Relatively high voltage precision DAC amplification

[Laval]

Hello everyone.

I'm working on a DC Voltage Reference/Calibrator project that would be comparable to Data Precision 8200 (hopefully better) in terms of specification i.e. about 0.01% accuracy (preferably better) with 6 digits resolution. The unit will be using a much more modern approach and components of course. I will receive my evaluation board for the AD5791 this week. That should allow me to experiment and prototype the DAC/Reference part.

I would like the unit to have a 100mV, 10V, 100V and 1kV or 1.2kV ranges. Since the DAC obviously won't output more then 2.5 to 10 volts (depending on vref), it will need some accurate amplification. This is not a big deal up to about 40V or so as there are many good precision op amps available. For the 100V range, I found this op amp:

https://www.analog.com/media/en/technical-documentation/data-sheets/adhv4702-1.pdf

This should be fine for 100V. An application note in the datasheet suggests a voltage range extension (basically bootstrapping) using an n and p MOSFETs on the positive and negative power rails respectively but it can only double the range. 200V is nowhere near 1kV. I'm also a bit concerned about stability since this is a rather fast amplifier for a DC application. It doesn't look like there is much choice anyway.

There is a 1kV op amp made by a company called Apex Microtechnology but it is very expansive (over 1k CAD). I guess I could go discrete and make an amplifier stage using an LTP feeding a cascode with negative feedback for linearity much like the VAS stage of classic audio amplifiers. This is what I'm considering at the moment.

I'm interested in any suggestion for the 1kV amplification (as well as the rest if you like). What would be a good strategy to get to 1kV while maintaining the precision and low noise.

Some complementary information:

• The device could be used to calibrate instruments that have a low enough precision but it would be used mostly as a precision voltage source.
• The device will be a benchtop unit with its internal power supply (one of the challenge because of high voltage).
• The voltage source doesn't have to be able to source much current as it will mostly be used with high impedance circuits. Maybe 10 or 20 ma max at 1Kv.
• I may eventually add a current source that could go to 100mA or 200mA, not sure though.

The complete and definitive specification is not yet determine as a lot will depend on what I can do with available components but accuracy is the main goal.

[PCB.Wiz]

The voltage source doesn't have to be able to source much current as it will mostly be used with high impedance circuits. Maybe 10 or 20 ma max at 1Kv.al.

That's still 20 watts...
One advantage you have is this does not need to be fast, it will be a challenge getting high precision high voltage resistors...

You could experiment with photovoltaic drivers as they allow slow and isolated drive into a high voltage MOSFET, and keep things simple ?
Your HV supply will need a soft start, as the capacitance of any MOSFET will tend to give a spike at power up.

[Laval]

That's still 20 watts...

20ma isn't really an important design goal. 5ma at 1Kv would be fine which is still 5 watts.

... it will be a challenge getting high precision high voltage resistors...

This is my main concern at the moment though, I assume, it could be possible to bodge some less precise resistors to get the desired values as long as they don't drift too much and have good TC. It looks like Caddock does make such resistors but I just don't want to look at the price for now 

You could experiment with photovoltaic drivers ...

I didn't know about these. I assume you are talking about these components:

https://www.mouser.ca/ProductDetail/Littelfuse/FDA117GRTR?qs=sGAEpiMZZMug9GoBKXZ75woEA5q4EdH8PW4%2FqvT8hLwjDiSWmxKgRA%3D%3D

That may be interesting. I can't open the datasheet PDF file for some reason. They would give isolation for sure but how do they compare to traditional opto couplers ?

Your HV supply will need a soft start, as the capacitance of any MOSFET will tend to give a spike at power up.

For sure. I have worked on inrush current limiting before and inrush current measure using zero crossing opto. I don't even know yet if I will use a SMPS or a linear PSU. Getting good transformers for linear PSU is a bit of a pain and they make the unit large and heavy.

Thanks for the thoughts.

