LMP7721: Unity gain stability of LMP7721

Hi Mohan,

yes, the LMP7721 is unity-gain stable. And yes, the phase margin is very small. But nevertheless the LMP7721 can work as voltage follower and can drive capacitive loads, provided a careful phase lead compensation is carried out to restore the phase margin. The following circuit could be used to drive a capacitive load of 1n.

The first plot shows the step response. The second plot demonstrates the frequency response and the third plot finally gives the result of a simplified phase stability analysis:

So, this circuit should work stably.

Of course, a simple scheme with isolation resistor can be used as well. But I wouldn't use an isolation resistor smaller than 47R:

The open loop output impedance can be found by simulation with TINA-TI. I took 30R for the simulations.

To your last question: I think that the LMP7715 and not the LMP7721 is used for the cable driver has only to do with cost efficiency: The LMP7715 is cheaper than the LMP7721.

Kai

Mohan,

Please see my answers below:

1.  Is it unity gain stable?  The datasheet indicates pretty poor phase margin (~ 35 degrees) when Av=+1.  The small signal step response has 40% overshoot.  Anything that can be done to improve this?

Yes, it is stable but as you said it does not have much phase margin.  You may improve the stability and capacitive drive by using a double feedback configuration-see below link:

https://training.ti.com/ti-precision-labs-op-amps-stability-1

2.  In unity-gain configuration, can putting a parallel R+C in the feedback path improve stability and reduce the overshoot?

This may help but it comes at the price of much lower bandwidth.  The best is to use an isolation resistor with a feedback capacitor from inverting input to output and feedback resistor from inverting input to right side of the isolation resistor (double loop configuration ).

3.  If I want to drive a 1.0 nF capacitive load on the output of LMP7721, I have to put a series isolation resistor.  What is the minimum value of series isolation resistor required?  (datasheet gives no information about open loop output impedance of the amp)

For an op amp like LMP7721, it's typically enough to use a small isolation resistor of 10-50ohm to drive 1nF load.  Use 10mV square waveform and verify the stability by looking at the small-signal overshoot (ideally less than 25 %)

4.  All of the guarding examples in the datasheet show the guard driven from a different amplifier, not from the LMP7721.  Can the LMP7721 output itself be used to drive the guard trace?  (using some series resistance to isolate the LMP7721 from the capacitance of the guard trace, of course).  Why is this approach not shown in the datasheet?

It is because the guard must be driven to a voltage present at the non-inverting input terminal and most op amps are used in a gain different than 1.  However,  in case of a buffer, you may drive the guard with the output but be aware that each 1 meter (1 yard) of coax  line has an equivalent capacitance of around 100pF.

Also, please be aware that if you are truly concerned about IB leakage, you must use a triax guard - see page 20 of LMP7721 datasheet for details.

Mohan,
A stability of the system is determined by the small-signal overshoot directly at the output of the amplifier and NOT on the right side of the isloation resistor. If you are driving high impedance or are not concern with the gain error, you may just use a buffer followed by the isolation resistor ( no need for double loop) to drive guard of the input line. Btw, capacitance of the guard may be just enough to filter out the signal before it's passed to the second stage so there may be no need for 1nF output cap. If you have any more questions, please show your proposed schematic so we better understand what you are trying to do.

Hi Kai, Marek and Thomas (thanks for all your replies, not sure who to address this to?)

Attached is a circuit drawing of what I am proposing to do.  You can assume the power rail is 3.3V.

I need a second stage amplifier because the LMP7721 is not Rail-to-Rail output.  I chose OPA376.  Is this a good choice?

It is driving a 12-bit SAR ADC.

The GUARD trace is derived from the LMP7721 output through a 1K/1000pF RC filter.  I am using the 1K to isolate the LMP7721 from

the capacitance of the guard trace.  I added 1000pF because this forms a nice filter which should hopefully eliminate any overshoot that appears in the LMP7721 output.  I use the same RC filter between STAGE 1 and STAGE 2 circuits.

The 604 ohm load is based on the LMP7721 datasheet, where the amp seems to show better stability with some kind of output load.

Please let me know what you think.

Thanks and Regards,

Mohan

Mohan,

What you propose should work BUT if the input signal is dynamic in nature and you are concerned about the input leakage current, I would remove C21 and lower R31 to avoid long RC delay causing the guard voltage to lag the input voltage resulting in a higher average leakage current - see below.

And if your input signal is dc or low frequency, I would drive the guard with the same RC fed to second stage - see below (you may adjust C27 1000pF to 610pF to account for 390pF guard capacitance).

Hi Marek,

Thanks for the reply and suggestions.  My input signal is not very dynamic.  You can consider my signal source to be a current source, it is outputting step function pulses of current every 200 milliseconds or so.  I am accumulating each current pulse on the 390pF capacitor and then taking ~100 milliseconds to measure it.  During that 100 milliseconds I do not want the current to leak off the capacitor, therefore the guard trace.  Please let me know if you think my original circuit will work for this case.

I am unclear what the advantages are of your second drawing, i.e. taking the guard trace from the signal going to the 2nd stage amplifier.  I separated it into two RC branches so I could select a different RC time constant going to Stage 2 if desired.  Is there a disadvantage to this approach?

Also I do not understand how we can reduce 1000pF-> 610pF.  Observe that the 390pF is not actually connected to my guard trace; the guard trace is surrounding it, but not connected.  So how can you subtract the 390pF from my original 1000pF?

Thank you,

Mohan

If signal is slow enough, I don't see an issue. But it is puzzling to me why would you want to have ANY extra delay in driving the guard line.  Having said that, you may need 100-200ohm in series with output to assure stability of driving the guard (you never said how long is the coax line).

Hi Mohan,

a guard makes only sense if it is driven by the guard driver as low ohmically as possible. Stray capacitance and the guard resistance form a voltage divider and the higher the guard resistance is the worse the mains hum can be suppressed:

That's why even in PH and ECG amplifiers the guard is driven very low ohmically. The guard resistance must only be high enough to sufficiently isolate the load capacitance from the guard driver output, in order to not erode the phase margin too much.

Kai

Hi Marek,

OK thanks.  Let me explain a bit more.  I thought the 1K/1000pF is a good idea because if there is any overshoot and ringing on the LMP7721 output, I do not want that to couple back to the guarded trace, and maybe cause oscillation (through positive feedback).  So I am trying to damp the LMP7721 output a little bit, to reduce ringing, before driving the guard trace.  Does that make sense?

Also... there is no coaxial line.  The input to the circuit is coming from a metal plate which is located nearby.  This metal plate forms part of a capacitive sensor.  We are picking up pulses of charge on this metal plate.  These are then partially transferred to the 390pF capacitor which forms a capacitive divider.  Here is a more complete drawing of the input stage:

Does this make more sense?

Unfortunately I am unable to surround the metal plate with a guard.   We may have problems with ambient 60 Hz pickup, but I am not sure yet.  Possibly the 390pF capacitor will help minimize this.

Thoughts?

Hi Dennis

Thanks for the follow up.  I was waiting for a reply from Marek on my last comment.  Following that we should be able to close the thread pretty soon.

Can you ping Marek for me?  Thank you.

Thanks,

Mohan