DSO Samplerates - From when they´re "enough" and why?

[Martin72]

Hi,
Surely an old question....
Samplerates on a DSO.
You must have "enough" samples to recreate a proper signal.
It also depends on the frequency/bandwith.
Right?
OK..
I´ve got a siglent scope with max. 2Gsa/s and 500Mhz bandwith.
At work we have older waverunner lt models from lecroy in addition to the new ones.
200Mhz and 200MSa/s, 500Mhz 500MSa/s, but also a model with 500Mhz and 1GSa/s.
I can not imagine that it was not enough at the time, with each model.
And yet it became more and more in the following years.
Even a cheap Rigol has 8GSa/s at 350Mhz.
In contrast, there are much more expensive models from other manufacturers, which have e.g. 4GSa/s at a bandwidth of 1Ghz.
(always max.)
Or our newer WR9054 which got 20(40) GSa/s and 500Mhz.
So the question:
How many samples do you really need, what does it depend on.
What is the minimum at which bandwidth, when do you need more at the same bandwidth.
Only in special cases or also in general.
Having more than needing is always better but in this case mostly a price decision.
Or not?
Carry on...

[nctnico]

If you have the headroom for a Gaussian roll-off: you'd like to have 10 samples per period. If there is a steeper anti-aliasing filter AND a good implementation of sin x/x reconstruction, then 2.5 samples per period is enough. Sin x / x is prone to producing pre-ringing (Gibbs ears) on sharp edges that are not supressed enough by the anti-aliasing filter. Using the selectable bandwidths you typically find on high frequency DSOs you can tune this depending on what you are looking for in a signal.

Likely the 500MHz / 1Gs/s scope you mention has equivalent time sampling. That way you can acquire a repetitive signal using many more samples per second than the samplerate of the ADC. Some older several GHz oscilloscope use samplers that sample at several 10's of kHz. For example the Tektronix TDS820: 6GHz bandwidth using a 50ks/s 14 bit converter.

[Fungus]

Pretty much what ntnico said. 2.5x is the minimum to avoid serious artifacts, more than 10x is probably a waste.

What I'd like to see would be dynamic bandwidth switching.

Most modern DSOs have software bandwidth upgrades so internally they have switchable filters, right? It would be nice if they could reduce the bandwidth when you turn all the channels on and avoid some artifacts. Bandwidth is cheaper than sample rate so it could make 'scopes cheaper.

eg. You could have a 4 channel, 1GSample/sec 'scope which has 200MHz bandwidth with one or two channels enabled and 100MHz bandwidth when you turn on three or four channels.

[tautech]

Hi,
Surely an old question....
Samplerates on a DSO.

How many samples do you really need, what does it depend on.

Needs.

Used wisely undersampling can be beneficial to needs.
https://www.eevblog.com/forum/testgear/siglent-sds2000x-hd-12bit-(published-for-chinese-domestic-market-only)/msg4320658/#msg4320658

[robert.rozee]

hi,
    this has been a topic of hot debate in the past. the simple answer is, it all depends on what you are looking for in the signal!

have a look at the below thread, link is to a post within the thread i made demonstrating the result of a square wave at a range of different sample rates relative to the fundamental (2x, 5x, 10x, 20x, 50x, 100x, 200x, 10,000x). this shows the results of just the oversampling, with an analog input bandwidth massively higher than the fundamenal frequency. ie, the effect of front-end bandwidth is excluded:
https://www.eevblog.com/forum/testgear/is-it-true-oscilloscope-must-reach-at-least-4x-observed-freq/msg4413487/#msg4413487

btw, the remainder of the thread may also answer some of your questions.


cheers,
rob   :-)

[JPortici]

At work we have older waverunner lt models from lecroy in addition to the new ones.
200Mhz and 200MSa/s, 500Mhz 500MSa/s, but also a model with 500Mhz and 1GSa/s.
I can not imagine that it was not enough at the time, with each model.

Of course it wasn't enough. That's why they usually had ET sampling (RIS on the waverunner), or accessories such as the 7291

[2N3055]

hi,
    this has been a topic of hot debate in the past. the simple answer is, it all depends on what you are looking for in the signal!

have a look at the below thread, link is to a post within the thread i made demonstrating the result of a square wave at a range of different sample rates relative to the fundamental (2x, 5x, 10x, 20x, 50x, 100x, 200x, 10,000x). this shows the results of just the oversampling, with an analog input bandwidth massively higher than the fundamenal frequency. ie, the effect of front-end bandwidth is excluded:
https://www.eevblog.com/forum/testgear/is-it-true-oscilloscope-must-reach-at-least-4x-observed-freq/msg4413487/#msg4413487

btw, the remainder of the thread may also answer some of your questions.


cheers,
rob   :-)

That post of yours is not really accurate.

