Author Topic: New Rigol DS1054Z oscilloscope  (Read 1446510 times)

0 Members and 4 Guests are viewing this topic.

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4225 on: September 08, 2018, 09:17:32 pm »
You see this trace, and I will save it as CSV screen data. What I should get? I get all 4 channels and 1200 points.

Correct.

When I change only the data source to memory, I get 15 points, ONLY channel 1 (when ch3 is active)

My copy of the manual says:



To me this suggests it's only going to save data for one channel in that mode.

« Last Edit: September 08, 2018, 09:24:33 pm by Fungus »
 

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4226 on: September 08, 2018, 09:20:22 pm »
looks like a bug. I shut off other 3 channels and saved memory (now 60pts), and the wave represents.

When you shut off 3 channels the 'scope will force sin(x)/x on and it will have more samples/MHz.

For this sort of work you should always try to maximize the sample rate, on a Rigol DS1054Z that might mean using one channel.
« Last Edit: September 08, 2018, 09:45:40 pm by Fungus »
 

Offline MrW0lf

  • Frequent Contributor
  • **
  • Posts: 921
  • Country: ee
    • lab!fyi
Re: New Rigol DS1054Z oscilloscope
« Reply #4227 on: September 09, 2018, 06:56:37 am »
In general, manual says that Sinc is "optional" and in the menu there is clear option Sinc=OFF. So it is fair assumption that it should be OFF, not replaced by god knows what.
Says who?

So are you saying that expecting that OFF is OFF is unfair assumption :-DD Seems working with so specific equipment has left its mark on you :)

Maybe it works like that $12,000 keysight 'scope I mentioned earlier:

This is ancient technology, also it can be assumed that on Keysight OFF is OFF, they do tend to have classical front end not needing special kinds of post-processing...

As for "screen sampling", even Rigol itself has moved on:



Maybe they'll give you one then can work with more than 1 channel ;)

« Last Edit: September 09, 2018, 07:15:07 am by MrW0lf »
 

Offline Karel

  • Super Contributor
  • ***
  • Posts: 1344
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4228 on: September 09, 2018, 07:55:07 am »
How is that done, Fungus?

There are many tools available to download the raw internal memory waveform data from the scope
to your pc.
The one I use is DSRemote. It saves the data in EDF format which can be opened in Octave, Scilab,
Matlab and EDFbrowser for further analysis.

Never use csv format.
The difference between theory and practice is less in theory than
the difference between theory and practice in practice.
Expensive tools cannot compensate for lack of experience.
 

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4229 on: September 09, 2018, 11:13:21 am »
So are you saying that expecting that OFF is OFF is unfair assumption

Haven't you seen any if the sigs around here?

"Lies, damned lies, and what's on the screen of your oscilloscope", etc.

Might be something to bear in mind when you sit down for days producing numbers then trying to draw "conclusions" from them.

As for "screen sampling", even Rigol itself has moved on:

Wisely investing the profits from their market-defining DS1054Z.

I can't wait to see that new ASIC in a $400 'scope later this year!
« Last Edit: September 09, 2018, 11:17:30 am by Fungus »
 

Offline metrologist

  • Super Contributor
  • ***
  • Posts: 1760
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4230 on: September 09, 2018, 01:46:52 pm »
looks like a bug. I shut off other 3 channels and saved memory (now 60pts), and the wave represents.

When you shut off 3 channels the 'scope will force sin(x)/x on and it will have more samples/MHz.

For this sort of work you should always try to maximize the sample rate, on a Rigol DS1054Z that might mean using one channel.

sinc was on all the time. It is not that issue. Just save memory to csv, and data is offset.

How is that done, Fungus?

There are many tools available to download the raw internal memory waveform data from the scope
to your pc.
The one I use is DSRemote. It saves the data in EDF format which can be opened in Octave, Scilab,
Matlab and EDFbrowser for further analysis.

Never use csv format.

What's wrong with csv? Don't the values have to be in some delimited text format, no matter what mechanism is used to get the data? Or do you mean do not use the scope's built in save to usb device functions? Perhaps we are working in the field or on antenna towers and having PC interface is difficult.

I am not talking about sinc or sample depth anyway. Just the save memory as csv format seems to not give you the data you see on the screen, when 3 or 4 channels are turned on...

So, I wonder what are those point data? It looks like the memory data are half a screen later than screen data.

