Author Topic: New Rigol DS1054Z oscilloscope  (Read 1435857 times)

0 Members and 1 Guest are viewing this topic.

Offline rf-loop

  • Super Contributor
  • ***
  • Posts: 3043
  • Country: fi
  • Starting with DLL21
Re: New Rigol DS1054Z oscilloscope
« Reply #4300 on: September 12, 2018, 11:48:54 pm »


If Photoshop a little rf-loop example it can be seen that averaged low-samplerate ack could not ever match true risetime:





What averaged?  There is nothing averaged. Bright sinc is just one random acquistion and scope draw (do)
 Sinc interpolation using just last acquisition data. It is not average. (btw, in scope there is selection, acquisition mode fast or slow) and then behind there is persistence what show all area where these single Sinc interpolated waveforms have been in persistence collecting time.  (and same principle in dots mode. Bright dots are just one (last) acquisition and persistence have collected more)

But it is not at all good to make this kind of analyze from these images. These are not made for analyze signals. These are made for explannate how scope works in basics and for this we need simplified situations before go to more high details in theory and real practice. You know... most easy way climb a tree is starting from ground.
« Last Edit: September 12, 2018, 11:50:47 pm by rf-loop »
If practice and theory is not equal it tells that used application of theory  is wrong or the theory itself is wrong.
It is much easier to think an apple fall to the ground than to think that the earth and the apple will begin to move toward each other and collide.
 

Offline MrW0lf

  • Frequent Contributor
  • **
  • Posts: 921
  • Country: ee
    • lab!fyi
Re: New Rigol DS1054Z oscilloscope
« Reply #4301 on: September 12, 2018, 11:57:15 pm »
What averaged?  There is nothing averaged.

I never said there is, but if it would be, then fall would be less steep than actual signal.
Never mind I'm already firing up most of my gear to do some checks.
Interesting stuff.

Disclaimer before anyone gets agitated again: Z is ok for tinkering with Arduino. Discussion is about (quite large) differences between scopes when start pushing it to the limit eg move far beyond Arduino.
 

Offline rf-loop

  • Super Contributor
  • ***
  • Posts: 3043
  • Country: fi
  • Starting with DLL21
Re: New Rigol DS1054Z oscilloscope
« Reply #4302 on: September 12, 2018, 11:58:45 pm »


Even at 100MSa/s there will be 99 more samples between each dot in the image, sinc will be fine with that.

No (if we stay here in entry or near entry level scopes segment)

If practice and theory is not equal it tells that used application of theory  is wrong or the theory itself is wrong.
It is much easier to think an apple fall to the ground than to think that the earth and the apple will begin to move toward each other and collide.
 

Offline TurboTom

  • Frequent Contributor
  • **
  • Posts: 544
  • Country: de
Re: New Rigol DS1054Z oscilloscope
« Reply #4303 on: September 13, 2018, 12:34:17 am »

Me? I'm not doing anything.


Me neither...almost  ;)

Quote
Edit: Maybe that's the true rise time of a pin on an Arduino Uno after all the capacitances, etc. have been taken into account. We need somebody with a better oscilloscope to measure one.

Things get even more complex and unclear when the rise times get faster, and since this thread turned into a completely academic discussion (with some not-so-academic reproaches once in a while...), let's put some more petrol on the fire  8)

First I tried to replicate Fungus's tests with the original Rigol probe, dot mode and infinite persitance. Since at the performance of the DS1000Z, Leo Bodnar's fast edge oscillator (<50ps) hasn't much advantage against the digital outputs of my rubidium source (<200ps), but the latter provides full 5V amplitude, I used that one as a source throughout these tests. I connected the Rigol probe (RP2200) set to 10:1 via a proper probe/BNC adapter to the 5V digital output of the Rb source. Screenshots "Quickprint 2~5" were taken this way. Rise times differ by about 1.5ns against Fungus's tests but the shape of the edge differs considerably. I'm very surprised by the sharp edges that my DS1000Z shows at 250Msps.

After that I went to cancel out the effect of the probe and connected the Rb source via a 40cm quality BNC cable and a 50Ohm feed-through terminator to the scope's input. Obviously, the amplitude is now reduced by half, hence the sensitivity was adjusted to 500mV/div (actually, it's 5 times as much signal at the scope's input since the probe in the first test had an attenuation of 10:1). Whatsoever, I ran the same tests again, but this time I also checked the results in vector mode. And now things really start to get interesting.

