Is the quest of replicating a Datron 470x calibrator totally foolish or crazy ?

[branadic]

One week later... any news on your project? Still haven't seen a schematic your are planning to build up.

-branadic-

[DC1MC]

True, I'm still mulling over my options, read and scratch my head, I've started to actually understand the difficulty of the project and why many of the "simple" versions with MCUs were really never taking off, and remains just concepts.
Even the original Datron TTL-diarrhea it's actually more than meets the eye at the first look, not to mention the Fluke schematic that I've barely had a look yet.
I kind of see why there is no agreed upon stable schematic that is reproducible and a newbie could be pointed to it, like for example the multitude of excellent voltage references projects.

It seem that there are two categories of people involved in this:

- Enthusiast dilettantes that didn't know/look into the problem enough, and eventually produces some "simple, quick" schematics, that of course are never implemented and like many simple obvious solutions mostly don't work .

- People that have actually tried to do it and said "screw this", tighten the belt and bought a profi instrument.

I'm momentarily in-between, I'm still waiting for a breakthrough but with not too much time at the end of the year, the progress it's not so great .

Whatever the final income will be, I really thanks to all of the members that contributed ideas and insights, they were really valuable to me (and probably for others) to better understand the actual difficulty of the task.

 Cheers,
 DC1MC

 P.S> TiN probably laughs like > !!!

 P.S2: If still a (semi-)polished schematic shows up and it's deemed acceptable by the metrology gurus I will try to implement it if I can afford and my efforts are continuing.

[Kleinstein]

I have a reasonable idea on how the circuit should look like. However there are a few points, where one has to decide what is really aimed for.  The main points to choose are:

1) Reference. This might be something like  1 x LM399, 2 x LM399, 2 x 1N825, ... 2 x LTZ1000 ?
 Here the main point is starting from about 7 V or 14 V. Starting from 14 V would be more stable as there would be no amplification set by resistors and still affordable with LM399 or 1N825

2) Output stage with gain, or without.  Without gain would be easier, but the maximum voltage would be some 11 or 12 V.
    With gain one could get something like twice the voltage as a second range.

3) How much output power is needed.  A higher current output stage might want some extra protection, like a variable current / power limit.

4) Especially with gain, there is the question on how to realize gain switching and maybe polarity reversal would be done.
   Switching could be with something like max327 (expensive CMOS switches), JFETs or relays or maybe mechanical switches.

5) For the version with gain, the question is if gain check needs an external reference or not.

6) Which µC to choose to create the PWM and how to enter the values. Not really critical - more like hard to get consensus because of personal preferences.

7) provisions for smaller voltages, especially below 1 V.  Should be something like a divider and buffer - so this could be separate.

how to do the negative side sensing: This could use an extra amplifier, which would be the circuit that is easier to understand. Alternatively one might use a floating reference circuit, which slightly depends on the reference used: The circuit would be a little more tricky, but could be less critical parts.

9) how much ADC resolution for zero sensing, maybe other uses. Here it's mainly µC internal (e.g. 10 Bit) or maybe an SD-ADC chip like MCP3421.

10) how much input / output protection: Especially the output is hard to protect very well against external higher voltage

My preference so far would be:
 - Starting from 2xLM399 or 2x1N825A (optional footprint, slightly lower voltage)
 - Output stage for some 100 mA max with limits at about 1 mA , 10 mA , 100 mA via mechanical switch, limited sink capability.
 - Output stage likely with optional gain (e.g. x 2), but no polarity reversal
 - Gain switching with JFETs  ( 2 Settings)
 - Sense lines with not very high input impedance - so kind of a slight compromise here.
 - Floating reference for negative side sense
 - PWM from AVR (e.g. Mega88), directly coupled. Control via isolated UART from PC. Only minimal function in the AVR , CAL etc. on the PC side.
 - PWM switching with 2x DG419 for the main part and 74HC4053 for the fine part.
 - external supply with some +25-30 V (if with gain), + 15 V, + 5 V, -10 V
 - not sure about zero sense ADC

The point however is that I currently don't have much time and an other project to go first.