Everything Fungus has said about this is correct, except the issue I am describing is not addressed.
 

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4231 on: September 09, 2018, 04:45:40 pm »
I just had a play with the 'scope and the issue is complex. I haven't figured out all the details yet.

The ASCII formats are a bit suspect when you have 4 channels on and maximum zoom. There seems to be a bug where the points at the right of the screen are missing. You can move the trigger to the left but there's definitely something wrong there.

Reading raw binary data over LAN isn't easy - you have to scale/offset it yourself and the manual isn't very clear on the details. Maybe the people at DSRemote already figured it out...

There are many tools available to download the raw internal memory waveform data from the scope
to your pc.
The one I use is DSRemote. It saves the data in EDF format which can be opened in Octave, Scilab,
Matlab and EDFbrowser for further analysis.
Perhaps we are working in the field or on antenna towers and having PC interface is difficult.

Battery powered DS1054Z up an antenna tower trying to look at four pulse generators simultaneously? You'd think that sort of job could budget more than a $350 oscilloscope.

Still, you could save the data in Rigol .wfm format then reload it back into the 'scope when you get back to the office. Use DSRemote there.

(I assume .wfm format preserves all the information)
 

Offline metrologist

  • Super Contributor
  • ***
  • Posts: 1760
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4232 on: September 09, 2018, 05:41:05 pm »
All of my tools are old and used, or inexpensive with limitations. They all have their idiosyncrasies, so what is important to me is to understand that and learn how to make the tool perform the intended function. I appreciate your time to explain things, Fungus.
 

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4233 on: September 09, 2018, 05:52:17 pm »
I figured out how RAW data over LAN in BYTE format works. It gives exactly the same numbers as downloading the data over LAN in ASCII format.

The values in the BYTE format map directly to vertical screen pixels so I believe these numbers are the raw sample data from deep inside the 'scope. These numbers are as accurate as it gets. 8)

(and I assume DSRemote is using this data)

The question is: Why are these numbers subtly different from the .csv numbers?
« Last Edit: September 09, 2018, 06:21:10 pm by Fungus »
 

Offline MrW0lf

  • Frequent Contributor
  • **
  • Posts: 921
  • Country: ee
    • lab!fyi
Re: New Rigol DS1054Z oscilloscope
« Reply #4234 on: September 09, 2018, 06:47:11 pm »
The values in the BYTE format map directly to vertical screen pixels so I believe these numbers are the raw sample data from deep inside the 'scope. These numbers are as accurate as it gets. 8)

Did you check for Sinc=ON|OFF @250MSa/s? Does raw data amplitude stay same or still changes like on screen? Of course you would need feed in 100MHz or so.
 

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4235 on: September 09, 2018, 06:57:04 pm »
Why are these numbers subtly different from the .csv numbers?

How subtle?

I've converted the differences between the two datasets into raw ADC values and they're all less than one ADC step (the majority are less than half an ADC step).

nb. One ADC step is exactly two pixels on screen so in physical terms the majority of differences between RAW and .csv are less than a pixel.

Seems like some sort of a rounding error to me. Maybe csv is done with fixed-point math.

 

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4236 on: September 09, 2018, 07:19:16 pm »
Did you check for Sinc=ON|OFF @250MSa/s? Does raw data amplitude stay same or still changes like on screen? Of course you would need feed in 100MHz or so.

Raw data seems to match what's on screen, ie. it obeys the sinc setting.

This probably means the 'raw' data is being read out through the FPGA reconstruction filter, there's no direct path to ADC memory.

FWIW I'm looking at a pulse from an Arduino pin, it has a measured rise time of about 4.5ns (and 20% overshoot thanks to a crappy dupont breadboard cable).
« Last Edit: September 09, 2018, 07:25:57 pm by Fungus »
 
The following users thanked this post: MrW0lf

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4237 on: September 09, 2018, 07:41:29 pm »
Conclusion: RAW data is slightly better but using CSV isn't a real problem.
« Last Edit: September 09, 2018, 07:45:09 pm by Fungus »
 

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4238 on: September 09, 2018, 07:51:20 pm »
One question that springs to mind while doing this is:

Does the trigger circuitry do sin(x)/x.

If it doesn't do it then there'd be a lot of horizontal jitter at these extremes - the trigger crossing would happen at different times depending on where the ADC happened to sample the wave.

