100VDC reference circuit

[king.oslo]

Hello there,

I am adjusting my 6 1/2 digit DMM, and I need to input several reference voltages. 100mVDC , 1VDC, 10VDC and 100VDC. My idea is to scale the voltages down from a 101V reference voltage to 10.1V, and from 10.1V to 1.01V and from 1.01V to 0.101V using voltage dividers. Then I would use potentiometers to fine tune the voltage to say 10.00000 or 1.00000.

In one of Dave's episodes (#210 Krohn-Hite DC Voltage Standard), he says it is possible to make a stable reference voltage from a zener and constant current source.

This is my attempt. I was hoping you could supply feedback to improve the design, or suggest a better idea if possible. I love to learn.

Attached is my LTSpice simulation and circuit schematic. As you can see, the voltage isn't completely stable. R2 is the multimeter. The two npn are the constant current source.

Thank you for your time.

Kind regards,
Marius

EDIT:

I see that I have put my capacitor the wrong way around.

[king.oslo]

Here is updated schematic with C1 with corrected polarity.M

[alm]

Weren't the DC voltage ranges OK according to your quick performance test?

I'm not so sure if a zener with constant current supply would be the most stable solutions, zeners tend to be fairly noisy. Their ripple rejection is worse than your typical three-terminal regulator, which is why you'll rarely see power supplies with just a zener and a pass transistor anymore. Dividing the output voltage and using a precision op-amp to compare the divided voltage with an IC reference (which is basically what a three terminal regulator does) should be more stable. I'm not sure if you need anything fancy, however. Just using a tree terminal regulator and make sure its adjust terminal stays close (< 30 V) to the input voltage might be enough (just don't short the output). With three lab supplies in series I got 100 V with a standard deviation of about 1 mV, and these were just basic linear bench supplies.

Make sure the current is very limited, since you don't need much current and 100 VDC can be dangerous.

Try searching the Volt nuts mailing list archive for information, although most only bother with a 10 VDC reference, since they can divide any higher voltage down to 10 VDC and compare them. They also aim for a very low long-term drift (something you don't need) and even lower noise.

[amspire]

Sorry to say, but the sort of circuit you are considering will not be even vaguely stable enough. If you connect your dvm to it, you will see the voltage just keeps drifting before your eyes.

First, high voltage zeners never used in precision references - way too much temperature coefficient. Zeners only can have a zero temperature coefficient somewhere in a range from about 5.1V to 6.2 volts, and so the precision reference zeners are specially manufactured and selected for low temp coefficient at a specified current. Even with a great zener, designing the matching constant current source takes experience too, so forget it.

Go instead for a good 10V IC reference like a a REF01, or something else from companies like Analog Devices or Maxim. Take a good look at the specs though. 5ppm / degC means that if you calibrate the DMM at 5 deg C different temperature from the temperature you calibrated the reference, you could be out .0025%.

Now the circuit will be a 9:1 voltage divider to divide the 100V down to 10V, a good opamp that regulates the divider voltage and the reference voltage to be the same, and a few transistors on the opamp output to allow it to control the 100V voltage. I can give you some tips there.

Next thing is the divider. With your meter, you will find nearly all resistors have a very noticeable temperature coefficient. The coefficient of metal film resistors is just huge, and the 90V across one resistor will heat it enough to start drifting very happily. One solution is to get a thick film 10:1 divider - there is a standard multimeter divider that is excellent.

Otherwise the best you can do is to get batch of 100K 1W metal film resistors all from the same manufacturers batch, and make the divider from 10 of them. This way, every resistor will be dissipating the same power, and typically the temperature coefficients will match to within a few PPM with luck.

Last to calibrate the reference is just a matter of calibrating the divider, and for that you add an extra switch to turn the divider in to a Hamon divider (see Conrad's page http://conradhoffman.com/HamonResistor.html ).

What you will do is close the switch and the divider will be 2:1 - ie the output will be 20V. Adjust the divider so that the reference to output voltage (10V) is exactly the same as the reference volts. Your meter can do this even uncalibrated. Then open the switch and the output is a precision 100V DC.

Last with reference circuits, even with a stable room temperature, you always have to turn them on and let the temperatures stabilize for at least 30 minutes before using them for precision work.

Richard.