[mawyatt]

Consider using a Switchmode Boost Regulator operating with the feedback resistor chain (use a few series Precision Rs for the HV) that will regulate to a few volts above the desired DAC set point. Using the same resistor divider chain with a different tap off as feedback with an op-amp and optical driver to drive a HV MOSFET as the output of a "Linear Regulator" to provide the desired output voltage. Since the HV Switchmode Boost output to the "Linear Regulator" is only a few volts above the desired output, power dissipation will be low.

Or use a separate low resolution DAC to control the HV Switchmode Boost Regulator, this is likely easier and cheaper as no need for precision DAC, Resistors or anything.

Be sure and provide proper overcurrent and short circuit protection, maybe a quick overcurrent/short circuit latching type circuit type protection which requires a user Reset to reactivate for user and component safety.

Best, 

[PCB.Wiz]

You could experiment with photovoltaic drivers ...

I didn't know about these. I assume you are talking about these components:

https://www.mouser.ca/ProductDetail/Littelfuse/FDA117GRTR?qs=sGAEpiMZZMug9GoBKXZ75woEA5q4EdH8PW4%2FqvT8hLwjDiSWmxKgRA%3D%3D

That may be interesting. I can't open the datasheet PDF file for some reason. They would give isolation for sure but how do they compare to traditional opto couplers ?

Yes, broadly. They are voltage clamped low current generators, so can provide gate voltage too. Traditional opto couplers are Diodes+transistor so need additional power.

The fine print matters for your application.
I find Toshiba photovoltaic drivers TLX9905 and TLX9906 and they differ slightly.
The TLX9905 sims like a simple clamped current source (needs external discharge resistor), while the TLX9906 has an active turn off helper, which I can sim as a diode+transistor+resistors.

For linear control of a FET, parts like the TLX9905 are what you need, as they will behave better in a feedback loop.
Spice indicates it can loop-regulate ok, with a slew response ballpark similar to the switching times.  Expected from uA and nF.
You have stable resistive loads and can tolerate slower response times.

[David Hess]

This is my main concern at the moment though, I assume, it could be possible to bodge some less precise resistors to get the desired values as long as they don't drift too much and have good TC. It looks like Caddock does make such resistors but I just don't want to look at the price for now

At high voltage, it is not only about drift and temperature coefficient; the voltage coefficient of resistance becomes a significant error term.

If I could not get a suitable divider network, I would consider using multiples of the same resistor in series to make a divider.  This way the same voltage is across each resistor and the voltage coefficient of resistance should at least partially cancel out.  With a proper thermally balanced layout, the temperature coefficients should also partially cancel out.

[jbb]

This is a totally sideways post, so feel free to disregard.

I saw a note once (I think pertaining to a TI part) that shunt type references can be series connected to form a higher voltage reference directly. This could perhaps be useable in the vicinity of 100 V. But probably impractical for 1 kV. Also may not be stable enough.

[Salitronic]

You could consider a cascaded amplifier configuration such as this: https://www.researchgate.net/publication/283661284_A_Cascaded_Linear_High-Voltage_Amplifier_Circuit_for_Dielectric_Measurement
and scale it to your required voltage range.

I have designed similar cascaded systems for high voltage (kV) signal generation but typically using cascaded H-bridge (switch-mode) converters which is not suitable for your very low noise requirement. It is a very convenient design topology as you have to deal with multiple modules but each one at a low voltage. Power is also split among all modules making thermal management a lot easier. Obviously it is a complex design with multiple modules each requiring independent isolation and synchronized control of all modules.

[RoGeorge]

100mv, 10v, 100v and 1Kv or 1.2Kv ranges
...
Maybe 10 or 20 ma max at 1Kv.
...
I may eventually add a current source that could go to 100ma or 200ma, not sure though.
...

Side note, kilovolt is small letter "k" followed by capital letter "V", so "2kV", not "2Kv".

"2Kv" is incorrect, and it means nothing.  Kilo prefix is always denoted by the small letter "k", never the capital letter "K".  Volt happened to be denoted by capital "V".  Other units might be with a small letter, for example meter is with a small "m", so 1.6km should be with small "k" and small "m".