In dot mode, the individual samples more or less group on a line with only noise-related jitter/stray width, regardless if sin(x)/x interpolation is selected or not. But as soon as vector mode is enabled, at the low sampling rates there's now considerable modulation in the over/undershoot areas present that gets better only after channels are turned off to increase sampling rate. This means that Rigol has got to be using different interpolation algorithms in dot mode and vector mode. This also raises the question what signal they use in the second stage (screen adjustment) trigger in the two display modes.

Actually, I'm even more suspicious about the performance of the analog front end of the DS1000Z. It appears that the front end is slew rate limited at high(er) signal levels at the gate of the first JFET. And I assume that Rigol uses some kind of digital high-pass filter after the ADC (maybe even a non-linear one, depending on the actual signal level and the sensitivity setting) to compensate for this. If time permits, I may actaully do some measurements inside my DS1000Z.

But as mentioned before, all this is more or less nonsense, it doesn't improve the scope at all and to be honest, it's not really worth spending that amount of time and effort on an instrument of that low price. But anyway, it's fun trying to understand what's going on inside and how Rigol managed to turn an apparently (less than) mediocre harware into something useful (more or less so, some may argue...).

Finally I attached a screenshot of the aliasing frequency of the Rb source signal at low sample rate to get an idea of the accuracy of the internal reference -- I'ld say it's also mediocre at 11ppm (at least for my DS1000Z), YMMV  :)

And sorry for the long post...
 

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 9655
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4304 on: September 13, 2018, 01:17:30 am »
Edit: Maybe that's the true rise time of a pin on an Arduino Uno after all the capacitances, etc. have been taken into account. We need somebody with a better oscilloscope to measure one.

Rise times differ by about 1.5ns against Fungus's tests but the shape of the edge differs considerably. I'm very surprised by the sharp edges that my DS1000Z shows at 250Msps.

This confirms what I was seeing (sort of). When I improved my probing, sin(x)/x started to win out over whatever it is that Rigol does when you turn it off.

With your much faster pulse and better probing you got this:

(which is hard to explain, I need to think about that one)


This (and other images) seems to show that your analog bandwidth is near 200MHz. Maybe you got lucky in the front-end component lottery! (do you get the same rise time on other channels?):


Finally: Your images make me suspect that the Arduino pulse is indeed around 4.5ns.

« Last Edit: September 13, 2018, 01:30:36 am by Fungus »
 

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 9655
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4305 on: September 13, 2018, 01:24:28 am »
Finally I attached a screenshot of the aliasing frequency of the Rb source signal at low sample rate to get an idea of the accuracy of the internal reference -- I'ld say it's also mediocre at 11ppm (at least for my DS1000Z), YMMV  :)

That's using the on-screen data to calculate it. What do you get if yo turn on the hardware frequency counter?

(Measure->Counter)

it's fun trying to understand what's going on inside and how Rigol managed to turn an apparently (less than) mediocre harware into something useful

There's definitely some 'impossible' performance numbers appearing in these tests.
« Last Edit: September 13, 2018, 01:29:09 am by Fungus »
 

Offline TurboTom

  • Frequent Contributor
  • **
  • Posts: 544
  • Country: de
Re: New Rigol DS1054Z oscilloscope
« Reply #4306 on: September 13, 2018, 01:49:55 am »
There's no noticable difference of the performance between the individual channels of my DS1000Z.

Oh yes, the hardware frequency counter more or less confirms the accuracy check of the aliasing interference frequency measurement, reading 10.0001MHz, that is if you want to believe in one least significant digit. Basically, the resolution of the hardware counter is just a little low to provide useful information. Two more digits (down to 1Hz resolution) would change this considerably... That makes the interference measurement so attractive (had been presented by @besauk here: https://www.eevblog.com/forum/testgear/how-to-quickly-determine-your-dso-timebase-accuracy/msg1748363/#msg1748363 ).

Quote

Finally: Your images make me suspect that the Arduino pulse is indeed around 4.5ns.