[king.oslo]

Weren't the DC voltage ranges OK according to your quick performance test?

Yes, they were okay, but I didn't think they were fantastic.

Dividing the output voltage and using a precision op-amp to compare the divided voltage with an IC reference (which is basically what a three terminal regulator does) should be more stable.

Alm, Richard, just a minute. Are you suggesting I amplify the 10V reference to 100V?

Perhaps using something like this: http://search.digikey.com/us/en/products/PA341CE/598-1918-ND/2700702  ?M

[vk6zgo]

No,Richard is saying to divide the 100 V down to 10V,then compare it with the reference.
The resultant control voltage,which is dependent on the error, is amplified & used to control the !00v supply (normally via a pass transistor).

The old way is to use a gas filled regulator tube,but these tubes are rare.difficult to use,& need to be chosen on test to 
obtain any reasonable degree of accuracy.

[IanB]

It's basically true, though, you are taking a low voltage reference and multiplying it up. This will also multiply any errors.

The ideal way would be to take a 1000 V reference and divide it down, but 1000 V references are hard to come by.

[amspire]

Weren't the DC voltage ranges OK according to your quick performance test?

Yes, they were okay, but I didn't think they were fantastic.

Dividing the output voltage and using a precision op-amp to compare the divided voltage with an IC reference (which is basically what a three terminal regulator does) should be more stable.

Alm, Richard, just a minute. Are you suggesting I amplify the 10V reference to 100V?

I am suggesting using a precision divider with an opamp to make in effect a 10x amplifier. I will do a quick sketch.

Perhaps using something like this: http://search.digikey.com/us/en/products/PA341CE/598-1918-ND/2700702  ?M

Way to expensive. You just want to able to use a standard op-amp with some extra circuitry to go to 100V. That will be in the sketch.

[ivan747]

Perhaps using something like this: http://search.digikey.com/us/en/products/PA341CE/598-1918-ND/2700702  ?M

Way to expensive. You just want to able to use a standard op-amp with some extra circuitry to go to 100V. That will be in the sketch.

They won't send him a free sample, right?

[amspire]

Attached is a sketch of a 100V ref concept.

This arrangement will get the best stability and accuracy that you can get from cheap parts (especially the resistors) and the 100V can quickly and easily be calibrated from the10V output.

It is only designed to give a few milliamps output, which is all a reference supply needs to do.

The op-amp should be a low offset FET-type opamp - the slower the better. The faster it is, the more likely it is you will have stability issues.

I added Q2 to eliminate the effect of the miller capacitance of Q1 (to help with stability), so Q1 can be any low voltage small signal NPN transistor. Q2 can also be a high voltage NPN transistor instead, and the lower the power rating, the better (about 0.5W maximum dissipation with 200V input) . Q2 just has to have a voltage rating greater then the maximum possible voltage from the transformer. The reference is inherently short circuit proof, thanks to the 22K 1W resistor.

The gate of Q2 ( or base if you use a NPN transistor instead) is connected to the 10V ref output, just because is was a handy 10V source, but any voltage between 10 and 15V is fine, so if you have a +15V supply for the reference and opamp, you can use that for the gate of Q2 as well.

For the 100K resistors, get the best quality 0.5W to 1W 100K metal film resistors that you can find at a good price, and you need to select them. The 9 top resistors are organised into 300K triplets, and you want each triplet to match within about 0.01%. The single lower resistor needs to be selected to be lower then any of the other 100K resistors, so we can add the series adjustment pot.

Stability issues will come from the fact that the Q1 is increasing the loop gain by about 22 times. This extra gain means the inbuilt opamp compensation probably will not be sufficient to get stability, and some extra compensation will probably be required.

Richard.

[king.oslo]

Thanks Richard,

I understand how it works (finally)! I will build this up within the next few days.

I am shopping parts, and an enclosure for it (perhaps with a heater? What do you reckon I could use? I can get 17W wire-wound resistors reasonably). I need to work out which parts to buy.

With two electrolytics of 22uF the V_pp will circa 5V. Is this fluctuation going to be a problem in order to get very stable 100V_OUT? Do you recommend different type of cap? Would it not be better with 2 x 4mF?