But there isn't. The trigger is perfectly stable no matter what the settings, even with all channels on and close to 2.5 samples per wave.

What's going on there?
 

Offline MrW0lf

  • Frequent Contributor
  • **
  • Posts: 921
  • Country: ee
    • lab!fyi
Re: New Rigol DS1054Z oscilloscope
« Reply #4239 on: September 10, 2018, 07:44:15 am »
Does the trigger circuitry do sin(x)/x.

It makes sense to do heavy pre-conditioning of the signal before trigger process, otherwise it would be indeed a complete mess. rig(x)/x => [trigger =>] sin(x)/x. So if there are two passes of heavy conditioning it explains why response becomes so aggressive at low sample counts.
 

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4240 on: September 10, 2018, 08:25:41 am »
Does the trigger circuitry do sin(x)/x.

It makes sense to do heavy pre-conditioning of the signal before trigger process, otherwise it would be indeed a complete mess. rig(x)/x => [trigger =>] sin(x)/x. So if there are two passes of heavy conditioning it explains why response becomes so aggressive at low sample counts.

That's an interesting theory. Can you expand on it a little?
 

Offline rf-loop

  • Super Contributor
  • ***
  • Posts: 3051
  • Country: fi
  • Starting with DLL21
Re: New Rigol DS1054Z oscilloscope
« Reply #4241 on: September 10, 2018, 10:00:39 am »
One question that springs to mind while doing this is:

Does the trigger circuitry do sin(x)/x.

If it doesn't do it then there'd be a lot of horizontal jitter at these extremes - the trigger crossing would happen at different times depending on where the ADC happened to sample the wave.

But there isn't. The trigger is perfectly stable no matter what the settings, even with all channels on and close to 2.5 samples per wave.

What's going on there?

Of course digital trigger "engine" do fine interpolation between true sample points for fine adjust waveform to trigger time position. I do not know any DSO what do not have this. There is differences how perfect it is done and how many data points are generated by interpolation (aka "oversampling") between true sampled points. In siglent there can use 500MSa/s and 1ns/div (true sample interval 2 div (2ns)). With 1ns/div every true sample interval is 50 pixel on screen (20ps/pixel).  Using fast mode dots, lines or Sinc I can not detect waveform time position displacement jitter in trigger position. (exept of course what come from possible noise).




First run using fast acquisition (normal) and dots only. When it is running dots mode is ok due to enough acquisitions with random time position dots in one TFT frame (looks around same as random time ETS). Here also turned persistence on. For show trace after stop. Then stopped. Now visible is only last acquisition dots highlighted.
« Last Edit: September 10, 2018, 10:09:47 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: 3051
  • Country: fi
  • Starting with DLL21
Re: New Rigol DS1054Z oscilloscope
« Reply #4242 on: September 10, 2018, 10:24:00 am »
Does the trigger circuitry do sin(x)/x.

It makes sense to do heavy pre-conditioning of the signal before trigger process, otherwise it would be indeed a complete mess. rig(x)/x => [trigger =>] sin(x)/x. So if there are two passes of heavy conditioning it explains why response becomes so aggressive at low sample counts.

I have not seen problems in Rigol1kZ digital trigger engine  fine interpolation system between true sample points for positioning waveform if also think its own performance and price category in this question. Fine adjusting acquisition time position works quite well when I have used it (note: it have fastest display time base 5ns/div). (very old FW / HW what I have used 2014-2015).
« Last Edit: September 10, 2018, 10:28:51 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 MrW0lf

  • Frequent Contributor
  • **
  • Posts: 921
  • Country: ee
    • lab!fyi
Re: New Rigol DS1054Z oscilloscope
« Reply #4243 on: September 10, 2018, 10:50:47 am »
I have not seen problems in Rigol1kZ digital trigger engine  fine interpolation system between true sample points for positioning waveform if also think its own performance and price category in this question.

Nobody is talking about problems with trigger engine. Question is on what it triggers. Fungus just reported that sample points downloadable from memory correlate with Sinc=ON|OFF in GUI. So they cannot be true "analog" sample points in principle. Yet trigger is working just fine. So can only conclude that substantial pre-processing by whatever means is taking place before trigger (which is digital AFAIK).
Pre-processing to compensate for frontend imperfections is not uncommon practice. What is uncommon is implementation at hand where instead of approximating ideal analog frontend response various weirdness takes place.