I think the RP2200 probe may also make some difference as my tests with the BNC cable show, even though the settings of the 'scopes input had to be changed to do the measurements in the different configurations. But yes, the arduino I/O line is most likely also not as fast as the digital outputs of my Rb source (I purposely designed them for high speed so I could directly use them for reflection measurements and the like...).
« Last Edit: September 13, 2018, 02:22:57 am by TurboTom »
 

Offline maukka

  • Supporter
  • ****
  • Posts: 84
  • Country: fi
Re: New Rigol DS1054Z oscilloscope
« Reply #4307 on: September 13, 2018, 02:35:25 am »
But this material where from images are is finnish language so you can not understand it due to fact that finnish language is like "enigma" for example google translator. And I have also explained previously why I pick up these images here. If you really do not understand then tell it directly that you did not understand instead if this indirect way tell same.

This (one part) of "teaching" material for peoples who do not have so much knowledge about digital oscilloscopes. Some kind of common bottom entry level information. Examples are made using Siglent but all these can do using nearly what ever this era entry level digital oscilloscope.

To be fair, that site also looks like it was translated to Finnish by Google.  ;D
 

Offline MrW0lf

  • Frequent Contributor
  • **
  • Posts: 921
  • Country: ee
    • lab!fyi
Re: New Rigol DS1054Z oscilloscope
« Reply #4308 on: September 13, 2018, 02:45:02 am »
Pretty crazy results by TurboTom :-+

I would comment on 250MSa/s Sinc=ON|OFF performance on these 2 images:





Look crazy overshoot when Sinc=ON and clear signs of Sinc or similar interpolation when OFF! In OFF state it looks more-less like it should in ON state.

Now look what I get on my scope with 50ns pulser, 20GSa/s ETS (blue) vs 250MSa/s RTS (cyan) averaged & Sinc'd.
Overshoot is quite modest.



Now here are all basic modes with measurements averaged & Sinc'd:



Same modes averaged & linear interpolation:



It refuses to measure rise with linear interpolation and low sample count, and of course no overshoot at all.
With Sinc interpolation rise times scale logically and there is overshoot but it is rather tiny, worst case 5% from amplitude.

Now I ask - math is math. How on earth Rigol manages to overshoot like crazy at 250MSa/s: 14% with Sinc=ON, 6% Sinc=OFF (there shouldn't be even mathematical source here!). Also at 500MSa/s it manages 7%. As can see at 1GSa/s there is no physical cause in sight.





« Last Edit: September 13, 2018, 02:53:34 am by MrW0lf »
 

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 9655
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4309 on: September 13, 2018, 02:52:50 am »
Now I ask - math is math. How on earth Rigol manages to overshoot like crazy at 250MSa/s: 14% at  Sinc=ON, 6% Sinc=OFF

The 14% is a sum of real electrical overshoot + Gibbs + aliasing.

Sinc = off? Electrical overshoot + only Rigol knows what.

at 500MSa/s it manages 7%

Gibbs/aliasing greatly reduced.
 

Offline MrW0lf

  • Frequent Contributor
  • **
  • Posts: 921
  • Country: ee
    • lab!fyi
Re: New Rigol DS1054Z oscilloscope
« Reply #4310 on: September 13, 2018, 03:02:52 am »
The 14% is a sum of real electrical overshoot + Gibbs + aliasing.

Sinc = off? Electrical overshoot + only Rigol knows what.

Look image below. There is no electrical / physical overshoot:

 

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 9655
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4311 on: September 13, 2018, 03:06:48 am »
The 14% is a sum of real electrical overshoot + Gibbs + aliasing.

Sinc = off? Electrical overshoot + only Rigol knows what.

Look image below. There is no electrical / physical overshoot:

So ... that parameter will be zero.  :popcorn:
 

Offline MrW0lf

  • Frequent Contributor
  • **
  • Posts: 921
  • Country: ee
    • lab!fyi
Re: New Rigol DS1054Z oscilloscope
« Reply #4312 on: September 13, 2018, 04:41:52 am »
The 14% is a sum of real electrical overshoot + Gibbs + aliasing.

Sinc = off? Electrical overshoot + only Rigol knows what.

Look image below. There is no electrical / physical overshoot:

So ... that parameter will be zero.  :popcorn:

Not much could be attributed to Sinc math either. Special content for mr Fungus produced by most academic scope I have. 50ns pulser hitting AD2. Cyan trace (R1) is 16x oversampling eg pretty much what frontend sees. Yellow tace (C1) is 100MSa/s RTS, Sinc'd, 1000x averaged. Effect on risetime is severe exactly like expected (8.9ns edge sampled at 100MSa/s turns to 17.3ns). Yet overshoot is 1% :D
So you cannot attribute math buzzwords to all the magic you see on the screen ;)



« Last Edit: September 13, 2018, 04:45:11 am by MrW0lf »
 

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 9655
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4313 on: September 13, 2018, 05:03:58 am »
...all the magic you see on the screen ;)

"Magic"? I think you're starting to like the DS1054Z.

