Some Important Points to Remember when Evaluating Waveform Update Rates

[marmad]

I can understand the advantage of DSOs with faster acquisition cycles being able to trigger in shorter times - but I'm also curious about what the practical limits are for the amount of visual data that can be conveyed by these DSOs (with high waveform update rates) to the display. I haven't been able to locate any oscilloscope-specific information, although I suspect one or more manufacturers have studied it.

I'm also curious why the Agilent X series only uses 64 levels of intensity gradation - it seems like a limiting factor in terms of getting information to the display. Perhaps it has to do with optimizing memory usage?

[robrenz]

IMO 64 levels from off to full brightness sounds like a lot to me.  I would guess that our eye has a hard time discerning a 1/64 incremental brightness change.

[marmad]

IMO 64 levels from off to full brightness sounds like a lot to me.  I would guess that our eye has a hard time discerning a 1/64 incremental brightness change.

I'm not so sure about that. I seem to recall that early monochrome displays had <= 64 levels of brightness - and it was slowly increased over time as a result of the non-smooth transition between gradations. But I can't really remember the details - perhaps I need to do some research.

Edit: A quick scan of literature seems to present varying information - but there seems to be some consistency to the data that the naked eye is sensitive up to 1% of intensity.

[Wuerstchenhund]

Perhaps this is just a problem with semantics? Certainly you don't mean that the display controller IC affects the acquisition cycle?

Yes, it does (on most modern DSOs). The wfm rate is essentially the speed the DSOs processing element reads out the acquisition memory. You can see this when using lots of additional processing as the wfm rate goes down somewhat on most scopes.

The image you brought up looks like it's a very old marketing drawing from Tek's first DPO scopes (which came up around a decade ago) and is not necessarily adequate for modern DSOs.

Could you elaborate on the better tools available for finding (or seeing) unknown glitches? I'm being serious - it's true that I'm reasonably new to DSOs - although I do think I understand how they work 

For example, if I want to find out if there are non-conformities (i.e. glitches, jitter) in a signal, I just set a mask and the allowed deviation and then set the scope to trigger on non-conformities. This works for repetitive and non-repetitive signals (i.e. a serial data bus). That means if there's a glitch then I'll see it. I can also segment the 1Gpts memory the scope has and capture multiple occurrences. Then depending on what the non-conformity is, I can run a series of analysing tools (i.e. jitter analysis, mathematical analysis) to quantify the problem, which makes it much easier to find what's causing it.

As I already said in this thread, I understand that for someone coming from an analog scope, a DSO is primarily a device to look at signals. However, for me it is (and always has been) a signal analyzer, a tool to examine and quantify signals. The fact that it has a screen just makes things easier. I am aware that you can't do that with an entry level (and even most lower midrange) DSOs, and all the high-end stuff at work does it because the company invested in most of the available options. But quite frankly, instead of this race for extreme wfm numbers that seems to start now thanks to Agilent, I'd rather see better signal analyis tools in a scope than a high wfm rate.

But YMMV of course, and I'm not the typical entry level scope customer (even for home use I'd rather buy older upper midrange/lower highend stuff than new entry level; funny enough, many of the analyis tools that are in my expensive work scopes I have now at home through the old LeCroy Waverunner I bought) so it's probably just me being awkward.

[Gunb]

Hi marmad,

independent from all the other things discussed here, could not follow any comment (seems, my personal update rate is too low ),
did you notice which requirements RIGOL has noted in their technical specs at the end of the manual?

For the DS4000 types they've required a sine of 10MHz, setting the scope to 10ns/DIV ...
Just for fun I've set up the generator & scope according to that data and noticed first, that wfm/s was about 80.000wfm/s. Then I've
noticed that they required to switch off interpolation, i.e. using "dots mode" for the display - result: the counter jumped to 115.000 wfm/s.

Logical that the drawing of connections between the samples may slow down the wfm/s, but didn't think about it before. So if maximum wfm/s
is required makes sense to switch off interpolation. Would say, that then the higher sample rate of 4GS/s makes sense also for a scope of 100MHz
bandwidth. There's not really more information, but you get more samples and the trace looks more continuously even with interpolation off
as with a scope for instance of 200MS/s, especially when zooming in.

But maybe you've done this test already and you're knowing about this details. Didn't notice that before, your video REVIEW of the DS2000
and measuring the wfm/s with the counter brought me to that. Seems to be interesting, that this test can't done with the Hameg HMO, its trigger
ouput is exactly the trigger of the signal, not the wfm/s.


Kind rgds,
Gunb

[Wuerstchenhund]

For the DS4000 types they've required a sine of 10MHz, setting the scope to 10ns/DIV ...
Just for fun I've set up the generator & scope according to that data and noticed first, that wfm/s was about 80.000wfm/s. Then I've noticed that they required to switch off interpolation, i.e. using "dots mode" for the display - result: the counter jumped to 115.000 wfm/s.

Which is understandable, since (like I said) the scope's processing limits the wfm rate.

I don't know what features your DS4000 has but you could try to to enable some more processing functions (i.e. FFT) and you will probably see that the wfm rate drops even more.

Seems to be interesting, that this test can't done with the Hameg HMO, its trigger ouput is exactly the trigger of the signal, not the wfm/s.

It's probably a different design choice where the acquisition does not stop after a cycle to wait for the scope processing to deal with the sample data. I haven't seen many scopes with such a design, though.

[Gunb]

Which is understandable, since (like I said) the scope's processing limits the wfm rate.

Exactly what I've mentioned before. I did not read your comment, without beeing inpolite, ony a  lack of time - sorry.

It's probably a different design choice where the acquisition does not stop after a cycle to wait for the scope processing to deal with the sample data. I haven't seen many scopes with such a design, though.

Of couse, what else?! You're right.


Kind rgds
Gunb

[marmad]

Yes, it does (on most modern DSOs). The wfm rate is essentially the speed the DSOs processing element reads out the acquisition memory. You can see this when using lots of additional processing as the wfm rate goes down somewhat on most scopes.

Well, this is certainly a problem of semantics - perhaps related to circuit design. Obviously, when you do things with the DSO that require processing time, you can affect the acquisition rate. But when I talk about display controllers, I'm talking about the actual silicon which drives the panel - which is unrelated to processing or acquisition. You were clearly referring to a block of circuitry encompassing  many components, including an ASIC or FPGA which is often shifting the display data to the LCD display controller. BTW, at least on some modern DSOs (such as the Agilent X series - and I'm guessing a few more as well), the main processor is not directly connected to acquisition memory.

For example, if I want to find out if there are non-conformities (i.e. glitches, jitter) in a signal, I just set a mask and the allowed deviation and then set the scope to trigger on non-conformities. This works for repetitive and non-repetitive signals (i.e. a serial data bus). That means if there's a glitch then I'll see it. I can also segment the 1Gpts memory the scope has and capture multiple occurrences. Then depending on what the non-conformity is, I can run a series of analysing tools (i.e. jitter analysis, mathematical analysis) to quantify the problem, which makes it much easier to find what's causing it.

Yes, the Rigol does all of these things - although it's analysis of segmented memory captures is limited to mask testing or fairly simple template comparisons. And these are certainly valuable tools for checking signal conformity. OTOH, they all take a certain amount of time - to set up and to run - which starts to add up when many signals are involved. Sometimes for me - and I imagine others that work with circuit design and testing - first looking at real or potential problem signals before more in-depth analysis is a quicker way towards the same point. That is where the higher wfrm/s rate and intensity grading is handy. And I have yet to hear any DSO users say that they didn't like the detail which the intensity grading provided and would return to using slower wfrm/s DSOs - but who knows, maybe they're out there?    Anyway, it seems we have different needs with our equipment - although I certainly agree with you that I'd like to see better analysis tools - especially since they could be provided as external software.

[marmad]

But maybe you've done this test already and you're knowing about this details. Didn't notice that before, your video REVIEW of the DS2000
and measuring the wfm/s with the counter brought me to that.

I never tested this with my DS2000 during the review - I was able to (just) get to the rated 50,000 wfrm/s with sin(x)/x interpolation turned on. But I just ran a very quick test right now using a 1MHz sine wave at 20ns with interpolation off and the rate was 52,780 wfrm/s.

[jpb]

The thing that struck me was that this very long discharge time  will mean that in the example given the next waveform cannot be displayed for 5 microseconds so
the waveform update rate will be limited to 200,000. OK that is quite a high upper limit but it is only 1/5 of Agilent's 3000X series. Also at slower sample rates the
charging time and thus discharging time would correspondingly increase.

Interesting. I wonder if the InfiniiVision X series use a digital trigger - I did a quick Googling but couldn't find a definitive answer - but I would guess they probably do. I know the Rigol UltraVision line does. According to this Rohde & Schwarz paper, one of the advantages of their digital trigger is "No Masking of Trigger Event":

"An analog trigger requires some time after a trigger decision to rearm the trigger circuitry before they can trigger again. During this rearm time, the oscilloscopes cannot respond to new trigger events - trigger events occurring during the rearm time are masked.
In contrast the digital trigger system of the R&S RTO oscilloscopes can evaluate individual trigger events with the Time-to-Digital-Converters (TDC) within 400 ps intervals (Figure 12) with a resolution of 250 fs."

A 400ps interval would put the limit of trigger events at 2.5 meg.

Thanks for the link.

I've modified my original post as I wrote it before reading the app note properly.

With a digital trigger only the samples (plus an extra one in-between) are looked at. The actual time of the trigger event is then got by interpolation. They claim this leads to less trigger jitter which seems odd to me as if the trigger points are either side of a narrow pulse I would have thought that there would be quite a lot of jitter introduce.

I can see the big advantage in the removal of blind time but I wonder if there are not some disadvantages which are not immediately obvious.


I find this fascinating as an engineer even though it probably has no practical relevance to my use of scopes.

[AndyC_772]

Here's an amusing observation...

My MSOX3054A arrived today, and I couldn't resist checking out the waveform update rate.

Fed with a 60 MHz sine wave to trigger on, it manages about 815,000 triggers/sec at 20ns/div.

Given that there are 10 divisions on the screen, that's a total displayed time of 200ns. Multiply by the waveform rate and we get 163msec, for a blind time of 83.7%.

...or do we?

It turns out that the 3000X is acquiring a lot more data than it's showing on the screen. If I stop the scope, I can zoom right out until the screen shows 500usec of data, and I can zoom in on that data and see any part of it in full resolution.

In other words, it's not triggering, acquiring, displaying and then overwriting - it's capturing all the time, and just picking out bits of the acquired waveform to display. It might not manage to show every possible trigger event on screen, but to say it's "blind" is being a bit harsh. *All* the data is there if you stop acquisition and search for it - and even that can be done automatically.

[marmad]

Sorry - I deleted my previous response because I thought some more about it and concluded you were right about the sample size.

It's a very interesting point that I hadn't considered before. I normally think of blind time as meaning the DSO is blind to data - but now that you mention this, it seems obvious that the strict definition is that it's blind to trigger events - the ADC and ASIC could continue to sample data even as display data is being processed or moved.

But I still think that 'blind' time is apt, because as I mentioned before, the main point is that you had to stop the DSO - so you weren't monitoring the wave real-time anymore (i.e. the DSO is 100% blind to any new triggers) - and if you didn't stop it, you would still only be able to monitor that small portion of non-blind time.

But I think the bigger question here is: how the hell can one plunk $12k down for an MSO? 

[AndyC_772]

I think you're right. If I put the scope in normal trigger mode and then stop the signal generator before stopping the scope, the extra data isn't available. What's on the screen is all you get. I still think it's a cool feature, though, and a nice touch that I wasn't expecting and wouldn't have missed if they hadn't bothered.

ps. just seen you've deleted your previous post, which is a shame because I think you got it spot on!

[AndyC_772]

But I think the bigger question here is: how the hell can one plunk $12k down for an MSO? 

Patience, keen observation, hard work, and since I do this stuff for a living, tax relief on capital equipment for my business

[grego]

But I think the bigger question here is: how the hell can one plunk $12k down for an MSO? 

Patience, keen observation, hard work, and since I do this stuff for a living, tax relief on capital equipment for my business

Wow because that used scope refurb from Agilent right now is listed at $8000. They accepted an offer like that?  Zoikes!

[marmad]

I think you're right. If I put the scope in normal trigger mode and then stop the signal generator before stopping the scope, the extra data isn't available. What's on the screen is all you get. I still think it's a cool feature, though, and a nice touch that I wasn't expecting and wouldn't have missed if they hadn't bothered.

ps. just seen you've deleted your previous post, which is a shame because I think you got it spot on!

Yeah, sorry, I've been editing and re-editing as my thoughts change about it. For any others who might read this later, my original response was:

The operative word here is 'stop'. I don't know what the Agilent does precisely, but it sounds as if when you tell it to stop acquiring, it fills the entire memory (500us / (1 / 4GSa/s) = 2M) on the final acquire. Why do I think it's the final acquire? Because the rated speed of the Agilent X3000 at 50us/div (500us) is 780 wfrm/s - with a blind time of 61% (at 50us/div).

But I'm not so sure about that anymore - I thought I remember reading or hearing that the X series used the full sample memory even when looking at a smaller subset. Any other owners want to chime in? But in either case, it's very interesting because it clarifies what blind time really means.

Patience, keen observation, hard work, and since I do this stuff for a living, tax relief on capital equipment for my business

Oooh..... unfortunately there's no emoticon for drooling - so I'll have to make do with 

[marmad]

@Andy - does the sample rate of the Agilent first drop to 2GSa/s when you set the timebase to 100us?

[AndyC_772]

Yes 

[marmad]

Yes 

I really think the MegaZoom ASIC has been designed to keep filling the entire memory post-trigger at the fastest sampling rate unless re-triggered (or unless you are using memory in segmented mode). During the processing and display of the current contents of the sample database (the 'intensity map'), it's blind to triggers - but perhaps it keeps sampling into a buffer.

[Bitstream]

FWIW.
I was curious how the trigger update rates and trigger blind time (aka dead time) compared for analog vs. digital scopes.  I ran tests on the following scopes:

*Rigol DS2202 200MHz digital scope
*Tek 647A 100MHz analog scope
*Tek 7104 1GHz analog scope

The Tek 647A and Tek 7104 both include a GATE_OUT signal so the trigger rate can be conveniently monitored with a counter.

In general, even analog scopes have dead time when the CRT beam is blanked and moved back to the left side of the CRT in preparation for the next trigger.  What's interesting is that the analog scopes (at least these 2) do not always have lower trigger blind time compared to the digital scope.  In general, the analog scope are better at the faster timebases but start to lose to the digital scope at around 20uS.  Newer analog scope may be quicker but I don't have others to test.

To calculate the blind time at each timebase setting, I measured the trigger update rate and used the formulas in this Agilent app note:
http://www.home.agilent.com/agilent/redirector.jspx?action=ref&cname=AGILENT_EDITORIAL&ckey=1374518&lc=eng&cc=US&nfr=-11143.0.00

[marmad]

To calculate the blind time at each timebase setting, I measured the trigger update rate and used the formulas in this Agilent app note:
http://www.home.agilent.com/agilent/redirector.jspx?action=ref&cname=AGILENT_EDITORIAL&ckey=1374518&lc=eng&cc=US&nfr=-11143.0.00

Very interesting! Thanks for your time and efforts - I always wondered about the comparison to analog scopes.

[Bitstream]

You're welcome.  Another thing I found interested while collecting the data is that the Tek 7104 scope from the '70s turns in a respectable 560,000 triggers/sec at <=10nS timebase.  It must be the magic in those Tek hybrid chips 

[electronics man]

I measured 25,000 wfm/s on my rigol ds1074z, will give it another go to see if I can get it to its max of 30000

[David Hess]

I've just read an application note on Random Interleaved Sampling and this refers to a component of the DSO called the Time to Digital Converter (TDC).

The TDC is used to place the trigger point very accurately between the actual samples (so that on multiple triggers you don't introduce a jitter equal to the sample rate).

The reason that I'm raising this in this thread is it suddenly struck me that the TDC will place an upper limit on the waveform update rate.

The TDC works by charging up from the start of the trigger to the next sample point, which will be a very short time - the maximum will be the sample to sample time which
in the app note is a maximum of 100 psecs for a 10GS/s scope.

The TDC then discharges over a much longer time which is measured digitally using a clock, 100MHz in the app note. The discharge time in the app note is a maximum of
5 microseconds - so 100 psecs to charge and 5 microseconds to discharge. This allows the time to be set to less than a picosecond (in the app note example).

The thing that struck me was that this very long discharge time  will mean that in the example given the next waveform cannot be displayed for 5 microseconds so
the waveform update rate will be limited to 200,000. OK that is quite a high upper limit but it is only 1/5 of Agilent's 3000X series. Also at slower sample rates the
charging time and thus discharging time would correspondingly increase.

I suppose it is possible to store and work on triggers and sample sets in parallel but this is added complication that most manufacturers wouldn't do.

It might also be that there is a trade off between wfps and timing accuracy.

This depends on the TDC implementation.  Even an integrating TDC can achieve better than 10 picosecond resolution at 50k+ waveforms per second but not by pulse stretching.  DSOs that use slow TDCs do so because they have other limitations.

[David Hess]

The thing that struck me was that this very long discharge time  will mean that in the example given the next waveform cannot be displayed for 5 microseconds so
the waveform update rate will be limited to 200,000. OK that is quite a high upper limit but it is only 1/5 of Agilent's 3000X series. Also at slower sample rates the
charging time and thus discharging time would correspondingly increase.

Interesting. I wonder if the InfiniiVision X series use a digital trigger - I did a quick Googling but couldn't find a definitive answer - but I would guess they probably do. I know the Rigol UltraVision line does. According to this Rohde & Schwarz paper, one of the advantages of their digital trigger is "No Masking of Trigger Event":

"An analog trigger requires some time after a trigger decision to rearm the trigger circuitry before they can trigger again. During this rearm time, the oscilloscopes cannot respond to new trigger events - trigger events occurring during the rearm time are masked.
In contrast the digital trigger system of the R&S RTO oscilloscopes can evaluate individual trigger events with the Time-to-Digital-Converters (TDC) within 400 ps intervals (Figure 12) with a resolution of 250 fs."

A 400ps interval would put the limit of trigger events at 2.5 meg.

The analog trigger rearm rate depends on the sweep reset time and not the trigger itself except insofar as it has to detect the other slope before arming.  It is quite practical to have triggers which operate into the 250 MHz+ range and some oscilloscopes did this because the trigger was used for other things as well like to drive a frequency counter.

Digital triggers are just a variation of a transition midpoint timing TDC.