New Analog Scopes?

[Gunb]

If you're on a limited budget, a cheap DSO plus a cheap old (originally expensive) analogue scope will give you the most functionality for your money.
A DSO needs to have high update rate and an intensity-graded display to fully replace an analogue one - I think the Rigol 2000 is  the cheapest new option unless you get a  very good deal on a used Agilent 5/6/7000 or tek DPO

Exactly, I agree. Finally someone who understands what we're talking about. 

[nctnico]

Of course, and I wouldn't buy a new analog scope today. But that was not the issue, it's a fact that there are special purposes where a former
expensive analog scope might be better than cheap digital scope of today. Or do you know a new 500MHz scope for a moderate price of less then 1000,- EUROs
that you can use for ham radio?

You won't find a new analog or digital 500MHz scope for less than €1000,- so that argument is pretty moot. However there are lots of TDS500 and TDS700 series scopes on Ebay in that price range which have 4 channels and 500MHz bandwidth. If you are handy you can buy a cheaper defective one with the well known and documented problems and fix it.

Aha! I can't use the zoom?! Really? I mean, when using a scope with an ADC of 1GS/s at max I assume that you understand that the time between two samples is 1ns, right? So you're zoom can't resolve more than showing you samples in a distance of 1ns, what happens between is not displayed, but interpolated if not switched off. An analog scope using 1ns can show any value between because there's no sampling but a continous steering of the beam. Disadvantage: you can't store it. Think about it.

That's only true for single shot signals. Even cheap Chinese are able to determine the timing relation between the sampling clock and the trigger to a resolution of 25ps, which gives you a timing resolution of 25ps when acquiring long enough.

No it's generally valid even for cheap Chinese scopes and it doesn't depend on single shot.

- ADC with a sample rate of 1GS/s enables 1ns time interval between two samples - as given in my example
- ADC with a sample rate of 4GS/s enables 25ps time interval between two samples - as you mentioned.

But it doesn't influence the fact that a screen of limited pixels for example 800 pixels on the abscissa cannot display more than 800 pixels at once. Either the scope displays only each - lets assume - 10th sample of the sample memory on screen by transferring it to the display buffer or you have to zoom in until you see the samples in the distance of your 25ps, in this case your timebase is very small.

You obviously never heard of peak detect which every decent DSO has. You really should learn how to use a DSO properly because you sound like someone trying to argue a vinyl record is better than a compact disk.

[Gunb]

Aha, so so. OK, after you let off the steam:

1st : because you seem to be unable to read right, I spoke about a used scope, not a new one

2nd: I spoke about an analog scope, not a digital one

3rd: don't know why you explain to me what I've tried to explain to the opener of that thread, but it's pretty much your opinion

4th: because you assumed - I did never claim that DSO are crap - you should read my comments more accurate

5th: you're argumentation with your high price scope has again nothing to do with my initial statement: buy a used analog scope for less than 1000,- EUROs where your digital might not have enough ressources and be happy. Your Tek is even used beyound 1000,- €, so it's not discussable.

6th: PEAK DETECT has nothing to do with the maximum sample rate of the ADC.

WOW! What do you want to do with "Peak detect"? Increasing the physical limit of the maximum sample rate of an ADC???
No?
Then again your comment is off-topic, because that was my last comment about.

Nearly any scope manual remarks to use peak detect when using slow timebases to reveal small fast glitches since the sample rate is also reduced.
And? It's just a different option, but cannot overcome the maximum sample rate of an ADC.

Don't know your problem, Mr. Recordplayer, but it seem's that you either have a problem to read things right or understand what they're about.

You're really a funny guy. First you started with "there's not dead-time" although anybody using DSOs knows that wfm/s is exactly about it, then
you compare cheap Chinese scopes with your nice still powerful high end device, that you still can't get for cheap, and now you try your luck with
the "peak detect" option  What comes next, the color of the probe?


By the way: Agilent's manual for the 54622D explains for people like you very precise when peak detect can be helpful. Maybe your biggest mistake is that you relate to your powerful Tek than to these cheap Chinese models.

Look at the attached diagram taken from Agilent. Even peak detect won't help you making the glitches visible, on an analog scope you might see it - and that's where the discussion started since the beam is usually much faster returned than the signal processing of a cheap & slow DSO. In this case an analog scope might be an advantage, in other cases not.