Second, I cannot get a 10.00000V reference. But I can get one which is 10V +-5mv: https://www.elfa.se/elfa3~eu_en/elfa/init.do?item=73-069-43&toc=0. What is your opinion?

Op-amp wise. The slowest I can get is 88Hz. Surely, this slow it will perform poor regulation. What do you reckon is optimal frequency? And how to power it?

Would you advice additional output capacitance?

Thanks.M

[amspire]

First the reference. All the precision reference IC's allow you to attach a potentiometer for calibration.  Add this with a link so you can go from the factory calibration to your calibration. There is every chance your meter calibration is very good. I would test the reference and if it looks like it reads within the manufacturers specs, then adjust it so it reads 10.00000V on your meter.  The thing is most of the time, an error of 0.01% is not a big deal. What you want is all the ranges of the meter to match, so if you had 10 of these reference chips all adjusted to be exactly 10V according to your meter, then if you put them in series, you want your meter to read 100V exactly.

Now in picking the chip, look more at temperature stability and long term drift rather then accuracy.  I would prefer a 1ppm/C 0.1% reference chip to a 5ppm/C 0.02% reference. I cannot go looking at data sheets right now - doing some hard disk recovery work, but Mouser and Digikey searches are a good starting place.

Secondly, you need to get a feel for how much resistors vary when you have a meter like yours. Grab a metal film resistor - perhaps 10K, but the value does not matter. Measure it on a handheld DMM. The reading will probably look absolutely rock steady. Now put it on your dmm. Watch the values drift.  Just the sub-milliwatt heating of the resistance test is enough to make the resistor change with a 6 1/2 digit meter. Touch the resistor with the tip of a finger, and you will really see the value change.  So This is why it is so important that every resistor in the 100V divider is dissipating the same power exactly. Now when you close the switches in the Harmon divider, this is no longer the case, so you have to change your technique. You let it stabilize first with the switch off. Close the switch and adjust the output to be exactly double the reference (as described before) and open the switch. Let it settle again. Quickly switch the switch on and adjust again (you have to do it in a few seconds). Keep repeating until the initial output voltage just after you turn the switch on is exactly double the reference. Then you will know that with the switch off, you have an accurate 100V.

Third, the opamp. By slow, I meant that a 1Mhz (or even 100K) opamp will probably be easier to get stable then a 10MHz opamp.  Here is the story about stability. In a feedback circuit, there is exactly a 180 degrees phase shift between the output and the opamp input connecting to the feedback at DC (ie you are connecting to an inverting input). In my circuit, we connect to the + input as the transistor is inverting the output. As the frequency increases, capacitances causes gain fall offs and at the same time, phase shifts. If the sum of all the phase shifts reaches 180 degrees and the gain is equal or great then 1 from the opamp input to the circuit output, the circuit will oscillate.

Basically if you have the output connected back to an inverting input, and the circuit has a 180 degrees phase shifts due to capacitances, then the signal the input sees is now in-phase (non-inverting) compared to the output and instead of negative feedback, you have positive feedback. Positive feedback is always unstable if the gain of the feedback loop is 1 or greater. Positive feed back means that if the output drops a little (you added a load) the signal the opamp input will see will say "drop the output even more", and this is the start of an oscillation.

So the trick they use in internally stabilized opamp is they add a single RC (resistor-capacitor pair) fall off that swamps out all other capacitive effects. So above a few Hz, the opamp has a 90 degree phase shift from this internal RC for AC all the way down to the opamp's unity gain point  and so it is stable with just about any load other then a straight capacitive load (which could add another 90 degrees shift and make it unstable).  The opamp designers do not care of the phase shifts below unity gain get really ugly as that does not affect stability. Now to boost the output to 100V, I have added an extra gain of 22 times, and the transistor circuit will have some phase shift of its own. This means that we can get oscillations even down to the point the opamp has a gain of 1/22, and as I mentioned before, the opamp can look really ugly for phase shifts down there. So you will probably have to add some extra frequency compensation to get the circuit stable. If you are into LTSpice simulations, it is a good exercise, but otherwise, playing with the 100P capacitor in my circuit is a good starting point.

Hope that makes some sense.

Richard.

[king.oslo]

It makes sense Richard! Will be back with some pictures or video when I get something done.M

[king.oslo]

Opamp has to go all the way to V_IN, right? In my case 325V?M

[amspire]

No, the opamp will run from a supply like a 15-20V single supply. The input should operate at less then that.

