high frequency capacitive divider - self resonance

Thread Starter

jonjames85

Joined Dec 13, 2017
16
Hi Everyone,

I am trying to design a measurement system (capacitive divider) for signals with frequency content up to 100MHz, however, I am struggling to find suitable capacitors with suitably low ESL. Even a few nH (10nH in LV part of divider C4 below) causes oscillations when excited by an exponential rise (due to self-resonance?) or a high initial spike when including a series resistor in an attempt to damp the oscillations. I have attached an image of the spice circuit and results. Does anyone have any suggestions on how to mitigate this issue, or suggest a very good high-frequency capacitor (require 50-100nF, very high dV/dt, voltage rating >100V).

View attachment 156330


Thank you in advance

Jon

upload_2018-7-17_16-21-20.png
 
Last edited:

DickCappels

Joined Aug 21, 2008
6,312
This is a little bit confusing because you are showing the results of a simulatioin and asking for suggestions for real world parts. If you can't get a ring-free simulation you would have to be very lucky to get the same circuit to work better in the real world. What is source of inductance in the simulation?

Do you have any description of your real world circuit that you can post for suggestions?
 

Thread Starter

jonjames85

Joined Dec 13, 2017
16
This is a little bit confusing because you are showing the results of a simulatioin and asking for suggestions for real world parts. If you can't get a ring-free simulation you would have to be very lucky to get the same circuit to work better in the real world. What is source of inductance in the simulation?

Do you have any description of your real world circuit that you can post for suggestions?
Hi DickCappels,

Thanks for your response, I am trying to include parasitic elements in an effort to predict unwanted oscillations at the design stage. I understand that I may not be able to account for all parasitic elements. For the circuit shown I have included capacitor C4 (100nF) equivalent series inductance and resistance. The oscillations are related to the capacitors self-resonance. I have searched for suitable capacitors (or bank of capacitors) according to the specifications I have listed and for those that will have a self-resonance above the 100MHz to no avail. I was hoping that someone could suggest a suitable capacitor, or a mitigation technique to minimise or even remove the oscillations?

Kubeek,

the transmission line is matched with a 50ohm resistor and terminated to the 1MΩ input of an oscilloscope, the oscillations appear to be purely from the self-resonance of capacitor C4. When the ESL of this capacitor is disregarded there are no oscillations.

Many thanks

Jon
 

Thread Starter

jonjames85

Joined Dec 13, 2017
16
Just realised you won't be able to see the inductance and resistance, these are accounted for in the capacitor properties dialog box.
upload_2018-7-17_15-55-3.png
 

kubeek

Joined Sep 20, 2005
5,718
the transmission line is matched with a 50ohm resistor and terminated to the 1MΩ input of an oscilloscope, the oscillations appear to be purely from the self-resonance of capacitor C4. When the ESL of this capacitor is disregarded there are no oscillations.
No it is not. The transmission line is loaded with 1MΩ+50Ω ohms (or in other words 1000050Ω), which is a long way from 50 ohms - the termination resistor needs to be to ground and not in series.
 

Thread Starter

jonjames85

Joined Dec 13, 2017
16
Kubeek,

Apologies you are right, in attempting to simplify the circuit I have deleted a few too many elements. I initially had this terminated with a Burch termination, this I believe compensates for the cable capacitance, see below. Even with the correct termination, oscillations still appear to be related to C4 inductance?

upload_2018-7-17_16-21-20.png

Thanks

Jon
 

kubeek

Joined Sep 20, 2005
5,718
I don´t think I understand why is C4 parallel to C2? Do C1 and C2 also have specified inductance?
You can mitigate inductance by using multiple capacitors in parallel, also I think there are some HF caps with special packages to get low inductance.
Can you also post your asc file so that we can play with it a bit?
 

Thread Starter

jonjames85

Joined Dec 13, 2017
16
C2 is the capacitance of my sensor its shield. I could increase to the required 100nF but would be better to include this in a separate 'LV arm'. C4 will form the LV side of the divider I have attempted several smaller capacitors in parallel (all identical), however with the details of the capacitors I have found, the inductance still causes an issue. I have attached the asc and .lib file for the cable, sorry I don't know how to make the .inc statement a relative file path so this will need changing. Any advice you have is appreciated.