To determine needed BW of the scope for good representation of squarewave signal, repetition frequency of squarewave is irrelevant.
Amplitude of harmonics will be determined by rise/fall times of the signal.

So for a squarewave signal that has 10ns rise/fall times, that is only that matters.
Repetition frequency can be 1 kHz, and if you sample with less than needed to show nicely that 10ns edge, you will se artefacts..

[magic]

Most modern DSOs have software bandwidth upgrades so internally they have switchable filters, right? It would be nice if they could reduce the bandwidth when you turn all the channels on and avoid some artifacts. Bandwidth is cheaper than sample rate so it could make 'scopes cheaper.

eg. You could have a 4 channel, 1GSample/sec 'scope which has 200MHz bandwidth with one or two channels enabled and 100MHz bandwidth when you turn on three or four channels.

AFAIK they have switchable digital filters.

Scopes which lose BW as more channels are enabled are already a thing. My cheapo Hantek behaves like that.
Nominally its BW is whatever it is (100M hacked to 200M) but I can clearly see loss of accuracy in 500MSPS mode.

[tautech]

Most modern DSOs have software bandwidth upgrades so internally they have switchable filters, right? It would be nice if they could reduce the bandwidth when you turn all the channels on and avoid some artifacts. Bandwidth is cheaper than sample rate so it could make 'scopes cheaper.

eg. You could have a 4 channel, 1GSample/sec 'scope which has 200MHz bandwidth with one or two channels enabled and 100MHz bandwidth when you turn on three or four channels.

AFAIK they have switchable digital filters.

Scopes which lose BW as more channels are enabled are already a thing. My cheapo Hantek behaves like that.
Nominally its BW is whatever it is (100M hacked to 200M) but I can clearly see loss of accuracy in 500MSPS mode.

Automatic.

SDS2054X HD and Plus have this where if you have a channel active on each ADC and enable any other channel the 350 MHz BW filter is automatically engaged and the BW indicator in each channel tab changes from Full to 350.
It maintains adequate sampling rates to minimise aliasing now an additional channel now shares the ADC's sampling rate.

[Fungus]

AFAIK they have switchable digital filters.

The RigolDS1054Z has capacitors in the analog front end that are switched in using MOSFETs.

[nctnico]

AFAIK they have switchable digital filters.

The RigolDS1054Z has capacitors in the analog front end that are switched in using MOSFETs.

That only gives you a 1st order filter. Good for getting a nicer roll-off but not to prevent aliasing.

edit: typo

[Fungus]

That only gives you a 1st filter. Good for getting a nicer roll-off but not to prevent aliasing.

Sure, but it's already in there and it's free. It would be nice to use it.

[vk6zgo]

Very early DSOs had abysmal memories, so for long settings of time/div, the sample rate was reduced, as there was no place to save the number of samples that would result from sampling, & displaying say, 2 cycles of a 50Hz sine wave, using the maximum sample rate, so the sample rate was reduced.

A sine wave of that order could be easily accommodated, using around 1kSa/s, but complex signals are "another Kettle of fish"
Analog video signals are particularly nasty, with components out to 5MHz, or sometimes more.

Not only are they wideband, but the component frequencies of that bandwidth are also constantly changing.
TV people were usually very enthusiastic to see their first DSO, but all their hopes were dashed by the unusable displays presented, as the instrument could not represent the real waveform.

All the discussion so far seems to be drifting into the problems of higher frequencies, where anti-aliasing filters come into play, but if sampling rates are still reduced at much lower frequencies, any alias components will still be present for those lower frequencies which 'scopes are very commonly required to observe,

[nctnico]

All the discussion so far seems to be drifting into the problems of higher frequencies, where anti-aliasing filters come into play, but if sampling rates are still reduced at much lower frequencies, any alias components will still be present for those lower frequencies which 'scopes are very commonly required to observe,

That is why a good DSO has peak detect. The ADCs run at full sample rate and the min/max of the sample interval are stored.