And because it becomes more & more interesting to me to get your expert's opinion: the second jpeg I've attached shows a screenshot taken from my 54622D where I admit that it does not have a brand new ADC, but it's good for an example.

The ADC has got a sample rate of 200MS/s at max, i.e. a resolution of 5ns/DIV. I switched off the interpolation between so you can see that samples appear in an exact time interval of 5ns according to the sample rate.

Now tell me please how peak detect can reveal what's between two subsequent samples? I doesn't matter if switched on or off  - there's simply no difference because working with the fastest timebase of the scope and the highest sample rate.

There's no gap between two samples when using an analog scope, since the beam is continuously deflected. Must not, but might reveal more details in this case. Only compromise with a DSO: take one with a faster ADC to reduce the gap between the samples to a smaller time inteval.

So, before telling me what I don't know, answer my questions, please.


Kind regards
Gunb

[alm]

Nobody denied that, but it's possible to investigate how good the amplifier is, of any scope. But that's an issue of analog as well as digital scopes.  There's still an analog input stage.

Analog input stages for DSOs are generally better. They don't have to have the (relatively) huge voltage swing to drive the plates of a CRT, just the inputs of an ADC. This reduces much of the thermal problems they had with analog scopes.

Aha! I can't use the zoom?! Really? I mean, when using a scope with an ADC of 1GS/s at max I assume that you understand that the time between two samples is 1ns, right? So you're zoom can't resolve more than showing you samples in a distance of 1ns, what happens between is not displayed, but interpolated if not switched off. An analog scope using 1ns can show any value between because there's no sampling but a continous steering of the beam. Disadvantage: you can't store it. Think about it.

This is a completely different argument, and is about bandwidth, not resolution. A digital scope with a sampling rate of 1GS/s will generally have a bandwidth of about 100 MHz. So the signal from the front-end will be pretty much identical between t and t+1ns. This is also what sampling theory tells us. Any signal you will generally use a 100 MHz scope for will be well below the Nyquist limit of 500 MHz.

In an analog scope, the beam will be bandwidth limited by the vertical amplifier and CRT. So it will produce a smoothed version of the signal, with the higher frequencies attenuated. Kind of like interpolation .

A signal with significant changes on time scales < 1 ns will have frequency content beyond 500 MHz. Are you really expecting your 100 MHz scope to give usable results at those frequencies?

[Lukas]

Aha! I can't use the zoom?! Really? I mean, when using a scope with an ADC of 1GS/s at max I assume that you understand that the time between two samples is 1ns, right? So you're zoom can't resolve more than showing you samples in a distance of 1ns, what happens between is not displayed, but interpolated if not switched off. An analog scope using 1ns can show any value between because there's no sampling but a continous steering of the beam. Disadvantage: you can't store it. Think about it.

That's only true for single shot signals. Even cheap Chinese are able to determine the timing relation between the sampling clock and the trigger to a resolution of 25ps, which gives you a timing resolution of 25ps when acquiring long enough.

No it's generally valid even for cheap Chinese scopes and it doesn't depend on single shot.

....
And we're not talking about random sampling, that's another construction and works for periodical signals only.

There isn't a big difference between random sampling and real time sampling + TDC for subsample trigger resolution. When switching to "dot mode" and enabling persistence a real time scope becomes equivalent to a random sampling scope, hence the time resolution is determined by the trigger TDC.

Take a look at the DSOX2000 datasheet: horizontal resolution 2.5ps Pretty much every DSO has a time resolution better than it's sample rate, no matter if real time or sampling.

[Gunb]

Analog input stages for DSOs are generally better. They don't have to have the (relatively) huge voltage swing to drive the plates of a CRT, just the inputs of an ADC. This reduces much of the thermal problems they had with analog scopes.

Ah, OK, that might be the reason why a friend of mine at R&S told me that they've developed a low noise input stage linear over the complete 500MHz bandwidth taking a few years to design a special ASIC for their digital scopes - funny, it could be so simple. Maybe they should have asked you?!

This is a completely different argument, and is about bandwidth, not resolution.

Wow! Now even you've understood what I'm talking about all the time! Great, now you should understand that there's also a horizontal resolution caused by the sample rate of an ADC! Fine!
Oppenheim/Schafer might help.