For current, it's "mA" for milliampere, not "ma".  Capital "A" stands for ampere, small "m" prefix stands for mili (10^-3) and capital "M" prefix stands for Mega (10⁶):  https://en.wikipedia.org/wiki/Unit_prefix

[moffy]

100mv, 10v, 100v and 1Kv or 1.2Kv ranges
...
Maybe 10 or 20 ma max at 1Kv.
...
I may eventually add a current source that could go to 100ma or 200ma, not sure though.
...

Side note, kilovolt is small letter "k" followed by capital letter "V", so "2kV", not "2Kv".

"2Kv" is incorrect, and it means nothing.  Kilo prefix is always denoted by the small letter "k", never the capital letter "K".  Volt happened to be denoted by capital "V".  Other units might be with a small letter, for example meter is with a small "m", so 1.6km should be with small "k" and small "m".

For current, it's "mA" for milliampere, not "ma".  Capital "A" stands for ampere, small "m" prefix stands for mili (10^-3) and capital "M" prefix stands for Mega (10⁶):  https://en.wikipedia.org/wiki/Unit_prefix

While I agree that 'mA' is technically correct, using a capital to abbreviate a proper name (Ampere, Volta), I think 'ma' has won by common usage.

[Hydron]

Personally I would not regard "ma" as ok just because some people haven't been educated as to what the proper capitalisation is.
On topic, maybe look at the schematics of older instruments which do have similar source ranges. I have a Keithley 237 which comes in handy as a high-ish voltage source (up to 1.1kV), though it's held back a bit by it's age - only 14 bit DACs are used so you only get 4 digit source resolution.

[SteveThackery]

While I agree that 'mA' is technically correct, using a capital to abbreviate a proper name (Ampere, Volta), I think 'ma' has won by common usage.

Oh, nooooo! The SI system of units is brilliant and we engineers owe it to ourselves, to learners and students, and to future generations to protect them against abuse.

[Laval]

You could consider a cascaded amplifier configuration such as this: https://www.researchgate.net/publication/283661284_A_Cascaded_Linear_High-Voltage_Amplifier_Circuit_for_Dielectric_Measurement
and scale it to your required voltage range.

That's an interesting read, thank you.

[Conrad Hoffman]

Long ago I worked for Burleigh Instruments who made HV amplifiers for piezoelectric devices. We used a fairly standard circuit (and simple) circuit with HV TV transistors. One trick was using series feedback resistors because voltage coefficient was a big factor. Nothing special, just RN70D resistors. We did find that MOX resistors were bad in this application. You'll find a HV amplifier circuit in The Art of Electronics; it's a direct rip-off of the Burleigh circuit and they hint as much in the text, though they did make one small but important improvement. Suggest you look at that. I've also attached the manual/schematic for one of the Burleigh products. It's a terrible scan but all I've got. We had only primitive op-amps back in the day and you'll do far better today as more open loop gain will improve linearity.

[voltsandjolts]

AoE chapter 9 has some ideas at page 695 onwards.

[Laval]

That is a lot of information and material to digest and thinker with. Thank you to everyone. Now, I have something to start with.

An interesting (and scary) thing. VCR really is a big problem with HV.

https://www.mouser.com/pdfdocs/Welwyn-VCR-Characteristics.pdf

[Conrad Hoffman]

Find an AoE 2nd edition (there are pdfs online) and look at page 169. That's the basic circuit we used to use for just about everything, both adjustable supplies and amplifiers. Back then we used Delco HV transistors and then Telefunken BU209 and various others.

[ZGoode]

Just want to join in here since I need something similar for my LCR bias fixture (although I only need up to 50V).  I was looking at the PA89A chip for this since it seems to be a nice ready to go amplifier for higher voltage.  Only downside is it is very expensive.  Getting a second hand one on eBay (a used pull that may or may not work) is still going to be at least $100 if I go this route for this.

[David Hess]

What about noise requirements?

There is something to learn about the design of the Data Precision 8200 from its specifications.  Notice that the noise is 10 times higher on the 100 volt range than the 10 volt range.  This means that the 100 volt output stage is operating at 10 times the gain of the 10 volt output stage, which makes sense.  The gain of the output stage is multiplying the input voltage noise of the output stage.

