Reducing the preshoot of inverting integrators

[Atomillo]

Hello all!

I am interested in finding out if there are any techniques out there to deal with the pre-shoot phenomena common of inverting integrators, especially for high bandwidth versions.
I understand that the classical analysis of the phenomena is that with a very fast step at the input, the feedback capacitor is a short, the opamp does not have any time to react and thus the output voltage is set by the divider formed by the input resistance and the output resistance of the opamp.

Thus, a simple way to reduce this could be to place a small capacitor between the inverting input and ground, which is what I have seen done with slower integrators (for example, the integrator used in the HP3456 ADC has a 470 pF connected from the input to ground). However for faster integrators (for example around 50 MHz or 100 MHz) the additional poles introduced by this capacitor seem like trouble.

I have done some SPICE sims with small inductances placed in series with the opamp input (to be obtained with small ferrites for example) with the idea of increasing the first impedance of the divider at the very high frequencies that the step signal has while leaving normal operation unaffected, but the results do not seem so great so far.

Has anyone encountered this problem before? If so, could you share any techniques you have seen to deal with it?

[dietert1]

Yes, opamp circuits have this limitation.
I'd guess a proper way to avoid deviations is to bandwidth-limit the input signal to a bandwidth where the opamp can still work well. There are many ways to do it. To implement a low pass you could divide the integrator input resistor in two and use a capacitor from their middle point to Gnd. Thus you combine a fast passive integrator with a slow active integrator, if you want so. I didn't simulate it, but it should work well.

Regards, Dieter

[jbb]

You could have a look at Deboo integrators. They avoid the problem at the cost of some more resistors.

https://www.analog.com/en/resources/technical-articles/consider-the-deboo-singlesupply-integrator.html

[David Hess]

You could have a look at Deboo integrators. They avoid the problem at the cost of some more resistors.

https://www.analog.com/en/resources/technical-articles/consider-the-deboo-singlesupply-integrator.html

Fast integrators often use that configuration, and there are even better ways.  An operational transconductance amplifier can drive a capacitor directly to make a very fast integrator.  The integrator in a fast function generator uses switched voltage controlled current sources, with the switching taking place at the current outputs which can be very fast with little disturbance.

I have seen capacitance added in a network at the input to an operational integrator to improve precision, and I have done it, but it always seems like a race against the devil.

[temperance]

An operational transconductance amplifier can drive a capacitor directly to make a very fast integrator.

Don't they suffer from the same problem when the current must reverse instantaneously?

[David Hess]

An operational transconductance amplifier can drive a capacitor directly to make a very fast integrator.

Don't they suffer from the same problem when the current must reverse instantaneously?

I do not think so, but there are not many choices for parts.  Analog Devices has one (LT1228) and TI has one (OPA860).  TI also has the old slower one, the LM13700.

I guess they do not solve your problem if you require an inverting integrator, but they are likely part of a solution.

https://www.ti.com/lit/ds/symlink/opa860.pdf?ts=1744498698320&ref_url=https%253A%252F%252Fwww.ti.com%252Fproduct%252FOPA860
https://www.analog.com/media/en/technical-documentation/data-sheets/1228fd.pdf

Usually the solution with operational integrators is to capture the fast part of the input current into a capacitor, with the operational integrator handling the lower frequency part.  Otherwise a fast enough current pulse would not even be captured.

[Atomillo]

Dear all:
Many thanks for your help! I will reply individually below:

Yes, opamp circuits have this limitation.
I'd guess a proper way to avoid deviations is to bandwidth-limit the input signal to a bandwidth where the opamp can still work well. There are many ways to do it. To implement a low pass you could divide the integrator input resistor in two and use a capacitor from their middle point to Gnd. Thus you combine a fast passive integrator with a slow active integrator, if you want so. I didn't simulate it, but it should work well.

Regards, Dieter

Dieter I think this is the way to go! I was already experimenting with placing a passive low pass filter to compensate for the opamp finite GBW, but now I can more aggressively reduce the bandwidth with the passive filter first and add a zero in the opamp circuit, which at high frequencies would effectively work then as an inverting amplifier. Initial simulations are promising and while I need to figure out the many details along the way this is very interesting. Thanks!

You could have a look at Deboo integrators. They avoid the problem at the cost of some more resistors.

https://www.analog.com/en/resources/technical-articles/consider-the-deboo-singlesupply-integrator.html

Jbb, I was aware of Deboo integrators, but the reason why I need an inverting configuration is because eventually I will need to add and integrate many identical signals. While in principle I see that adding signals can be done with a non-inverting integrator, the lack of a virtual ground makes me sceptical of possible “crosstalk” between inputs.

Fast integrators often use that configuration, and there are even better ways.  An operational transconductance amplifier can drive a capacitor directly to make a very fast integrator.  The integrator in a fast function generator uses switched voltage controlled current sources, with the switching taking place at the current outputs which can be very fast with little disturbance.

I have seen capacitance added in a network at the input to an operational integrator to improve precision, and I have done it, but it always seems like a race against the devil.

David Hess, you have come very close to the present prototype! I currently have an OPA860 connected as a common base which adds the input currents and are then integrated into an output capacitor. DC stabilization of the network is not trivial, but besides that, the issue is the relatively high current noise of the OTA at 1 kHz.

I have considered designing my own bipolar common base stage, but then you run at the classical tradeoff of Zin vs current noise, which I can’t really afford (since I need the very low Zin to add the signals in the first place). Furthermore, the value of the input resistor is a compromise between low enough to have high gain and high enough to no be affected by the finite Zin of the topology.

It is true that open loop designs do not have to worry about preshoots and other errors caused by loop dynamics but I have been rudely remembered why feedback is so important when trying to exploit those advantages hahaha

Once again, thanks to all for the discussion!

[Andree Henkel]

An operational transconductance amplifier can drive a capacitor directly to make a very fast integrator.

Don't they suffer from the same problem when the current must reverse instantaneously?

I do not think so, but there are not many choices for parts.  Analog Devices has one (LT1228) and TI has one (OPA860).  TI also has the old slower one, the LM13700.

I guess they do not solve your problem if you require an inverting integrator, but they are likely part of a solution.

just switch the inputs of the OTA configured as integrator to get inverting type.

[David Hess]

just switch the inputs of the OTA configured as integrator to get inverting type.

I was thinking in terms of the inverting input not shifting because of delay of feedback.  An OTA operating open loop allows its input to shift without feedback.