High voltage reference in an original way :D

[daqq]

http://www.ti.com/lit/an/sbaa203/sbaa203.pdf

[EmmanuelFaure]

Original, or pure stupid?

[JS]

Only TI could build a TI voltage reference using 10000 TI's voltage references. 

JS

[zlymex]

I once build an 100V reference(with 10V output as well) by stacking 10 LT1021BCN8-10 plus a constant current source. This IC is much better in term of drift and noise.

I also had a plan to stack 14 LTZ1000 to obtain a super 100V. For this purpose, I have selected 14 LT1000 that added up to exact 100.000V. The 14 power supplies have to be isolated.

[daqq]

I also had a plan to stack 14 LTZ1000 to obtain a super 100V. For this purpose, I have selected 14 LT1000 that added up to exact 100.000V. The 14 power supplies have to be isolated.

I'd love to see this approach applied to the 100kV version

[zlymex]

I also had a plan to stack 14 LTZ1000 to obtain a super 100V. For this purpose, I have selected 14 LT1000 that added up to exact 100.000V. The 14 power supplies have to be isolated.

I'd love to see this approach applied to the 100kV version

I'm trying to squeeze one unit as small as possible, for a starter

[retrolefty]

Only TI could build a TI voltage reference using 10000 TI's voltage references. 

JS

 Pretty unusal app. note: 

Currently, the voltage source for 100 kV is built using this method. 10000 REF5010 devices are used to build this device. This design probably hits the record of simultaneous use of TI parts in a single device. Its electrical parameters are under characterization.

[JS]

I once build an 100V reference(with 10V output as well) by stacking 10 LT1021BCN8-10 plus a constant current source. This IC is much better in term of drift and noise.

I also had a plan to stack 14 LTZ1000 to obtain a super 100V. For this purpose, I have selected 14 LT1000 that added up to exact 100.000V. The 14 power supplies have to be isolated.

Do they? Why not to supply one from the reference one before, in some ladder fashion? Maybe separate supply for the reference, which could be 3 or 4 not insulated supplies for all the references.

JS

[zlymex]

I once build an 100V reference(with 10V output as well) by stacking 10 LT1021BCN8-10 plus a constant current source. This IC is much better in term of drift and noise.

I also had a plan to stack 14 LTZ1000 to obtain a super 100V. For this purpose, I have selected 14 LT1000 that added up to exact 100.000V. The 14 power supplies have to be isolated.

Do they? Why not to supply one from the reference one before, in some ladder fashion? Maybe separate supply for the reference, which could be 3 or 4 not insulated supplies for all the references.

JS

I didn't quite catch you.
LTZ1000 is not operated in shunt mode(2-terminal mode? or zener diode mode) like REF5010 or LT1021, where the supply and the output are the same.
The 7.1V output of a LTZ1000 circuit comes from two sense pins which are not the same as the supply rails. Ideally there should be no current flow through these sense pins in either direction(in or out). All the 28 sense pins should be connected in series fashion in order to get a perfect 100V. If any of the 14 power supply is not floated, there would be a current flow through some of the sense pins and degrade the performance to say at least. That is also the reason why devices like REF5010 can be cascaded in the simplest way.

[JS]

I didn't quite catch you.
LTZ1000 is not operated in shunt mode(2-terminal mode? or zener diode mode) like REF5010 or LT1021, where the supply and the output are the same.
The 7.1V output of a LTZ1000 circuit comes from two sense pins which are not the same as the supply rails. Ideally there should be no current flow through these sense pins in either direction(in or out). All the 28 sense pins should be connected in series fashion in order to get a perfect 100V. If any of the 14 power supply is not floated, there would be a current flow through some of the sense pins and degrade the performance to say at least. That is also the reason why devices like REF5010 can be cascaded in the simplest way.

  I was thinking on the full configuration for each one, with the output buffer which can source current. Use the output of the opamp to power the next stage, not much juice needed as the previous stage should balance the current. The problem, as usual, would be the heaters. In this config, the ground of the first stage, the pin 1 of LT1013 (or after the diode) of the second and the V+15 of the third should be tied together. Some tweak of the circuit should probably be used to manage the zener - sense terminal. I'm not saying is an easy solution but seems there may be a way, and once solved for one stage would take the need of 14 clean floating voltage supplies.

  Isn't a better approach (cheaper and more reasonable) to use an amp stage to take that 7V and turn them into 100V? There are really stable resistor networks and you could ovenize them to get even better stability. A pain LT5400 can't handle 100V but can handle 60V without trouble, so two of them and keeping the LTZ in the middle would do it. To complete the circuit two low drift opamps and a few high voltage transistors would be needed.