Maybe your previous numbers didn't tell you everything.  :popcorn:
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 9533
  • Country: us
  • DavidH
Re: New Rigol DS1054Z oscilloscope
« Reply #4314 on: September 13, 2018, 02:18:13 pm »
Also can see that without Sinc there is not so much wobbling in corners and edges upside and downside from trig level. Also this is normal ok. This is one reason why it is good that oscilloscope can always turn to true sample dots without any interpolation (poormans partially emulated RIS mode what is good for some kind of waveforms where Sinc is not best  and vice versa.) (of course separate trigger engine interpolation is working if or when it exists independent of displayed wfm interpolation) endependent of what example Keysight say or advertise or produce.

The wobbling (1) is because even with an input signal completely within the Nyquist bandwidth, the non-linearity in the digitizer mixes the signal and sample frequencies producing mixing products above the Nyquist frequency and then sin(x)/x reconstruction fails because essentially there are multiple solutions.  Just the aliasing of the input signal by itself will not do that (2) although I think it would screw up digital triggering.  It can be thought of as looking as a lightly modulated AM signal.

This is why I much prefer ETS which raises the sampling rate so high that the mixing produces no aliasing and as a bonus, there is none of that pre-shoot and post-shoot nonsense.  Averaging helps if you have nothing better but not as much as ETS.

Your later example on the Siglent shows it better and it is exactly what I see on ancient DSOs with sin(x)/x interpolation and no ETS.  HP/Agilent/Keysight likes to make fun of Tektronix's interleaved digitizers which tend to suffer more from this due to poorer linearity.  Even Rigol seems pretty good in this respect except for that 100 to 200 MHz overload problem in the analog signal conditioning.

(1) I've been calling this wobulation.  If you see it, then your have exceeded the capabilities of your DSO.
 

Offline rf-loop

  • Super Contributor
  • ***
  • Posts: 3043
  • Country: fi
  • Starting with DLL21
Re: New Rigol DS1054Z oscilloscope
« Reply #4315 on: September 13, 2018, 04:49:38 pm »
Also can see that without Sinc there is not so much wobbling in corners and edges upside and downside from trig level. Also this is normal ok. This is one reason why it is good that oscilloscope can always turn to true sample dots without any interpolation (poormans partially emulated RIS mode what is good for some kind of waveforms where Sinc is not best  and vice versa.) (of course separate trigger engine interpolation is working if or when it exists independent of displayed wfm interpolation) endependent of what example Keysight say or advertise or produce.

The wobbling (1) is because even with an input signal completely within the Nyquist bandwidth, the non-linearity in the digitizer mixes the signal and sample frequencies producing mixing products above the Nyquist frequency and then sin(x)/x reconstruction fails because essentially there are multiple solutions.  Just the aliasing of the input signal by itself will not do that (2) although I think it would screw up digital triggering.  It can be thought of as looking as a lightly modulated AM signal.

This is why I much prefer ETS which raises the sampling rate so high that the mixing produces no aliasing and as a bonus, there is none of that pre-shoot and post-shoot nonsense.  Averaging helps if you have nothing better but not as much as ETS.

Your later example on the Siglent shows it better and it is exactly what I see on ancient DSOs with sin(x)/x interpolation and no ETS.  HP/Agilent/Keysight likes to make fun of Tektronix's interleaved digitizers which tend to suffer more from this due to poorer linearity.  Even Rigol seems pretty good in this respect except for that 100 to 200 MHz overload problem in the analog signal conditioning.

(1) I've been calling this wobulation.  If you see it, then your have exceeded the capabilities of your DSO.

Yes but ETS is not real time scope at all. ETS can do even with 1Hz sampling speed for reconstruct 1GHz signal... if we look only alone fully repetitive signals... (who need scope only for these) ETS is ok.

