EEVblog #601 - Why Digital Oscilloscopes Appear Noisy

[EEVblog]

Why do digital oscilloscopes appear noisier than traditional analog oscilloscopes?
Dave busts the myth that digital scopes are noiser than analog scopes, and demonstrates what inherent advantages digital scopes can have over analog scopes in terms of true waveform capture. And also why your analog scope may be hiding important signal detail from you.
Demonstrations of how memory depth, analog bandwidth, averaging, and intensity graded displays can all effect the signal detail you see on your digital oscilloscope.
And how long exposure camera shots on analog oscilloscopes can reveal detail you can't see with your eyes.

Demonstrations are done on the new Tektronix MDO3000, the Rigol DS1052E, the Tektronix TDS220, and Tektronix 2225 analog oscilloscope.
Previous video on common mode noise measurement:

[BravoV]

Noob question, so how do you tell if the noise is came naturally like your example there vs noise generated by crap/inferior scope front end or it's ADC ?

[c4757p]

Noob question, so how do you tell if the noise is came naturally like your example there vs noise generated by crap/inferior scope front end or it's ADC ?

Coin toss?

[EEVblog]

Noob question, so how do you tell if the noise is came naturally like your example there vs noise generated by crap/inferior scope front end or it's ADC ?

No input = noise from scope front end/sampling system.
When viewing a signal, you'd need to know how much your scope contributes at the given settings.

[electronics man]

I Alwase knew digital scopes were better than Analog ones, but don't digital scopes produce a lot more crap than analog ones

[valentinc]

       Dave, you are right about all you said in the video, BUT there is an inherent problem with digital scopes, that is not on an analog and that is: almost always the vertical resolution of the LCD screen is higher than the ADC resolution (8 bits, 256 samples)... And those pixels going all over the place are exactly from that: the vertical resolution is larger than 256 pixels, and there is some "noise" shown on the screen that there isn't present there in reality...

       I've seen some USB scopes that have 10 bit ADCs, but that is rare...

[Tek_TDS220]

Informative and entertaining as always.  I agree with Valentinc that analog scopes have higher resolution on all axes: x, y, and z. 

It hurts to hear you describe the TDS220 as 'ancient' - I still use one every day...

[madires]

Actually you have to take the vertical screen resolution combined with the quantization noise also into account for DSOs. If the signal wiggles between two quantization steps of the ADC you could get up to 4 steps noise (in a simplified case). Now, multiply that with the ratio of the screen resolution to the ADC resolution. Take a small 800x600 screen for example. Substract some space for menus and stuff and you got about 500 pixels vertical resolution for the signal. The ADC is 8 bit. That makes 2 pixels per bit or 8 pixels for 4 bits. That way the noise looks worse than it is. On a CRO you'll see a smooth signal, because you won't notice such a small change in the signal.

[Rigby]

Dave, I absolutely love it when you tell everyone they're wrong, then you go and prove it.  I absolutely love it.

It's always the things everyone "knows" to be true that are false, because no one ever challenges them.  No one challenges them because they're "known."

Serial contrarians like yourself (and myself, though I'm not as skilled) don't get a lot of love when we prove the opposite of common knowledge, but you do it anyway.

I love it.

Thank you.  Bravo.

[GK]

If an analogue oscilloscope "appears" less noisy then, for all sakes and purposes, it is less noisy. I guess it all boils down to how anal you want to be over the use of the word noise in this context.

I often find myself firing up an analogue scope when there are details to a waveform that are just too fine to properly discern on a DSO. In such instances the "ability" of the analogue scope to hide what Dave refers to as "uncorrelated noise" in the video is actually a blessing, not a negative. The effective resolution of a good analogue CRT display is less than the actual width of the trace. Changing the DSO's memory depth setting only works to a degree and averaging is completely useless on any signal that isn't completely periodic and stable. Hitting the BW limit button is also no help if the waveforms you are viewing are higher in frequency than the BW limit!

It wouldn't be too hard to set up some waveforms for demonstration purposes.

[EEVblog]

       Dave, you are right about all you said in the video, BUT there is an inherent problem with digital scopes, that is not on an analog and that is: almost always the vertical resolution of the LCD screen is higher than the ADC resolution (8 bits, 256 samples)...

Not on many of the low end scopes. The Rigol for example is 320x240 for the entire window.

[EEVblog]

If an analogue oscilloscope "appears" less noisy then, for all sakes and purposes, it is less noisy.

So can be a digital scope if you lower the memory depth, turn on hi-res mode, and limit the bandwidth.

In such instances the "ability" of the analogue scope to hide what Dave refers to as "uncorrelated noise" in the video is actually a blessing, not a negative.

Yes, of course, it can be useful. But it can also be a hindrance like I showed in a previous video, and that is the point I made last time and tried to make again here.

[EEVblog]

Dave, I absolutely love it when you tell everyone they're wrong, then you go and prove it.  I absolutely love it.

I should clarify though that I'm not saying analog scopes are crap and useless, and that all digital scopes are as good as an analog, and hence everyone is plain wrong. I just wanted to point out that there is much more to the story than most people realise.

[EEVblog]

It hurts to hear you describe the TDS220 as 'ancient' - I still use one every day...

It's coming up to 20 years old. That's older than a lot of people on this forum

[GK]

If an analogue oscilloscope "appears" less noisy then, for all sakes and purposes, it is less noisy.

So can be a digital scope if you lower the memory depth, turn on hi-res mode, and limit the bandwidth.

