Settling Time Measurement Assistance

[Tdeball]

Hello,

I have been having issues measuring the settling time of a pulsed 16 DAC (LTC2641) used in a piece of precision test equipment I am designing. The DAC has a supposed settling time of 1us, but settling to what - 99%, 99.9%, 99.99%? If you zoom out and look at the waveform from 1V/division, it looks great and matches the datasheet image. If you offset and zoom, the DAC value is 10s-100s of counts off and the reading does not level off and get lost in the measurement noise until ~100us - far from the 1us spec. I have quite a bit of equipment and get different results from different sources. I presume the differences are coming from different filtering and/or loading. I have also tried buffering the DAC output to assist with the probe loading and I get the same readings across instruments.

When measuring with a 14-bit or 12-bit oscilloscope, I can compensate the 10x probe but observe the same issue as the DAC - if you offset and zoom, even the probe compensation looks terrible. How can you trust a scope measurement if you can't confirm that a "good" square wave is "good". I know that I will quickly get answers like "a perfect square wave is impossible" and I understand that, but I would have thought generating and measuring a square wave with a settling time of a few or tens of microseconds would be an easier task. I have also tried the DP0011A differential probe but it was too noisy to resolve the top corner of the rising edge.

I then used the digitizing feature on my 7.5 digit DMM7510 at 500,000sps, but I believe the front end is heavily filtering the signal and not showing the true rise time.

I realize I am outside of the normal use case for this equipment and that this may be a harder issue than I give it credit. This DAC pulse is used as an input to my control system for highly sensitive test equipment that I am trying to tune to get the sharpest edge I can. My edge is sharper than I can measure so I cannot tune it further - which I believe I can do if I was able to measure it. I also cannot tell if the waveform is "clean" in this region if the scope probe is filtering out any oscillations.

I would love some guidance or suggestions.

Thanks,
Tom

[iMo]

Schematics, pictures?

[tggzzz]

Time and amplitude?

What exactly do you mean by "offset and zoom"?

Oscilloscopes take a long time to recover from input overloads, i.e. where the input signal is a long way "off" the top or bottom of the display.

The traditional way to avoid that is with a sampling scope, where the input signal is ignored except at the sampling instant.

A high resolution scope could be used if the waveform is captured so that the entire waveform is visible on the screen, then post-capture enlargement is used to examine the settling. That can't be done with an 8-bit scope.

[macboy]

...
Oscilloscopes take a long time to recover from input overloads, i.e. where the input signal is a long way "off" the top or bottom of the display...

That is usually true but not always. Some scopes have front ends specifically designed to handle such situations, so that offset measurements of that kind can be made without the usual issues. The venerable Rigol MSO5000 series is an example.

[tggzzz]

...
Oscilloscopes take a long time to recover from input overloads, i.e. where the input signal is a long way "off" the top or bottom of the display...

That is usually true but not always. Some scopes have front ends specifically designed to handle such situations, so that offset measurements of that kind can be made without the usual issues. The venerable Rigol MSO5000 series is an example.

How does it avoid the problem?

I've just skimmed the 8-page data sheet. It make no mention of that ability, which I would have thought would be a useful selling feature. (Mind you, it doesn't mention the ADC resolution let alone ENOB!).

[Tdeball]

Time and amplitude?

What exactly do you mean by "offset and zoom"?

Oscilloscopes take a long time to recover from input overloads, i.e. where the input signal is a long way "off" the top or bottom of the display.

The traditional way to avoid that is with a sampling scope, where the input signal is ignored except at the sampling instant.

A high resolution scope could be used if the waveform is captured so that the entire waveform is visible on the screen, then post-capture enlargement is used to examine the settling. That can't be done with an 8-bit scope.

I think this solved my issue. Capturing the full waveform and then zooming with a high resolution scope (tested with HD304MSO and SDS2204X HD) works better than trying to set the offset and vertical scale prior to capturing the waveform. This let me better calibrate my probe and then I captured my waveform and the settling time of my waveform was less than my probe's - something I have not been able to confidently say to date.

Thank you!

