Rigol HDO1000 and HDO4000 12bit oscilloscopes launched in China

[David Hess]

The waveforms on those Agilent/Keysight scopes are drawn to the screen with an ASIC in hardware, so the CPU is just doing UI stuff. But even the Keysight stuff can be a little slow to respond some times and will stop acquisitions while moving traces. Other brands have done the same path with ASICs (R&S) and while the FPGA platforms that Rigol and Siglent use dont quite have the same throughput capabilities it is just a programming/system design choice that they feel slower/lag more, they could be very snappy with some work on the software side.

Fast response has more to do with the software design than the hardware.  Old DSOs had very limited processing power and the designers knew it, so the software was designed to keep the user interface updated in real time.  Wordstar played the same game to provide real time response over a serial link to a terminal.  Commands entered during screen updates interrupt the screen update to update characters and lines, and then the screen update continues where it left off.

One method I have used is to implement the user interface functions in real time, while the processing tasks use the remaining processor resources cooperatively.  This gets a little weird, and useful, when the main processing task crashes, yet the user interface continues to operate as usual allowing diagnostics to be performed.

If  modern instrument has poor user interface performance, then it is because of sloppy programming.  The designers just did not care.

[David Hess]

Out of curiosity: Where is the noise floor (dB) in FFT mode on this 'scope?

I would rather know the SFDR (spurious free dynamic range) which would reveal the linearity.  Noise can be removed but spurs from non-linearity hide the true signal components.

Settling time and overload recovery of a 12 bit signal path would be useful to know, but these are difficult to test without a suitable signal source.

[David Hess]

No, sin x/ x is not a filter!  sin x/x is a method for constructing a visible signal from samples. By definition, the trace you get from sin x/x reconstruction goes through all the sample points. There is no phase mangling, no adding fictional information or whatever.

Sin(x)/x is implemented as a finite impulse response filter with coefficients from the Sinc function applied to a sample record which has been sample stuffed.  It is a low pass filter which removes the high frequency components from the sample stuffing.

[Martin72]

Hallo,

If I remember correctly, you have the SDS2104X HD.

Yepp, but with 500Mhz bandwith.

We just have to consider the difference in bandwidth when assessing it.
(According to tautech's hint, perhaps 200 MHz bandwidth would also be possible?)

Got bandwithlimiter on 20Mhz AND 200Mhz, should activate the 200 for this.

All 8 measurements with 200 MSa memory and 2 GSa/s sampling rate

No problem, but didn´t Dave used 1M ?

1 mV/div 50 Ohm maximum bandwidth (100 MHz or 200 MHz)
1 mV/div 50 Ohm 20 MHz bandwidth
1 mV/div 1 MOhm maximum bandwidth (100 MHz or 200 MHz) open input (deepl say open entrance)
1 mV/div 1 MOhm 20 MHz bandwidth open input

1 V/div 50 Ohm maximum bandwidth (100 MHz or 200 MHz)
1 V/div 50 Ohm 20 MHz bandwidth
1 V/div 1 MOhm maximum bandwidth (100 MHz or 200 MHz) open input
1 V/div 1 MOhm 20 MHz bandwidth open input

Open input is clear, but what´s the other, switched to GND ?

Martin

[nctnico]

No, sin x/ x is not a filter!  sin x/x is a method for constructing a visible signal from samples. By definition, the trace you get from sin x/x reconstruction goes through all the sample points. There is no phase mangling, no adding fictional information or whatever.

Sin(x)/x is implemented as a finite impulse response filter with coefficients from the Sinc function applied to a sample record which has been sample stuffed.  It is a low pass filter which removes the high frequency components from the sample stuffing.

Technically correct but in the end it doesn't matter for the original signal. From a filter theory standpoint creating a visible waveform is an upsampling operation which needs a filter to remove the steps indeed. But this filter isn't removing any frequency content from the original signal. So for the acquired signal itself, nothing changes. That is what I wanted to point out (without going deep into implementation details).

[EEVblog]

The manual that was posted earlier appears unfinished, I wouldn't rely on it for a complete+accurate list of features.

I am not sure if it's the manual or the scope that is unfinished. For example, the manual talks about a header in the binary file. But the files Dave posted contained no such header (assuming they come directly from saving on scope).

Yes they came direct form the scope using the Quick save button

[EEVblog]

Hallo,

If I remember correctly, you have the SDS2104X HD.

Yepp, but with 500Mhz bandwith.

We just have to consider the difference in bandwidth when assessing it.
(According to tautech's hint, perhaps 200 MHz bandwidth would also be possible?)

Got bandwithlimiter on 20Mhz AND 200Mhz, should activate the 200 for this.

All 8 measurements with 200 MSa memory and 2 GSa/s sampling rate

No problem, but didn´t Dave used 1M ?