To a degree. Such settings still don't make the DSO's that I either own or use at work effectively less "noisy" in the display or higher in perceived resolution under all circumstances than my best analogue scopes. That's just my practical experience. DSO's are a Lecroy LT364L Waverunner, Lecroy WaveJet portable and a Rigol 1202CA.
 

[Ketturi]

I still think that old high-end analog scopes have much lower noise floor compared to cheapest DSOs. Scope front ends are still as critical as before and you can't just make them dirt cheap. I can't see how that rigol could be better than equipment sold at many times higher price.

[Tek_TDS220]

Ketturi, I guess your point is that for some applications, it might be better to buy a used analog scope vs a new digital scope due to a lower 'noise floor'? 

On the surface, this makes some sense.  When you need to digitize a signal at 1 GS/sec and display it, you are going to make a lot of RF noise in the same box.  The more expensive digital scopes have better isolation between the digital circuitry and the analog front end.  However, consider the cheapest Rigol DS2000 (~$1300 US) with a 500 uV/div range.  They must be doing a good job of isolation.  Also, digital scopes have a lot of other advantages.

[pickle9000]

Knowing your tools is very important. Every brand type will have it's high points and issues. What's critical to to know they exist.

Certain tools just will not have the functionality required. Again you need to know your tools and the limitations. 

[max_torque]

What might have been interesting, and shown off the ability of the expensive digital scope, would have been to generate a 1khz (say) 500Mvpkp sine wave signal, and then "swamp it" under say 1Vpkp white noise, and use the display graduation adjustment to show the base signal!

[mike1305]

Food for thought. A video that explains how update rate can also effect the visual appearance of noise. Basic statistics really (since noise is Gaussian), but nice to see live.

[Tek_TDS220]

Nice video mike1305, but I was worried the entire time that he was going to knock over the expensive Agilent scope onto the floor. 

How do you know the noise is Gaussian?  I guess it doesn't matter as long as it is not correlated with the sweep rate (as Dave pointed out).

[David Hess]

Even with equivalent input referred noise, the quantization noise of a DSO limits performance.

I ran across this a couple years ago when I modified one of my function generators to produce a fast transition sync pulse output.  One thing I noticed during testing was a tiny little "blip" getting into the main function generator output which coincided with the sync output and probably had to do with the original design not using a single point ground.

The blip was plainly visible on all of my analog oscilloscopes and none of my DSOs even with averaging with one exception;  it was plainly visible on my ancient 7854 which has a 10 bit digitizer and display to match.  It was also plainly visible on my high bandwidth analog sampling oscilloscope.  High resolution mode on a modern DSO should have revealed it though assuming the display resolution was high enough.

[BravoV]

Noob question, so how do you tell if the noise is came naturally like your example there vs noise generated by crap/inferior scope front end or it's ADC ?

No input = noise from scope front end/sampling system.
When viewing a signal, you'd need to know how much your scope contributes at the given settings.

Great ! Just want to hear this cause worry after watching this video, many "young players" will just blindly shout out loud & proudly claim that their One Hung Low digital scope is "DEFINITELY NOT" noisier than a high end analog scope that probably has a cherry picked low noise components at it's front end, just because Dave said so in this video without reading this kind of _{small foot print note}. 

[Noise Floor]

I've had a conversation on this topic at least once a year for the past 5 years.  Thank you for making a video I can point people to.       

[99tito99]

Hi Dave:  As expected (and so you should never disappoint) a great primer on oscilloscope noise (or should I say data resolution).  I think you do a good job by providing a wide swath (but not noisy) when describing these type of principles and I always find them applicability.  Although I don’t own a digital, I do have an analog (TEK 2213A), and so, I now see what I could not see.  Cheers, Mark

[leppie]

It hurts to hear you describe the TDS220 as 'ancient' - I still use one every day...

I have TDS340 with a TDS360 mainboard. Dave will probably call this prehistoric!

But I am a notch under 40, so I am probably "qualified" to use one 

[waspinator]

the closeup of the $10K Tektronix 3000 screen is really disappointing. Looks just as bad as the cheap Rigol. An HDPI+ screen is probably less than $100, why haven't they bothered upgrading that?

[Rigby]

the closeup of the $10K Tektronix 3000 screen is really disappointing. Looks just as bad as the cheap Rigol. An HDPI+ screen is probably less than $100, why haven't they bothered upgrading that?

they optimize for display speed over display quality, and rightly so.  Oscilloscopes are not meant for precise measurement, they're meant to be used to observe the shape of a waveform.

The display included lets one easily see the shape of the waveform.

[senso]

It will always look "bad", its only a 8 bits ADC.

[Legion]

For the Rigol 2000 series is the intensity grading option controlled by the "Waveform Intensity" or the "Persistence Time" setting? I'd check it myself but mine is in for warranty repair.

[KedasProbe]

Like shown in the video it's just a matter of representing the (noisy/extra) data.
Like this image below: (the gray or colour levels on the hameg are limited though.)


It can also help if you select to show dots only so these isn't a long line drawn.

edit: Notice that hameg decided to draw the grid on top of the noisy part. (blue part)

[M0BSW]

 I will no doubt buy a digital Scope, however, nothing smells as good as my  Telequipment 61A or my New scope a Tektronix 465B,beautiful smell when there warmed up, no reason given accept I just love the old birds.

