Hacking the Rigol DHO800/900 Scope

[Mechatrommer]

Why do the phases of the signals between channels 1 and 2 diverge? The higher the signal frequency, the greater the discrepancy.

My generator produces a maximum sine wave of 70 MHz, so at 70 MHz the phase difference is 20%, and it doesn’t matter between which channels, between 1 and 2, between 1 and 3, between 1 and 4!

I assume what you are observing is "channel skew".

probe's cable length and model/attenuation type difference also can give effect. try to match length and same probe model as much as possible. or use bnc splitter same length everything is typical skew verification. yeah if anything, report it in bug thread, not hack thread. ymmv.

[Fungus]

Hello everybody!

I recently purchased a Rigol DHO804 oscilloscope. Everything works fine, I haven’t noticed any bugs yet, but..
This question is: Why do the phases of the signals between channels 1 and 2 diverge? The higher the signal frequency, the greater the discrepancy.

My generator produces a maximum sine wave of 70 MHz, so at 70 MHz the phase difference is 20%, and it doesn’t matter between which channels, between 1 and 2, between 1 and 3, between 1 and 4!
But a nuance was noticed that between 2 and 3, 3 and 4, 2 and 4 channels there is no such phase discrepancy, only relative to 1 channel and any other.

Can someone check on their oscillation whether this is the same nonsense or not?

How are you connecting the signal to the 'scope? Are all the paths from the generator to the BNC connectors the same length?

At those frequencies you should be able to measure differences in the lengths of the wires.

You can adjust these skews in the scope software. See section 5.11 of the manual. The adjustment range is -100 to + 100 ns, so your 3 ns skew is easily within the range.

Hence an option to tweak it manually.

[Mechatrommer]

You can adjust these skews in the scope software. See section 5.11 of the manual. The adjustment range is -100 to + 100 ns, so your 3 ns skew is easily within the range.

Hence an option to tweak it manually.

not acceptable imho if for no good reason. all CH should be in phase or 0s skew in setting by default, we set non zero skew value only if for special purpose, maybe say an example we cannot match length in DUT or anything out of dso control etc. another suggestion is redo dso calibration...

[AndyBig]

According to my measurements, the discrepancy between the phases of the channels does not exceed 5% (18°) at 60 MHz, this is less than 1 ns.

[ebastler]

How are you connecting the signal to the 'scope? Are all the paths from the generator to the BNC connectors the same length?
At those frequencies you should be able to measure differences in the lengths of the wires.

True in principle. But a time difference of 3 ns corresponds to about 0.6 m cable length, so it should be an obvious difference, right?

I was also assuming that @Slavius had swapped the same probes or cables across the channels when he tested for inter-channel delays in the various channel combinations. If that was not the case, then indeed he should double-check whether the difference is due to internal skew in the scope or due to external connections.

[Slavius]

I tried both the native oscilloscope probes from the kit and from the Rigol DS1054Z kit (this oscillator has slightly better probes both in frequency-amplitude and phase characteristics).

All probes are connected in parallel to the generator output, ground probes are connected at one point of the generator ground.

Something else I noticed:
1) First 4 photos: Default setting (Ch-Ch Skew = 0.00s) in channel 1 settings. The difference in phase shift between 1 and 2, 1 and 3, 1 and 4 channels is about 20 degrees, but if you turn on three or four channels at the same time, the shift is eliminated.
 
 
 
 

2) Last 4 photos: When setting (Ch-Ch Skew = 750ps) in the settings of channel 1. The phase shift difference between 1 and 2, 1 and 3, 1 and 4 channels is about 1 degree, but if you turn on three or four channels at the same time, the phase shift becomes about 20 degrees.
 
 
 
 

Why this happens is unclear.

[AndyBig]

The phase shift difference between 1 and 2, 1 and 3, 1 and 4 channels is about 1 degree, but if you turn on three or four channels at the same time, the phase shift becomes about 20 degrees.

This is probably due to the change in sampling rate.

[Fungus]

if you turn on three or four channels at the same time, the phase shift becomes about 20 degrees.

Why this happens is unclear.

There's only one ADC in the 'scope, shared between all channels.

When you turn on more than one channel you can see the sample rate drop.

[Fungus]

How are you connecting the signal to the 'scope? Are all the paths from the generator to the BNC connectors the same length?
At those frequencies you should be able to measure differences in the lengths of the wires.

True in principle. But a time difference of 3 ns corresponds to about 0.6 m cable length, so it should be an obvious difference, right?

The differences being discussed are sub-ns.

[Fungus]

All probes are connected in parallel to the generator output, ground probes are connected at one point of the generator ground.

Does it vary if you swap the probes around? At these frequencies the tiniest difference in capacitance between probes will show up.

[Slavius]

The phase shift difference between 1 and 2, 1 and 3, 1 and 4 channels is about 1 degree, but if you turn on three or four channels at the same time, the phase shift becomes about 20 degrees.

This is probably due to the change in sampling rate.

Perhaps yes, I agree that this is not a completely correct measurement, which is associated precisely with lowering the sampling frequency.

Roughly speaking, when 2 channels are turned on, the sampling frequency drops by half, and when 3 or 4 channels are turned on, the sampling frequency drops by 4 times.

If we take the initial maximum measured frequency for this oscilloscope = 100 MHz, then for two channels it will be = 50 MHz, for channels 3 and 4 = 25 MHz.

At such frequencies I think everything will be fine.

p.s. I'll try to do the same experiment with Rigol DS 1054Z. Just out of interest, whether it will behave the same way or not.

[ebastler]

if you turn on three or four channels at the same time, the phase shift becomes about 20 degrees.

Why this happens is unclear.

There's only one ADC in the 'scope, shared between all channels.

When you turn on more than one channel you can see the sample rate drop.

Yes, it will probably have to to with the discretisation due to the limited sampling rate. The sampling interval is 1.6 ns or 3.2 ns (625 or 312 MHz sampling rate), larger than the observed skew of 0.75 ns.

Still, the scope should know if the samples are not acquired at exactly the same time, and correct for any offsets. And it is strange that CH1 is different from the others. Smells like a firmware glitch... Do other DHO800s show the same effect?

[Mechatrommer]

Adc is interleaved 0.8ns apart (1.25GSps) so it seems dso doesnt do built in deskew. Best to verify in 1X match length 50ohm environment, otherwise just deskew in setting whatever probes you have from same source.. older ds1054z you have to do mental deskew imagination in brain if it matters. Ymmv.

[ebastler]

Adc is interleaved 0.8ns apart (1.25GSps) so it seems dso doesnt do built in deskew.

Right, the observed phase shift may be exactly the ADC sampling interval. (Give or take some "analog" tolerances.) But the thing is that the scope does apparently compensate the skew across channels 2..4, and channel 1 too when running at 312 MSa/s. Only when running at 625 MSa/s, and then only for channel 1, the compensation is missing.

That would imply a firmware bug. Hence I am curious whether all DHO800 show this effect.

[Fungus]

Still, the scope should know if the samples are not acquired at exactly the same time, and correct for any offsets.

Easy to say, but ... in the real world there's probably many little hardware things adding up.


Question: Are the probes properly compensated?

Overlay all four traces on top of each other and compensate them to match each other exactly.

[ebastler]

Easy to say, but ... in the real world there's probably many little hardware things adding up.

Question: Are the probes properly compensated?

Overlay all four traces on top of each other and compensate them to match each other exactly.

Please take another careful look at Slavius' reply #2605. He did manually compensate the skew carefully -- but then it's off again when he enables more channels.

How would "little hardware things" and probe compensation change when you enable a third or fourth channel, cutting the sampling rate in half? But the observed skew does change then. In my opinion this clearly points to a firmware glitch, not doing the skew compensation properly.

[Fungus]

Question: Are the probes properly compensated?

Please take another careful look at Slavius' reply #2605. He did manually compensate the skew carefully -- but then it's off again when he enables more channels.

Probe compensation uses the compensation signal on the front panel.

Edit: It's probably not that though. That matches the probe capacitance to the input capacitance but it doesn't tell you anything about the total capacitance of probe+input.

[Fungus]

In my opinion this clearly points to a firmware glitch not doing the skew compensation properly.

More likely they're not even trying to because it's a fools errand.

Connect up the probes, compensate the skew manually, take the measurements. Don't fiddle around with the setup once you've got it dialed in.

[ebastler]

In my opinion this clearly points to a firmware glitch not doing the skew compensation properly.

More likely they're not even trying to because it's a fools errand.

Connect up the probes, compensate the skew manually, take the measurements. Don't fiddle around with the setup once you've got it dialed in.

You still don't get it. The skew is quite stable over time, according to Slavius. But it changes depending on how many channels are enabled. Do you expect users to redo the skew compensation whenever they switch a third channel on or off?

I would really be curious whether all DHO800s behave this way -- they should, if my assumption of a firmware bug is correct. But nobody seems interested enough to try and put this in the bug report thread if confirmed?

[Fungus]

You still don't get it. The skew is quite stable over time, according to Slavius. But it changes depending on how many channels are enabled.

 I got it...

Do you expect users to redo the skew compensation whenever they switch a third channel on or off?

Oh, the humanity...

Who knows what switching the sequence of inputs at the ADC can do when there's a different capacitance on each path.

I would really be curious whether all DHO800s behave this way

Aren't you more curious to see if any of your oscilloscopes do this?

[Mechatrommer]

btw, this skew thing should be in bug report thread, not here.. Slavius should organize the test as suggested to prove this is a FW bug or just test setup mismatch... i suggest moving the last chunk of posts here starting from his post to bug thread.

[Fungus]

btw, this skew thing should be in bug report thread, not here.

We don't know if it's a bug or a curiosity yet.

It's not a hack though, I'll give you that.

[ebastler]

Aren't you more curious to see if any of your oscilloscopes do this?

No, I am no longer curious since I checked this on the SDS800X HD. Worst measured phase deviation between channels is 0.3° with a 100 MHz sine signal, i.e. 0.3/360 * 10 ns = 8 ps. I did not do any skew adjustments on the scope, and the skew is independent of the number of channels I enable.

Those measured values may be lucky; they are significantly better than the guaranteed datasheet spec of 100 ps for interchannel skew. Which in turn is much better than Rigol's "spec" (they give a typical value only) of 2 ns (!).

[Mechatrommer]

..Worst measured phase deviation between channels is 0.3° with a 100 MHz sine signal, i.e. 0.3/360 * 10 ns = 8 ps. I did not do any skew adjustments on the scope, and the skew is independent of the number of channels I enable.

maybe measurement in term of phase is less relevant, maybe we need measurement in time delay... and delay changes when all or only 2 CH actives, maybe we can find the reason why it changes.

It's not a hack though, I'll give you that.

thats why for certain, it should not be here..

[ebastler]

We don't know if it's a bug or a curiosity yet.

Getting the answer to that question is exactly why I suggested that other DHO800 owners check whether it's a systematic issue. I was not interested in a "mine is better than yours" competition -- although I am pleased enough with the result now that you asked for it.

It's not a hack though, I'll give you that.

Yep; I mentioned that this should be in a different thread in my very first reply to Slavius. But it's near-impossible to move a discussion elsewhere once it has been started; people (myself included) just keep replying as long as it's interesting.

maybe measurement in term of phase is less relevant, maybe we need measurement in time delay... and delay changes when all or only 2 CH actives, maybe we can find the reason why it changes.

I agree, a delay time is what we are really after. But I don't have a source for very steep pulse edges (and the scope would bandwidth-limit them anyway). Letting the scope measure the phase deviation between two sine waves has the advantage of using the full waveform, so it should be reasonably accurate.