Some outtakes from literature:
https://community.keysight.com/community/keysight-blogs/oscilloscopes/blog/2017/02/22/real-time-or-sampling-oscilloscope-key-comparisons
Quote
Frequency response is another key consideration in your selection criteria. Sampling oscilloscopes do not use digital signal processing (DSP) correction, so the frequency response rolls off slowly and looks more Gaussian in shape. Real-time oscilloscopes can implement DSP to correct their frequency response. For instance, Keysight’s S-Series oscilloscope has a very flat frequency response across its bandwidth, which means its gain will not vary by more than 1 dB across the entire band.

And what we see with Rigol? Response clumsily approximating flat on higher voltage ranges. Also "screen sampling" technology is clumsy copy of Keysight tech*. It would be all ok it frequency response would be uniform and predictable across all vertical ranges and post-processing modes.

Extra content:
http://www.diva-portal.org/smash/get/diva2:7513/FULLTEXT01.pdf
Post-Correction of Analog-to-Digital Converters

Quote
Post-correction of A/D Converters
Error correction of ADCs has received increasing attention during the
last two decades. Several methods have been proposed and evaluated
during this time, e.g., [HSP00, LASH02a, IHK91, Mou89, TL97, Hum02,
RI87, Iro86]. These methods have in common that the ADC to be corrected
is treated as a closed entity, i.e., internal signals and states of the
ADC are not available, and the calibration and correction methods must
operate outside of the converter. Moreover, the correction is dependent
on the output signal x(n) of the ADC to be corrected. That is, the
correction is an operation incorporated after the ADC, hence the name
post-correction.

In this chapter we will first review some of the ADC post-correction
methods that have been proposed in the past. This will then lead to the
introduction of a generalized post-correction method.
We are mainly concerned with look-up table correction methods.
These are, as the name suggests, methods that produce a corrected ADC
output value through the use of a look-up table (or possibly several tables
– see Chapter 5 for a scheme incorporating two look-up tables),
where pre-calculated values are stored. A distinction between static and
dynamic correction is made. If the correction for a sample x(n) is a function
only of the value x(n), i.e., not depending on past or future samples,
signal derivatives, signal frequency, etc., then the correction is said to be
static. Else, it is said to be dynamic.

* *SOX3000T seem to deliver about 20kpts horizontal for calculus
« Last Edit: September 10, 2018, 11:43:11 am by MrW0lf »
 

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4244 on: September 10, 2018, 01:01:03 pm »
Fungus just reported that sample points downloadable from memory correlate with Sinc=ON|OFF in GUI. So they cannot be true "analog" sample points in principle.

That doesn't mean they aren't "true analog sample points" in ADC memory, only that we haven't found a way to access an unprocessed version of them.

can only conclude that substantial pre-processing by whatever means is taking place before trigger (which is digital AFAIK).

Nope, completely wrong

Trigger units work in parallel with the ADC to generate a horizontal timing offset for aligning waveforms on screen. That's it.

It's easy to demonstrate, here's a pulse in dot mode:


Here's another capture of the pulse. Notice that the dots aren't in the same horizontal positions - they've been offset in that axis using timing information from the trigger unit:


Finally, here's a whole bunch of them, overlaid. There's captures at many different times but they all line up and overlay beautifully, no horizontal difference between each capture:


If the signal was being reconstructed as you claim there would always be a dot exactly on the horizontal trigger position.

The third image (dots at every possible horizontal position) would also be impossible.
« Last Edit: September 10, 2018, 01:28:49 pm by Fungus »
 

Offline rf-loop

  • Super Contributor
  • ***
  • Posts: 3051
  • Country: fi
  • Starting with DLL21
Re: New Rigol DS1054Z oscilloscope
« Reply #4245 on: September 10, 2018, 01:30:57 pm »
Fungus just reported that sample points downloadable from memory correlate with Sinc=ON|OFF in GUI. So they cannot be true "analog" sample points in principle.

That doesn't mean they aren't "true analog sample points" in ADC memory, only that we haven't found a way to access an unprocessed version of them.

can only conclude that substantial pre-processing by whatever means is taking place before trigger (which is digital AFAIK).

Nope, completely wrong

Trigger units work in parallel with the ADC to generate a horizontal timing offset for aligning waveforms on screen. That's it.