When we need real time scope what can solve signals with full BW using one shot... we need also accept some interpolation methods produced not so nice things (exept if we use very high samplerate ADC vs BW what do not need interpolation at all or just tiny vectors but who want spend 10Gsa ADC and huge memory to 50M scope.. ). Btw, I have not seen so much corners wobbling if signal really meet Sinc rules. But then we need also rtemember that Sinc in usual scope can not go up to fNyquist. Not at all. Usually need stay under 0.8*fNyq for minimize sinewave amplitudce wobbling and other waveforms harmonics what ADC can see need also keep least under fNyq (of course amount they affect depends harmonics levels.) if not we see edges corners wobbling. It is one displayed form of aliasing.  As we know oscilloscopes Sinc can not be ideal... no one can wait until this calculus is ready. Because samples position related to signal is random then example with square wave etc sample position may be optimal or less optimal.  Averaging sequential acquisitions of course helps but.. anly for long time continuous repetitive waveforms what can solve using nearly what ever.  Sinc interpolation is filter and it is not perfect Sinc even when it draw via true sample points. If we separate one harmonic what is near fNyq this harmonic start show amplitude wobbling (bit like AM) and depending this Sinc reconstruction performance things (many compromises) also these are good to keep under around 0.8*fNyg. If we need very perfect result all what ADC can see need keep even more below fNyq exept if we start compute lot of sequantial samples ... up to fNyg we need infinite..
But then, some modern advanced scopes may have more inbulid DSP operations for reduce and "makeup" these problems, example Gibbs...  do that finally reconstruction is more perfect.

Then also there is other thing...  in digital trigger "engine".  Its own interpolation methods between true samples for fine adjust wfm trigger position between two samples. 

And then, if scope is using decimated samplerates when slow t/div and memory start reject "samplerate" what is pushed to acquisition memory  and from there collected (how ever it is done) to display memory. There, inside trigger sub system, it still can use example true ADC stream before decimation.
Example Siglent do it. So it is possible to run in SARI mode (Sequential Acquisitions Random Interleave). It can reconstruct signal far over fNyquist if run dots mode with decimated samplerates.

Perhaps even Siglent do not know there is this mode implemented. I have not seen any advertisement...

Here 3 images.




1. original signal with full samplerate and interpolations off.. just only true dots (if your browser/computer run these full speed also speed is around same what can see on scope screen irl.
Here sample dots are dense in every acq and also then lot of acqs overlaid and randomly interleaved so no need any interpolations.



2. Samplerate dropped low. Decimated samplerate 50MSa/s  (fNyq 25MHz) what is far below signal first harmonic aka base frequency 45MHz. This signal of course have lot of harmonics...example 5. is 225MHz
Display total mess and can not trig of course. There can see running some alias freq products like example 5MHz




3. Same as 2 but now... dots mode.  And this is fun. Acq memory samplerate 50MSa/s (fNyq 25MHz)
It can reconstruct  and show 45MHz 4ns risetime square. 50MSa/s! And this oscilloscope do not have ETS mode at all!
Just not advertised SARI mode. (yes also some other scope may have this)
Of course if this can do other way and with full ADC speed then it can be like LeCroy RIS mode or emulate it.
With further develop and finishing this can be useful when scope is working in decimated samplerates.

Data points time interval in acquisition memory is 20ns (one point 2div)
What happen. Acquisition and trig position detected between memory data points and quite accurate. Moved to display memory.
As soon as possible next acquistion positioned (note that signal position to samples is time random) and moved to display memory together with previous... now dots are in random time position... then next and next and next until display moved to TFT and whole new turn...
In this case with current acquistion speed is quite low and it can do around 13 - 15 acquisitions for one image where now these Sequential Acquistions are Randomly Interleaved (due to fact that signal and ADC is not in time sync).
If like it can call SARI mode. How many sequential acquistions (note1) is interleaved (overlaid) depends current  wfm/s speed. This is strong evidence that Siglent full digital side trigger engine use ADC true samples for trigger and fine position when operate under true ADC speed (1GSa or 500MSa/s).

Note1. These all sequential acquisitions can also find in waveform history buffer  separately and time stamped (time is not true real time due to lack of RTC what can set and what keep time - what a shame they did not implement battery back up RTC)
Just with this setting every last 3912 acquistion with true raw ADC data is stored to history buffer fifo for analyze with scope and if want look with Sinc, Vector or dots mode because both interpolations are fully post processed.