  I'd guess the REF5010 approach as a transfer and a KVD to calibrate it to an LTZ makes an easier world to live in.

JS

[Kleinstein]

Stacking 10 references for a 100 V reference makes some sense, if you have a setup (e.g. switches) to measure / calibrate the individual parts too. So you get a 100 V reference and a way to a 10 V to 100 V transfer by adding the individual voltages. As with a 10 V to 100 V transfer this is more about short time stability, the Ref5010 might not be such a bad choice.

The problem with a resistive divider is that you have to rely on linearity of the resistors - at 100 V you can't just take it for granted at the sub ppm level. So the resistor way (e.g. hamond divider) is that easy either.

[Alex Nikitin]

Stacking 10 references for a 100 V reference makes some sense, if you have a setup (e.g. switches) to measure / calibrate the individual parts too. So you get a 100 V reference and a way to a 10 V to 100 V transfer by adding the individual voltages. As with a 10 V to 100 V transfer this is more about short time stability, the Ref5010 might not be such a bad choice.

The problem with a resistive divider is that you have to rely on linearity of the resistors - at 100 V you can't just take it for granted at the sub ppm level. So the resistor way (e.g. hamond divider) is that easy either.

Also with a resistive divider the voltage noise of the reference is multiplied directly (plus the divider would add some noise), but for a stacked 10V references the resulting noise is increased by a square root, so for 10 references the total noise would be only about 3.16 times of the noise from one reference and the relative noise is actually reduced.

Cheers

Alex

[Alex Nikitin]

Uhmmm...  No, I don't think so.  If the 10 references were in parallel, then yes, there would be an improvement in noise by:

\$\sqrt{N}\$  ;

but when you place the references in series then the noise adds geometrically according to the RSS formula:

\$\sqrt{\Sigma{X_n^2}}\$

It is the same thing .

Cheers

Alex

[Alex Nikitin]

Not if the references have different amounts of noise.  If they are all the same (which is unlikely) then yes, it *is* the same.  I know it's splitting hairs, but it'd good to be more precise I think.  In practical terms, it means you would want to hand-select the references in your chain for low noise.

It is exactly the same for a parallel configuration.

Cheers

Alex

[Alex Nikitin]

Not if the references have different amounts of noise.  If they are all the same (which is unlikely) then yes, it *is* the same.  I know it's splitting hairs, but it'd good to be more precise I think.  In practical terms, it means you would want to hand-select the references in your chain for low noise.

It is exactly the same for a parallel configuration.

Cheers

Alex

Uhmmm... no.   Assuming the noise of each reference is the same, the noise of paralleled references is reduced by \$\sqrt(10)\$ ; but if the references are in series, then the noise is increased .

OTOH, perhaps we are miscommunicating, and I'm not understanding what you are saying?

The relative noise is reduced in the same way for both cases. For the series connection the noise voltage increases by \$\sqrt(10)\$ but at the same time the DC "signal" voltage increases 10 times.

Cheers

Alex

P.S. - say, a single 10V reference has 10uV p-p noise. WIth a resistive divider in a feedback loop of an amplifier we'll get 100V and 100uV p-p noise. With 10 references in series we'll get 100V and only about 31.6uV p-p noise.

[Alex Nikitin]

OK, I think I was taking your statement (incorrectly) out of context.  I would agree with you that placing them in series would be better noise-wise than taking one reference an multiplying it up by 10.  Even if that if that one reference were (say) an LTZ1000 boosted to 100V, the series configuration would probably be less noise.  (I didn't do the calculations-- just "back of the napkin" right now).  However, that 100V series reference would be very drifty thing-- only good for very short term transfers.

Strictly speaking, if the series connected references are thermally coupled and protected from air movements, than the resulting drift (relative again!) would be not worse that for a single reference and should actually be somewhat better as the references would have a distribution of drifts.

The LTZ1000 based 100V reference, while more noisy, would be stable as a rock.

It would be as stable as the resistive divider  .

Cheers

Alex

[Kleinstein]

For the series configuration it is not that practical to use chips that are made for a relatively exact 10 V value each, as those tend to have more drift and cost  than those that directly output the raw zener voltage. So the more sensible version might have been 14 pieces of LM329 (ideally in a metal case if you can still get them) or those Chinese low noise zeners (2DW..) and than one adjustment to get the exact value of 100 V if needed.

In a way the 10 references REF5010 also provides a reasonable accurate 100 V without adjustment. Though at that price I would not mind an adjustment with a good DMM.

[EmmanuelFaure]

I checked on pricing from Hong Kong Components

http://www.hkinventory.com/?