The LTFLU (aka SZA263) reference zener diode circuit

[Kleinstein]

The two resistors R1/R2 set the voltage over R3.  So depending on R1/R2 a slightly different, especially larger value might be needed. A voltage slightly higher than the base voltage might have a slight advantage. Transistor gain slightly depends on the emitter - collector voltage, but this should not be a very large effect - so I doubt it is used to also compensate higher order TC.

Similar to the LTZ1000 circuit, the OP is not that critical. The transistor will have a gain of something around 100. So 10 µV of drift of the OP would result in 0.1 µV of drift for the 7 V reference, or 0.15 µV at the output. The rather high impedance at the R3 side can be an issue. So don't forget the bias current.

[Dr. Frank]

Anybody checked what proposed by (z)Lymex few posts ago:


that's to eliminate R1/R2?

I'm interested in using only this 4 resistors as I already have some Vishay metal foil resistors (R401/R402) in the right value to get something like 9.999XXXV but I do not have the second divider (R1/R2).

My idea is to check this SZA with this schematic and than if it behaves than order some vishay VHP goodness for (R401/R402) (and eventually some others for R1/R2).

How they decided the ratio of the second divider (R1/R2)? I suppose must be in something slightly more (or equal) to achieve Vzener+Vbe?
How this missing two resistors affect the temperature compensation current when you trim output with R401/R402 ratio variation?

Should work, didn't test that. That's the same mode as the LTZ1000 circuit, as the base-collector voltage U(BC) =0.
That modus operandi is also used in the 5440 and 5720 reference circuit, see schematic on previous page.

As the collector voltage  is a bit lower, you need a bigger collector resistor, for zero T.C. compensation.

Frank

[AG7CK]

Although the "(z)Lymex" circuit might work, it is imo not in the spirit of the original use of the "refamp".

In order to understand the refamp one should imo think of it as a transistor amplifier with a TC-matched reference emitter offset.

This simple linear regulator block diagram describes the common use of the refamp:



In the context of the SZA263 / LTFLU-1 refamps *both* the roles of Voltage Reference and (first) Error Amplifier are played by the refamp. The "(z)Lymex" circuit (with the transistor in "soft saturation") is more in the spirit of LTZ1000 or maybe an actively controlled 1N829A circuit.

HP 3450A (revised manual dated 1969 - the instrument itself or a refamp-based predecessor is referenced in HP Journal year 1963) shows this very clearly:



Note the "low" resistor values (high current) in the output voltage divider, the base series resistor for the refamp, the high value refamp collector resistor and the superimposed collector bias supply 16.2v. All this for high gain combined with low loading of the 10v reference sampling resistor divider.

The following differential stage (with the the operating-point setting "second" voltage divider discussed in posts over) is just a 2nd error amp gain stage and nothing but that. It is this second stage role that is played by op-amps in more modern versions of the circuit. [The driver and/or series element (Control Element in the first figure) sometimes is and sometimes is not added to the op-amp.]

A similar circuit from Fluke 341A (manual dated year 1969):



This circuit might help answering a question raised in a post above concerning series diode for zener bias current.

---
PS Simpifying the two circuits "as much as possible" will be recognized by many as one of the ways to make simple but good linear regulators often used a few decades ago (and by some people even today):



---

[Kleinstein]

The  "(z)Lymex" circuit with one less divider is not that much different from the others. The main difference is working at a emitter to collector voltage of around 0.6 V instead of something in the 1 V range. At around 3 V the voltage for the collector resistor is not that high and transistor gain is limited, but it is still not that bad. Using a low voltage at the transistor is attractive as this leaves more to the resistor and thus a higher gain. This a little different with a 15 V reference circuit or when an extra higher auxiliary voltage is used. So with just the 10 V circuit the version with the extra divider is not that attractive: less voltage gain for the transistor  (and thus a slightly more sensitive resistor and more sensitivity to OPs drift) and two extra parts. The only advantage I see would be not having the OPs bias at the critical divider. However with modern OPs with bias levels below nA, this should not be a bigger issue. The more critical point to OPs bias would be the collector anyway.

One thing one might consider would be adding another capacitor and a non critical resistor to improve tolerance to capacitive loading.

[MK]

it would be fun to try a 10V ref based on this, but when one potential supplier according to google search says they can provide 10,000 per month of the chip you just know it will be difficult to source...

[mimmus78]

According to my ltspice simulation, I noticed that the base of the SZA reference (7V ish) is way more stable than the 10V node.
Apart for (z)lymex did anybody wired this out the 732a or verified it?

[Kleinstein]

The base is expected to be more stable - this is the raw 7 V reference point. The 7 to 10 V amplification (or should I say 10 to 7 V division) only adds errors to this.

[mimmus78]

The 7 to 10 V amplification (or should I say 10 to 7 V division) only adds errors to this.

I think by calibrating the 7V to 10V divider ratio in two different times you can compute what was 7V and 10V drift eliminating most uncertainty from this weak point.
I made some math on a xsl with simulated values in LTSpisce and it seems to works, don't know if this works also in reality.
Maybe the better is to use a pot to set ratio back to original divider ratio after a drift.
Only thing is that you need a 3458a to do this calibration and the measurement can be quite tricky to do.

[Kleinstein]

One usually does not need a reference of exactly 10 V. The important part is usually to have a voltage with approximately (like 1 or 10%) the right size and know precisely the value.

The high end zeners (LTFLU and LTZ1000) tend to have very low drift. You have to pay quite a lot to get resistors more stable. The simulation can show you how much a change in resistance will effect the final voltage. This sensitivity is than multiplied with the estimated drift rate of the parts. Chances are that drift of the 2 critical resistors for 10 to 7 V division will be a major contribution, followed by the LTFLU and than the other resistors and the OP.

The stability does not always follow a simple linear relationship. It is quite often a combination of a linear part, a few exponential contributions and also effects from the environment (temperature, humidity, pressure, magnetic field, radiation, gravity).

[Dr. Frank]

I think by calibrating the 7V to 10V divider ratio in two different times you can compute what was 7V and 10V drift eliminating most uncertainty from this weak point.
I made some math on a xsl with simulated values in LTSpisce and it seems to works, don't know if this works also in reality.
Maybe the better is to use a pot to set ratio back to original divider ratio after a drift.
Only thing is that you need a 3458a to do this calibration and the measurement can be quite tricky to do.

If I understand you right, you got an error in reasoning.. your argument obviously contains a circular reasoning..

You cannot extract the drift of the raw RefAmp (7V) voltage, simply from measuring this 7V / 10V several times.

The 7V drift independently from the 10V, so the 10V drift consist of the 7V drift PLUS the divider resistors drifts, which is then logically higher than the 7V drift.
This 10V/7V ratio divider is the really weak point.

The ratio calibration itself is very easy to do, and it's not tricky at all.. you can use a 3458A, or a Fluke 720A, for 0.1ppm accuracy, and to remove this ratio uncertainty instantly.
You can even use any good 6 1/2 digit DMM with ratio function to measure the 10V drift with about 1ppm accuracy.

If the 732A/B would provide an additional buffered RefAmp output, this would give much better uncertainty than these 10V.

That's a hint for the development of the 732C, probably. 

Frank

[mimmus78]

Hi Frank, thank for your response, I try to better explain my idea.

I'm not talking to null out the reference drift (we all know this is not possible) but only the drift that is directly caused by the 7V to 10V ratio change, that we know is the major contributor to the overall drift.

What I plan is to follow this protocol  to do this:

at beginning of the observation period:

  - I can determine what is the suppression ratio of the divider on the 10V and on the 7V nodes.
  - I can also determine with sufficient accuracy what is the divider ratio.

Than let say that after 1 year:

  - Check that suppression ratio is not changed
  - Check how much divider ratio is changed

At this point:

  - If divider ratio and suppression ratio is not changed (significantly) you can assume that the divider has not changed (significantly) and so have more confidence that the hole reference is quite stable.
  - If you find that suppression ratio is changed a lot, you know something strange happens and you cannot assume anything.
  - If suppression ratio is constant but divider ratio is changed a few ppm, you should be able to calculate what correction factor you can apply to the two outputs to calibrate out those changes. This should be totally doable because you know that divider ratio has not change that much and you know also suppression ratio remained constant during this observation period.

After this you still remain with:

  - drift caused by the current setting resistor (very high suppression ratio)
  - drift caused by the nulling TC resistor drift and generally with eventual changes in TC (I suppose also this are almost negligible if you used good resistor)
  - drift of the Zener or other components (OP Amp offset, etc)
  - drift of other external causes (that can be mitigated some way)

but because those drift have less impact on the REFENCE you end up with a better predictable reference.
So you moved from a concept of totally stable reference to a more predictable one.

Than I think having a stable 10.0000X V is still a great advantage for comparing multiple references or also for averaging purposed without having to be worried too much for the drift of the averaging resistors.

PS: I only make "serious" electronic from a couple of years ... hope I pass DR. Frank "peer" review and if not I hope to have learn something more.

[Dr. Frank]

Mimmus78,

now I understand your idea.
You're mainly interested in the influence of the different resistors of the circuit, or how their individual drifts influence the overall drift of the RefAmp reference voltage, and the 10V.
The suppression ratio is then the relation between the individual drifts and the drift of the outputs.

These parameters have been determined experimentally also on the LTZ1000 circuit by several people, but far easier.

If you simply vary each resistor by a small amount, say a few percent, by adding another resistor in parallel, you can directly determine these suppression ratios.

They will be in the same ballpark, like between 100 and 2000, as both circuits are similar.
The 7V / 10V stepup resistors have a suppression ratio of about 1, which makes them very critical.

So you can calculate from the known T.C.s and the estimated timely drifts the overall changes over temperature and time directly.

Btw.: I think, as these suppression ratios are partly quite high, that you can't really measure them, if you make comparing measurements 1 year apart.. the changes are probably too small, compared to the stability of your measurement equipment.
The only thing you have to measure over 1 year, or so, is the absolute drift of the RefAmp, and the 10V output.

Frank.

[Kleinstein]

One could also determine suppression factor from a simulation of the circuit, if a suitable model for the zener is found (correct differential resistance). The factor should be rather higher for the 2 resistors to set the currents and rather low (more like 3, as about 1/3 of the output is set by the resistor ratio) for the 10 to 7 divider.

[mimmus78]

Well I think in reality my method is applicable only to the divider where you can easily check with very good accuracy divider ratio by two relative measurements.
Fortunately this divider is also considered the main cause of the drift on the 10V node, so it still makes sense to focus just on it.

Measuring the others resistor is unpractical and than they have a so high suppression factor that their effects cannot get out of the hole system drift.

I think is better to determine suppression factor of the divider sample by sample and not by simulation.
This suppression factor need to stay fair constant in the period of observation as only in this case your calculated 10V drift can be considered valid.

[Kleinstein]

There is no real need to get a very accurate suppression ratio. This number is mainly important in when you decide on which resistors to spend money and which can be slightly cheaper ones. So simulated numbers are OK. Especially for the 10 to 7 divider the number should be very close to the calculated one. It has 2 Contributions: the main part is the divider directly which should be factor of a little under 1/3 depending on the resistors only and a smaller effect (expected to be about 1/100-1/500) of the 10 V value on the 7 V reference. One might need to measure the second part to get a good number, as here parameters like the zener resistance enter. So a simulation could only give a rough number.

It makes some sense to make the raw 7 V reference available externally, so the 7 to 10 V step can be checked independently and if needed corrections can be us. However as the ratio should be rather stable and thus only a small correction, there is likely no real need to also do a correction of the influence of the divider to the raw reference. The current setting resistor is expected to have the same effect and is usually not more stable than the critical divider.

[chuckb]

When did Linear change from the LTFLU-1CH Zener to the newer -1ACH?

I have seen photos of the -1CH Zener with date codes from 8915 to 9052.
I have seen photos of the -1ACH Zener with date codes of 9515 to present day

Does anyone have LTFLU parts with date codes between 9052 and 9515 and which part number is it?

[mimmus78]

I found another interesting implementation for this reference. This should avoid other two important resistors so it should be more stable than "usual 10V design". Off course you online get 6.7V (negative)...

Inviato dal mio ONEPLUS A5010 utilizzando Tapatalk

[Pipelie]

I found another interesting implementation for this reference. This should avoid other two important resistors so it should be more stable than "usual 10V design". Off course you online get 6.7V (negative)...

Inviato dal mio ONEPLUS A5010 utilizzando Tapatalk

it looks similar to this shematic here, posted #94 https://www.eevblog.com/forum/metrology/the-ltflu-(aka-sza263)-reference-zener-diode-circuit/75/

[mimmus78]

No, this is just a simplified version of the classical circuit where R1/R2 ratio is still very critical as this scale the reference out to 10V.

In the classical version you can get the 6.7V out at the base of the reference transistor, this node should be very stable but in my 731 this point is more nosier o susceptible to disturbance than the "direct output" so I don't know how much is useful other than a sanity check.

In this alternative version you get the 6.7V just out of the buffer ... if I'm not wrong it should be better and more stable.

[Kleinstein]

This version should give you a stable -6.7V and a not so stable +10 V.  The 10 V value depends on R5/R6 even more than the usual version that starts with +7 V and than amplifies to + 10 V (here only about 1/3 of the output depends on the resistors).

There is no magic way around the scaling from 7 to 10 V. Starting with 6.7 V might allow to use a capacitive divider (e.g. charge pumps with LTC1043) and only do the fine adjust from a nominal 10.1 to 10 V with resistors. However this can add noise.

[mimmus78]

This version should give you a stable -6.7V and a not so stable +10 V.  The 10 V value depends on R5/R6 even more than the usual version that starts with +7 V and than amplifies to + 10 V (here only about 1/3 of the output depends on the resistors).

There is no magic way around the scaling from 7 to 10 V. Starting with 6.7 V might allow to use a capacitive divider (e.g. charge pumps with LTC1043) and only do the fine adjust from a nominal 10.1 to 10 V with resistors. However this can add noise.

Why they used trimpot there? Just to trim the current compensation a little bit?

Inviato dal mio ONEPLUS A5010 utilizzando Tapatalk

[Kleinstein]

The trimm-pot would adjust the upper voltage (around 10 V) and also the Zener current. This will have a slight effect one the TC of the -6.7 V output. So maybe It a TC adjustment and the possible 10 V output is not really used.

[mimmus78]

The trimm-pot would adjust the upper voltage (around 10 V) and also the Zener current. This will have a slight effect one the TC of the -6.7 V output. So maybe It a TC adjustment and the possible 10 V output is not really used.

Well most probably is a crude 10V adjust to get the voltage up to 10V.

This may be required by the pre-characterization of the references as every chip has his own compensation resistors and maybe this coupling was done at 10V.

I think that if you lower this 10V to something like 7.2V this should bring even more rejection factor to this "divider" so that any good resistor couple will be practically ininfluent.

In this case you end up with just 2 resistors with already very good rejection factor.

Only cons is that you need to source other two custom resistors.

Inviato dal mio ONEPLUS A5010 utilizzando Tapatalk

[mimmus78]

Anyway as the lower side of the reference is at negative -7V ... I guess there should be +3V at the top and not 10V.

[AG7CK]

There is neither 3v nor 10v "on the top". There is something between 9.9v or so (6.5X15.4/10.05) and 11.9v or so (6.9x17.4/10.05) on the left leg of CR1 - depending on vref, pot setting and component values. (No sane person would adjust a refamp to 10.000xx.. volt with a 2 kohm potentiometer).

The voltage arises because U1A has its non-inverting input grounded and therefore serves as an inverting amplifier of the output reference voltage. U1A is the (very stable) power supply for the refamp. R1 is only a pull-up resistor in order to guarantee positive start-up.

This circuit is similar to some older Fluke circuits and others using appr. +- 7v ref. I am sure you did not "find" it laying around somewhere, so maybe you could give information or link to where it originates.