A digital scope with a sampling rate of 1GS/s will generally have a bandwidth of about 100 MHz. So the signal from the front-end will be pretty much identical between t and t+1ns. This is also what sampling theory tells us. Any signal you will generally use a 100 MHz scope for will be well below the Nyquist limit of 500 MHz.

Oh, now we start with Nyquist, why not! I assume you mean 50MHz, not 500MHz in your example above. And yes, when you relate to a simple sine, 50MHz is maximum, but if you want to distinguish a sine from a ramp, then even about 20MHz is allowed for the signal.

But independent from Nyquist you're allowed to oversample a signal with more then twice of the highest frequency, right? And what else could be the advantage even if you're scope is limited to 100MHz bandwidth? Right Sir, the gap between two samples becomes smaller, so probality to catch a small pulse between subsequent samples is growing!

And that's exactly the reason why a scope as for instance the Rigol DS4012 with 100 MHz bandwidth and a 4GS/s ADC reveals more details then a scope with only 1GS/s and the same bandwith.
Nyquist is not the only criteria.

In an analog scope, the beam will be bandwidth limited by the vertical amplifier and CRT. So it will produce a smoothed version of the signal, with the higher frequencies attenuated. Kind of like interpolation .

   Oh no, how many times do I have to emphasize that I'm not talking about the VERTICAL resolution ? Is it really so difficult to understand?

A signal with significant changes on time scales < 1 ns will have frequency content beyond 500 MHz. Are you really expecting your 100 MHz scope to give usable results at those frequencies?

Hä? Just for you: take an 80MHz signal, sample it with 1GS/s and again with 4GS/s, both scopes have got 100MHz bandwidth. Then follow my steps above: switch off the interpolation and take a screenshot as I did above with the Agilent.

I tell you, what you'll get:

- one screenshot where samples will have a distance of 1ns with the 1GS/s ADC
- one screenshot where samples will have a distance of 0.25ns with the 4GS/s ADC

And now you're allowed to guess where the probality to see a small glitch of 0.5ns is higher? Correct, the scope with the 4GS/s ADC!!! Well done, boy!

And that has nothing to do with the bandwidth or Nyquist!

Just interfere a sine with small glitches and use both scopes to see the difference. Difference between theory and practice - my eyes are not too bad to see the difference.

 

[nctnico]

You really have no clue about digital signal processing and what bandwidth actually means...

[Gunb]

You really have no clue about digital signal processing and what bandwidth actually means...

We've got you to be teached by clever answers like this without proving the opposite, guy.

Be glad not to know what I'm doing every day and where I work  Maybe you've got one
of the products I'm designing on your desk 

But: there are lots of well made Agilent app. notes that you can use for the beginning
where many things you don't understand are described very well.

Rgds
Gunb

[Gunb]

There isn't a big difference between random sampling and real time sampling + TDC for subsample trigger resolution. When switching to "dot mode" and enabling persistence a real time scope becomes equivalent to a random sampling scope, hence the time resolution is determined by the trigger TDC.

Take a look at the DSOX2000 datasheet: horizontal resolution 2.5ps Pretty much every DSO has a time resolution better than it's sample rate, no matter if real time or sampling.

And also: wrong again.

If you would understand the datasheet right, you could have noticed that the 2.5ps have nothing to do with the sample rate provided by the ADC.

If you would have read the manual of the DSO2000X series well, you could have noticed that Agilent specifies a sample taken each 500ps provided by the 2GS/s ADC, German manual, page 158, above. Sounds logical to me 

And there's a tremendous difference between ETS and real time sampling. First implies a repetitive signal to take samples from several periods with a small delay from period to period. Putting them all together results in a higher horizontal resolution. Can't be done with non-periodic signals.

 

[PA4TIM]

Befor you digital guys kill each other, there is a reason I still use my real scopes ( my DSO is a Hameg 350 MHz and it it great, i had a 100 MHz Rigol and that was crap)
This reasons for using my 547 and 7603/7704 are
- they will probably outlive me
- are easy to repair
- heat my lab
- the plugin system is great, i have a transtistor Tr plugin, slide-back plugin, uV plugins, differential plugins, TDR, 1 GHz, opamp plugin ect.
- supersharp trace ( 547) with fery low noise
- what you see is there, not the games a DSO can play
- very sturdy and foolproof, things like 500V max input using a 1X probe instead of the pitty 50V of many DSOs.
- for RF, like modulation they are great
- when repairing high voltage stuff like tube amps I use analoge scopes
- they (547) are pieces of pure analoge art
- there are dual beams ( to bad I do not have a 556 yet)
- easy to operate, knobs for everything
- weight, you do not pull the scope from the bench that easy
- XY mode, multitrace like curvetracing

Disadvantages
- big to huge
- powerhungry
- no on screen info, i like the measure data on my dso, beats counting devisions and doing the math.
- the movement of the horizontal base is more easy then using the trigger, holdoff, delayed timebase ect.