It's easy to demonstrate, here's a pulse in dot mode:


Here's another capture of the pulse. Notice that the dots aren't in the same horizontal positions - they've been offset in that axis using timing information from the trigger unit:


Finally, here's a whole bunch of them, overlaid. There's captures at many different times but they all line up and overlay beautifully, no horizontal difference between each wave:


If the signal was being reconstructed for display (as you claim) then there would always be a dot exactly on the horizontal trigger position and the third image (dots at every possible horizontal position) would be impossible.

And now, turn off Sinc and this Gibbs phenomenon disappear (because it is not in real input signal and produced by Sinc interpolation. And in these cases Sinc do not reconstruct true input signal right (even in theory it can not, rules are included in theory) . (Gibbs "ear" is not error, it is Sinc interpolation normal "feature" specially if input signal violate rules))
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.
 

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4246 on: September 10, 2018, 02:01:26 pm »
And now, turn off Sinc and this Gibbs phenomenon disappear (because it is not in real input signal and produced by Sinc interpolation. And in these cases Sinc do not reconstruct true input signal right (even in theory it can not, rules are included in theory) . (Gibbs "ear" is not error, it is Sinc interpolation normal "feature" specially if input signal violate rules))

The wobble on screen isn't just the Gibbs phenomenon, it's mostly ringing from looking at an Arduino pin through a crappy piece of breadboard wire.

I can turn sinc on/off and there's some difference on screen. It's most visible in the preshoot, here's a couple of captures:





Maybe you've accidentally hit on the reason why Rigol decided to enable the "sinc on/off" button when you get to the extremes, ie. to see how much Gibbs is on screen.  :-+

And, yes, Wolfie, we know it's not truly "off" - no reconstruction at all would display a horrible mess on screen.

There's a pretty curve on screen so they must be doing a different reconstruction (ie. not sin(x)/x) that produces a nice display but minimizes the Gibbs Effect.

I think we might finally be getting to the bottom of the Rigol "sinc" mystery (and why there's no truly "raw" data available).



Edit: Just for completeness, here's the same pulse with only 2 channels enabled on the 'scope. The sin(x)/x button is disabled now (no longer needed - we have twice as many samples available so Gibbs phenomenon isn't a problem).

« Last Edit: September 10, 2018, 05:53:25 pm by Fungus »
 

Online Fungus

  • Super Contributor
  • ***
  • Posts: 9725
  • Country: 00
Re: New Rigol DS1054Z oscilloscope
« Reply #4247 on: September 10, 2018, 02:23:33 pm »
I think we might finally be getting to the bottom of the Rigol "sinc" mystery (and why there's no truly "raw" data available).

Having said that, there might be genuinely raw data inside 'wfm' files. I haven't explored those yet - it would be a lot of work.

Those files have to load and redisplay exactly what was on screen previously.
 

Offline MrW0lf

  • Frequent Contributor
  • **
  • Posts: 921
  • Country: ee
    • lab!fyi
Re: New Rigol DS1054Z oscilloscope
« Reply #4248 on: September 10, 2018, 04:33:08 pm »
Heres trivial dynamic ADC post corrector simulation:

A-(A[0.000000003]-A[0.000000002])-(A[-0.000000003]-A[-0.000000002])

This formula works for 1GSa/s. A is active channel. Brackets contain sample location in time.
For this test I physically suppressed scope analog bandwidth for more dramatic effect. Blue is original signal. Cyan is simulated ADC post corrector. Linear interpolation (excuse lots of environmental noise).

Relatively sharp edge: Can see something reminiscent of sin(x)/x interpolation, nothing too dramatic. Better risetime.


Relatively blunt edge: Effects even more subtle.


Square with blunt edges: Leaves impression of better "reconstruction", level somewhat rises.


Full bandwidth sweep with sine: Dramatic "bandwidth boost" effect in analog response rolloff region.


All left to do is apply 100MHz LowPass filter (indigo trace). Note measured risetimes.


« Last Edit: September 10, 2018, 04:49:16 pm by MrW0lf »
 

Offline MrW0lf

  • Frequent Contributor
  • **
  • Posts: 921
  • Country: ee
    • lab!fyi
Re: New Rigol DS1054Z oscilloscope
« Reply #4249 on: September 10, 2018, 05:49:40 pm »
Here less exaggerated example:

A-(A[0.000000001]-A)-(A[-0.000000001]-A)

Tuned response more-less flat from 0 to 100MHz on non-suppressed scope. Visual cost is those little sin(x)/x like corners.




 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf