AC Attenuator for DMM

[nano_user]

I have a problem with design of AC frequency compensated attenuator for hi-precision DMM. As far as I know, these frequency compensated attenuator are made either by a parallel connection of capacitors and resistors to reduce the impact
of parasitic capacitances, or only on inductances and capacitances. Inductive attenuators have a small impedance at low frequencies and are not suitable for a voltmeter. Capacitance vary greatly from temperature. A parallel resistive-capacitive
attenuators will have different coefficients depending on the frequency (the coefficient of resistance ratio is not equal to the coefficient of capacitance ratio), since the capacitances have large tolerances. How to be? This attenuator is designed
to work with voltage up to 1000 V and from 1 kHz up to 10 MHz. The accuracy of the coefficients themselves is not important, the main thing is their stability with respect to temperature, frequency and time.
I will be grateful for literature and advice about this problem

[David Hess]

High impedance oscilloscope dividers face the same problem.

Discrete designs use trimmer capacitors to adjust the compensation.  Older hybrids have laser trimmed dividers with surface mount discrete trimmer capacitors.  Newer hybrid designs might use laser trimming of both the resistor network and capacitor network.

Another issue to beware of is "hook" generated by the printed circuit board substrate.  At low frequencies, the dielectric constant may change significantly with frequency and this can depend on humidity.

[TiN]

To make discussion constructive, there should be more information of what stability and temperature range is required. Design for 1% attenuator for -40C to +105C is very different to the one that is stable to 10ppm and +18..+28C. Perhaps good example of later ones would be study of AC front end in Wavetek 4920 and Fluke 792A. Fluke 5790A have some interesting aspects too, but it is covered in less details.

[Gyro]

Welcome to the forum nano_user,

I'm not sure how much it helps, but here's a writeup of an attenuator I built recently for 10kV peak with AC compensation...

https://www.eevblog.com/forum/testgear/high-voltage-x10-attenuator-for-my-gay-mta-memory-voltmeter/msg1723094/#msg1723094

Building one for 1kV should be at least 10 times easier, although you still need to be very careful of insulation and overload tolerance.

Things that I learned:

- Voltage coefficient of resistors is at least as important as capacitor TC. You need to go for higher voltage parts than you think you need, and pick low voltage coefficient specified ones.
- When capacitance trimming, beware of the voltage rating of trimmer caps. I used trimmed cable length to adjust compensation.
- Safe air-wiring avoids PCB leakages and parasitics.
- You need an accurate way of measuring the capacitance of assembled RC assemblies.
- If you can avoid screening then you can avoid more parasitics, not sure how practical in your case.
- NPO or C0G capacitors should be more available for 1kV use, series connection could help with voltage rating.

Work out how you are going to achieve consistent capacitance connection to your DMM (banana plugs and leads won't be sufficient) and check how consistent the DMM's input capacitance is across different voltage ranges.

[nano_user]

Thanks for the help 
I need about 5-15 ppm/C divider ratio temperature stability at 15-28 C. I can not take a attenuator from high-precision DMM, because many of the components for them were made on special order and are not available.
At the moment I'm going to design a resistive-capacitive divider on PTFE-like PCB with PTFE trimmer capacitors in parallel with constant caps.

[coppercone2]

Why even use resistors for AC attenuation? Why not just a capacitive divider?

nmv, I think the answer is input impedance consistancy. You could buffer it with a high voltage follower and then use a capacitive divider though. Would only work on low ranges, or you would need some kind of crazy high voltage amplifier. For AC some of your biggest bastards would be handled, like DC offset means nothing really. Biggest problems would be distortion, noise and rail  voltage.

[Magnificent Bastard]

I have a problem with design of AC frequency compensated attenuator for hi-precision DMM. As far as I know, these frequency compensated attenuator are made either by a parallel connection of capacitors and resistors to reduce the impact
of parasitic capacitances, or only on inductances and capacitances. Inductive attenuators have a small impedance at low frequencies and are not suitable for a voltmeter. Capacitance vary greatly from temperature. A parallel resistive-capacitive
attenuators will have different coefficients depending on the frequency (the coefficient of resistance ratio is not equal to the coefficient of capacitance ratio), since the capacitances have large tolerances. How to be? This attenuator is designed
to work with voltage up to 1000 V and from 1 kHz up to 10 MHz. The accuracy of the coefficients themselves is not important, the main thing is their stability with respect to temperature, frequency and time.
I will be grateful for literature and advice about this problem

Do you need the entire 1KHz to 10MHz range from just ONE divider, or can you get away with multiple dividers (maybe 3?) that are optimized for low-medium-high frequencies?

I assume you need 1:10:100 -- is that correct?

[nano_user]

High voltage amplifier it's exotic, which is not exactly necessary here. In addition, a high-voltage amplifier with a bandwidth of tens of MHz... This at least is very expensive, but may be generally does not exist in nature.
About several attenuators - yes, most likely there will be two of them - purely resistive and resistive-capacitive, the frequency of switching from one to another will be determined empirically.
Just while the best that I've seen - it's resistive-capacitive dividers in DMM, until nothing is learned (by me) better. Until recently, I had a lot of questions about using trimmer capacitors because they are not very stable, but then I saw PTFE trimmer capacitors and they meet the requirements.

[e61_phil]

1000V and 10MHz?

That will give more than 6A peak even if you only connect one meter of coax cable

[Gyro]

1000V and 10MHz?

That will give more than 6A peak even if you only connect one meter of coax cable

Hence the OP's wish to build an attenuator?

Yes, he needs to minimize input capacitance, which implies a 'probe' type construction, rather than an input connected via a metre length of coax.

[Kleinstein]

The 10 MHz and 1000 V are likely not at the same time. If at such a high power level would have to keep a close look at dielectric losses only low loss materials would be possible.

A high value purely resistive divider starts to behave odd already at relatively low frequencies (e.g. 100 Hz). So even this would need at least approximate capacitive adjustments: usually some parallel capacitance to the lower resistor to compensate for the parasitic capacitance in parallel to the high value resistor. Air based trimmers can be rather stable - it's usually only a few 10s of pF to adjust. The problem is less with the trimmers but with parasitic capacitance that changes with something like humidity.

Because of the input impedance of the following amplifier, one might have to include a 1st buffer stage after the divider.

[Magnificent Bastard]

About several attenuators - yes, most likely there will be two of them - purely resistive and resistive-capacitive, the frequency of switching from one to another will be determined empirically.

OK, in that case, you might look into coaxial dividers.  You might see this as kind of like thermal converters, but instead of a thermal converter as a load, you would instead have a well behaved (RF rated) resistor as a load.  There are many metal film resistors that are RF rated (up to several GHz) that would be excellent for this application.  This could all be built as many RF-rated resistors in series on a long-thin PCB (Rogers material), with the lower part of the divider on the far end of the board.  This PC board would be placed (axially) inside of a large cylindrical enclosure, with a Type-N connector on each end for "input" and "output".  The first half of the divider would be surrounded with a smaller cylinder, to act as a "guard"-- one end is connect to the input voltage, and the other end floating.  The length of this smaller cylinder would be tuned in length to compensate for stray capacitance.  A similar guard is placed on the low side of the large resistance-- open at the center, and the other end connected to the center of the divider on the output connector.  These compensation guards would be tuned for flat response up to 10MHz.  This would provide a divider that would easily exceed your specs.  Oh--- and (probably) only a primary calibration lab (or NIST) would have the equipment and expertise needed to properly calibrate this divider.

For some hints on design-- search the scientific literature for "coaxial thermal converters" and "RF resistors".  This problem may have already been solved, so also search for "coaxial voltage dividers".