Square signal level > 20ms not looking good - Siglent SDS800X HD

[bdunham7]

At the moment, i would like to know, why there is a fluctuation of that signal, that is quite long. Also why it is depending on how the signal was milliseconds before.

This appears to be because the input amplifier has dual paths, one for DC/LF and one for everything else.  It is the DC/LF path that is causing the issue, it has nothing to do with probes or input capacitance.  I noticed this issue on my SDS1104X-E a while back and posted about it here:

https://www.eevblog.com/forum/testgear/siglent-sds1104x-e-anomaly-on-what-should-be-a-simple-task/msg3453152/#msg3453152

https://www.eevblog.com/forum/testgear/dmmcheck-plus-multimeter-calibration-reference-experiences/msg4955995/#msg4955995

Originally I thought is was an overload recovery issue, but the second link showing a 100Hz square wave clearly illustrates that you don't need an overload to see it.  They never addressed this issue in the SDS1000X-E series and now it appears that  Siglent has gone ahead and released another scope with the same issue.  More expensive Siglent models don't seen to have the same issue. I'm not sure why this shows up in the lower-priced designs, but it does appear to be consistent.

[David Hess]

This appears to be because the input amplifier has dual paths, one for DC/LF and one for everything else.  It is the DC/LF path that is causing the issue, it has nothing to do with probes or input capacitance.  I noticed this issue on my SDS1104X-E a while back and posted about it here:

Tektronix used DC coupled dual path input amplifiers going back to at least 1982 once JFET operational amplifiers became available.  The advantage is no requirement for a matched pair of JFETs because the operational amplifier can control the DC offset.

In the old Tektronix design, there is a medium frequency compensation control to make sure that the gain of the fast path matches the gain of the slow path.  Modern designs still have this requirement, but somehow try to get away with no adjustment.

[thm_w]

Originally I thought is was an overload recovery issue, but the second link showing a 100Hz square wave clearly illustrates that you don't need an overload to see it.  They never addressed this issue in the SDS1000X-E series and now it appears that  Siglent has gone ahead and released another scope with the same issue.  More expensive Siglent models don't seen to have the same issue. I'm not sure why this shows up in the lower-priced designs, but it does appear to be consistent.

Dave did point out in the teardown that the front end was different between 1000 and 2000. SDS800 and SDS1000 HD appear to have the same front end. Maybe not a direct copy of the 1000x but its similar with the two trim pots.
Would be disappointing to see on a $1400 scope if this is the case, not as bad when its a $400 unit.

youtu.be/Pf1TSnMRh7E?t=777
youtu.be/XLC1CPWznTE?t=660

[eTobey]

Is there any video, that puts some light on this issue? At least it would be good to know, that you have a tolerance of +-25mV on the signal in this case.

[eTobey]

More measurements about this:

The fluctuation depends on a few circumstances i have figured out so far:
- the vertical scale
- the position of the level (signal) in the vertical range (without extending out of screen)
- size of the step of the signal
- probe att. / coax (less important)

Too much things, that one could account for roughly. The total deviation is about +- 40mV, without that initial spike. (picture "t1" -"t3")

Not something i would have expected from such a scope. If it has 12 bits, and a zoom function, then it should show the signal more accurately (a bit more at least).

Some more pictures that show what happens when overdriving the signal a added too. I named the pictures so that one could make a slideshow with them, and see how it behaves.

[eTobey]

I was curious, how my cheap, old and tiny DSO would perform. Never thought, that they would look so much better. Notice, that the DSO Nano V2 has a 10mV scale.

Its a 6.7V squarewave with the same frequence as above.

[joeqsmith]

First noticed the long ground lead, unterminated output, then wondered if you were saturating the input and there was a recovery time.  Not that these aren't contributing but then followed bdunham7's links, looks like some of the low end Siglent scopes have a problem. 

https://www.eevblog.com/forum/beginners/probe-compensated-but-level-of-square-signal-gt-20ms-not-looking-good/msg5523046/#msg5523046

This one is just wow!
https://www.eevblog.com/forum/testgear/siglent-sds1104x-e-anomaly-on-what-should-be-a-simple-task/msg3451402/#msg3451402

[tautech]

The OP one day after gaining additional experience will discover the meaning of Step Response and how that might or not, impact on real measurements.

Until that day this can be summed as, chasing ghosts.

Anyone that has repaired and calibrated scopes should know about step response and now adjustment for best is always a tradeoff.

[David Hess]

If I had an oscilloscope with that transient response, I would consider it broken.  The response below is from the fast rise output of my PG506, which may be considered perfect, into my Tektronix 2232 which is well within its specifications.  The digital storage mode trace is practically identical.  There is nothing to see past the time shown.

[thm_w]

The OP one day after gaining additional experience will discover the meaning of Step Response and how that might or not, impact on real measurements.

Until that day this can be summed as, chasing ghosts.

Anyone that has repaired and calibrated scopes should know about step response and now adjustment for best is always a tradeoff.

No one expects a poor step response at such low frequencies (100Hz).
Why not admit its a cost saving measure?

[tautech]

The OP one day after gaining additional experience will discover the meaning of Step Response and how that might or not, impact on real measurements.

Until that day this can be summed as, chasing ghosts.

Anyone that has repaired and calibrated scopes should know about step response and now adjustment for best is always a tradeoff.

No one expects a poor step response at such low frequencies (100Hz).
Why not admit its a cost saving measure?

Maybe it is but I have personally seen sub-par step response adjustment from the production line.
Also in several old CRO's I've repaired.

In real use cases it not need be absolutely perfect and in most cases adjustment is a tradeoff.
Then we can introduce different probes, cable assemblies and terminations to deliver all manner of results.
Which is correct to the uninitiated ?
How can they know they might be doing it wrong, something that is not industry standard ?

The rabbit hole is deep for those with limited experience.

[David Hess]

No one expects a poor step response at such low frequencies (100Hz).
Why not admit its a cost saving measure?

Poor lower frequency settling time is now a more difficult problem in the sense that it less understood now than in the past.  I expect it in modern designs, but not old ones.  Those old discrete transistor designs used lots of circuit tricks to achieve it, which are now lost.

Operational amplifier settling time is a good example.  In the short term, it only depends on slew rate and bandwidth, and the math is straightforward, however over a longer period of time, it depends on thermal or electrostatic effects, which is why if you want fast precision settling time, factors other than slew rate and bandwidth are important.  In datasheets you just have to rely on the 0.1% and 0.01% settling time specifications, and notice that they have very little to do with slew rate and bandwidth.

I keep laughing at the newer 12-bit DSOs, because they never settle to 12-bits, and do not present 12-bit noise performance.  Such performance might be possible, maybe, but requires at least a heroic effort involving hybrid construction and lots of discrete parts.

[eTobey]

I could have given some real use case measures, but the Siglent SDG1032 is not a device i would trust. It sometimes does things that would ruin your day, as it is buggy and sends a few negative volts down the line when going 1mV below 0V with the knob.  And from experience, i cant tell, that there will be more bugs. This is why i never connected it to my circuit. 

Is a 4ms rising edge still considered a step response? It does perform as bad in this case. That is quite dissapointing.

[bdunham7]

Poor lower frequency settling time is now a more difficult problem in the sense that it less understood now than in the past.  I expect it in modern designs, but not old ones.  Those old discrete transistor designs used lots of circuit tricks to achieve it, which are now lost.

Operational amplifier settling time is a good example.  In the short term, it only depends on slew rate and bandwidth, and the math is straightforward, however over a longer period of time, it depends on thermal or electrostatic effects, which is why if you want fast precision settling time, factors other than slew rate and bandwidth are important.  In datasheets you just have to rely on the 0.1% and 0.01% settling time specifications, and notice that they have very little to do with slew rate and bandwidth.

I keep laughing at the newer 12-bit DSOs, because they never settle to 12-bits, and do not present 12-bit noise performance.  Such performance might be possible, maybe, but requires at least a heroic effort involving hybrid construction and lots of discrete parts.

It's a matter of degree.  Nobody actually expects a 12-bit DSO to have 12 bits worth of noise (LF or HF) or DC gain performance.  It's been discussed elsewhere, but an ENOB of 10 would be spectacular.

However, even an 8-bit DSO can do much better than what we're looking at here.  This is an SDS2354X+, an 8-bit scope with a "10 bit mode" (using an averaging algorithm) that seems pretty much perfect at first glance:



If you use the vertical zoom to look at the top you see a very slight dip in 10 bit mode, but the noise in 8-bit mode masks it entirely.  The 10-bit mode does wonders for HF noise and resolution but really can't improve the performance of the front end and 8-bit ADC when it comes to LF noise issues.





The previous examples were all "on-screen", but if you increase the vertical sensitivity so that only the top is on the screen--but not overloading the input amp beyond the stated DC offset--you start to see more of the same LF dip.  So this model is much better, probably acceptable to most, but still not perfect.

 

[eTobey]

However, even an 8-bit DSO can do much better than what we're looking at here....

What alternative 8-bit scope would there be for the SDS800X HD? Lets say to 800$? I dont mind it having fewer features.

That Siglent SDS800X is like having a Ferrari, that has cheap tires. It looks good, sounds good, and accelerates fast, but you cant go fast on the curves. 

[eTobey]

I have now made the use case tests with the SDG1032, and the scope. For voltage levels and timing (1.65V to 3V; 34kHz; 10%-50% duty). Coax is of course terminated, but with cheap termination)

The fluctuations are now "hidden", but i guess they will confuse me one day, and lead me to look for problems in my circuit that are not there.

The green trace is a 10x probe, and the red is a coax (50Ohm term).

This fluctuation alone would add an error of +- 0.9% in any of my calculations.

I wonder if a 350MHz probe would look better here (Green trace at "Use case levels an timings.png")

[shapirus]

I have now made the use case tests with the SDG1032, and the scope. For voltage levels and timing (1.65V to 3V; 34kHz; 10%-50% duty). Coax is of course terminated, but with cheap termination)

Well that sure doesn't look good. So is it the same with other signal sources, have you finally ruled out the signal generator?

OTOH it's "only" about a 1.5% deviation in the 10Hz signal fed via coax... but still not good. More in the 34 kHz signal.

It'd be interesting to know how it would compare to the Rigol DHO800/900 in the same test. Can't really check because it can't zoom in like that.

What's especially weird is how different the transitions look for the 10x probe. Did it have the variable compensation cap in the same position in both tests?

[tautech]

Clint Eastwood:
A man needs to know his limitations equipment.

2 screenshots from a perfectly compensated 10x probe.





And they are perfect when the LED downlights above are set to OFF.

[eTobey]

It'd be interesting to know how it would compare to the Rigol DHO800/900 in the same test. Can't really check because it can't zoom in like that.

What's especially weird is how different the transitions look for the 10x probe. Did it have the variable compensation cap in the same position in both tests?

But you can use math channels and scale these?

What both tests?

Clint Eastwood:
A man needs to know his limitations equipment.

I would like to know that equipment you are using. Cant be an SDS800, because it has a time without LAN connection. 

I guess its "perfect" at 15 Hz too?

[tautech]

I would like to know that equipment you are using.

This question explains a whole lot, the man just cannot read or take note of any fine detail. 

[shapirus]

But you can use math channels and scale these?

Yes. I believe noise will make it useless, though, but I'll still try it some time later.

What both tests?

Both screenshots.

[eTobey]

I would like to know that equipment you are using.

This question explains a whole lot, the man just cannot read or take note of any fine detail. 

That man is surely not alone in this regard. 

But you can use math channels and scale these?

Yes. I believe noise will make it useless, though, but I'll still try it some time later.

What both tests?

Both screenshots.

You can use averaging?

What both screenshots? I posted a lot ;-)

[shapirus]

What both screenshots? I posted a lot ;-)

I was referring to a specific post that I quoted. It had exactly two screenshots.

[eTobey]

What both screenshots? I posted a lot ;-)

I was referring to a specific post that I quoted. It had exactly two screenshots.

These were from the same setup without any changes, only the freq/duty was changed.

[joeqsmith]

I would like to know that equipment you are using.

This question explains a whole lot, the man just cannot read or take note of any fine detail. 

I couldn't find what you used as well in your four posts.