Thanks
Jon
 

Attachments

Thread Starter

jonjames85

Joined Dec 13, 2017
16
Looking at it some more, are you really measuring a 700kv pulse with a rise time of 50ns?
actually the aim is to measure 700-850kV peak, with rise times down to 10ns. It is proving quite a challenge ;), many issues due to magnitude, frequency, radiated fields etc. but plenty of literature to suggest it has been done before. I have done lots of modelling (FEM) for the high voltage side but neglected the LV side until now, my mistake.
 

kubeek

Joined Sep 20, 2005
5,718
That sounds really interesting, may I ask what is the source of such signal?
I am still thinking about solving the inductance issues in the divider in a physical way, and not compensating it later. Could you do the divider at the source side, something like a planar pcb capacitor with three layers? That would I think make the inductance so low that it should not have any effect on the signal.
 

Thread Starter

jonjames85

Joined Dec 13, 2017
16
It is interesting but quite the challenge. The transients are normally associated, but not restricted to the breakdown of pressurised insulating gases, such as SF6. I had considered a planar PCB capacitor very briefly but as I am trying to keep the LV side components as physically small as possible to be mounted close to the point of measurement, I assumed that it would be impossible to get the kind of capacitance value that I'm looking for (~100nF). Maybe I will have a more in-depth look into it.
 

kubeek

Joined Sep 20, 2005
5,718
Also come to think of it, you could do the division in a cascade, say 1:10 in each step, 10pF to 100pF should be doable on a pcb. When you do that four times, you won´t need a 100nF cap at the end, if the only purpose of that value is the division ratio.
 

Janis59

Joined Aug 21, 2017
1,148
About the general task: I had tried to realize this capacitative diveder for decades lonmg, and never got any better than +/- 100%.
My problem is ca 10...20 kV 100 MHz high impedance circuit where I want to know all the time voltage. I had tried a 0,1 pF toward 100 pF however it the real life it shows maybe 1000% error or larger. The best result is got by SMD components, especially those designed for work in GHz range. The PCB material is having ultimative role, ordinary FR4 is giving short-circuit being 10 and more ~fold of that impedance the capacitors has. The only materials having rather good conduct are teflon base Rogers company DUROID(TM) - there are three alternatives at their website for this. At least tan(fi) to that is 0,0004 instead of ordinary 0,15.
Next, logically are wire lengths. The geometry MUST be the wise enough let the wire loops are less than one or maximum two MILImeters in each portion. That is very difficult to organize.
The all my trials to mount the cascade just between the copper bars on ordinary teflon platelet, was unsuccessful, as the capacitors are expanding in high voltage, thus the system was mechanically cracked and sequentely exploded burning up to centimeter black hole. Its the 100 pieces of 10 pF 2 kV switched in series and lower part is the same capacitors 10 pieces in parallel. All kaput.
 

Thread Starter

jonjames85

Joined Dec 13, 2017
16
About the general task: I had tried to realize this capacitative diveder for decades lonmg, and never got any better than +/- 100%.
My problem is ca 10...20 kV 100 MHz high impedance circuit where I want to know all the time voltage. I had tried a 0,1 pF toward 100 pF however it the real life it shows maybe 1000% error or larger. The best result is got by SMD components, especially those designed for work in GHz range. The PCB material is having ultimative role, ordinary FR4 is giving short-circuit being 10 and more ~fold of that impedance the capacitors has. The only materials having rather good conduct are teflon base Rogers company DUROID(TM) - there are three alternatives at their website for this. At least tan(fi) to that is 0,0004 instead of ordinary 0,15.
Next, logically are wire lengths. The geometry MUST be the wise enough let the wire loops are less than one or maximum two MILImeters in each portion. That is very difficult to organize.
The all my trials to mount the cascade just between the copper bars on ordinary teflon platelet, was unsuccessful, as the capacitors are expanding in high voltage, thus the system was mechanically cracked and sequentely exploded burning up to centimeter black hole. Its the 100 pieces of 10 pF 2 kV switched in series and lower part is the same capacitors 10 pieces in parallel. All kaput.
Thanks for sharing your experience with this. I will look into the teflon base products.I agree that SMD components may be my best option and are the only components I have come across with sufficiently low ESL/ high enough series resonance.

One of the other possibilities I have come across is to compensate for the inductance using a coupled magnetic winding. Does anyone have any experience of this technique or any other techniques that they might have come across?
 
Top