[Kleinstein]

Here is a still crude circuit.  There are some resistor values given, but they still need some final adjustments / check.

The reference part is still missing and one might want to consider a reference + buffer - that is the main still open point.
The reference would be relative to AGND = sense low.
Another part still not included is the zero detection. This might well need another JFET switch to separate it from the sense input, if not needed.

For the filter one could consider a 3rd OP for higher order filtering. It kind of depends on the PWM frequency used.
The OP IC8 is there for current compensation for the negative sense pin = analog ground. This can include the reference current.
The OP IC9 is compensating most if the input current for the sense input.

IC3 should be an ADA4522 or similar AZ OP. The other OPs are less critical.
The JFETs for switching should be a different type, more like MMBFJ202 or similar.

For protection one might use fuses and high power clamps for the sense inputs.

[DC1MC]

Thanks a lot @Kleinstein  , you're keeping my hope alive.
I see that the reference buffering seem to be a sensitive topic, but it also seem to be unavoidable, as long as we're "sucking" some electrons from it, that is drawing some current.

Could you kindly explain to this neophyte why it's such a big deal, even with the modern OP, do they really degrade so much the quality of the reference (noise/drift/tempco/??) that people are reluctant do buffer their voltage references ?

 Cheers,
 DC!MC

[Kleinstein]

Even a good amplifier is less accurate than a piece of wire. So one avoids amplifiers not really needed.  For a BJT base Buffer, noise is usually less of a problem, but long term drift can be a downside. With an AZ buffer, noise often is a problem and so are the extra current spikes from the AZ part of the OP. It is mainly the rather old circuits where we get schematics and in these days amplifiers where still rather expensive and not as good.

The difficulty is that die CMOS switches give some extra charge injection current peaks, and these peak could upset the reference. In addition the settling time of the reference might be relatively long and this could cause INL or require more minimum time for the PWM signal. It depends on the reference circuit too: With 2 times LM399 one could use the bridge type circuit with one OP to drive the output and the reference currents. This would probably be OK as the output is already buffered. With a rather fast OP, like OP27, OPA209 the OP should also settle relatively fast after the pulses. The drift and noise of an OP27 is lower than that of most references at 14 V so it is not such a big deal to have a buffer. The higher current 1N825 might prefer the circuit usually used with 7 V refs : an amplified voltage to provide the current source and than a separate buffer. For the2x LM399 one might  even get away without a buffer and use just some capacitance.

There is also the "buffering" problem at the ground side. I think one can get away with just sufficient capacitance (e.g. 1 µF+10 µF) here - so there would be quite some capacitance from the sense low to drive low. The voltage would be limited by protective diodes anyway. I think this would be acceptable for normal use, where there is just a wire connection.

For the zero detection amplifier one could consider a differential input (e.g. AD620 or similar) and extra MUX (e.g. DG409) so that one could use this also for a gain check / measurement. The 4 possible inputs could be sense against zero, sense against an auxiliary 10 V (e.g. divider) and one amplifier zero or reversed polarity. There could be one optional external input.

[Vtile]

Isn't that 1n82x series be pretty similar in specs, only that 1n829A is most suitable for production as the zero TC point is closer to the nominal 7.5mA.

Ie. I had a few 1n823 zeners where the zTC point were closer to 4.5 mA than nominal 7.5 mA, with wild variations between individual parts what comes to higher order fluctuations from temperature. The cost is friendlier for a hobbyist though, you can get 10 1n823 for a cost of one 1n829A to sort and pick. A null bridge is a good way to pick the best part(s) for the reference, in that way you can use another zener as a ref to get in the mV differential range, even when starting without stable reference, like I did.

As going to derail more the topic, is my dilettante thinking in somewhat right track that with higher current reference the part is less susceptible to noise from load and stray voltages around the circuit, but slower to stabilize because of self heating?