At one of my employers we had a brand new 500MHz Lecroy DSO sitting on top of a cabinet. Looked like a really nice oscilloscope but nobody ever used it. The lack of peak detect made it completely useless for the work we where doing. I don't think I have ever seen it being switched on. When I visited a couple of years later (after I left), it was still sitting there.

[alm]

All the discussion so far seems to be drifting into the problems of higher frequencies, where anti-aliasing filters come into play, but if sampling rates are still reduced at much lower frequencies, any alias components will still be present for those lower frequencies which 'scopes are very commonly required to observe,

Unless you are zooming in a lot after acquisition, the "resampling" by the display's limited horizontal resolution, which unless you're using an ancient Tek DSO is many orders of magnitude lower than the sample memory, is probably a bigger worry. But I believe most scopes, including Lecroy scopes, do some sort of peak detect when they resample data for display?

That is why a good DSO has peak detect. The ADCs run at full sample rate and the min/max of the sample interval are stored.

Having enough sample memory so the sample resolution is much higher than the display resolution also helps a lot. I used to use peak detect all the time back when my (Tek TDS-220) scope had 2.5 kS of sample memory, and so it was constantly under-sampling. On modern deep memory scopes I can easily get by without, since sample rates are generally fast enough to capture the data I need. Except in extreme cases where the duty cycle is approaching the sample memory depth, like a 1ns pulse with a repetition rate of 1s, which would need multiple GS of memory depth to properly sample. But this is not something I encounter often, plus if I only care about the pulse timing, I might well be using a logic analyzer instead.

[nctnico]

All the discussion so far seems to be drifting into the problems of higher frequencies, where anti-aliasing filters come into play, but if sampling rates are still reduced at much lower frequencies, any alias components will still be present for those lower frequencies which 'scopes are very commonly required to observe,

Unless you are zooming in a lot after acquisition, the "resampling" by the display's limited horizontal resolution, which unless you're using an ancient Tek DSO is many orders of magnitude lower than the sample memory, is probably a bigger worry. But I believe most scopes, including Lecroy scopes, do some sort of peak detect when they resample data for display?

Yes and no. In order to make the process fast, the acquired data is decimated using only a subset of the samples (one out of X). If you feed a slow sweep into a DSO with deep memory, you can get all kinds of funny aliasing effects. In practical situations this doesn't really matter though.

[alm]

Yes and no. In order to make the process fast, the acquired data is decimated using only a subset of the samples (one out of X). If you feed a slow sweep into a DSO with deep memory, you can get all kinds of funny aliasing effects. In practical situations this doesn't really matter though.

I just tried it on two Lecroy scopes (one VxWorks and one X-Stream) which definitely didn't have peak detect enabled , with a signal with a 1 ms period and 2 ns pulse width, and as long as the time base and memory depth were set so the sample rate was at least 500 MS/s, the scopes would reliably display each peak. If the signal was decimated before display, wouldn't you expect this 1 sample peak to disappear intermittently? Of course the amplitude is not very stable with only one sample somewhere within the pulse, but all four pulses are shown consistently.

Do other brands work differently?

[nctnico]

There is a bit more to it, but there is definitely some optimisation going on. Try a frequency sweep.

[alm]

Here's a logarithmic 100 Hz to 15 MHz sweep. Looks like a decent approximation within the limited screen resolution to me.

[markone]

-snip
At work we have older waverunner lt models from lecroy in addition to the new ones.
200Mhz and 200MSa/s, 500Mhz 500MSa/s, but also a model with 500Mhz and 1GSa/s.

I would say that nowadays those specs, only usable with equivalent time acquisition mode, are simply laughable. 

[nctnico]

Here's a logarithmic 100 Hz to 15 MHz sweep. Looks like a decent approximation within the limited screen resolution to me.

I wrote slow which means a long sweep over a small frequency span.

[alm]

I would say that nowadays those specs, only usable with equivalent time acquisition mode, are simply laughable.

Back then you could choose either sample rate (e.g. Tek TDS3000) or memory depth (Lecroy and HP). The former is better at short time bases, but all the extra sample rate is useless as soon as you go to slower time bases, because they don't have the memory depth to sustain their maximum sample rate. And the latter is not very useful at short time bases unless you can use equivalent time sampling, but is much better at longer time bases. Below 100 MHz those scopes should still be fine.

Of course these days you can easily get both.

I wrote slow which means a long sweep over a small frequency span.