[Richard W.]

At the moment i'd NOT readjust the meter. The specs aren't too bad.
I'd redo all measurements with improved enviornmental conditions, doing exact measurements isn't easy at this level.

~ Let the meter warm up. Many many hours, the longer the better.

~ Let the meter in place while do the measurements. The displayed values will drift, if you move the meter. Why?
1. Temperature distribution will change
2. The boards inside the meter will bend a little bit. Also bad.

~ Clean the plugs and sockets; contact resistance and thermoelectric effects are evil 

Here is a really nice brochure how to do it right: http://www.keithley.de/data?asset=55835
 

But if you want to build a reference, here are several appnotes to pick the right chip:
http://cds.linear.com/docs/Application%20Note/an42.pdf
http://pdfserv.maxim-ic.com/en/an/AN2879.pdf

Here you can see what happens if you mechanically stress components (especially voltage references):
http://www.amplifier.cd/Technische_Berichte/Spannungsreferenzen/Spannungsreferenz.html
(the google-translator link would have been too long...)


regards

[king.oslo]

Thanks Richard W. I will compare the meter with another reference before I decide to adjust it. I will post information in the other thread when I have it. Still, this is a great exercise, and nice to have a reference for both DC and AC!

Richard (without W ), I am probably mistaken: The 20V opamp will only open the npn transistor to conduct 20V-0.65V = 19.35V away from the 100V rail?  So the 100V rail can only be about 120VAC?

[amspire]

Richard (without W ), I am probably mistaken: The 20V opamp will only open the npn transistor to conduct 20V-0.65V = 19.35V away from the 100V rail?  So the 100V rail can only be about 120VAC?

The NPN connected to the opamp output works at low voltages. It functions up to aboutn the voltage on the gate/base of Q2. Q2 handles all the high voltages and is the only device needing a high voltage rating. I would probably aim at a supply of 130V to 200V,

[alm]

~ Let the meter in place while do the measurements. The displayed values will drift, if you move the meter. Why?

Some voltage references, including the LM 199/299/399 used in the later HP 3456A (and most other 6.5 digit) multimeters, are also sensitive to position. Its output voltage will vary slightly depending on orientation. Not sure what the exact mechanism is, I believe there was some speculation on the volt-nuts mailing list.

[king.oslo]

Here is a really nice brochure how to do it right: http://www.keithley.de/data?asset=55835
 

But if you want to build a reference, here are several appnotes to pick the right chip:
http://cds.linear.com/docs/Application%20Note/an42.pdf
http://pdfserv.maxim-ic.com/en/an/AN2879.pdf

Here you can see what happens if you mechanically stress components (especially voltage references):
http://www.amplifier.cd/Technische_Berichte/Spannungsreferenzen/Spannungsreferenz.html
(the google-translator link would have been too long...)

Richard W, these were very interesting. Thank you.

Richard (without W ), I am probably mistaken: The 20V opamp will only open the npn transistor to conduct 20V-0.65V = 19.35V away from the 100V rail?  So the 100V rail can only be about 120VAC?

The NPN connected to the opamp output works at low voltages. It functions up to aboutn the voltage on the gate/base of Q2. Q2 handles all the high voltages and is the only device needing a high voltage rating. I would probably aim at a supply of 130V to 200V,

Ok. There is certainly something which I have misunderstood about the circuit.

Say the supply is 150VDC at the drain of Q2. But Q2 will always be conducting current, because the 10V ref connected to the gate is always on. Therefore, almost all of the 150V will be conducted to the collector of Q1. When the opamp output conducts higher than 0.65V, Q1 starts to conduct current away from the 150V supply. If the opamp has 20V supply, Q1 will only be able to conduct maximum ~19.35V away from the supply.

The rest of the current is lead away by the hamon divider. Perhaps ca 11V?

Is this incorrect? As far as I can tell, the circuit cannot reduce the voltage of the 100V rail to lower than 150V-19.35V-11V = ~120V

I think I am about to learn something new 

Thank you for your time.

Kind regards,
Marius

[amspire]

The voltage on Q1 cannot get higher, then the Q2 gate voltage as if that happened, Q2 would be turned completely off. Basically all the current in Q1 has to also go through the drain of Q2 (nothing flows out the gate) and the collector voltage of Q1 will settle at exactly the voltage required by Q2 to turn on enough to conduct the current from the Q1 collector. The result is the Q1 collector voltage is always less then the Q2 gate voltage.

As I said this circuit works just as well if Q2 is a NPN transistor instead of a MOSFET, and it might be easier getting a small high voltage transistor then a low current high voltage MOSFET. If you have any old TV's or CRT monitors to scrap, they almost certainly have some suitable low power high voltage transistors.

What this circuit does for you is it means that the collector voltage of Q1 is only changing by a small amount in in its full range of regulation - perhaps a fraction of a volt, rather then by 200V (or whatever the supply voltage). This greatly reduces the effect of Q1's collector to base capacitance.

Now you could just use a high voltage NPN for Q1 and eliminate Q2 totally. Probably worth a try too just to see what happens.

The reason I called it a concept circuit is I haven't done any breadboard, simulation or detailed calculations, but the idea is sound, and I am leaving it to you to breadboard it and try it out.  There are circuits that have been built and tested - I seem to remember seeing one somewhere in one of the Jim Williams Linear Technology application notes for a high voltage regulator, and all you have to do is replace the divider in the circuit with the Harmon divider. But if you an make this circuit work, it is a good one for supplying a low reference current as the current is limited by one resistor which makes it totally short circuit proof. Even if the circuit doesn't work, nothing disastrous should happen (as long as you keep your fingers off the high voltage).

It is the kind of circuit you can experiment with. 

To breadboard the circuit. you can start off without the voltage reference IC - just have a 15V supply for the opamp, and use a 470 ohm and a 1K resistor as a voltage divider for the 15V to supply about 10V to act as a reference. Add an electrolytic cap to ground. That is all you need to debug the regulator for the output.

Richard.

[Richard W.]

~ Let the meter in place while do the measurements. The displayed values will drift, if you move the meter. Why?

Some voltage references, including the LM 199/299/399 used in the later HP 3456A (and most other 6.5 digit) multimeters, are also sensitive to position. Its output voltage will vary slightly depending on orientation. Not sure what the exact mechanism is, I believe there was some speculation on the volt-nuts mailing list.

My "why?" was a rhetoric question. 
 As i said before, i think it's caused by mechanical stress to the semiconductor itself and temperature variations.
Thats also the reason why there are often slots in the pcb around the reference chips.

if i finally have a new oscilloscope and some time, i'll check the theory. A LM399 is about 15€ at digikey, i'll order two or three and try it 

[king.oslo]

What this circuit does for you is it means that the collector voltage of Q1 is only changing by a small amount in in its full range of regulation - perhaps a fraction of a volt, rather then by 200V (or whatever the supply voltage). This greatly reduces the effect of Q1's collector to base capacitance.

Latest update: I built the circuit in LTSpice. I used a 50nF + 70ohm as compensation. That seems to make it stable (no idea why).

it seems to regulate some current away from the 100V rail, but not much. Perhaps 20V. No idea why. At the minute I regulate the output voltage by regulating the supply. I will look at it in the morning. I am too tired now to make sensible decisions.

Thank you for your time.

Kind regards,
Marius

[king.oslo]

In an other thread Richard helped me overcome stability issues. Thank you.

I made three tweaks. My goal is to be able to use 230VAC without a step-down transformer to ~115V. I moved the 1k resistor and swapped the 200pF C_out to 470u. The simulation indicate very little fluctuation.

I would like to proceed to build this up (and perhaps post photographs or video), unless you have concerns with the big output cap.

Thank you for your time.

[amspire]

You definitely do not want to move the 1K resistor. Even the leakage current of the capacitor will cause a significant error.

Don't get too worried about getting the AC filtered out completely, as the DC range on the meter is very good at rejecting mains AC noise.

About running straight off the 240V.  Yeah it can be done, but I don't like it.

The mains neutral is still live mains, and the neutral will have a lot of noise to ground. If a socket has the neutral and active leads swapped, then you are connecting your earthed meter straight to an active mains.

It is easy to use a voltage multiplier rectifier, so basically you can use a transformer with any secondary voltage. Do you have any transformer at all?

Richard.