LM399 based 10 V reference

[dietert1]

Looks good. I'd have five proposals:
- Select a metal enclosure and make board to fit the mounting holes
- LM7818 instead of LM7812 for better LM399 oven performance, mount to enclosure
- Solder or screw terminals instead of supply socket
- Add separate output Gnd terminal
- If PTC heater won't be used, move temperature sensor closer to the 10 V to 7 V divider resistors

Regards, Dieter

[Birb]

Thanks for your feedback!
Perhaps I was slightly unclear - the PTC heater will remain in use, but I do not need the 5V regulator to be placed together with it.
However, using a higher voltage might cause the 5V regulator to dissipate a larger amount of energy and that may cause heating issues - I'll see.
Actually, the temperature sensor will have some wires to extend it to be inside the enclosure for the resistors (I have some insulating fiber/ foam material)
As for a metal case, I will probably use an extruded aluminium one.
The ground terminal is likely connected through external wires.

The socket was chosen because some 24V lithium battery packs, as well as most power plugs use this type of connector to connect.
I could consider to break it out and use wires directly, I suppose.
Though I haven't figured out a good way to keep it on continuously.

[iMo]

It has been discussed here many times - the voltage regulator inside the box or close to the reference circuitry is not the best idea in case your input voltage will vary (ie a battery or mains transformer). That is because the heat dissipation of your 7812/18 will vary with the input voltage changes..

[Birb]

Then perhaps I will separate the PCB into two, to try and isolate the heat better.
Or I may need to figure out a different power solution?
Though using a higher voltage rated PTC may be able to allow me to continue to heat the resistors without adding the 12V to 5V conversion, at the expense of higher temps for the resistors and current draw.

[iMo]

There are two options (when using linear regulators, what is usually recommended with Vrefs):

1. to leave your voltage regulator on the Vref board as-is in your design, AND to add the second one in front of it and have it mounted outside the Vref board
2. to have a single voltage regulator outside the Vref board (thus not thermally coupled to the Vref circuitry).

Mind the chip temperature of ANY linear voltage regulator is aprox:

Tchip = (Vin-Vout)*Iout*Rthja+Tambient, where the Rthja is the thermal resistivity of the junction to ambient (a constant).

[Birb]

Thanks for the feedback.
By second one, do you mean the 12V to 5V linear regulator? Thanks.
At 12V the temperature is around 60C
I haven't figured out how to effectively reduce the setpoint, but it may include adding a layer of insulation between the PTCs and the resistors, so the resistors are kept at ~50C or lower (preferably)
Is this more optimal?

[iMo]

Thanks for the feedback.
By second one, do you mean the 12V to 5V linear regulator? Thanks.

The second option means your "input to 12V" regulator (the 7812) should be off of the voltage reference board.
Also the 7805 powering your PTC will change its temperature as well - its in/out voltages will be "constant" but the current will change.
Simply look at the above equation and try to apply it to your design accordingly.. 

[Birb]

Random new question / idea / whatever:
I just realized that it is possible to use one single octal resistor array to achieve the necessary resistances:
Using 6 resistors, it's possible to make a 1:2 1/3 resistor ratio, equivalent to a 3/7 divider ratio.
Then, using the two remaining resistors, if chosen correctly, we can have an approximate 3k resistance to set the zener current.
8x 5k resistors work for this, as two 5ks in parallel give 2.5k, which would mean a zener current of 1.2mA.
The question is, how much impact does the 0.2mA have on the LM399?

However, something like this is inheritly non-trimmable, but it would probably have a tiny footprint and be relatively stable.
One more thing: Is it advisable to insulate the resistor array together with the LM399 to achieve better thermal stability for both, or would this be detrimental to the resistors, due to the high oven temperature of the LM399?
Thanks!

[Andreas]

The question is, how much impact does the 0.2mA have on the LM399?

Mhm 0.2 mA with max 1 Ohm dynamic resistance gives (constant) 0.2 mV higher output than nominal.
Besides this the ageing rate will sligthly increase (ageing parameter is given based on 1.000 mA)

However, something like this is inheritly non-trimmable,

No, look e.g. at the AD587 datasheet how external trimming can be done by a pot and a additional "RT" resistor.

One more thing: Is it advisable to insulate the resistor array together with the LM399 to achieve better thermal stability for both, or would this be detrimental to the resistors, due to the high oven temperature of the LM399?

I personally would isolate heat sources like voltage regulators and LM399 heaters from the rest of the precision cirquit. (OP-Amp, voltage divider)
Otherwise you have large heat gradients depending on the distance to the heat source.
I would probably also use an extra resistor for the Zener diode to remove heating from the voltage divider.
And would use less than 10% of the specified power rating.

The voltage divider itself should be a high stable one (e.g. thin film resistors like TDP1603 series).
From my ADR1000#01 I know that it needs about 2 years until the resistor array gets stable.
See diagram with divider drift here:
https://www.eevblog.com/forum/metrology/lowest-drift-lowest-noise-voltage-reference/msg5617061/#msg5617061

with best regards

Andreas

[iMo]

The question is, how much impact does the 0.2mA have on the LM399?

Mhm 0.2 mA with max 1 Ohm dynamic resistance gives (constant) 0.2 mV higher output than nominal.
Besides this the ageing rate will sligthly increase (ageing parameter is given based on 1.000 mA)
..

The zener current itself in the 399 is set fixed to 250uA afaik, it had been discussed here many times in past. The same with 1399 (but different current). With higher currents (like 2mA) the 1.75mA is diverted off the zener (plus minus)..
Thus you cannot set the zener current in 399 or 1399. The small differences you see while measuring the voltage vs. current come from a nonlinear characteristics of the circuitry around the zener (see the schematics in the datasheet).

[Alex Nikitin]

The small differences you see while measuring the voltage vs. current come from a nonlinear characteristics of the circuitry around the zener (see the schematics in the datasheet).

Not as the result of a nonlinearity, only of a limited open loop gain.

Cheers

Alex

[Birb]

Hmm, I think my real question should be about how the 0.2mA affects drift.
Is it really a big difference?
Also, how does it affect settling times of the references? Thanks
I mainly want to try to avoid any extra resistors as this can keep the entire setup simple and component count minimal.

Also, is it better to have multiple references in parallel or put the LM399s themselves in parallel with each other?
(The advantage of small units is several could be placed together to form a more accurate(?) unit.)
(Though I note the ratio of 3:7 is really idealized and as far as my "testing" is the voltage is more like 7.04ish, so that would be a bit higher than wanted.)

[iMo]

Hmm, I think my real question should be about how the 0.2mA affects drift.
Is it really a big difference?
Also, how does it affect settling times of the references? Thanks
I mainly want to try to avoid any extra resistors as this can keep the entire setup simple and component count minimal.

Also, is it better to have multiple references in parallel or put the LM399s themselves in parallel with each other?
(The advantage of small units is several could be placed together to form a more accurate(?) unit.)
(Though I note the ratio of 3:7 is really idealized and as far as my "testing" is the voltage is more like 7.04ish, so that would be a bit higher than wanted.)

1. No, 2. No, it does not affect,  3. you may put N 399s in "parallel" (it does not mean the zeners are all parallel, they have separate zener cathode rezistors) via N summing resistors into a single common 7->10V amplifier (if required), 4. the output noise will drop sqrt(N) times down, where the N is number of "paralleled" refs (like 4 refs will do aprox 2x less noise).

[Kleinstein]

Instead of in parallel, I would consider the new ADR1399 for about half the noise, comparable to 4 x LM399 in parallel.

The actual voltage of the LM399 can vary quite a bit between units. They can be above or below 7 V. A parallel version would need extra resistors for averaging and separate paths for the current. One could do the averaging at the 7 V level or the 10 V level to also average over amplifiers and resistors.

[Birb]

My idea is to average at the 10V level, which, while increasing costs, does allow for post selection of the better references, allowing for better combinations at the end.
I am using lm399 because it's what I have on hand, and should work well if chosen well.
If I'm not mistaken, output protection should come after averaging correct?
This might increase the risks of mishaps during testing, but oh well.
As for external trim, I might try the AD587 method.

[Birb]

Other thing I noticed: While the LT5400 is not capable of producing the 1:2.333 resistor ratios, it can produce 1:2.5 and 1:2.25, the former creating around 9.8V and the latter 10.111V.
Therefore, if three voltage references are averaged with two in the 1:2.25 configuration and one in the 1:2.5 configuration, then the LT5400 is also capable of producing 10V with a small remainder. (At the expense of cost!)
(Though one question is if averaging is fine if the voltage references are just connected directly?)
                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                             
In addition, while 1:14 is not available, the LT5400 can also produce 1:16 which could be used to step up 7V to 7.5V, which can then be divided by an LTC1043, to produce 2.5ish volts, for ADCs
(I might try this in the future, though that depends on time)

[Kleinstein]

Averaging a 1:2.25 and 1:2.5 from different references does not work that simple. It would add quite some importantance to the resistors, about as much as doing the 9.8 to 10 V or 10.11 to 10 V step. It could be still simple averaging with 3 equal resistors and thus easy matching.

It depends one the actual voltage of the LM399 (or ADR1399) how much step there would actually be left from 10.11 V nominal. It would make sense to first check a few references. If they are on the high side the 9.8 V to 10 V step could be smaller than the 10.11 to 10 V step. With only some 2 or 1.1 % nominal (and possibly better) one should get an attenuation for some 50 or 90 (or better) for resistor drift. That could be good enough for a LM399 / ADR1399. AFAIR the LM399 has a tolerance of 2% - so look at the actual voltage first.

The going down version could be a little simper, using only 1 amplifier, with 1 divider at the 7 V and than the gain.

[Birb]

Something minor I note is that the LT5400 in the 1:2.5 ratio configuration uses the 1:10 variant, which means the 1:2.5 ratio has a generally higher resistance than the 1:2.25 (Variant 8 or 1:9), so the Johnson noise would be higher
(Though I'd assume this is very negliglible)

Anyways, I think this config might be better suited to use ADR1399 due to lower drift and tempcos.

Here's a schematic showing the conceptual 1x NOMCA resistor network LM399 voltage reference.

[dietert1]

There was a thread about statistical arrays, where the conclusion was to better use NOMCT instead of NOMCA. They have an improved passivation and generate less excess noise. They can be used for  < 1 ppm applications, see e.g. ADR1001 long term study by Andreas.

Regards, Dieter

Edit: Sorry i meant ADR1000. Andreas' most recent post:
https://www.eevblog.com/forum/metrology/lowest-drift-lowest-noise-voltage-reference/msg5662371/#msg5662371.
The resistor array part used is TDP16031002AUF, see:
https://www.eevblog.com/forum/metrology/lowest-drift-lowest-noise-voltage-reference/msg4547699/#msg4547699.

[Echo88]

https://cds.cern.ch/record/2814429/files/2109.02448.pdf
https://www.researchgate.net/publication/367220764_Review_on_Excess_Noise_Measurements_of_Resistors

[Birb]

When it’s stated the NOMCT has better passivation does this apply to the NiCr NOMCTs or does it mean the TaN NOMCTs?
Thanks!

[Kleinstein]

The difference seems to be from TaN based resistors to NiCr based resistors. From what I understand it is not so much the passivation, but more the substrate roughness and film thickness. TaN tends to be thinner and NiCr is easier to etch to a fine line pattern.

With only a LM399 reference the extra 1/f noise from a bit noisy TaN resistors in a 7 to 10 V gain stage would be hardly noticable (maybe comparable to an OP07 OP-amp).  It is still easy to choose the lower noise NiCr version.

[Birb]

The (probably not so good) artifact 10V reference:
Should achieve 10V with relatively ok accuract: +- a few mV
Stability is hopefully good?
Tempco is probably very low because LT5400s are used instead of NOMCT (Ratio drift is lower)
No clue about op amps though (LT1012 has a drift of ~0.2uV/K)
Also not exactly sure the best way to design a battery charger, so there is a very naive implementation - probably need to change that
(And the battery voltage has to be ~18V to accomidate the LDO regs to work at 15V (The LM317 is there to regulate to around 16V in case a 24V battery is used, at the expense of wasted power!)
Entire thing is best separated into 4 boards (3 references + 1 battery charger), but I show it as a complete schematic.

(this entire thing could be not very good (or cost effective!) but it's an idea.)
(Not shown: AC input capabilities, transformer)

[Kleinstein]

There is no need for separate regulators for the references. A single reference for mupliple units should be fine.

The idea with mixing different gain setting to get closer to 10 V only works if the references are relatively close to nominal. So one would have to select the refenrence with a suitable voltage. The actual voltage of the reference would determine which type of circuit is more suitable. For the fine trim one would very likely still need some extra adjustment points.

With 3 reference in parallel one also needs to take care at the ground side. The ADR1399 ref. current should be more like 2 mA and this may already add a bit drop at copper traces, especially if one would use separate PCBs.

The LM399 and likely also the ADR1399 usually has little hysteresis and no real need to have it powered 24/7. So battery power would only be for a realatively short time when actually needed. So no need for hot shipping and similar things as with the really high end reference.

[Birb]

Thanks for the advice.
I added three regulators mainly because of the initial inrush current of ~100mA, which is quite a lot (And 100ma is the upper limit of the LT1761)
On that note, how can this current be easily delivered during startup? Or does it not really matter and even a lower current will startup just fine given sufficient time?
(Do I need a startup resistor? Are there methods that affect this less? Thanks)

As for the battery thing, I guess it's just "best practise" or something, though I might try it in a future project when I need to.
On that note, how important are the three resistors that do the averaging?
(Actually i can just do the math - but i'm too lazy so desmos it is)
(Based on a numerical estimate, assuming the tempco of only one of the 10.111V reference's resistor is different, the total tempco of the ratio is some 0.0689 of that)
(I.e. if one had a tempco of ~500ppm/K and all the others at 0ppm/K the total tempco would be ~35ppm)
(Thus it seems that even with mediocre matching the difference might be almost immeasurable with what I have)