[Kleinstein]

With 1N823 the scattering is quite high - one might not find a suitable current in all cases. So there is a certain risk to get a batch of diodes that can not get zero TC. How expensive the 1N829A is depends on the source. For 1N825 chances are good to find zero TC at an acceptable current.
The price also varies quite a lot between sources. So it might still be ok to get 1N829A.

The higher current zener diodes tend to have lower noise. The zener diodes are rather low impedance, so noise pickup or stray (leakage) current is not a problem at all. It may still need some sorting to get low noise units, as there can be quite some scattering in the noise level.
Self heating is higher, but the time constant for stabilization is about the same.

[Echo88]

How much noise does such a Zener produce compared to a fast and easy solution like the LM399 or the more expensive LTZ1000?

[Kleinstein]

The zener noise depends on the type and sample. The 1N82x are supposed to be relatively low noise, especially if selected. So noise wise they may come close to the LTZ1000. Except for the bad samples they should be lower noise than LM399.

The downside of the 1N82x zeners is that they need an selected current to adjust the TC and still get a low TC only for a limited temperature range (like +-5 K).  Noise performance can also vary and might need a check. Especially with 1N823 the current for low TC might end up rather high.

The LM399 is kind of easy to use and low TC even over a large range -  the downside is quite some noise.

One could design the board to alternatively accept LM399 or 1N82x , with just a few resistor values to change because of the different current. So the choice of reference is not that critical yet. 

[Andreas]

The 1N82x are supposed to be relatively low noise, especially if selected.

The 1N829 are normal temperature compensated zeners, and not low noise buried zeners.
Due to the high current (7.5 mA) against the LM399 zener current (0.4 mA) there is some compensation.

On volt-nuts you find a longer thread of WarrenS who had experience with selecting them.
But as far as I understood the selection was mostly for low popcorn-noise.
And you need a good pre-ageing with high current (50mA for one year) before selection.
https://volt-nuts.febo.narkive.com/9QIiUfpN/some-questions-to-zeners-1n823-1n829

My own measurements 0.1-10 Hz 1/f noise on some samples (some NOS + some fresh devices)
showed that there is a large dependancy regarding the manufacturer and also a certain stray from diode to diode.

Motorola 4 samples (NOS) 2 - 2.5 uVpp
ST 3 samples (NOS) 11.5 - 22.5 uVpp
APD 5 samples (new from Farnell) 2 - 2.8 uVpp

LM399 typical 4 uVpp (but partly with a large stray).
LTZ1000 typical 1.2 uVpp (2uVpp max).

And of course in every case you have to do a screening for popcorn noise.

With best regards

Andreas

[Vtile]

My dilettante ideas again. Have anyone tested (definedly, but which were the results.  ) what happens if you bake these TC zeners with close to 180 deg. C for a few tens of hours and then cool them back with slow slope ie. 10..20 hours.

To partly answer to myself yes relaxation heat treatments seems to be used but at lower temperatures and longer times at least for whole units.

[z01z]

There is a paper on how Datron did the aging and selection, here's the link from the 2dw23x topic.

REDUCING OF THE TIME DRIFT OF ZENER STABILIZATION VOLTAGES, famous DATRON, Thanks zlymex !
https://www.eevblog.com/forum/metrology/t-c-hysteresis-measurements-on-brand-new-lt1027dcls8-5-voltage-reference/?action=dlattach;attach=239750

[DC1MC]

There is a paper on how Datron did the aging and selection, here's the link from the 2dw23x topic.

REDUCING OF THE TIME DRIFT OF ZENER STABILIZATION VOLTAGES, famous DATRON, Thanks zlymex !
https://www.eevblog.com/forum/metrology/t-c-hysteresis-measurements-on-brand-new-lt1027dcls8-5-voltage-reference/?action=dlattach;attach=239750

So now I can buy 50 of these guys:
https://www.ebay.de/itm/5x-1N823-Voltage-Reference-Diode-Zenerdiode-6-2-V-0-5-W/332779932314
And start the burn-out procedure and in a month I'll have some cool references  ?
I can do a simple rezistive oven, do the diodes really need to be under power, and at what current ? If yes it's some kind of ion/doppin migration happening or what ?
Is it worth doing it ?

 Cheers,
 DC1MC

EDIT:
Actually is it something technologically special for the 1N82X series, I've seen on evilbuy 6.2V Zeners form Vishay that costs 6EUR/100
https://www.ebay.de/itm/Zener-Diode-VISHAY-500mW-10-25-50-100-Stuck-Spannung-2-4V-75V-Dioden/192070809110

Are there any other similar diodes that can be bought in bulk and aged ?

[Kleinstein]

The 1N82x are somewhat special: it is not just a 6.3 V zener, but a zener and normal diode in series. So more like a 5.6 V zener and 0.7 V drop from the forward biased diode. A normal 6.3 V zener would have a considerable positive TC (e.g. 2-3 mV/K).  Within the class of normal surface zeners the 1N82x are also relatively low noise.

There are similar diodes as ZTK6.8 - though about as expensive and I don't know about noise or drift.
Then there are Chinese 2DW232 reference diodes. Form some sources they are very low noise. However there seem to be quite some scattering - so more like 1N823 not 1N829. The voltage is a little lower and it's a transistor like case (TO-39) with access to the point between the 2 diodes.

For aging there are sereral effects going on: dopant / impurity diffusion, relaxation of stress from the case, charge build up on isolated traps in the surface oxide (a little like an EPROM). So it might help to do heating by the diode current itself.

[DC1MC]

The 1N82x are somewhat special: it is not just a 6.3 V zener, but a zener and normal diode in series. So more like a 5.6 V zener and 0.7 V drop from the forward biased diode. A normal 6.3 V zener would have a considerable positive TC (e.g. 2-3 mV/K).  Within the class of normal surface zeners the 1N82x are also relatively low noise.

There are similar diodes as ZTK6.8 - though about as expensive and I don't know about noise or drift.
Then there are Chinese 2DW232 reference diodes. Form some sources they are very low noise. However there seem to be quite some scattering - so more like 1N823 not 1N829. The voltage is a little lower and it's a transistor like case (TO-39) with access to the point between the 2 diodes.

For aging there are sereral effects going on: dopant / impurity diffusion, relaxation of stress from the case, charge build up on isolated traps in the surface oxide (a little like an EPROM). So it might help to do heating by the diode current itself.

 -  So you say that technologically speaking, the 1N82X are having this extra junction and similar models, at least from this poin of view,  are ZTX6,8 and 2DW232, others are not known ?
 -  Also is it really possible to raise the junction temperature to around 100C just by injecting current, without destroying  the diode due to side effects ? That would be cool, because will make the life of "Zenerzüchter" so much easier than using an  external heater ?


 Thanks for sharing this cool information 

 Cheers,
 DC1MC

[Andreas]

And start the burn-out procedure and in a month I'll have some cool references  ?

Perhaps in your dreams:
- did you read about the yield of WarrenS for zero TC current for 1N823 against 1N825 in the linked thread?
- did you read about my results for noise especially for "ST" parts that I have measured? Maybe you have more luck but I fear: no.

with best regards

Andreas

[DC1MC]

And start the burn-out procedure and in a month I'll have some cool references  ?

Perhaps in your dreams:
- did you read about the yield of WarrenS for zero TC current for 1N823 against 1N825 in the linked thread?
- did you read about my results for noise especially for "ST" parts that I have measured? Maybe you have more luck but I fear: no.

with best regards

Andreas

Well, we all have to dream, don't we, it's Christmas soon ?

Now, joking aside, of course I did read your post, ST were having the largest noise, BUT only 3 !!! samples (were they pre-selected or random), what about 200 pieces, I can seralize them in sets of 10 and wipe out a 72V PS and let them burn with a limiting resistor, either cyclic or continuously.
I'm still curious if the burn-in cycles (heating of the junction) can b made with just the current or one needs an external heater ?
In the end now and then we should keep an eye on this diode guys, who knows when a cool lot may show up, of course there no lack of expensive stuff, but the proletariat needs also precision ,

 Cheers,
 DC1MC
I

[Kleinstein]

With zeners from one batch chances are good most of them will behave similar and a few are bad. So if 3 units from one source are high noise one has to expect something like 199 noisy ones and maybe 1 that is worse if you test 200.

If low noise without the price-tag of the LTZ1000 is the priority, the Chinese 2DW232 maybe the best bet. Here starting from something like 10 units might make sense to find some 4 good ones. Still it may be a batch thing so one could get 10 good ones or none at all. Here selecting would be for suitable current at zero TC and checking for not too much noise and maybe also check for drift after a few 100 hours. The big unknown with these is the long term drift.

If it is just about getting a good TC, the LM399 is a good bet. The big advantage is that it's easy to use: no current to adjust and essentially always a low TC.

[Vtile]

Don't get my ramblings get your way of getting LTZ1000 etc. proven units. My measuring capabilities ends to 10ppm resolution, but it just interesting to follow and wonder in thoughts in this area, more than regular electronics. 

The discrete zeners are definitely satisfying to work with though, when no walkers are breathing your neck.

The heat threatment idea just popped to my mind from working with class, metals and gemstones (not this metaphysic jimba-jamba superstition) where all areas the heat treatments do give extreme results, extending the qualities far beyond regular. So why not a device which is constructed from metal, class and gemstone. 

[z01z]

-  So you say that technologically speaking, the 1N82X are having this extra junction and similar models, at least from this poin of view,  are ZTX6,8 and 2DW232, others are not known ?

My understanding is that 1N82X is "just" a regular Zener and a regular diode in one package. However the Zener voltage was selected such that is has the same temperature coefficient as a diode, but with opposite sign. This makes it possible to find a current that gives zero TC.

Do they worth the hassle? I suppose it's a telling sign that Datron started using LTZs instead. If they had smaller noise than an LTZ, then maybe, but it looks like they don't.

[Echo88]

Since the invention of the LM399 and LTZ1000 basically nobody used Discrete Zeners as references for high accuracy DMMs/Calibrators, since the time spent on selecting suitable zeners is not worth it. I still have a testboard with two populated LM399 and a few populated KX-boards which only needs precision resistors and LTZ1000. You can have them for free if you want.

[DC1MC]

I do , how should we do it ?

 Cheers,
 DC1MC

[Echo88]

Sent you a PM.

[Kleinstein]

For the first board I would go with the LM399: they are easy to use and still affordable, even when using 2 of them.  At least the last circuit I showed has the reference directly at the low side sense terminal and thus could be a little tricky with an external reference board. At least one would need a special board made for a separate negative side supply. So one would probably have the reference as part of the PWM-DAC circuit. One could still have the option to choose between 1 Reference and amplifier x 2 or 2 references in series and a buffer. This would be just 2 precision resistors or 1 simple one. So one could still decide later.  For the LM399 it might be cheaper to use 2xLM399 instead of 1xLM399 and 2 precision resistors. A first test could even use just LM329 to seen if the DAC works.

One point still open is the choice of µC. In principle something like an AVR should be sufficient to create the PWM signals.

Another point might be the idea of not using PWM for the fine part, but a ready made (e.g. 12 Bit) DAC chip. The advantage would be to avoid possible interactions between the 2 PWM signals, that might cause DNL errors as some specific codes, if the isolation between the 2 PWM signals is not very good.
Depending on the µC and frequency the coarse part could provide some 15-16 Bit of resolution. Something like 1 bit might be lost for overlap to simplify the calibration. So even 15 bit coarse - 1 bit lost and 10 Bit fine resolution would give 24 bits of total resolution.