10/100 microvolts RMS over 10 kHz is not particularly demanding if the adjustable reference source is 10 volts.  Any of the proposed high voltage output stages should meet this requirement.

Given your output current requirements, I would be tempted to use a high voltage class-a instead of class-ab output stage, with an NPN emitter follower driving the output positive and an NPN current sink pulling the output negative with 5 milliamps or whatever.  Or high voltage n-channel MOSFETs could be used just as easily.  Many oscilloscope high voltage z-axis amplifiers operate somewhat like this with a transimpedance configuration, which is equivalent to an inverting amplifier.

[Laval]

Just want to join in here since I need something similar for my LCR bias fixture (although I only need up to 50V).  I was looking at the PA89A chip for this since it seems to be a nice ready to go amplifier for higher voltage.  Only downside is it is very expensive.  Getting a second hand one on eBay (a used pull that may or may not work) is still going to be at least $100 if I go this route for this.

Yes, PA89A is the one i was referring to in my original post. I don't know your complete specification but would this amplifier do ?

https://www.mouser.com/ProductDetail/Analog-Devices/ADA4522-2ARMZ?qs=UfUFg%2FkmHHHprAoiiQKXwA%3D%3D

[ZGoode]

Oop, don't know how I missed a zero.  I meant +/-500V, not 50V

Just want to join in here since I need something similar for my LCR bias fixture (although I only need up to 50V).  I was looking at the PA89A chip for this since it seems to be a nice ready to go amplifier for higher voltage.  Only downside is it is very expensive.  Getting a second hand one on eBay (a used pull that may or may not work) is still going to be at least $100 if I go this route for this.

Yes, PA89A is the one i was referring to in my original post. I don't know your complete specification but would this amplifier do ?

https://www.mouser.com/ProductDetail/Analog-Devices/ADA4522-2ARMZ?qs=UfUFg%2FkmHHHprAoiiQKXwA%3D%3D

[Laval]

What about noise requirements?

There is something to learn about the design of the Data Precision 8200 from its specifications.  Notice that the noise is 10 times higher on the 100 volt range than the 10 volt range.  This means that the 100 volt output stage is operating at 10 times the gain of the 10 volt output stage, which makes sense.  The gain of the output stage is multiplying the input voltage noise of the output stage.

10/100 microvolts RMS over

Indeed, the relative amount of noise is the same in terms of ppm for both ranges. I could probably live with that noise specification unless I can make better.

[ZGoode]

The Systron Donner M107 service manual might be a good starting point for a high voltage reference/amplifier.  I was looking at this as well.  The service manual has the full schematics for a high voltage system, I just haven't had the time to do a deep dive into the schematic yet.

[David Hess]

What about noise requirements?

There is something to learn about the design of the Data Precision 8200 from its specifications.  Notice that the noise is 10 times higher on the 100 volt range than the 10 volt range.  This means that the 100 volt output stage is operating at 10 times the gain of the 10 volt output stage, which makes sense.  The gain of the output stage is multiplying the input voltage noise of the output stage.

10/100 microvolts RMS over

Indeed, the relative amount of noise is the same in terms of ppm for both ranges. I could probably live with that noise specification unless I can make better.

Low frequency noise from 0.1 to 10 Hz is usually more relevant.

Reference noise almost always overwhelms amplifier noise by at least an order of magnitude.  In this case, the noise from the high voltage divider is also a significant source.  The feedback resistor of the high voltage divider can be bypassed with a capacitor to reduce high frequency gain, lowering high frequency noise, at the expense of settling time and bandwidth which probably do not matter.  If this is done, then the amplifier's feedback point needs to be protected from high voltages, but this is not difficult.

Under those conditions with an inverting high voltage amplifier, I think the noise could be improved by 10 times, at which point the reference noise will dominate.

In the intended applications, 10 or 100 microvolts of noise over 10 kHz was probably insignificant, so I would not worry too much about achieving lower noise.

[Conrad Hoffman]

We used to use the Apex parts when we needed a complete amplifier in a small space, or several of them. They make great parts but I'd never use them for a personal project because if I pop it, I can't fix it, only replace it. A discrete circuit can be fixed at low cost.