[mawyatt]

I think this solved my issue. Capturing the full waveform and then zooming with a high resolution scope (tested with HD304MSO and SDS2204X HD) works better than trying to set the offset and vertical scale prior to capturing the waveform. This let me better calibrate my probe and then I captured my waveform and the settling time of my waveform was less than my probe's - something I have not been able to confidently say to date.

Thank you!

Recall Jim Williams at LT had an app note about this very subject. The solution to observe the long tail of settling was employing limiting diodes to keep the signal from saturating the scopes input. As the signal starts to settle towards final value the level drops below the diodes conduction and become observable. The diodes have a DC offset bias which is roughly the final value of the signal under test, thus limit the waveform as seen from the scope input to +-Vd until the signal approaches final value and below the diodes limit of +-Vd.

We've also found the Siglent SDS2000X+ and SDS800X HD to have very good front ends and offset range

Best

[David Hess]

That is usually true but not always. Some scopes have front ends specifically designed to handle such situations, so that offset measurements of that kind can be made without the usual issues. The venerable Rigol MSO5000 series is an example.

The oscilloscope performance should still be verified.  For instance the old Tektronix 7A13 vertical amplifier uses active clamping for fast recovery, and is advertised as such, but what they meant by "fast recovery" is still pretty slow.  It was just faster than oscilloscopes without active clampling.

Incidentally, a "high resolution" 12 bit oscilloscope is unlikely to ever settle to 16 bits.

Recall Jim Williams at LT had an app note about this very subject. The solution to observe the long tail of settling was employing limiting diodes to keep the signal from saturating the scopes input. As the signal starts to settle towards final value the level drops below the diodes conduction and become observable. The diodes have a DC offset bias which is roughly the final value of the signal under test, thus limit the waveform as seen from the scope input to +-Vd until the signal approaches final value and below the diodes limit of +-Vd.

Jim Williams wrote a couple of application notes (attached) discussing settling time measurements, and he was not the only one.

[Tdeball]

I’ll dig into those. I watched the Jim Williams YouTube video and read the accompanying app note earlier and it was very informative. Hopefully I will not have to make a custom measurement circuit but it would be awesome to have the capability showed in the video/app note.

Thank you!

[David Hess]

The false summing node circuit shown below is what PMI recommended, and is a very simple version of what Jim Williams suggested except for lacking a way to prevent oscilloscope overload.

[Tdeball]

Interesting that there is not a product or open source project/design for settling time measurement. Is this a niche measurement? Is there any interest or demand in settling time measurement equipment?

[David Hess]

Interesting that there is not a product or open source project/design for settling time measurement. Is this a niche measurement? Is there any interest or demand in settling time measurement equipment?

It is a very difficult measurement to make unless the false summing node method is used, so it requires building the measurement into the circuit which is not suitable for a general test instrument.  The difficulty comes about because without the false summing node method, the test source and measuring instrument must themselves have a better settling time than the DUT, which is somewhere between difficult and impossible.

Practically every pulse generator available does not produce a suitably flat pulse.  A reference level pulse generator like that used for oscilloscope calibration might work, but how common are those?  I do not know of any currently produced.

Common oscilloscopes used to make the measurement will overload at the sensitivity required.  Sampling oscilloscopes are immune to overload so can do it, but how many sampling oscilloscopes are currently in production?

So a project or design would be complex and expensive and not worth the trouble for such a rare measurement.  Building a custom test circuit to use a common pulse generator and oscilloscope is the easier and less costly solution.

[tggzzz]

I don't think the OP has given a numerical statement of their UUTs rise time. I haven't followed every reply in this thread.

Perhaps a well-implemented 74lvc1g* step generator or Leo Bodnars step generator would be sufficient.

[David Hess]

I don't think the OP has given a numerical statement of their UUTs rise time. I haven't followed every reply in this thread.

Perhaps a well-implemented 74lvc1g* step generator or Leo Bodnars step generator would be sufficient.

Based on the original post, the measured settling time is in microseconds.

It does not require anything so fast unless measuring settling time of a similarly fast device, like a fast current feedback operational amplifier, and very fast sources may still have long settling time tails making them unsuitable for testing slower devices at higher precision.

An 74lvc1g* driving a bipolar or diode current switch (like Jim William's design) would be a good choice and could achieve high precision to below a nanosecond, but any tri-state or open collector/source would work for a faster than 1 microsecond step.