Rigol DS 1054Z behavior: normal or defective?

[ballen]

Hi All,

I've been doing electronics as a hobby since I was a teenager 40+ years ago, and decided it was finally time to retire my 1970s Tektronix 2-channel CRT scope.  After a few minutes of on-line research I ordered a Rigol DS 1054Z (with 24 Msamples memory and all possible options enabled the price was still low) which arrived earlier today.  I've been playing around with it for a couple of hours, and am impressed by what it can do.

One odd thing I have noticed: if I look at the built-in calibration signal (1kHz square wave, 3v amplitude) the horizontal traces at 0 and 3v look broader than in the manual.  If I zoom in on them, I can see that they are each made of a pair of lines of equal brightness, separated in amplitude by tens of mV.  I'm not sure how many bits are used internally, but I guess this might be quantization noise from the input ADC.  I can also see this if I (electronically) ground the input.  I would have hoped that the electronics noise at the input is not enough to dither the ADC.  I've also noticed that as I shift the position of the trace vertically, the relationship between the two lines shifts and distorts.

My question: is this behavior "normal" or did I get a scope with a bad input amplifier or noisy ADC?  Easy to send it back now, if that's the case.

Cheers,
   Bruce

Edit: when I shift the vertical gain from 500mV per division to 200mV, I can hear a relay click shut internally.  That's when the shape of trace near zero distorts and becomes dependent on the Y-axis offset.  I don't think this should happen, is it "normal"?  Also, I can see the doubling of the horizontal trace lines when the gain is set so that 3V is full scale, so it's got to be at least 30 mV, meaning that it's in the 8th or 9th bit of an ADC.

Here's a photo of the calibrator signal, I have turned off the display grid to make the structure more visible:

[Fungus]

My question: is this behavior "normal" or did I get a scope with a bad input amplifier or noisy ADC?  Easy to send it back now, if that's the case.

Yes, it's normal. It's an aliasing effect from decimating a small sine wave.

https://www.eevblog.com/forum/testgear/help-my-new-rigol-ds1054z-shows-a-weird-double-trace/msg1363145/#msg1363145

[ballen]

Thank you for the link -- it is exactly the issue that I see.

I read through the thread, and it appears that you and the original poster see this behavior, but that another person with a DS1054Z (emocabras) did not.

I don't see how aliasing (in the traditional sense of sampling a signal with power beyond the Nyquist frequency of half the sample rate) would produce this.  And if I insert a 20MHz filter (here Nyquist is 500 Mhz) the problem persists.

A simpler explanation would be if the scope use two interleaved ADCs and they have slightly different zero points.  Do you know if that's the case? That would also explain why the size of the effect is different in the two cases reported in the thread.

[Fungus]

I don't see how aliasing (in the traditional sense of sampling a signal with power beyond the Nyquist frequency of half the sample rate) would produce this.  And if I insert a 20MHz filter (here Nyquist is 25MHz) the problem persists.

It's not aliasing on the ADC/signal, it's aliasing together with quantization errors in the screen display algorithm.

Whatever it's doing in the display subsystem is  get a disproportionate number of dots at top/bottom. At some zoom levels it looks like two lines on screen.

A simpler explanation would be if the scope use two interleaved ADCs and they have slightly different zero points.  Do you know if that's the case?

The 'scope only has one ADC so that's not it.

[2N3055]

Try performing Self calibration.
Any time temperature changes more than few degrees C from last call, you need to perform Self Cal.
It should align any misalignments.

Also, you should try doing same measurement using external, independent signal source.  Relaxation (RC) oscillator with 74HCT14 will do and can be made in 10 minutes, dead bug style.

The 'scope only has one ADC so that's not it.

Not quite accurate.. It does have 1 A/D chip, but... HMCAD1511 consists of 4 interleaved A/D converters internally,  250 MSPS each.  They have fine gain adjustments for each individual converter.
Calibration is not automatic inside chip, scope performs it during self cal, and then sets gain coefficients for HMCAD chip. .

[ballen]

It's not aliasing on the ADC/signal, it's aliasing together with quantization errors in the screen display algorithm.  Whatever it's doing in the display subsystem is  get a disproportionate number of dots at top/bottom. At some zoom levels it looks like two lines on screen.

I'm afraid that I don't understand that either.  The display has a resolution of 600 pixels in the vertical direction.  The ADC (I just googled, Analog Devices HMCAD1511 with 8 bits) has 2^8=256 different output values and with built-in digital gain it should be able to resolve this 0 to 3V signal into (say) 200 distinct values, which would be at most 3 pixels wide.  The two lines I am seeing are separated by much more than 3 pixels.

Another way of stating my ignorance: when the calibration signal is at 0V with (say) a mV of noise, the 8-bit ADC should be producing the same value for each sample.  The scope should display those values at the same height.  If there is dither, one ought to see two values displaced by a few pixels.  So I don't understand what "aliasing together with quantization errors in the screen display algorithm" means.

Edit: I just saw the reply that the ADC is actually four interleaved ones.  An offset there would explain it.  I will try running the calibration signal into all four channels.

 

[ballen]

Try performing Self calibration.
Any time temperature changes more than few degrees C from last call, you need to perform Self Cal.
It should align any misalignments.

Did that, made no difference.

Also, you should try doing same measurement using external, independent signal source. 

OK, I'll hook it to the calibration signal of my Tex scope 

HMCAD1511 consists of 4 interleaved A/D converters internally,  250 MSPS each.  They have fine gain adjustments for each individual converter. Calibration is not automatic inside chip, scope performs it during self cal, and then sets gain coefficients for HMCAD chip.

I ran the full auto-calibration yesterday (took quite a long time) but that made no difference.

I tried running all four channels together, and that indeed changes the "double-valued" behavior. So I think the issue is that the four A/D have different offsets and these are not correctly compensated.  When each A/D is assigned to its own channel, then the double-trace goes away for certain values of the vertical position offset, and is present (dithering?) for other values of the position offset.  Fungus, could you try this on your scope, to see if using all four channels make a difference?

One last question: it appears that the ADC is 8 bit but the Rigol specs say "up to 12 bit resolution".  They could get the additional four bits by summing sets of 16 samples, which could be done at speeds of up to 64 Msamples/second.  Is this done in software?  I could see two ways of doing this. One is "postprocessing": acquire at 1Gs/second, then afterwards sum groups of 16 samples.  The other is "realtime", so sum groups of 16 samples as the ADC is running, and only record those samples into memory at 64MHz.  Does the scope do either/both of these?

[Fungus]

The two lines I am seeing are separated by much more than 3 pixels.

Yes, there's noise on the input. It's not just the lowest bit of the ADC.

[Fungus]

I'm afraid that I don't understand that either. 

Look at this image:



The un-zoomed version (top half of the screen looks horribly wrong. The zoomed version (bottom of the screen) shows the samples.

Where did the two separate lines come from? From whatever algorithm is being used to generate the un-zoomed image.

You want to fix it? Push the Acquire button and turn on "Average" in the Acquisition Mode.

Will it affect you in reality? No, because you'll hardly ever look at a signal that way, see next part.

Edit: I just saw the reply that the ADC is actually four interleaved ones.  An offset there would explain it.  I will try running the calibration signal into all four channels.

a) Note that you're really zoomed in and most of the signal is off screen. This means you're pushing the outer limits of the vertical amplifiers, etc., which is never a good idea because you get into a whole class of problems with amplifier overloading and recovery times on any oscilloscope (although I'm sure those $10,000 'scopes will have better opamps!)

b) By pushing the limits of the opamps you might also be revealing tiny differences in calibration between ADCs. This would explain why everybody sees something different when they try this.

In practice it won't affect you. Most owners never even noticed their 'scope does this. I never noticed it until it was pointed out in that other thread.

[Miti]

Nope, I don't see this. Not even when the top side is way out of screen.

[tom66]

I'm unable to replicate this on my old DS1074Z running 00.02.03.SP5 (not upgraded since I bought it)

[2N3055]

I'm afraid that I don't understand that either. 

Look at this image:



The un-zoomed version (top half of the screen looks horribly wrong. The zoomed version (bottom of the screen) shows the samples.

Where did the two separate lines come from? From whatever algorithm is being used to generate the un-zoomed image.

You want to fix it? Push the Acquire button and turn on "Average" in the Acquisition Mode.

Will it affect you in reality? No, because you'll hardly ever look at a signal that way, see next part.

Edit: I just saw the reply that the ADC is actually four interleaved ones.  An offset there would explain it.  I will try running the calibration signal into all four channels.

a) Note that you're really zoomed in and most of the signal is off screen. This means you're pushing the outer limits of the vertical amplifiers, etc., which is never a good idea because you get into a whole class of problems with amplifier overloading and recovery times on any oscilloscope (although I'm sure those $10,000 'scopes will have better opamps!)

b) By pushing the limits of the opamps you might also be revealing tiny differences in calibration between ADCs. This would explain why everybody sees something different when they try this.

In practice it won't affect you. Most owners never even noticed their 'scope does this. I never noticed it until it was pointed out in that other thread.

When I had 1000Z I never had that problem at any range.. it was 2016 vintage...

[Fungus]

In practice it won't affect you. Most owners never even noticed their 'scope does this. I never noticed it until it was pointed out in that other thread.

When I had 1000Z I never had that problem at any range.. it was 2016 vintage...

Like I said, it takes a particular combination of signal/noise/zoom. You might have had it but never noticed.

(or lived in a place where it doesn't happen).

[tooki]

In practice it won't affect you. Most owners never even noticed their 'scope does this. I never noticed it until it was pointed out in that other thread.

When I had 1000Z I never had that problem at any range.. it was 2016 vintage...

Like I said, it takes a particular combination of signal/noise/zoom. You might have had it but never noticed.

(or lived in a place where it doesn't happen).

But even assuming a particular combination of conditions enabled this bug, the conditions shown in the image aren’t it: that’s a simple view (auto, then scaled back one detent of the horizontal scale) of the built in calibration signal, which most definitely should not produce double traces.

[Fungus]

But even assuming a particular combination of conditions enabled this bug, the conditions shown in the image aren’t it: that’s a simple view (auto, then scaled back one detent of the horizontal scale)

Nope. It's set to 20mV vertical. The full signal is way off screen (150 grid squares tall!)

most definitely should not produce double traces.

It's not a double trace when you zoom in, it just looks like one when you zoom out due to a quirk in the decimation for that particular signal.

[2N3055]

But even assuming a particular combination of conditions enabled this bug, the conditions shown in the image aren’t it: that’s a simple view (auto, then scaled back one detent of the horizontal scale)

Nope. It's set to 20mV vertical. The full signal is way off screen (150 grid squares tall!)

most definitely should not produce double traces.

It's not a double trace when you zoom in, it just looks like one when you zoom out due to a quirk in the decimation for that particular signal.

We are talking about OP. His photo shows simple 500 mV/DIV with scope probe calibration source connected. Nothing off the screen, no zoom, nothing.. And double lines clearly visible. That is wrong.

[Fungus]

We are talking about OP. His photo shows simple 500 mV/DIV with scope probe calibration source connected. Nothing off the screen, no zoom, nothing.. And double lines clearly visible. That is wrong.

Oh, duh! I hadn't spotted that.     

Yeah, that's not good. I retract all posts.

Send it back!

[ballen]

In the photo attached to my original starting post in this thread, there are three "non-standard" settings, intended to make the effect clearer and to prevent my camera from saturating.  Miti, Tom66, could you please try this one more time but with exactly the settings I have used?  To ensure this, first set the scope to default settings by putting

Utility->System->Power Set=default and power cycle.

Then input the calibration signal on channel 1, use "Auto" to pick trigger/timing/gains, and set

Display->Type=dots
Display->Grid=off
Display->Intensity=25%

Last step, shift the vertical position up and down.  I am starting to think that what I am seeing here is just the discrete quantization of levels from the 8-bit ADC.  Meaning: I have the vertical position set so that the max and min square wave values split between two adjacent ADC output steps, and the display algorithm has each ADC output step illuminate only half the pixel height to the next output step. Hence a black band appears between adjacent ADC levels.

[tautech]

Display->Type=dots
Display->Grid=off
Display->Intensity=25%

Last step, shift the vertical position up and down.  I am starting to think that what I am seeing here is just the discrete quantization of levels from the 8-bit ADC.  Meaning: I have the vertical position set so that the max and min square wave values split between two adjacent ADC output steps, and the display algorithm has each ADC output step illuminate only half the pixel height to the next output step. Hence a black band appears between adjacent ADC levels.

I doubt this.
SDS1104X-E with those settings ^ and imperfectly compensated probe.



The Rigol pusher in this forum Fungus has told you to quit it...take his advice.

[ballen]

OK, another theory, the ADC has a broken bit or bit combination.  For example here is a screen shot showing some power supply noise (a few mV) AC coupled on the output of a power supply.  Notice the horizontal black striped bands -- it's as if some ADC output combinations are missing.  Hard to imagine that this could get past QC at either the chip fab or at Rigol.

[ballen]

Here is my DS1104Z-S from 2014, right after power up with the settings of ballen. In the vector display the stripes are not visible.

Peter, thank you, this would indicate that my scope is "normal" rather than "defective".  The stripes seem to be discrete ADC output levels, normally hidden in vector mode.

[Fungus]

OK, another theory, the ADC has a broken bit or bit combination.  For example here is a screen shot showing some power supply noise (a few mV) AC coupled on the output of a power supply.  Notice the horizontal black striped bands -- it's as if some ADC output combinations are missing.  Hard to imagine that this could get past QC at either the chip fab or at Rigol.

That's normal in dots mode. The ADC only has 200 values and the screen has 400 pixels. There's also all sorts of amplification in between.

Bottom line: Don't use dots mode. The display is only valid when the dots have been mathematically reconstructed.

(A lot of more expensive 'scopes don't even have dots mode...it's an anachronism)

[rf-loop]

Here is the MSO2302A, there the stripes are slightly present. You can see them more when you change the settings.

Peter

This your image Ch2 is because afaik it do not have at all true 500uV. Its is just "pixel" zoom. There is roughly only 100 steps in screen vertical area. Other v/div have 200 steps.

[TurboTom]

It's true, the highest sensitivity that the DS2000 series of scopes features is 1mV/div, the 500µV/div is just a magnification witout additional sampling data.

The funny thing is: these "double trace effects" on the Rigol scopes appear to be an artefact of the way the intensity grading on these machines is processed. It seems that almost right from the beginning after release of their "UltraVision" intensity grading engine, there's been a bug (feature??  ) in it that as yet didn't get addressed: When the sampling memory gets decimated (horizontally) to fit the screen resolution, the samples that represent a position on screen get summed up / averaged, which makes sense of course. It's just that if this summing up overflows the size of the summing memory that rigol chose, the corresponding pixel gets set dark instead of maximum brightness. So what's visible on screen when displaying a (gaussian) noisy signal, is two bright bands where the sum just doesn't overflow, and a dark region in between, where it does overflow. This effect can be easily demonstrated when taking a single shot of a noisy signal at slow time base where the effect is visible, and then zooming in with the timebase. The effect disappears and the trace looks completely normal, with a proper density distribution as one would expect from a noisy signal. See the attached screenshot to get an idea of what I mean. I guess Rigol should look into their display engine's basic algorithms... But to just correct the "intensity overflow clipping" shouldn't be too difficult.

Edit: Added another screenshot with a noise signal from an AWG fed to the input of the scope. In order to demonstrate this effect, a high sampling memory depth is required.

[Miti]

In the photo attached to my original starting post in this thread, there are three "non-standard" settings, intended to make the effect clearer and to prevent my camera from saturating.  Miti, Tom66, could you please try this one more time but with exactly the settings I have used?  To ensure this, first set the scope to default settings by putting

Utility->System->Power Set=default and power cycle.

Then input the calibration signal on channel 1, use "Auto" to pick trigger/timing/gains, and set

Display->Type=dots
Display->Grid=off
Display->Intensity=25%

Last step, shift the vertical position up and down.  I am starting to think that what I am seeing here is just the discrete quantization of levels from the 8-bit ADC.  Meaning: I have the vertical position set so that the max and min square wave values split between two adjacent ADC output steps, and the display algorithm has each ADC output step illuminate only half the pixel height to the next output step. Hence a black band appears between adjacent ADC levels.

Yes, I see the two lines now.