]
BW and sampling rates are everything.

good! thats "more accurate" but thats not apple to apple anymore with DHO800/900... cheers.

[Fungus]

good! thats "more accurate"

This still needs explaining though:

[Mechatrommer]

good! thats "more accurate"

This still needs explaining though:
[img]

thats where turning Sinc off, or plot in dot mode, and dial down to proper timescale can give extra clue as to why. circuit debug is not simply looking at one screenshot and be happy. if ringing is same freq or more as sampled dots, thats probably interpolation artifact. but if one cycle of ringing sampled by 10 points or more and you see almost straight lines between dots then that probably real.

[Fungus]

good! thats "more accurate"

This still needs explaining though:
[img]

thats where turning Sinc off, or plot in dot mode

That is in dot mode...

(according to tautech)

[Mechatrommer]

good! thats "more accurate"

This still needs explaining though:
[img]

thats where turning Sinc off, or plot in dot mode

That is in dot mode...
(according to tautech)

then that probably causal... a real signal. who knows whats inside semiconductors IC? that one is not as fancy as other examples presented before. someone mentioned high order "physical/analog/real" filter that can cause apparently non-causal ringing, but i have not be presented with one real life example, and i dont know if the scope has it. or what it does internally in DSP. probing with other scopes, compare and peer reviews may help.

[Mechatrommer]

here talking about improperly designed scope? 6GHz BW SDA6000 @ 10GSps, what? (bought used probably phased out from company and out of cal) feel free to pick which one to believe, i would say... close enough... if you still got concerned, really you need more BW and sample rate, and a new one!

[tautech]

good! thats "more accurate"

This still needs explaining though:
[img]

thats where turning Sinc off, or plot in dot mode

That is in dot mode...

(according to tautech)

No.

The scope menu clearly shows the scope is in Dot mode.
However this was a running capture, not one from Stop mode where many fewer dots get shown.
Yet, even with a capture we can toggle to Vector or Sinc etc modes.
There's good reason why we show full menus so to accurately display scope settings.

[Mechatrommer]

However this was a running capture, not one from Stop mode where many fewer dots get shown.

analyzing those fewer dots may reveals whether the scope showing sampled data only, or sampled data along with interpolated data, like Lecroy snapshot is combination of vector and dots, but the dots are interpolated data.

since at it, Rigol DHO800/900 cannot turn off Sinc, i searched last night.. this is point on them (not to forget the other  points to be fair, but thats OT, Fungus starts to cringe
i suspect ) so i have to download and analyze data in PC as screenshot shows... lets analyze this "improperly designed" scope..

1.jpg = the dso respond to a "pulse" (a reality approximation to a theoretical "step") here we dont see Gibbs, only under-damped front end respond
2.jpg = reducing sample rate by enabling all channels, bingo! Gibbs Phenomenon. newbies need to be educated to avoid this delusion. or even long time users can usually forgot the fact. if not careful
3.jpg = increasing time scale to see how Gibbs represent differently
4.jpg = increasing time scale more, the "imaginary" part of Gibbs disappear, maybe got folded back (aliased) during interpolation process?
5.jpg = this is where downloadable data to PC comes to rescue. vector plot without fancy Sinc shows how the signal was sampled. no Gibbs! this what i tried to explain.
6.jpg = look closer at sampled data vectored

with PC data available, one can do further fancy interpolation (or countless amount of analysis types) if needed, but i'm not a fancy guy, so vector plot is enough for me. i'll do interpolation and judgement in my head myself. imagine even if raw/sampled data pulled from scope is the one with interpolation/Gibbs, user will not be able to further analysis correctly, such as when plotting our own FFT (7.jpg.. there's bug or misconception i know Fnd'tl should be the 1st Hm ) imagine a process... sample data -> fft -> sw bw limiting -> ifft -> pull data to PC -> fft -> display.... the resulting end FFT will looks funny i suspect.

i've been searching for siglent programmability and sw/program example in pc to no available. since ds1000e or ds1000z i think. but recently i see siglent is following the path of using ni-visa driver as its API, now its a good step imho for me long time using ni-visa in rigol. lets hope there will be good sw to post process raw data from siglent's scope. as for now, i got cheap dho800/900 to play with, and its 12bits is the main reason i bite the bullet getting it, to further push boundary down the noise floor of my FFT. one funny thing i realized is i eventually hit the limit of my fft library memory points. it cannot do more than 25Mpts FFT analysis, maybe Windows is limiting memory quota for my program/library thread. so i cannot utilize fully the 50Mpts DHO900 memory. even more so for siglent with much larger memory.

since its the lowest price and already familiar with the workflow thats why DHO800 is the chosen one. siglent? maybe later when i'm rich enough to spend money for minor minor things. back to my sw, i was planning to add LA/serial decoder (from long memory data) functionality 7 years ago, which the main drawback in DS1000Z toyish serial decoding function. but then i lost enthusiasm during development and leaved my VisaDSO as it be. since i dont use serial decoder much, and i already have cloned salae 8 channels if i really need the job done. if minor upgrade/fix thats not requires very long time to do, then i come back to VisaDSO.. like few weeks ago.

with diy PC SW, there is no limit to where your imagination can go... (ahaks the marketing tone BS)... fwiw.

[tautech]

However this was a running capture, not one from Stop mode where many fewer dots get shown.

analyzing those fewer dots may reveals whether the scope showing sampled data only, or sampled data along with interpolated data, like Lecroy snapshot is combination of vector and dots, but the dots are interpolated data.

Want some special screenshot ?
Just ask.