[Hydrawerk]

Have you seen this old thread? http://www.home.agilent.com/owc_discussions/thread.jspa?threadID=17549&start=90&tstart=0

[David Hess]

I thought of this discussion thread where I linked a video and included a pair of Tektronix articles which discuss using analog oscilloscopes for noise measurement:

https://www.eevblog.com/forum/chat/will-your-scope-lie/msg420557/#msg420557

A good DSO can make this sort of noise measurement directly but not all of them can do it accurately.

[R_G_B_]

Granny bashing the TDS210 lol still a handy scope better than no scope I guess.

[miguelvp]

It will always look "bad", its only a 8 bits ADC.

Vertical resolution 8 bits (11 bits with Hi Res)

I know 11 is not that much but in reality it's 3 orders of magnitude bigger.

And don't forget the 2.5 Gsps, show me a DAC that's 0.4 ns per sample, say you get the 500 MHz scope that's 2ns per point with 5 samples to determine that point.

As others mention, the purpose is not precision but to be able to see the waves so say you get this:
DAC5670IGDJ (14bit, 2.4 GSPS ) That's 0.417 ns per sample so you will have a 480 MHz scope with more vertical resolution, but it will cost way more because you have to add support logic for the extra 6bits or extra 3 bits for the high resolution.

But what does the vertical resolution give you? more precise measurements? sure but it comes at an exponential cost. Each bit doubles the resolution and math functions etc, so 2^6 =64 times the detail that you have to support, that's going to cost quite a bit of money, power and resources, and for what?

Then consider that because of the extra resolution your noise gets bigger so you have to deal with that as well.

Edit: by bigger I mean, it renders the lower end bits useless.

Edit: think about this, how come we don't have 10 based digital chips, heck we can measure 0.1 volt increments right? so 0-1V at 0.1V and we can have decimal computers. But it's really not that simple.

Edit: also consider the setting time of that fast 14 bit chip at 3.5ns, so you couldn't do better than 285 MHz.

Argh, edit again: The thing is that oscilloscopes are tools for the time domain. You want to use them to see what is happening in the smallest time possible for high speed communications or what have you.

I mean what is more important anyways, being able to see your signals changing states or to measure the voltages missing the signals?

[VK3DRB]

(In the following, "CRO" is Australian slang for oscilloscope. It means "Cathode Ray Oscilloscope" but these days it means any oscilloscope - cathode ray tube or LCD.)

I came across a bloke working in the Australian army at an Agilent test equipment show. He said to me "Digital CRO's are for pooftas. Anyone who knows what they are doing will only use an analogue CRO. I only use analogue because I know a lot more than any engineer." He was a bad advertisement for the military. (I have come across others just as stupid and arrogant.)

Sure, analogue CRO's can have advantages. But these days, the digital CRO is a FAR more a useful instrument to me. I have one that is a combined digital CRO and logic analyser. Try to get that in an analogue CRO.

 

[rigrunner]

Very informative video 

Bearing in mind that i'm a hobbyist and not a professional. I have basic scopes - Hantek hacked to 200Mhz and an old Iwatsu 5710.
Where i find the digital scope lets me down is lag.

If i'm looking at an RF waveform and tuning an RF stage the lag on the DSO gets annoying and i switch to the analogue every time,

I'd be interested to see how the lag has changed through the different generations of DSO.

[Rigby]

"lag" is a very fuzzy, often misused term.

Are you talking about latency between signal acquisition & signal display?  Are you talking about the screen update rate?  Are you talking about control responsiveness?  Can you be more specific?

[miguelvp]

Update lag is really not that important either, I mean if it was between two channels then yeah, you don't want that. But your eyes can't sync with an external device and the output of the scope so that doesn't matter.

The important thing is that when you trigger a signal you can inspect all those 10 million points in 1/50th of a second knowing that it didn't skip a beat. And I don't know if you get all 10 M at the higher BW, but just to illustrate.

[rigrunner]

> Can you be more specific?

I don't know exactly where the perceived lag lies.  I suspect it is latency between signal acquisition & signal display due to my DSO being low end, but i can't be certain.

>Update lag is really not that important either,

It is to me when i'm tuning an RF filter stage for max peak.

[miguelvp]

Average human reaction time (and trained at that) is about 200ms at best. Say that from your optics to your brain it's only 10ms and the 190ms is from the brain to the muscle.

10ms, that means that you can see at 100fps and notice if you skip a frame and that will be a very highly trained video game player, like in a fighting game that they can tell if a game glitches at 60fps. (16.666 ms per frame)

So the perceive lag is like the audiophoolery thread for sound. The thing is, that yeah, it's important that the scope is capable to display the smallest runt possible and your persistence of vision will detect it. Not because you can see that fast but because the display doesn't update so fast that you will miss it.

Edit: like subliminal messages, you are not aware of them even if your eye/brain detects them.

[rigrunner]

The perceived lag on the DSO causes me to overshoot when tuning for peaks. The delay i see is way more than 10ms, closer to 250ms i would think. I have to wait for the DSO to catch up and then retune. I switch to the analogue scope from that point onwards in the tuning process.
It's the only thing that keeps me holding on to my analogue scope. The DSO works fine for everything else.
 

[miguelvp]

Anything under 30 to 50ms as a runt might not be perceived, the 16.66ms (60Hz) is that you can notice a frame being gone from a sequential play, but showing a single frame that doesn't have anything to do with the normal sequence (therefore I use the runt as a sample) must be displayed for longer than 50ms for you to be aware of it.

Not sure what scope manufacturers use for displaying runts, but if the scope frame rate is higher than 30Hz you might not see the runt. Your eye/brain might detect it but your conscious mind wont.