This is the worst I could get. Is this the reason you can't live without peak detect on deep memory scopes?

[Fungus]

200Mhz and 200MSa/s, 500Mhz 500MSa/s, but also a model with 500Mhz and 1GSa/s.
I can not imagine that it was not enough at the time

It really wasn't...

(Maybe if you change the persistence to "infinite" and use dot mode)

How old are they? How much sample memory? How much did they cost? 

[nctnico]

I wrote slow which means a long sweep over a small frequency span.

This is the worst I could get. Is this the reason you can't live without peak detect on deep memory scopes?

It is not about peak detect but the decimation used by DSOs with deep memory can be fooled by some signals. You just didn't find the weak spot yet.

[hpw]

My opinion(s) changed on that matter....

So the question is how many samples you like to have on a Rise/Fall edge... 5 or 10 or 20 samples?

At the end of the day, some Ferrari DSO is required as having RIS (LeCroy) on repetitive signal or than TEK for none repetitive.

I do not now know, how the new LeCroy has the repetitive signal limitation as using RIS.

On Siglent 2K+ did not found any RIS functions or I am missing something or may newer once as HD 2K or ??

[Martin72]

Elder siglent scopes got equivalent samplerate, the newer ones not.

https://int.siglent.com/products-overview/sds1000cfl/


Tek doc:

https://www.tek.com/en/documents/application-note/real-time-versus-equivalent-time-sampling

[David Hess]

As far as what sample rate is sufficient, it depends on the rise/fall time of the signal.  As a rough conversion, use the 0.35 rule to get MHz and then base sample rate on that.  So for example 100 nanoseconds is 3.5 MHz and the sample rate should be several times higher than that.

I am used to DSOs which support equivalent time sampling where the sample rate ends up being orders of magnitude higher than strictly necessary, so I almost never need to consider it.  I have never really considered a sample rate of 2.5 times to be sufficient.

But I believe most scopes, including Lecroy scopes, do some sort of peak detect when they resample data for display?

I think LeCroy's older DSOs were descendants of the transient digitizers they made for physics experiments, including nuclear weapons testing, so they had big sample memories and relied on post processing rather than processing during decimation where peak detection is done.  So for a long time, LeCroy eschewed peak detection.

I think the first "modern" DSO with peak detection was the Tektronix 2230 which was first available in 1986.  It has a 100 MHz bandwidth and 20 MS/s sampling to support peak detection down to 100 nanoseconds.  It was quickly replaced by the 2232 with a sample rate of 100 MS/s and peak detection down to 10 nanoseconds.  Both supported equivalent time sampling at 2 GS/s.

[2N3055]

My opinion(s) changed on that matter....

So the question is how many samples you like to have on a Rise/Fall edge... 5 or 10 or 20 samples?

At the end of the day, some Ferrari DSO is required as having RIS (LeCroy) on repetitive signal or than TEK for none repetitive.

I do not now know, how the new LeCroy has the repetitive signal limitation as using RIS.

On Siglent 2K+ did not found any RIS functions or I am missing something or may newer once as HD 2K or ??

It was repeated many times, when in dot mode it behaves as RIS...

[nctnico]

My opinion(s) changed on that matter....

So the question is how many samples you like to have on a Rise/Fall edge... 5 or 10 or 20 samples?

At the end of the day, some Ferrari DSO is required as having RIS (LeCroy) on repetitive signal or than TEK for none repetitive.

I do not now know, how the new LeCroy has the repetitive signal limitation as using RIS.

On Siglent 2K+ did not found any RIS functions or I am missing something or may newer once as HD 2K or ??

It was repeated many times, when in dot mode it behaves as RIS...

Not by design. If your signal is an exact multiple of the oscilloscope's sampling frequency, you won't get a signal on screen. The key to RIS / equivalent time sampling is that the sampling points are shifted in time so the signal is sampled at different points in time to make sure that you always get something on screen.

[David Hess]

Not by design. If your signal is an exact multiple of the oscilloscope's sampling frequency, you won't get a signal on screen. The key to RIS / equivalent time sampling is that the sampling points are shifted in time so the signal is sampled at different points in time to make sure that you always get something on screen.

The only time I have had that be a problem is when using the DSO to sample a signal from the inside of the DSO itself.

Some DSOs, and I know HP did this, randomly dither the sample clock and then measure the resulting sample time to avoid synchronization with the signal source.