2

I am designing a guard drive for a coaxial cable, to drive its shield as a guard for low leakage measurements. After seeing some issues in the sim, I arrived at the following minimum circuit which reproduces the issue:

enter image description here

![enter image description here

The lines from right to left show a source impedance variation from 100 ohm to 100 Mohm. The desired bandwidth is DC - 50kHz with a gain of ~0dB. The bandwidth can be lower for very large input impedance as they roll off sooner even with driven guard. But the amplification is a problem.

C1 is the inevitable capacitance between the signal and its guard. C2 is the inevitable capacitance between the guard and ground. These caps are inevitable because they represent what the cable and EMC caps/diodes would contribute. Individually, neither C1 nor C2 lead to oscillations when R1 is sufficient. However, when both are present, any value of R1 will lead to oscillations for some input impedances at some frequencies. I also tried more complex impedances instead of R1 but it didn't improve substantially.

So I am now puzzled. How are driven cable guards implemented in practise, to prevent oscillations?

EDIT:

The only remedy, I found so far is adding capacitance from the signal to ground (essentially adding to the Opamp input capacitance). However, in practise, adding ~1 nF of capacitance from signal to ground can also be a concerning source of leakage. And this also limits bandwidth, so I hope that there is a better solution.

tobalt
  • 18,646
  • 16
  • 73
  • 1
    Surely there must be dozens of articles online about this? – Andy aka May 08 '21 at 09:41
  • @Andyaka I thought as much but couldn't find anything meaningful, beyond "use an opamp to buffer the input and drive the guard electrode". Could you link one please, maybe I am indeed searching wrong – tobalt May 08 '21 at 10:11
  • @user287001 C1 and C2 are representative for the cable. C2 would also include some EMC capacitance. That is why I called these inevitable. A guard drive with no input cable isn't much good. The frequency range is DC - 50kHz – tobalt May 08 '21 at 10:13
  • @tobalt The purpose of the guard is clear but does it oscillate as this model with no real cable, having only 2 capacitors as a substitute for the real cable? –  May 08 '21 at 10:23
  • @user287001 This is a minimum model. I did more extensive sims before with better models for cable (lumped LCs) and actual diodes etc and noticed this problem. When tracing the problem, I arrived at this minimum circuit which considers only the cable and EMC capacitances. The positive feedback is an inherent property of guard drives, but no issue in itself with stronger negative feedback. The problem arises due to the combined presence of C1 and C2. – tobalt May 08 '21 at 10:28
  • 1
    @Andyaka nevermind. I found a much better search term. "Driven guard" as wikipedia has it turn up nothing. "Driven Shield" seems to be the name for it in the tech literature. – tobalt May 08 '21 at 10:34

1 Answers1

2

Force with opamp the guard have the same voltage as the signal wire:

enter image description here

This surely must be tuned for certain open loop freq response of the opamp, signal source resistance and capacitances. Adjust R2. I used a generic opamp which has GBW=10MHz.

This stands well some leakage resistance inserted in parallel with the capacitors. No idea what happens when there's a real cable.

ADD: The questioner said he tried my circuit in time domain simulation and found it oscillates. He's right. Good looking small signal frequency response proves nothing of the stability of feedback circuits, because it cannot reveal poles in the right half plane nor the effect of saturation or other non-linearities.

The problem is interesting because if a solution exists the high-Z signal source may be usable without a signal transmitter. I made another test circuit:

enter image description here

This time it's stable. The signal is rectangular 200us long 1V DC pulse. The circuit is tuned for 3 megaohm source resistance, the given "inevitable capacitances" and the generic opamp which has GBW=10MHz.

I'm afraid the circuit is too capacitive load for the source. But it's remarkably less than the capacitors alone would be.

ADD2 self-adaptive compensation of leak and capacitive loading which does not depend on the internal resistance of the signal source sounds ambitious. Seems like nobody else is interested than me. Harmful. But the next one is not a guard driver. It's a compensator and it needs no tuning for different source impedances:

enter image description here

It has an opamp circuit which is known as negative impedance converter. Known load C1 is compensated by inserting in parallel with it negative capacitance - the negative of C3. The opamp has GBW=10Mhz and trying too tight compensation with it makes the circuit unstable. It stands C3=298pF. C4 is inserted to push the high frequency roll-off of the opamp a little further. Without it the tendency to oscillate can be seen with smaller C3.

The signal is 200us long rectangular 1V pulse. The simulation skips the initial and shows the steady state result.

It compensates also known leakage resistance. It's R4 in the next image:

enter image description here

If the capacitance and leakage resistance are unknown they must be measured somehow. Making it automatically can be tricky. I skip it.

  • Thanks for your answer. I tried this (feedback from after R2). It has the same problem as my circuit above. For certain Rsource and certain frequencies, it will always amplify. Yes guard and signal will be held at the same voltage, but this voltage will be larger than the signal source. E.g. your circuit for 3 Mohm source will amplify. Another problem: even at 100k source, if you simulate time domain it doesnt work, because AC analysis ignores opamp saturation. – tobalt May 08 '21 at 13:19
  • It is a general spice problem. If you build opamps with large feedback resistors, they can look benign in the AC analysis, because they are still stable (using 10s of Volts..), but once plotting the transient analysis it becomes apparent that this wont work – tobalt May 08 '21 at 13:22
  • OK. It was a bad quess. Your original circuit maybe actually doesn't oscillate continuously. Is it possible to equalize the frequency response with a tuned filter or do you expect something which needs no tuning? Maybe a proper signal transmitter at the sensor end of the cable? I inserted another circuit to the answer. It's at least stable in my time domain simulation. –  May 08 '21 at 19:59
  • Thanks for your continued investment. The circuits for a generic preamp. Impedances can range from kOhms to way above GOhm. For kOhm-MOhm range the guard mainly improves bandwidth. Beyond that it enables any measurement at all. So no transmitter at the DUT side is possible. As you noticed, good solutions exist for particular Rsource-frequency combinations. So Im hoping that it's possible to make the driver impedance somehow auto-adaptive with clever arrangements of R,C,L. – tobalt May 09 '21 at 05:43