This feature have many limits but it can use and is also useful in some special cases when user is bit experienced with it.
It may work also in some other same kind of oscilloscope where user can select true Dots mode without interpolation.
« Last Edit: September 13, 2018, 07:05:14 pm by rf-loop »
If practice and theory is not equal it tells that used application of theory  is wrong or the theory itself is wrong.
It is much easier to think an apple fall to the ground than to think that the earth and the apple will begin to move toward each other and collide.
 
The following users thanked this post: tautech, MrW0lf

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 9533
  • Country: us
  • DavidH
Re: New Rigol DS1054Z oscilloscope
« Reply #4316 on: September 13, 2018, 11:52:12 pm »
Yes but ETS is not real time scope at all. ETS can do even with 1Hz sampling speed for reconstruct 1GHz signal... if we look only alone fully repetitive signals... (who need scope only for these) ETS is ok.

Averaging is not real time either but it is recommended for correcting the errors produced by sin(x)/x interpolation.

My preference for dealing with decimation for whatever reason is to capture a real time histogram and display that like DPO style DSOs do.  Rohde & Schwarz implemented a lessor version of this in their RTO series of DSOs.
 

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 9655
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4317 on: September 14, 2018, 12:23:53 am »
Example Siglent do it. So it is possible to run in SARI mode (Sequential Acquisitions Random Interleave). It can reconstruct signal far over fNyquist if run dots mode with decimated samplerates.

So can Rigol. This image was produced at full sample rate earlier in the thread:


« Last Edit: September 14, 2018, 12:25:29 am by Fungus »
 

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 9655
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4318 on: September 14, 2018, 12:31:42 am »
There's no noticable difference of the performance between the individual channels of my DS1000Z.

I've seen other people measuring close to 200Mhz measured bandwidth on their DS1054Z in the past but others have measured around 135MHz.

Maybe the higher numbers are more accurate and it all depends on the probing technique and signal generator used. I don't know.

I don't have to gear to test mine properly so I usually throw the number "130MHz" around just in case I get accused of being a crazy deluded Rigol fanboy (it happens!)
 

Offline MrW0lf

  • Frequent Contributor
  • **
  • Posts: 921
  • Country: ee
    • lab!fyi
Re: New Rigol DS1054Z oscilloscope
« Reply #4319 on: September 14, 2018, 12:59:55 am »
So can Rigol. This image was produced at full sample rate earlier in the thread:


Yea but no. It look like dot mode allright... But it sort of isnt, otherwise same signal would look just like here:



ETS, oversampling, SARI etc are about giving true analog scope like representation of (semi-)static signal. Here we see signal completely different between 2 sampling rates. Also we see degradation of rise time at 250MSa/s like it should, yet corners are ~90 degree, yet smooth corner seen at 1GSa/s is lost :-//
« Last Edit: September 14, 2018, 01:02:15 am by MrW0lf »
 

Offline rf-loop

  • Super Contributor
  • ***
  • Posts: 3043
  • Country: fi
  • Starting with DLL21
Re: New Rigol DS1054Z oscilloscope
« Reply #4320 on: September 14, 2018, 01:15:09 am »
Example Siglent do it. So it is possible to run in SARI mode (Sequential Acquisitions Random Interleave). It can reconstruct signal far over fNyquist if run dots mode with decimated samplerates.

So can Rigol. This image was produced at full sample rate earlier in the thread:


Totally different thing.
You did not note that in my example whole 45MHz signal whole cycle is near sample interval (decimated samplerate as you can see was 50MSa/s and square wave is 45MHz. So whole cycle have only bit over one sample in one trigged acquisition.

Perhaps you forget thinking or reading what you comment.

It looks like I have not told and show this special case things so that also you can understand.
 
« Last Edit: September 14, 2018, 01:21:26 am by rf-loop »
If practice and theory is not equal it tells that used application of theory  is wrong or the theory itself is wrong.
It is much easier to think an apple fall to the ground than to think that the earth and the apple will begin to move toward each other and collide.
 