Edit: Of course on CROs, the phosphor persistence will make sure you see it. On digital displays the faster they can really switch off the worse it would be for you to actually see the runt. But as a manufacturer, they can decide to display it longer so you can see it.

[rigrunner]

Runts and display persistence don't enter into what i'm observing.  It's a distinct delay between my hand turning a tuning core and what is displayed on the DSO screen.  The analogue scope appears to be much more real time than the DSO.

As i mentioned earlier - i'm happy to attribute this to my DSO being peasant specification but i would be interested in seeing how it has evolved though the generations of DSO.

[miguelvp]

So because of what I said before, maybe digital scope manufacturers choose their display so that a pixel will ghost the value for at least 50 or more milliseconds so that you can see it. I can see (no pun intended) the advantage of having displays that by nature lag their output for this purpose.

[miguelvp]

Runts and display persistence don't enter into what i'm observing.  It's a distinct delay between my hand turning a tuning core and what is displayed on the DSO screen.  The analogue scope appears to be much more real time than the DSO.

As i mentioned earlier - i'm happy to attribute this to my DSO being peasant specification but i would be interested in seeing how it has evolved though the generations of DSO.

There is no storage on analogue scopes, that's why they are more responsive to inputs. Analogs have to do computations before showing the results. But some high end ones can update that pretty quick, but since it's not analogue they have an inherit delay from input to output.

And by storage, even if they do have storage, they display the output instantly, the storage happens later.

[BravoV]

Miguelvp, you sound very experienced in this scope thingy, just curious, what scopes you've used In the past or own now ?

[miguelvp]

Miguelvp, you sound very experienced in this scope thingy, just curious, what scopes you've used In the past or own now ?

Not experienced on scopes at all, just common sense of what I would do to make sure a scope will work as intended.

I've been a game developer for many years so I know about perception and human interface. Did some racing and drag racing games (well simulations) so I know about the human aspect of reaction time. Just many years of experience in user interaction with computers. Before that I was on medical imaging and had to deal with a lot of custom hardware and research, meaning programming OS9 based 68000 custom hardware to deal with a lot of data back in the 90's. Again dealing with perception and how radiologists interact with systems.

Now I work for a AAA game studio and I'm pretty sure most of you have played games that I've worked on, but my engineering side always makes me dig deeper into how things work so I've always been keeping up with technology and what is possible.

As far as electronics I'm a veteran noob because I don't deal with electronics directly but I do know what they are capable of doing. In my work you either keep up or well work, making flash games

[kizzap]

This is possibly a stupid question, but wouldn't a lot of that "noise" on the DSOs come from the INL and DNL of the ADCs? 

[David Hess]

Runts and display persistence don't enter into what i'm observing.  It's a distinct delay between my hand turning a tuning core and what is displayed on the DSO screen.  The analogue scope appears to be much more real time than the DSO.

As i mentioned earlier - i'm happy to attribute this to my DSO being peasant specification but i would be interested in seeing how it has evolved though the generations of DSO.

There is no storage on analogue scopes, that's why they are more responsive to inputs. Analogs have to do computations before showing the results. But some high end ones can update that pretty quick, but since it's not analogue they have an inherit delay from input to output.

And by storage, even if they do have storage, they display the output instantly, the storage happens later.

My Tektronix 7834 would like a word with you.
 
There *are* (or were) analog CRT based storage oscilloscopes and they had the same responsiveness as the familiar analog CRT oscilloscopes.  They coexisted with the early DSOs up to about 1991 but DSOs were more economical and displaced them quickly.

[miguelvp]

Oh, and to answer your other question, I've been using a second hand tektronix 7603 for the last 15+ years, I still have it but I recently purchased a Rigol DS2072, yeah nothing fancy, but that doesn't mean that I don't know what is out there (that I can afford but my wife won't let me get it).

My dad was an electronic engineer and I used to read all his manuals (he used to work for GTE) as a kid, I did turn to software because in the mid 80's programming was king (and still is) that doesn't mean I didn't keep up with my other side. I'm a strong believer that if you don't understand the electronics behind of what you are programming you are not a good programmer at all.

[miguelvp]

Runts and display persistence don't enter into what i'm observing.  It's a distinct delay between my hand turning a tuning core and what is displayed on the DSO screen.  The analogue scope appears to be much more real time than the DSO.

As i mentioned earlier - i'm happy to attribute this to my DSO being peasant specification but i would be interested in seeing how it has evolved though the generations of DSO.

There is no storage on analogue scopes, that's why they are more responsive to inputs. Analogs have to do computations before showing the results. But some high end ones can update that pretty quick, but since it's not analogue they have an inherit delay from input to output.

And by storage, even if they do have storage, they display the output instantly, the storage happens later.

My Tektronix 7834 would like a word with you.
 
There *are* (or were) analog CRT based storage oscilloscopes and they had the same responsiveness as the familiar analog CRT oscilloscopes.  They coexisted with the early DSOs up to about 1991 but DSOs were more economical and displaced them quickly.

And that's why i said, and you quoted as well:

And by storage, even if they do have storage, they display the output instantly, the storage happens later.

meaning that they do store it without altering their analogue response, maybe I wasn't clear on that, sorry.

Edit: but yeah, I did say that analoges have to do computations, but I really meant digital scopes have to do computations, sorry again maybe the wine is affecting my typing

[David Hess]

Runts and display persistence don't enter into what i'm observing.  It's a distinct delay between my hand turning a tuning core and what is displayed on the DSO screen.  The analogue scope appears to be much more real time than the DSO.

As i mentioned earlier - i'm happy to attribute this to my DSO being peasant specification but i would be interested in seeing how it has evolved though the generations of DSO.

This problem definitely exists even on some high end DSOs.  Waveform acquisition rates have gone up which is good but the latency from acquisition to display and from user interface to display can be very noticeable.  On the interface side, the modern low and high end DSOs I have evaluated are little or no better than my ancient real time DSOs except in acquisition rate.

DSOs will never achieve the display and interface latency that analog vector CRT oscilloscopes had but for all practical purposes they should get close enough and they can certainly perform better than they do.  Latency includes:

1. Waveform record fill time - use a short record length to minimize this.  Early DPO style DSOs wrote directly to the display memory and effectively processed acquisitions in real time so they would not suffer from this.  I do not know if any modern ones operate this way.
2. Trigger rearm time - some DSOs take a long time to rearm their trigger.
3. Processing time - the waveform record has to be transferred and processed for display.
4. User interface latency - changing the operating parameters may have considerable latency.  This seems to have gotten a lot worse with oscilloscopes that use embedded desktop operating systems.
5. The display refresh time - Not much can be done about this but 50 or 60 Hz should be plenty fast.  Some LCD displays have a latency of more than 1 frame increasing their latency beyond what the refresh rate implies but I do not know that they are used in embedded systems.

[miguelvp]

But at the end of the day what do you prefer?

Acquisition rate or Display rate.
I will go for the former.

I rather capture the waveform and see all the detail than see it in (almost) real time.

Also, can you blink an led fast enough that you can't see it? a high speed camera will see it. heck, there are cameras that can track light moving in short distances nowadays, actually pretty cool tech that you can see in non real time how light travels (yeah, just like you expect as a wave more than as a particle)

but speaking of dualities:

Programmers we are way behind of what is possible hardware wise. Let me rephrase that: programming languages are way behind of what is possible, I know that. There have been so many tech advances and the only thing holding back is really the programming paradigms side of things. But I do blame "us" programmers for that because most of us are creatures of comfort and don't want to learn a new way of thinking.

But I think it's getting better even if we are still lagging by at least 9 years behind what is achievable. I also blame managers that want to keep things "safe" so don't rock the boat.

I know the whole disrupting technology terminology has been used way too much, but what it really means is that someone finally had the time to mix together common sense technologies that no one wanted to spend the effort doing before. Even if the amount of effort is minimal nowadays.

And it's disruptive because they wake us (programmers) up from our nap.

[David Hess]

Runts and display persistence don't enter into what i'm observing.  It's a distinct delay between my hand turning a tuning core and what is displayed on the DSO screen.  The analogue scope appears to be much more real time than the DSO.

As i mentioned earlier - i'm happy to attribute this to my DSO being peasant specification but i would be interested in seeing how it has evolved though the generations of DSO.

There is no storage on analogue scopes, that's why they are more responsive to inputs. Analogs have to do computations before showing the results. But some high end ones can update that pretty quick, but since it's not analogue they have an inherit delay from input to output.

And by storage, even if they do have storage, they display the output instantly, the storage happens later.

My Tektronix 7834 would like a word with you.
 
There *are* (or were) analog CRT based storage oscilloscopes and they had the same responsiveness as the familiar analog CRT oscilloscopes.  They coexisted with the early DSOs up to about 1991 but DSOs were more economical and displaced them quickly.

And that's why i said, and you quoted as well:

And by storage, even if they do have storage, they display the output instantly, the storage happens later.

meaning that they do store it without altering their analogue response, maybe I wasn't clear on that, sorry.

Edit: but yeah, I did say that analoges have to do computations, but I really meant digital scopes have to do computations, sorry again maybe the wine is affecting my typing

Wine will do that.   I have run across people who had no idea analog storage oscilloscopes ever existed though.

Where it gets weird are the fast transfer mesh storage oscilloscopes (the 7834 supports this) where storage occurs before display.  This is done to raise the writing rate but in practice, variable persistence mode is usually fast enough and a lot easier to use.  The persistence is typically adjustable over a range of 10s of milliseconds to minutes.  For oscilloscopes that support bistable storage mode which is much slower, persistence is hours to days and the display may even be preserved when the oscilloscope is turned off but bistable storage does not have graded intensity like variable persistence mode.

For myself, I elected to refurbish a 7834 for my variable persistence needs and get an old but maintainable DSO for everything else instead of paying for a modern DSO which supports both.  Just duplicating the capability of a single old differential comparator input would cost more than a new DSO.

[David Hess]

But at the end of the day what do you prefer?

Acquisition rate or Display rate.
I will go for the former.

I rather capture the waveform and see all the detail than see it in (almost) real time.

Why not both?  There is no conflict between the two unless you use segmented memory which trades one for the other.  Latency only needs to be "video" speed anyway and most DSOs do not even achieve that.

Also, can you blink an led fast enough that you can't see it? a high speed camera will see it. heck, there are cameras that can track light moving in short distances nowadays, actually pretty cool tech that you can see in non real time how light travels (yeah, just like you expect as a wave more than as a particle)

The Tektronix 7104 oscilloscope can sweep its screen comfortably faster than the speed of light:

http://www.oregonlive.com/silicon-forest/index.ssf/2011/09/a_tektronix_oscilloscope_that.html

[miguelvp]

maybe it's the same tech they use to capture light faster than the speed of light.



But my point is, that you want the image to persist so you can detect it with your eye. That doesn't contradict anything, you can capture faster than light, but at least display slower that you can see it.


Edit: let me rephrase what I said, not capturing faster than light, but capturing light at 1 millimeter travel per frame, or at a femtosecond interval. I do want a scope that can do femtoseconds, nah, it will take a lot of memory to get anything useful.

[David Hess]

Edit: let me rephrase what I said, not capturing faster than light, but capturing light at 1 millimeter travel per frame, or at a femtosecond interval. I do want a scope that can do femtoseconds, nah, it will take a lot of memory to get anything useful.

My fastest oscilloscope only gets down to 10 picoseconds per division.  I am so behind the times.

[miguelvp]

Edit: let me rephrase what I said, not capturing faster than light, but capturing light at 1 millimeter travel per frame, or at a femtosecond interval. I do want a scope that can do femtoseconds, nah, it will take a lot of memory to get anything useful.

My fastest oscilloscope only gets down to 10 picoseconds per division.  I am so behind the times.

a 200 MHz scope is 5 nanoseconds per division so your's is 2,000 times faster than that.
They are talking about 10,000 times faster than yours, or 20 million times faster than a 200MHz scope. In binary it's 24 orders of magnitude better than a 200MHz scope. And they are doing a whole 2D image not just one pixel.

[hikariuk]

My Tektronix 7834 would like a word with you.
 
There *are* (or were) analog CRT based storage oscilloscopes and they had the same responsiveness as the familiar analog CRT oscilloscopes.  They coexisted with the early DSOs up to about 1991 but DSOs were more economical and displaced them quickly.

As would my Tektronix 5115

[R_G_B_]

Use the FFT with a good adc resolution if you looking for hidden anomalies.
Scopes are good for time varying signals e.g checking if a waveform is there has the correct amplitude, time, phase, frequency.

Use the right tool for the job.

[waspinator]

the closeup of the $10K Tektronix 3000 screen is really disappointing. Looks just as bad as the cheap Rigol. An HDPI+ screen is probably less than $100, why haven't they bothered upgrading that?

they optimize for display speed over display quality, and rightly so.  Oscilloscopes are not meant for precise measurement, they're meant to be used to observe the shape of a waveform.

The display included lets one easily see the shape of the waveform.

at 10K they shouldn't need to compromise. not many would care if it was 10.1K, but I'm sure they would appreciate not looking at pixels.

[David Hess]

Edit: let me rephrase what I said, not capturing faster than light, but capturing light at 1 millimeter travel per frame, or at a femtosecond interval. I do want a scope that can do femtoseconds, nah, it will take a lot of memory to get anything useful.

My fastest oscilloscope only gets down to 10 picoseconds per division.  I am so behind the times.

a 200 MHz scope is 5 nanoseconds per division so your's is 2,000 times faster than that.
They are talking about 10,000 times faster than yours, or 20 million times faster than a 200MHz scope. In binary it's 24 orders of magnitude better than a 200MHz scope. And they are doing a whole 2D image not just one pixel.

To borrow a phrase from Jim Williams - I'm sorry, but 14GHz is the fastest scope in my house.

The first time I used my sampling scope as a TDR, I was amazed at what could be seen.  It literally can display propagation of light, albeit RF photons in a transmission line, down to millimeters limited in my case only by my signal source.

[David Hess]

the closeup of the $10K Tektronix 3000 screen is really disappointing. Looks just as bad as the cheap Rigol. An HDPI+ screen is probably less than $100, why haven't they bothered upgrading that?

they optimize for display speed over display quality, and rightly so.  Oscilloscopes are not meant for precise measurement, they're meant to be used to observe the shape of a waveform.

Doesn't the TDS3000 series, not to be confused with the THS3000 series which is completely different, use an anti-aliased and index graded display?

I remember a comment over on TekScopes@yahoogroups.com that the TDS3000 series was the lowest end of oscilloscopes from Tektronix worth getting in connection with its DPO functionality but I wonder if the less expensive DSO2000 series would be better.  I have considered both at one time or another.

[rf-design]

Very good demonstration Dave but I missed some background explaination on the board.

To my understanding the best demonstration of the visibility of the noise distribution is to show a gaussian distribution on the board and how the eye makes a threshold on the visibility of the more rare events that the voltage or trace is 2x times the noise RMS value away from the mean. This threshold depend a little bit on the persistance of the analog scope phospor.

There are analog scopes built in the past which use analog physical backprocessing of the electron trace with a microchannel plate. This plate have much higher vertical resolution than most DSOs and amplifies the electrons from the trace with a logarithmic function. Then the amplified electrons are displayed with phospor.

Do you know a DSOs which have a logarithmic grading function?

It will be fair to show the best analog art in terms of a MCP scope in comparison to DSOs.

To me DSOs manufactures have still a way to go because most of them does not give numbers to there products. I did not see a SNR over sensitivity and only some give me a ENOB at some point. Also the ADCs resolution seem to contribute to SNR more than the frontend. True matured ADCs are noise vs. power limited, not by resolution.

The scope vendors offer +3 bit resolution with averaging, but did not tell if it is true resolution or done with additional, not specified, dither noise.

The scopes did not lie but the vendors hide the critical informations about there product because they better know how fare they are away from matured products for professionals.

[David Hess]

To my understanding the best demonstration of the visibility of the noise distribution is to show a gaussian distribution on the board and how the eye makes a threshold on the visibility of the more rare events that the voltage or trace is 2x times the noise RMS value away from the mean. This threshold depend a little bit on the persistance of the analog scope phospor.

I have linked an example in this forum of doing this a couple of times.  This is one of those measurements which is esoteric but trivial on an analog oscilloscope while being difficult or impossible on a DSO unless it can calculate it for you.  In theory calculating the RMS value will work and that is how sampling voltmeters do it but somehow the DSOs I have tested return the wrong results.  A DSO which can generate a histogram and return the standard deviation should work better but that is one of those features used for market segmentation.

This is also one of those measurements where record length and aliasing are irrelevant.  Neither alters the standard distribution (or RMS value) of the signal so if it changes, the oscilloscope has failed by lying.  This includes oscilloscopes which have an anti-aliasing filter linked to their sample rate and is a good example of why anti-alias filtering has no place in a DSO.

Do you know a DSOs which have a logarithmic grading function?

It would be interesting to know which produce an index graded display accurate enough (or at least not flawed enough) to make a dual trace RMS noise measurement by mark I eyeball.  My informal test of a high end Tektronix oscilloscope failed this spectacularly although it should be an easy thing to get right unless the display sacrifices functionality to look pretty.

The scope vendors offer +3 bit resolution with averaging, but did not tell if it is true resolution or done with additional, not specified, dither noise.

The input referred noise is so high that dithering should not be needed.

The scopes did not lie but the vendors hide the critical information about there product because they better know how fare they are away from matured products for professionals.

The most common might be conflating vertical sensitivity settings with digital magnification which just throws away bits.  Even my ancient DSOs can do this but at least they do not lie about it.

[rf-design]

To my understanding the best demonstration of the visibility of the noise distribution is to show a gaussian distribution on the board and how the eye makes a threshold on the visibility of the more rare events that the voltage or trace is 2x times the noise RMS value away from the mean. This threshold depend a little bit on the persistance of the analog scope phospor.

I have linked an example in this forum of doing this a couple of times.  This is one of those measurements which is esoteric but trivial on an analog oscilloscope while being difficult or impossible on a DSO unless it can calculate it for you.  In theory calculating the RMS value will work and that is how sampling voltmeters do it but somehow the DSOs I have tested return the wrong results.  A DSO which can generate a histogram and return the standard deviation should work better but that is one of those features used for market segmentation.

This is also one of those measurements where record length and aliasing are irrelevant.  Neither alters the standard distribution (or RMS value) of the signal so if it changes, the oscilloscope has failed by lying.  This includes oscilloscopes which have an anti-aliasing filter linked to their sample rate and is a good example of why anti-alias filtering has no place in a DSO.

I imagine a gray level grading where a logarithmic function is applied to the distribution and the positive and negative single sigma values get a red line. I think I simply recall what I have seen in some user plotting matlab code over the years.

[Wim_L]

It would be interesting to know which produce an index graded display accurate enough (or at least not flawed enough) to make a dual trace RMS noise measurement by mark I eyeball.  My informal test of a high end Tektronix oscilloscope failed this spectacularly although it should be an easy thing to get right unless the display sacrifices functionality to look pretty.

I suspect almost all, perhaps all, will fail at that test.

In an analog scope, the two beams share a screen. In a DSO, each channel has its own memory where points are accumulated. The data is then added up into the display buffer in a separate colour for each trace. When traces overlap, the usual result is that the trace drawn last casts a shadow over the previous traces.

[miguelvp]

One thing no one mentioned is that the Rigol DS1052E that Dave used in the video is probably the noisier (because of the fan that is).

[David Hess]

It would be interesting to know which produce an index graded display accurate enough (or at least not flawed enough) to make a dual trace RMS noise measurement by mark I eyeball.  My informal test of a high end Tektronix oscilloscope failed this spectacularly although it should be an easy thing to get right unless the display sacrifices functionality to look pretty.

I suspect almost all, perhaps all, will fail at that test.

In an analog scope, the two beams share a screen. In a DSO, each channel has its own memory where points are accumulated. The data is then added up into the display buffer in a separate colour for each trace. When traces overlap, the usual result is that the trace drawn last casts a shadow over the previous traces.

I think the older monochrome CRT DSOs that I have used could have done this with or without variable persistence since not all of them supported it.

It is not as accurate but this test works on analog sampling oscilloscopes where the display relies on dot density instead of grading.

[David Hess]

To my understanding the best demonstration of the visibility of the noise distribution is to show a gaussian distribution on the board and how the eye makes a threshold on the visibility of the more rare events that the voltage or trace is 2x times the noise RMS value away from the mean. This threshold depend a little bit on the persistance of the analog scope phospor.

I have linked an example in this forum of doing this a couple of times.  This is one of those measurements which is esoteric but trivial on an analog oscilloscope while being difficult or impossible on a DSO unless it can calculate it for you.  In theory calculating the RMS value will work and that is how sampling voltmeters do it but somehow the DSOs I have tested return the wrong results.  A DSO which can generate a histogram and return the standard deviation should work better but that is one of those features used for market segmentation.

This is also one of those measurements where record length and aliasing are irrelevant.  Neither alters the standard distribution (or RMS value) of the signal so if it changes, the oscilloscope has failed by lying.  This includes oscilloscopes which have an anti-aliasing filter linked to their sample rate and is a good example of why anti-alias filtering has no place in a DSO.

I imagine a gray level grading where a logarithmic function is applied to the distribution and the positive and negative single sigma values get a red line. I think I simply recall what I have seen in some user plotting matlab code over the years.

Just adding a indicator for the standard distribution of the vertical signal at each horizontal point would be enough.  I am going to add that to the feature list of my imaginary DSO design.  At least then the false color display could be informative.

[katzohki]

I really appreciated this video, as I was one of the ones questioning how well my digital scope really was. Thank you for explaining it very well.

My Instek has other weird idiosyncrasies where if you zoom in really close on the noise (when paused) it looks like sin(x)/x pattern. Pretty sure it's a software glitch because it disappears when you change the timebase. Yikes!

[David Hess]

My Instek has other weird idiosyncrasies where if you zoom in really close on the noise (when paused) it looks like sin(x)/x pattern. Pretty sure it's a software glitch because it disappears when you change the timebase. Yikes!

Could it be aliasing with the horizontal resolution on the display side?

[Zbig]

My Instek has other weird idiosyncrasies where if you zoom in really close on the noise (when paused) it looks like sin(x)/x pattern. Pretty sure it's a software glitch because it disappears when you change the timebase. Yikes!

Well, no, it's intentional. It is using sine interpolation in order to present you a nice smooth waveform no matter how far apart the actual consecutive samples are. Most scopes also offer an option to connect the dots linearly or not to connect them at all.

[pickle9000]

My Instek has other weird idiosyncrasies where if you zoom in really close on the noise (when paused) it looks like sin(x)/x pattern. Pretty sure it's a software glitch because it disappears when you change the timebase. Yikes!

Turn on the dot mode, to look at the unprocessed points.

[ornea]

Since we are getting nitty gritty ... the top and bottom of the waveforms/line appear to be slightly different. At 7:15 the top seems has 1-2 pixel height  notches missing while the bottom appears to have 1-3 pixel spikes.  I would have expected that the bottom and top of the line to look similar. It is very subtle but curious if it is explainable.

Justin

[David Hess]

I now wonder if the problem katzohki describes and excessive apparent noise on DSOs may be related to aliasing produced by nonlinearity and jitter in the ADC which will be worse if interleaving is used.  I brought this up recently in another discussion thread which includes a link to an Agilent application note mentioning the issue:

https://www.eevblog.com/forum/testgear/rigol-ds1074z-oscillosope/msg465619/#msg465619

My hypothesis is that the newer DSOs which accumulate multiple waveform acquisition records for the display record to yield a high waveform update rate look noisier than they should because in effect the envelope of the signal is expanded by the interaction of aliasing with the sin(x)/x reconstruction.

This effect is very apparent on my old 500 MSample/second DSO in which the digitizers use interleaving.  Fast edges or clean sine waves which are significantly below the Nyquist frequency show the effect but I never thought much of it because I normally use equivalent time sampling mode raising the sample rate into the 10+ GSamples/second range where aliasing is insignificant.

It might be possible to check this on a DSO which interleaves when only one channel is used by enabling the other channels and watching for a change in the display of the original waveform.  Alternatively make the DSO display just one acquisition record at a time and watch for the "wobbulation".

I do not think the noise tests that Dave did would have revealed this if it was happening.

[vokars]

I honestly regret, being so late with my comment here  .

I have an analog oscilloscope that is really nothing special: A rebranded GW Instek GOS-630 (Voltcraft 630-2). And nevertheless I love it. Compared to the 8 Bit DSO's I have worked with (yes I intend to buy the RTB2000 10Bit scope  ) it not only seems to show less noise. It has "no" noise. Why?

The basic message of David's video is at 5:29: "The bottom line is: Any noise on the analog amplifier down here or on the signal you are feeding in, any noise which is uncorrelated to your sweep speed or your trigger it just randomly appears, it's not synchronized with the sweep then it's goining to appear quite dim or non existent, because it is not gonna show up all the time."


a) Of course any noise (which is per definition not correlated with anything especially the trigger, sweep, ..) on an analog scope is visualized as beam width due to the screen luminescence /phosphorescence of the screen. The luminescence/phoshorescence layer of the screen is basically a memory storing excitation energy resulting in the typical persistance of scope- or (former) TV-screens. Depending on the material of the luminescence layer it typically stores the excitation energy for several milliseconds.

Because also the beam intensity has an impact on the beam width I prefer to adust the intensity as low as possible. Every change in beam width with fixed intensity /beam focus is: noise. And vice versa: If there is noise, the beam width increases. This can also be verified on a digital scope by switching on the persistance mode.

b) To really compare the noise produced by the measurement device: Using 1mV/div would make more sense instead of using 1V/div for a comparison as in the video. And with 1mV/div the noise shows up, if it would be hidden in the beam width of the 1V/div setting. With my analog scope it does not (older analog scope models I have worked with and which have worse amplifiers show a broad noise floor on the beam without hiding it). No increase in beam width. No "hidden data". That is my definition of "no" noise (of course an approximation that makes sense for practical usage). Crosscheck: Apply a noisy signal and watch the increase in beam width. Yes: Dave does that in the video and I am really astonished that he cannot see the increase in beam width as I do  .

c) Theoretical viewpoint: 8 bit scopes produce additional noise by their quantization process. An 8 bit scope can never be better than 48dB dynamic range (without Hi Res). Averaging can be applied but reduces the noise also from the signal, which is what you want to see. Means: With equal amplifier noise a digital scope has always more noise per design than an analog scope. Fully in line with experience.