I use my Hameg about 80 % of the time, for the rest the 547 or 7603.
Last week I repaired a satelite tuner, there was so much crap on the psu signal ( high switching peaks, digital garbage, ripple, the Hameg had troubles catching it. It was capable but needed a ot of fiddeling with the settings. Then I took the 547 and in 30 minutes all bad caps were found and replaced. I had that in the past with the Rigol, i made a CW decoder but it would not work right ( the analog part) the Rigol showed perfect pulses. Then I took the 7603 and then I did see the problem at once. Some pulses were lower in amplitude or chopped. The Hameg is a lot better at that point.
The thing that made me trust dso's less however was a Trr diode measurement. The Rigiol shreenshot I had saved, later I used the same setup with the hameg and a 1 GHz analg sample plugin and the 7603. The Rigol was so far off it was almost scary.


Both are 1n4148 reverse recovery time shots using the same setup

But I think a good DSO outperforms a good analog scope on most points and is general perfect usable.  But then you talk out of the hobby budget. I rather have my 50 MHz acurate 547 as a 100 MHz lying Chinese toy scope. ( and i have the gear to check, most pro-toy teardowns are not against references, hell, some not even have a signal generator of function generator and still tell the toy performs well. Yeah, whit the standard crappy not enough BW probes but nice colors, a great giftbox and it powers up.

[nctnico]

No wonder the Rigol is so far off. It doesn't have 50 Ohm inputs (which you use on the Hameg) and its bandwidth is probably too limited for what you want to measure. You just choose the wrong tool for the job. If you measure a 50MHz square wave with a 100MHz scope you'll get something that looks line a sine wave but you can't tell what the signal really looks like. Its like trying to measure millimeters with a ruler which only shows centimeters. There is always a limit to the usefulness of a tool but that doesn't make it a bad tool.

[PA4TIM]

Dream on, ever heard of inline termintors ? I have done the same measurement also using a 100 MHz 7603 and that was more like the Hameg, a bit slower but as clean.
The bandwidth has nothing to do with it. I have done a lot of BW testing so I know the differences, what you tell about squarewaves that look lile sines ? Look at my website, the pictures are there showing wath you tell. This has to do with things like for instance the phase offset between channels, the huge difference in Trr, much more as BW should give and difference in the trace going all over the screen. I have the same hoorible differences while doing load tests, had to use filtering ect to get a usable trace that still looked like shit from the noise, the Hameg draws it as sharp as a rothring pen. The analoge scopes too.

I May not be an expert on how the digital stuff calculates and translates the signals but I think I can say I do know scopes and for most how to correct use them.

Measured even things like input impedance over a few hundered MHz with a vna, measure that on a Rigol and you do not need a 50 Ohm input, unless you want to increase the imput impedance >:-), i have scoes to 5 GHz, tested probes to the limit ( that way I  found out the Rigol 100 MHz probes are indeed 100 MHz, the scope is too, so do your math....A 100 MHz tek probe is more then 100 MHz, it is made for a 100 MHz scope and made to be sure a 100 MHz scope can use that 100 MHz.

[Gunb]

Yes,

and that's what I've also noticed: the RIGOL DS4012 isn't a cheap scope and it has got 1mV/DIV sensitivity, also my HMO2524 and the analog HM407. Both Hamegs have got extreme low noise levels, but the Rigol seems to be connected to a noise generator. It has got 1mV noise floor, so the 1mV/DIV can't be used for small signals. It's a good scope, but not usable for that purpose. So I prefer the HM407 in analog mode or even the HMO with its reference class ADC rather than the RIGOL.

The trace width of both Hamegs does not even become bigger when changing to 1mV/DIV, can't see noise at all.

That's why my initial recommendation was to extend a cheap DSO with an older analog scope of good quality if people have a limited budget only.

[madshaman]

I may not know much but I know this, if someone made a new 10Ghz crt-based analog-only scope I could afford, I'd cry tears, then I'd happily pull out my wallet.

In fact, I'm pretty sure that's what I want to try and build.  Know idea how long the learning process and the attempts at construction of each part would take me, but that seems like a happy road. (And yeah, extending it by adding a high speed ADC, writing software, is logical next step)

That aside, I use my analog scope (400Mhz) more than my digital one (200Mhz, 8-bit, 1GSa/s single shot), that might change over time.  Right now it's mostly a subjective choice, I like using it more.  Only one single problem: can't really scope circuit startup/shutdown or other kinds kinds of single shot events with an analog scope.

[Lukas]

There isn't a big difference between random sampling and real time sampling + TDC for subsample trigger resolution. When switching to "dot mode" and enabling persistence a real time scope becomes equivalent to a random sampling scope, hence the time resolution is determined by the trigger TDC.

Take a look at the DSOX2000 datasheet: horizontal resolution 2.5ps Pretty much every DSO has a time resolution better than it's sample rate, no matter if real time or sampling.

And also: wrong again.

If you would understand the datasheet right, you could have noticed that the 2.5ps have nothing to do with the sample rate provided by the ADC.

Where did I say that?

If you would have read the manual of the DSO2000X series well, you could have noticed that Agilent specifies a sample taken each 500ps provided by the 2GS/s ADC, German manual, page 158, above. Sounds logical to me 

500ps sample resolution vs 2.5ps time resolution

And there's a tremendous difference between ETS and real time sampling. First implies a repetitive signal to take samples from several periods with a small delay from period to period. Putting them all together results in a higher horizontal resolution. Can't be done with non-periodic signals.

 

Of course, my statement was based on repetitive signals. In single-shot applications the time resoultion is equivalent to the distance between the samples. When acquiring repetitive signals a real time scope behaves identically to a random sampling scope (with a high sample rate)

To clarify my point: Time resolution of DSOs acquiring repetitive signals is determined by the trigger TDC, not by the sampling rate.

[nctnico]

Dream on, ever heard of inline termintors ? I have done the same measurement also using a 100 MHz 7603 and that was more like the Hameg, a bit slower but as clean.

A Trr measurement basically sends a step into the oscilloscope which is nasty. Depending on the frequency response you can get the effects shown on the Rigol. Agilent has an appnote on this:
http://cp.literature.agilent.com/litweb/pdf/5988-8008EN.pdf

In short: it comes down to how the input filter is implemented. A sharp filter will cause ringing but prevents aliasing due to the sampling frequency.

In most practical cases a scope (analog or digital) shows an interpretation of a signal anyway. A 1kHz 1Vpp sine wave will show fine on any scope but at higher frequencies and/or low amplitudes all bets are off. If you measure a signal with a scope you have to factor in its limits. IOW you have to know what the signal looks like and how your scope shows such a signal given its limits. In some cases a signal will show just garbage which means you need to get a better scope.

[Gunb]

Where did I say that?

Why do you mention it then - makes no sense to discuss off-topic?!

500ps sample resolution vs 2.5ps time resolution

And? Tell me how you get more samples than the ADC can provide at max - that's the issue I'm talking about - has nothing to do with your 2.5ps, absolutely nothing.

Of course, my statement was based on repetitive signals. In single-shot applications the time resoultion is equivalent to the distance between the samples. When acquiring repetitive signals a real time scope behaves identically to a random sampling scope (with a high sample rate)

To clarify my point: Time resolution of DSOs acquiring repetitive signals is determined by the trigger TDC, not by the sampling rate.

And? Has nothing do to with the issue of that thread, and has nothing to do with my previous comments. Can't use ETS for most of my purposes since the signals I'm processing are seldom periodic. Again: that has nothing do to with my previous statements.

[alm]

I may not know much but I know this, if someone made a new 10Ghz crt-based analog-only scope I could afford, I'd cry tears, then I'd happily pull out my wallet.

Not going to happen. Try making a CRT with that kind of bandwidth. They were approaching the limits of what CRTs could do at the end of the analog scope era. The highest end analog scopes went up to a few hundred MHz, with a select few scopes going up to 1GHz. You can go higher, but then limited deflection becomes a problem. I believe a Soviet company designed a digitizer (basically a CRT with a video camera in front) with a few GHz bandwidth. The CRT was too long to fit into a standard 19" rack, something like 1m long. This would be extremely expensive, and also kind of bulky. I don't think much progress has been made in CRT technology since then, since the only analog scopes that have been designed since are low-end scopes.

[alm]

Oh, now we start with Nyquist, why not! I assume you mean 50MHz, not 500MHz in your example above.

The Nyquist limit states the maximum bandwidth of the signal (it assumes bandwidth limited signals) that you can completely (!) reconstruct at a given sampling rate. This is half the sampling rate, so 500 MHz for 1GS/s.

But independent from Nyquist you're allowed to oversample a signal with more then twice of the highest frequency, right? And what else could be the advantage even if you're scope is limited to 100MHz bandwidth? Right Sir, the gap between two samples becomes smaller, so probality to catch a small pulse between subsequent samples is growing!

If the front-end limits the signal to 100 MHz, there is no pulse going to be smaller than 1 ns. You can't construct a 1 ns pulse from sine waves (Fourier series) with frequencies up to 100 MHz. Of course the front-end won't have a brick-wall response, which makes the issue slightly more complex.

  Oh no, how many times do I have to emphasize that I'm not talking about the VERTICAL resolution ? Is it really so difficult to understand?

Horizontal resolution is linked to bandwidth. What good is horizontal resolution of the high frequency components of your signal (necessary to actually see changes at those time scales) are attenuated?

Hä? Just for you: take an 80MHz signal, sample it with 1GS/s and again with 4GS/s, both scopes have got 100MHz bandwidth. Then follow my steps above: switch off the interpolation and take a screenshot as I did above with the Agilent.

Why would you switch off interpolation? Interpolation is an essential part of sampling theorem. It says you can reconstruct (with interpolation) a signal.

And now you're allowed to guess where the probality to see a small glitch of 0.5ns is higher? Correct, the scope with the 4GS/s ADC!!! Well done, boy!

If there is a small glitch of 0.5 ns, it is not an 80 MHz signal, and the glitch most likely won't make it through the front-end unless it had a very large amplitude.

[PA4TIM]

A Trr measurement basically sends a step into the oscilloscope which is nasty. Depending on the frequency response you can get the effects shown on the Rigol.

In short: it comes down to how the input filter is implemented. A sharp filter will cause ringing but prevents aliasing due to the sampling frequency.

Read the appnotes from Williams about scope mesurements. He often shows how to get it right and how to reognize faults in screenshots/setups.
A step is not nasty, it is a way to seperate the boys from the men. But I know what causes the things the Rigol shows and that is not why I showed that pictures, the point is, my other scopes did not make such a mess out of it. Gunb tell it right, the Rigol has a build in noise generator 

In most practical cases a scope (analog or digital) shows an interpretation of a signal anyway. A 1kHz 1Vpp sine wave will show fine on any scope but at higher frequencies and/or low amplitudes all bets are off. If you measure a signal with a scope you have to factor in its limits. IOW you have to know what the signal looks like and how your scope shows such a signal given its limits. In some cases a signal will show just garbage which means you need to get a better scope.

I agree, and the Rigol showed a very bad interpretation. Because I know what to expect while measuing, most signals showed as garbage and I measure a lot against the limmits, I knew I needed a better DSO and that became the Hameg.

Do not forget we often compare horses and cows when it comes to scopes. Our beloved analoge sopes were often state of the art at that time and build with no cost spared. Specs were worst case numbers and you could trust then.
They still make quality like this, but only a few members here have acces to DSO's that have that level. They costed a fortune back then, they cost a fortune now. I'm talking integrety off measurements, optimalision of input and other stages, quality of used components, ect. But because BW has gone skyhigh on the still very topmodels now, we wrongly compare the low BW budget models from today with the topmodels from back then, the only thing they have in common is that BW. The rest is at a totally different level. The toys do not even have real specs, they are just "avarage" or " typical" or "best" and often pure theoretical or just picked by marketing.

Madshaman:
Forget building a 10 GHz (analog crt) scope. There has been a 1GHz realtime Tek, the CRTs for those are extreme and very rare. You will not find a CRT that does 10 GHz without sampling. An analog sample scope only needs a 10 to 20 MHz baseunit.
Try building a 10 MHz scope first, including triggering, several timebases and calibrated V/div input section. Then the right delay network, low noise amplifiers flat over 10 MHz and the high voltage part, 100 MHz scopes go over 10kV for the CRT, and defelection voltages go up too because they are related. Vertical and horizontal amplifiers from analog scopes are state of the art designs and even today used as examples in modern analog design literature. ( i have two books where the writers use Tektronix amplifiers from the 60/70's  for this)

[Gunb]

You really have no clue about digital signal processing and what bandwidth actually means...

So, too tired of your bullish behaviour yesterday I take a last try to examine what I've already implied when joining the discussion at the beginning and where you
still got stuck:

Requierements:

- Scope with 100 MHz bandwidth
- Square wave of 7MHz , 50% duty cycle

Let's assume we want at least 5 harmonics:

f0 = 7MHz                       (1. harmonic)
f1 = 3*7MHz = 21MHz    (2. harmonic)
f2 = 5*7MHz = 35MHz    (3. harmonic)
f3 = 7*7MHz = 49MHz    (4. harmonic)
f4 = 9*7MHz = 63MHz    (5. harmonic)

- Since we've got 50% duty cycle we only will get odd multiples of the base frequency.
- Since we're 37MHz away from the 100MHz cut-off frequency we are on the safe side getting all 5 harmonics with no/less than 3dB attenuation.

Concludingly, our 100MHz scope is still sufficient concerning these facts. Are we still friends at this point? 

Now we go on with the ADC and the required sample rate. Nyquist says, at least sample rate has to be 2x of the highest frequency the signal consists of.
So, with f4=63MHz we should have at least:

- fsample = 2*f4 = 126MHz = 126MS/s

The old Agilent I'm using has got 200MS/s, so sufficient for the example above and exactly twice of the scope's bandwidth.
Means also, a 100MHz sine will be attenuated with 3db at the scope's cut-off frequency and if sampled not at the zero crossings you will even see something on screen.
Of course, that's not sufficient for a square wave where the harmonics are of higher frequency than 100MHz. In this case - as you also mentioned - you'll only see something
that looks like a sine, since harmonics are missing.

Think, we're still friends at this point, right? 

BREAK: all these thing I've implied when joining the discussion, these are basics that everybody should know using a DSO But that's where you've started and what I've alreay passed.


With all that basics in mind, now the point where I've started this thread:

- 3 scopes: all with 100 MHz bandwidth,
- supposed that the maximum signal frequency we want to use them for is 100MHz (cut-off, 3db attn.)

scope 1 with ADC: 200MS/s
scope 2 with ADC: 1GS/s
scope 3 with ADC: 4GS/s

The 200MS/s ADC is absolutely sufficient concerning Nyquist for a scope with 100MHz bandwidth.

So the question comes up: what advantage can there be to use a 1GS/s or even a 4GS/s ADC for a 100MHz bandwith limited scope?

-> They're completely oversized according the requirements of Nyquist.
-> but, if we don't think of Nyquist at this point and search for further advantages we notice that the gaps between two subsequent samples is
only determined by the ADC's sample rate, so:

- the 200MS/s ADC will generate samples with a time interval of 5ns
- the 1GS/s ADC will generate samples with a time interval of 1ns
- and last but not least the 4GS/s ADC will generate samples with a time interval of 250ps

3x sampled the same signal:

- interpolation leads to smoother signals concerning higher sample rates
- more details are aquired because of smaller gaps the higher the sample rate, independet of Nyquist.

These facts you can try yourself with any DSO, are mentioned in nearly any app. note from Agilent, Tek, Hameg, ... whatever
and are even part of the books about the basics of how to measure right with DSOs.

Don't know, why it's so difficult to understand.

Believe me, my friend, I'm busied with signal processing more than 20 years now, especially the hard cases where disturbed signals
have to be processed, but the issue we're talking about in this thread are of quite common nature, nothing special.

Sometimes it might be useful trying to understand what other participants want to explain before keeping them for stupid.


However, have a nice day,
Gunb

[Gunb]

If there is a small glitch of 0.5 ns, it is not an 80 MHz signal, and the glitch most likely won't make it through the front-end unless it had a very large amplitude.

But you might see parts of it, and that has got an influence on the signal you sample.

Why switching off interpolation? Only for making the argumentation clearer and to see that the gap between subsequent samples is smaller with higher sample rates
than with ADCs of lower ones. You simply get closer to the original analog signal then.

Some people seem not to know that. Interpolation is always approximation of the real analog signal, when you want to see the samples
itself for researching the samples in memory it can be useful to switch it off. Needed that a few times for DSP processing of weak disturbed signals for synchronisation
of a digital matched filter algorithm. In this case interpolation is not useful because I'm calculating with the real samples in memory that I've to see and not the "falsified"
connection between samples added by the scope for improving visualization.


Kind regards
Gunb

[alm]

- 3 scopes: all with 100 MHz bandwidth,
- supposed that the maximum signal frequency we want to use them for is 100MHz (cut-off, 3db attn.)

scope 1 with ADC: 200MS/s
scope 2 with ADC: 1GS/s
scope 3 with ADC: 4GS/s

The 200MS/s ADC is absolutely sufficient concerning Nyquist for a scope with 100MHz bandwidth.

An important point is that the definition of bandwidth in sampling theorem (the highest frequency component in the signal) does not match the definition used in filters and test equipment (-3 dB point). 200 MS/s is sufficient to sample a bandwidth-limited signal with no frequencies beyond 100 MHz. If there are frequencies beyond 100 MHz, it will screw up the interpolation. This is why usually figures like 4-10x bandwidth are quoted for the minimum sampling rate of a scope with a Gaussian (-6dB/octave) response. Only about 2.5x is necessary for scopes with a sharper roll-off, the so called brick-wall response (which is not really like a brick wall, but much sharper), since less higher frequency components will make it past the front-end. Another factor is that interpolation algorithms are not perfect, so having some head room beyond the Nyquist frequency makes interpolation easier. The 100 MHz/1GS/s example comfortably meets these limits.

So the question comes up: what advantage can there be to use a 1GS/s or even a 4GS/s ADC for a 100MHz bandwith limited scope?

None if the scope is really bandwidth-limited in the Nyquist sense, and you use perfect sin(x)/x interpolation. In the real-world case the 1 GS/s scope will perform better for some signals, but 4 GS/s is unlikely to make much of a difference unless you have specific signals with a large amount of frequency content beyond a few hundred MHz that will make it past the front-end somehow.

- more details are aquired because of smaller gaps the higher the sample rate, independet of Nyquist.

There are no more details to acquire in a bandwidth-limited signal. And it's hard to see how components even approaching the 500 MHz nyquist frequency of a 1GS/s scope will make it past the 100 MHz front-end. Responses tend to fall off much sharper than 6 dB/octave if you go well beyond the rated bandwidth, in my experience.

Yes, but even a glitch consists of lower frequency parts according to Fourier, that might be within the range of your scope's bandwidth. That has got an influence
on your samples - the more you get, the more you see.

The low frequency parts will not need > 1 GS/s to reconstruct.

Why switching off interpolation? Only for making the argumentation clearer and to see that the gap between subsequent samples is smaller with higher sample rates
than with ADCs of lower ones. You simply get closer to the original analog signal then.

Interpolation also gets you closer to the analog signal. That's what the whole Nyquist-Shannon-Whosthatthirdguy (Whittaker?) sampling theorem is about.

[Gunb]

@alm:

I find it always funny that most of your statements exactly say the same I've mentioned before 
So, what do you try to explain to me, then? Cut-off frequency and limited devation of filter curves?
Think, it's not necessary to waste time with the basics.

I agree that I don't go into the details as you do, that would blast that thread completely with more
and more discussions.

I also agree in all of your above explanations, but especially that statement:

... won't make it through the front-end unless it had a very large amplitude.

is one of the major issues which fits my case and where the higher sample rate
even of the 100MHz version was helpful a few times.

The application: few kilometers of long harness wire in the open field with signal replies of a about 100mV and disturbed by external
noise and glitches much stronger then the signal itself.

[tinhead]

Let's assume we want at least 5 harmonics:

f0 = 7MHz                       (1. harmonic)
f1 = 3*7MHz = 21MHz    (2. harmonic)
f2 = 5*7MHz = 35MHz    (3. harmonic)
f3 = 7*7MHz = 49MHz    (4. harmonic)
f4 = 9*7MHz = 63MHz    (5. harmonic)

your statement is a little bit misleading, something like "5. harmonic" could be read as 5th harmonic,
but here it is 9th harmonic you taking about. It is easier, even in the 50% duty case, to call them what they are,
multiplied fundamental frequency.