1 mV/div 50 Ohm maximum bandwidth (100 MHz or 200 MHz)
1 mV/div 50 Ohm 20 MHz bandwidth
1 mV/div 1 MOhm maximum bandwidth (100 MHz or 200 MHz) open input (deepl say open entrance)
1 mV/div 1 MOhm 20 MHz bandwidth open input

1 V/div 50 Ohm maximum bandwidth (100 MHz or 200 MHz)
1 V/div 50 Ohm 20 MHz bandwidth
1 V/div 1 MOhm maximum bandwidth (100 MHz or 200 MHz) open input
1 V/div 1 MOhm 20 MHz bandwidth open input

Open input is clear, but what´s the other, switched to GND ?

The other is switched to 50ohm internal.

[rf-loop]

No, sin x/ x is not a filter!  sin x/x is a method for constructing a visible signal from samples. By definition, the trace you get from sin x/x reconstruction goes through all the sample points. There is no phase mangling, no adding fictional information or whatever.

... from the sample stuffing.

Öhh... Can you explain perfectly  - just with your own words - what is this: "...removes the high frequency components from the sample stuffing.". 
Do you mean it remove or destroy sampled data - yes Rigol can do it, it is proofed in history -  but I think you do not mean it. So what is this.
Afaik, in scopes what I have used (except two old Rigols) Sinc draw throung every true raw sampled point - and exactly - least every time what I have inspected these..

[switchabl]

In its most basic form, sin x/x interpolation works by inserting ("stuffing") zeroes between each of your samples. In the frequency domain this creates multiple copies of your original spectrum which you then remove using (an approximation to) a brickwall filter, i.e. filter it with a sin x/x kernel. The resulting signal retains the original samples and interpolates between them in such a way that no higher frequency components are added.

Note: If you look at a practical implementation, you may not actually see the zero stuffing explicitly. But that is only because the equations have been rewritten in a way that is faster to compute but less clear to understand (polyphase representation).

[MathWizard]

So whats the difference between a DSO that can trigger on, and decode logic protocols, and an MSO again ??

[tautech]

So whats the difference between a DSO that can trigger on, and decode logic protocols, and an MSO again ??

Total number of channels.
A mix of analog and digital channels, typically 4 and 16.

[egonotto]

Hello,

Martin72: "Got bandwithlimiter on 20Mhz AND 200Mhz, should activate the 200 for this."

Yes.

Martin72: "No problem, but didn´t Dave used 1M ?"

Dave post the 50 Ohm Data with 1 MSa and 250 MSa but the 1 MOhm only with 1 MSa. I hope he will post the 1 MOhm files with 250 MSa too.

Martin72: "Open input is clear, but what´s the other, switched to GND ?"

Sorry my mistake, somehow I implicitly assumed that the input is already terminated by the 50 ohms. At 50 ohms, of course, also open.

Best regards
egonotto

[rf-design]

CMOS image sensor ADCs do not need GS/s speed. Instead they rely on special architectures which minimize area and use less power because they need to fit into the remain border space of the image area. Stacked is the new more technology game to give high performance and sell high.

[2N3055]

Hello maxwell3e10,

How do you make your graphics?

Best regards
egonotto

Yes, I'm curious too. @maxvell3e10, would you be so kind to quickly explain how you calculate and get the plots?
Looks nice and maybe we could replicate results..

Thanks!!
Sinisa

[maxwell3e10]

I was using MATLAB for these, specifically pwelch function. The precise window function does not matter. To reduce fluctuations in the spectrum at high frequency one can either decimate the fft power spectrum or use a shorter window in pwelch. Also I subtracted the mean before doing the fft.

[David Hess]

Öhh... Can you explain perfectly  - just with your own words - what is this: "...removes the high frequency components from the sample stuffing.". 
Do you mean it remove or destroy sampled data - yes Rigol can do it, it is proofed in history -  but I think you do not mean it. So what is this.
Afaik, in scopes what I have used (except two old Rigols) Sinc draw throung every true raw sampled point - and exactly - least every time what I have inspected these..

Switchabl covered it as well as I could.  The sample stuffing creates copies of the original spectrum which the sin(x)/x filter removes, leaving the original sample points and the new ones connecting them with the original frequency spectrum.

Rigol on their earlier DSOs messed it up somehow, but most DSOs manage to do it correctly.

So whats the difference between a DSO that can trigger on, and decode logic protocols, and an MSO again ??

Usually the difference is that the MSO inputs have only 1 bit resolution, produced using a comparator as a 1-bit ADC, and are lower input impedance, typically 100 kilohms.  The MSO inputs often but not always has lower timing resolution than the oscilloscope inputs; modern instruments tend to use the FPGA clock for the MSO inputs.

[JeremyC]

The HDO4000 doesn’t have bode plot option, nor sin(x)/x

It's got Sinx/x, but you can't turn it off.

Referring to the user guide this scope doesn’t have sin(x)/x interpolation and the display mode is vector only, to my knowledge it’s the same as in the MSO5000 scopes.
Check section 19.1, page 238 in the user guide.
(Attachment Link)

Have you watched my video?
The zoomed in waveform in not linear vector based, it's sinx/x
I asked Rigol if I could turn off Sinx/x and they said no I can't.
I do know what that manual is showing, but that's not how the scope i have actually works.

Yes, I did, twice before I posted my message.
After when you pointed me to “The zoomed in waveform in not linear vector based, it's sinx/x” I watched the video few times (min 18:05 – 25:00) ... I do agree with you, for sure there is interpolation involved. I was suggested by the user guide and it’s the reason why I pointed the sin(x)/x topic.
Is it possible to switch to dot mode before “stop” and later zoom it and repeat the test?

Dave, thank you very much for sharing your knowledge and experience with the community.
I was not trying to start an argue/war conversation, I’m only trying to understand this topic.
Personally, I don’t care much for sin(x)/ interpolation, sometimes it comes handy. If I analyze anything besides RF, I use dot, or vector mode and I prefer min 20 samples per period.
I truly appreciate your contribution to the EE community.

[MathWizard]

I want to get a SDS2104x+, then I see these rigol 12bit models and the sds HD 12bit scopes. But they would cost over 2x as much, at least the Siglent would.

I'd love if the rigol are closer to 2k, than 3-5k

[JeremyC]

I want to get a SDS2104x+, then I see these rigol 12bit models and the sds HD 12bit scopes. But they would cost over 2x as much, at least the Siglent would.

I'd love if the rigol are closer to 2k, than 3-5k

I believe the 2104X+ it's great choice, probably you will be satisfied with it for next decade. If you work with audio, you should check Rightmark + descend ADC/DAC solution like the Scarlet or MOTU, both below $200.

[EEVblog]

I want to get a SDS2104x+, then I see these rigol 12bit models and the sds HD 12bit scopes. But they would cost over 2x as much, at least the Siglent would.
I'd love if the rigol are closer to 2k, than 3-5k

The HDO1000 series will start at $699, still 12bit, but doesn't have the 100uV-500uV software expanded ranges.

[EEVblog]

I was using MATLAB for these, specifically pwelch function. The precise window function does not matter. To reduce fluctuations in the spectrum at high frequency one can either decimate the fft power spectrum or use a shorter window in pwelch. Also I subtracted the mean before doing the fft.

A short tutorial video on doing this would be cool.

[tautech]

I want to get a SDS2104x+, then I see these rigol 12bit models and the sds HD 12bit scopes. But they would cost over 2x as much, at least the Siglent would.
I'd love if the rigol are closer to 2k, than 3-5k

The HDO1000 series will start at $699, still 12bit, but doesn't have the 100uV-500uV software expanded ranges.

Different class of instrument Dave, you are comparing apples and oranges.
SDS2000X Plus is a 500 MHz design vs 200 MHz, it also has 2x 2GSa/s, 200 Mpts x2 mem depth, inbuilt AWG, MSO, 10x probe sense.....need I go on ? 

[maxwell3e10]

I was using MATLAB for these, specifically pwelch function. The precise window function does not matter. To reduce fluctuations in the spectrum at high frequency one can either decimate the fft power spectrum or use a shorter window in pwelch. Also I subtracted the mean before doing the fft.

A short tutorial video on doing this would be cool.

Could be useful to have a video on FFT and power spectral density. There is also scipy.signal.welch package that does PSD estimation in Python. In Matlab it is just a few commands:

Code: [Select]

fileID = fopen('RigolDS3-50Ohm1V20M.bin')
data3 = (fread(fileID,'uint16')-2^15)/2^16*10;
data3=data3-mean(data3);
[fdata3,fl3]=pwelch(data3,[],[],[],50e6,'psd');
loglog(fl3,sqrt(fdata3),'r');


[Fungus]

The sample stuffing creates copies of the original spectrum which the sin(x)/x filter removes, leaving the original sample points and the new ones connecting them with the original frequency spectrum.

Rigol on their earlier DSOs messed it up somehow.

Citation needed.

[tautech]

The sample stuffing creates copies of the original spectrum which the sin(x)/x filter removes, leaving the original sample points and the new ones connecting them with the original frequency spectrum.

Rigol on their earlier DSOs messed it up somehow.

Citation needed.

David was replying to rf_loop and if you look back at his earliest posts rf_loop brings to this forum the errors Rigol makes bringing sample points to the display…..now over a decade ago.

Like rf_loop, sample stuffing is the first I have heard of such a term.