Offline rf-loop

  • Super Contributor
  • ***
  • Posts: 3043
  • Country: fi
  • Starting with DLL21
Re: New Rigol DS1054Z oscilloscope
« Reply #4321 on: September 14, 2018, 02:01:59 am »
Here we see signal completely different between 2 sampling rates. Also we see degradation of rise time at 250MSa/s like it should, yet corners are ~90 degree, yet smooth corner seen at 1GSa/s is lost :-//

It need understand how trigger engine works and how it fine adjust dots stream time position in every single trigged horizontal  acquisition (aka horizontal sweep if think analog scope)

facts and fiction:
This "triger engine" is perhaps partially guilty who produce fake signal shape. (250MSa/s there is 4ns sampling period! Remember it and then think what happen in trigger engine when it is waiting and then... next some sample cross over trig level.  First thing it do is trying find when signal goes over trig level then interpolate using other nearest sample(s). After then all samples need shift (or keep unshifted in rare cases) for adjust so that "interpolated part of wave" (between sample points) trigger position is in right place what is interpolated but not necessary true right. But of course also all samples are now shifted  they still keep this level what they have in these places where they are sampled...even when interpolated trigger position is not  but because do not know exactly details how Rigol trigger engine works can not know exactly what happen. Only can tell that this image where is   nearly sharp 90degree corners is false (it can know because these corners are totally impossible in practice and theory with its analog front end.  (and I suspect it is jut because nature of trigger engine what can not even in theory jump over this problem without very different design) But This is more or less close true. I do not have Rigol for do some special tests so that can find more truth or flush out wrong imaginations about how it works. 
(btw also there in Siglent can find this phenomena. Also @Performa01 have previoulsy little scratched the surface of this matter and wondered what it was all about long time ago with SDS2000X and later in SDS1000X-E review)
« Last Edit: September 14, 2018, 02:13:03 am by rf-loop »
If practice and theory is not equal it tells that used application of theory  is wrong or the theory itself is wrong.
It is much easier to think an apple fall to the ground than to think that the earth and the apple will begin to move toward each other and collide.
 
The following users thanked this post: MrW0lf

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 9655
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4322 on: September 14, 2018, 03:51:04 am »
I love how you assume RIGOL must be wrong.

Maybe this image is truly showing RAW samples, just like the Siglent (eg. the 'scope might know that it's on the extreme limit and switch off reconstruction in dots mode). It's the only explanation that makes sense. This image is impossible otherwise:


With 1GSa/s it might decide that there's enough samples for the analog bandwidth and turn that off. Sin(x)/x gives us this image:


It sucks that we don't have more options to play with to characterize the front end but Rigol doesn't design for the volt-heads. The way it works gives good results for everybody else.  :-//
« Last Edit: September 14, 2018, 03:52:36 am by Fungus »
 

Offline MrW0lf

  • Frequent Contributor
  • **
  • Posts: 921
  • Country: ee
    • lab!fyi
Re: New Rigol DS1054Z oscilloscope
« Reply #4323 on: September 14, 2018, 04:23:31 am »
This "triger engine" is perhaps partially guilty who produce fake signal shape.

This is very useful lead. I tried to "attack" trigger on my AD2 and soon enough it was very "broken" and oversampling mode was rendered incapable:



Now AD2 is honest academic scope and does not try to hide that it cannot accurately trigger. On persistence can see huge amount of jitter. Grey trace is averaged one and blue is oversampling (800MSa/s). Signal fed in is 30ns pulse with 8.4ns edges. Scope sampling rate 50MSa/s.

If some clever but little clumsy programmer would now try to hide this jitter it could end in some puzzling visual phenomenon :P

Edit: If trigger engine would run at full ADC speed (100MSa/s) it would be all much better. Now if you say Siglent trigger actually uses full ADC speed under similar condition then it kicks some serious ass :-+ They'd better get busy polishing that SARI thing ASAP...
« Last Edit: September 14, 2018, 04:33:19 am by MrW0lf »
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 9533
  • Country: us
  • DavidH
Re: New Rigol DS1054Z oscilloscope
« Reply #4324 on: September 16, 2018, 12:47:14 pm »
I've seen other people measuring close to 200Mhz measured bandwidth on their DS1054Z in the past but others have measured around 135MHz.

Maybe the higher numbers are more accurate and it all depends on the probing technique and signal generator used. I don't know.

TurboTom's test results here show that bandwidth depends on vertical sensitivity which should not happen nearly to the extent shown.  (1) My own observations of tests done at high signal levels show slew rating limiting which would cause variation in full power bandwidth which is a different problem but it would also affect measured bandwidth under certain conditions.

Neither should happen in a properly designed oscilloscope.  Rigol screwed up.

(1) Some oscilloscopes have switched gain stages at their highest sensitivities and a few of these may drop in bandwidth significantly but it was well known limitation and the specifications included it.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf