[SOLVED] Fast summing amplifier ringing when inputs are multiplexed

[Materiewelle]

Hello everyone!

I have a project where I need to create a series of voltage pulses with high accuracy and low noise. The pulses consist of two components, which are each supplied by a DAC. The DACs can achieve >~ 1us settling time, but it's desirable that the pulses settle faster, and that the pulse timing can be controlled independently from the DACs by one digital line per component.
Integrated, simple, and compact solutions are preferred, and it's also preferred to not inflate the BOM with additional ICs. Power and thermal budget are available in abundance. There is already the fast, accurate, and low noise difference amplifier LT1995 and the dual switch ADG5436 in the design. The very low noise and relatively fast OpAmp AD8597 is used as a DAC buffer. The DACs are bipolar and can be modeled by their 6.8k output resistance and equivalent feedback resistance (on chip), respectively.

When the DAC outputs are just added directly, the pulses are smooth and don't overshoot or ring, which makes sense because the DACs are quite limited in bandwidth and the dynamic behavior is optimized with the 10p caps.

No switches used: Pulses limited in BW, but don't ring. Settling is limited by the DAC and timing control is via DAC interface, which is not preferred. Pulse timings and magnitudes are typical for the application.

Now, when I use switch ICs to construct the pulses, I see rather horrible ringing:

Switch ICs added for potentially faster settling and simple / independent digital interface. Summing A+B amplifier now overshooting and ringing.

Detail.

Now my obvious question: Do you have any suggestion how I could reduce the ringing of the summing amp? Is the amp simply under-damped or do you think this is excited by some sort of glitch that the switches introduce (break-before-make characteristic, charge injection)?

I have tried the following:

• RC lowpass behind each switch. Works great but degrades BW and initial accuracy.
• Switch IC with lower charge injection and larger R_on. Does not seem to change much and large R_on also degrades accuracy.

As mentioned earlier, I would like to keep the ICs used here. A simple fix is of course preferred.

Looking forward to your suggestions!

All the best

 

[danadakk]

Your confident your probing is correct when taking measurements ?

ABCs of Probes

Learn about the fundamentals of probes and discover how to choose the right probe for your oscilloscope and get reliable measurements every time.

www.tek.com

Look at settling time and overshoot graphs in datasheet.

Bypassing on OpAmp, C types matter.

Both bulk and .01, .01 ceramic disk on supply pins.....look at their ref design board bypassing.
Evaluate the ceramics in a 50 ohm jig for response, not all ceramics equal ESR and effectiveness.

Layout and stray C at input pins to OpAmp can have large detrimental effects.

https://www.ti.com/lit/an/sloa046/sloa046.pdf

POST a well focused pic of your pCB layout and component placement.

Some significant C associated with that switch :

Look at G peaking due to Cload, another consideration, eg. phase margin :

Regards, Dana.

 

[KlausST]

Hi,

amplifiers / OPAMPs are not good in reproducing high dV/dt input signals.

* Thus the first step should be to low pass filter the input. This helps the amplifier not to get out of regulation (maybe even get into input stage saturation) .. thus always being able to follow the input signal

* next step would be to see if it is necessary (after #1) to reduce ringing at all. A local C as feedback or RC as feedback should solve this.

* next step would be to chose devices (amplifiers) that can fulfill your requirements.

*****
Your circuit:

I can´t see why you choose the differential amplifier at all ... while you don´t use it as differential amplifier.
It would be cheaper and less critical (signal) to use just a simple OPAMP in non-inverting configuration.

Also an OPAMP driving a "switch" should be considered like driving a capacitance.
Many Opamps are bad in driving capacitance, thus datasheets often tell how to drive high capacitive loads.
Also switching a load always causes the regulation loop to get upset .. resulting in ringing.

Additionally if you expect an OPAMP to react fast ... you need to provide a stable power supply to them. This means local power supply bypass capacitors.
I know ... some simulators don´t need them. But ony because they "expect" the power supply to be perfect. In a real world the power supply is not perfect.
Thus I recommend to add them in the simulation .. this also helps to not forget them in the real circuit.

*****

To be able to help you in more detail .. you first need to do your job: Provide specifications regarding input signal and expected output signal behaviour regarding timing, precision, and so on. Give numbers with units, not "blury" informations like "as fast as possible". Math does not work with texts.

Klaus

 

[danadakk]

Don't despair, wide bandwidth OpAmps a big challenge to tame.

Even R parasitics matter. And don't underestimate the variation in C's used
for bypassing, especially ceramics. I did tests on .01 & .1 uF's many years ago between
different vendors. Real eye opener. Just do a simple 50 ohm jig to test.

Many fast opamp datasheets have supply bypassing and layout recommendations.
Not this one it seems. Look at some other fast OpAmp datasheets as the general
principles apply on layout and bypassing.

Look at ringing frequency, clues given about LC values that can sometimes point
to root of contributor.

See attached, Jim Williams several discussions in these ap notes on fast OpAmps.

find this on web :

Jim_Williams_The_Art_and_Science_of_Analog_Circuit_Design.pdf

Regards, Dana.

 

[Materiewelle]

Thanks, Dana and Klaus.

This is only a simulation, which is why decoupling caps are not present in the schematics. It is therefore not a problem of layout at this point.

Low-pass filtering the input has drawbacks, such as reduced accuracy. This is also the reason why no standalone Op-Amps are used. The difference amplifiers come with precisely matched resistors. Such precise resistors as standalone parts would actually be more expensive than the integrated solution. (Cost of components is anyways secondary in this project).

That the switch has fairly high capacitance is actually a very interesting comment from both of you. I think this would rather effect the amps BEFORE the switches though, which don't show problematic ringing, see the switch_out_A/B traces in the second graph. It's really the amp AFTER the switch that is struggling.

"next step would be to see if it is necessary (after #1) to reduce ringing at all": That is an interesting thought... After all, the signal settles after only about 300 ns, which is sufficient in my application. I am just wondering if the overshoot could be reduced below 1V and the ringing to less than one period... Is there a simple way to reduce the bandwidth of the amp by placing a passive in the feedback path or similar?

Some specifications that are relevant:

• All amps supplied with +-15V.
• DAC output voltages A & B:
  • Output range (each) -13.5...13.5 V (limited by software).
  • < 80 nV / rtHz white noise.
  • < 500 ppm full scale initial accuracy.
• Sum A+B:
  • Abs(A + B) < 13.5 V (guaranteed by software).
  • < 100 nV / rtHz white noise.
  • < 1200 ppm full scale initial accuracy.
  • Period of pulse train ~20 us.
  • ON time A: ~5 us (typical).
  • OFF time B: ~2 us (typical).
  • Settling time < 500 ns.
  • Overshoot < 1 V, no more than ~1 period of ringing.

Best regards

 

[danadakk]

External compensation of OpAmps :

https://web.mit.edu/klund/www/papers/ACC04_opcomp.pdf

https://www.ti.com/lit/an/slva662/slva662.pdf

Regards, Dana.

 

[crutschow]

That the switch has fairly high capacitance is actually a very interesting comment

It's not the capacitance so much, as the charge injection from the switch changing state that may be the problem.
That's difficult to compensate for.

Since the load appears to be a high impedance, perhaps you could just use the output of the switch directly, rather than going through that final op amp.

 

[Materiewelle]

It's not the capacitance so much, as the charge injection from the switch changing state that may be the problem.
That's difficult to compensate for.

Since the load appears to be a high impedance, perhaps you could just use the output of the switch directly, rather than going through that final op amp.

Thanks, Zapper!

The output signal of this circuit is used as reference input to another circuit, and the input impedance of this one is in fact not so large, only a few kOhms. So the amp is needed. The fact that the signal is used further downstream is what makes the overshoot and ringing so worrisome.

I think that with the constraints that are given in this project, it is probably best to see if the overshoot / ringing is acceptable.
If not, simple RC low-pass filters behind each of the switches can be a consideration, even if it will reduce the accuracy that can be expected before calibration.

With 100 Ohms / 360 pF it looks rather smooth. Of course, another downside is added delay and longer rise / fall times, but I guess you can't have everything...


Thanks for your help you all. If you have any further suggestions, I'd be glad to hear them!


All the best