LM399 based 10 V reference

[Kleinstein]

The LM399 is fine with less current (e.g. series resistor or current limit from regulator) during start up. It would just take a little longer (e.g. another 30 seconds) to reach the stable temperature. A weak regulator with not much eat sink could still be effected a little from the temperature rise causing a little voltage drift. So good cooling for the regulator may still help. I have not done the math how much the effect really is.

The TC of the averaging resistors are as important as the difference in the voltages is compared to the 10 V. So it depends on the actual ref. voltages and may be attenuated by about a factor of 100. So the resistors should be resonable matched if there is a significant difference. It is similar to a divider from the 10.11 V to 10 V, just with equal resistors and this way a bit easier to get good matching.

[Birb]

Hmm it seems like another LT5400 can be used, though I’d assume that there is attenuation on the actual tempo of the resistors, so possibly even 25ppm/K may be ok (numerical testing is ~2ppm/K).
Though I wonder: Should there be another buffer on the averaged output?
(But that adds another few uV/K which is not that desirable)

Edit: Possibly some buffer can be used like the LT1010?

[Andreas]

The LT1010 has too much input bias current. You need a additional driver for it.
But why not use a low noise chopper like ADA4522 which delivers enough current?

Take care on your cirquit.
The ADR1399 needs more than 1 mA for the zener.
At the LM317 the resistors are not according to data sheet (minimum output current + voltage divider interchanged).
I am not shure wether the LM317 survives back powering by the battery when the input voltage is off / shorted to gnd.
But in any case the battery will be discharged by the voltage divider. (which should carry 5 mA according to data sheet).

with best regards

Andreas

[Birb]

yeah, the battery circuit is not particularly good and needs changing, i’ll fix it when i can.
(Though one issue is the battery voltage, 18V is not as easy compared to 12 or 24V of gel batteries which are usually easier to use.)
(Perhaps NiMh trickle charge?)
As for the output buffer, something like the ADA4523 might be good, but i also want to know if filtering should be added in the feedback loop, similar to that in the datasheet of the ADA4523?
thanks!

[dietert1]

For our ADR1399 setup there is a 21 V 2 Ah Li-Ion battery pack i got from Aldi (meant for battery tools). The power supply it buffers runs at 20 V (2x 10 Vref). Pass transistor and battery are outside of the oven while op-amp and 2:1 voltage divider are inside the oven.

Regards, Dieter

[Kleinstein]

As a LM399/ADR1399 does not need to be powered 24/7 a separate battery pack like the power tool battery makes sense. The alternative is than direct power from mains for the initial ageing and less critical use.

The ADA4523 is overkill. It is anyway a rather extreme part with limited use: very low voltage noise, but high current noise and high power consumption that makes it less suitable for high precision (may still be OK with a 5 V supply). The OP-amps to consider are more OPA205, ADA4522, LTC2057, MCP6V51. The question there is if one needs an extra in loop driver to reduce the loading to the OP-amps.

[iMo]

I've been using mains+trafo+uA723 w/ aprox 15V for the 1001. And an external post for a 21V 1AH LiIon battery pack when transferred the reference to the metrology offices. BTW., both sources could be wired at the same time (ie. the battery wired via a simple diode with a "trickle charging" resistor in parallel, wired to the input of the 723). Simple and it works fine.

[Korchahin]

...The ADA4523 is overkill. It is anyway a rather extreme part with limited use: very low voltage noise, but high current noise and high power consumption that makes it less suitable for high precision (may still be OK with a 5 V supply). The OP-amps to consider are more OPA205, ADA4522, LTC2057, MCP6V51. The question there is if one needs an extra in loop driver to reduce the loading to the OP-amps.

I have some OPA2187, can I use them? Maybe it even makes sense to use both at once?

[iMo]

Almost any opamp with less than 1uV/C input drift and less than 1uVpp 0.1-10Hz input noise is ok. When using a large input R (in an RC filter after the 399/1399 for example) the input current noise and its drift should be low as well (not much important when directly coupled to the zener as its output impedance is less than 1ohm). Afaik the OP07 has been recommended here since ever..

[Birb]

Hi, when you state large input R, how large do you mean? (Also, what about the capacitor for the RC filter? Does it matter?)
Thanks!

[iMo]

The max input R in an RC input filter depends on the opamp's input bias current and its drift and part's overall V/I noise conditions. Like with input current I the voltage drop on the R will be R*I with several consequences (ie the input current has got a noise and a drift and a TC, as well as the R has some intrinsic noise and a long term stability and a TC).
The capacitor - similar - it has got some leakage current which may fluctuate (thus creating a voltage drop on the R which will fluctuate as well), and some DA, TC, etc. There are pretty long discussions here on those topics..

PS: like the OP07 - with say +/-2nA inp bias current and a "large R" 100k resistor the drop will be +/-200uV, with its say +/-15pA/degC TC it adds up +/-1.5uV/degC. Not counting  I/V noise effects.

With the OPA189 the inp bias current is typ +/-70pA (DS), thus with 100k it creates +/-7uV offset. I cannot find the TC of the inp bias current in the DS. It seems to me this opamp could be used with for example 100k as ie. its input voltage spectral density is equal to that 100k resistor (DS graph). Not counting resistor's TC and its long term stability, and opamp's chopping artifacts on its input..

And the OPA205 has got typ +/-100pA inp bias current, with aprox 5pA/degC (DS graph). With 100k that adds +/-10uV to its +/-4uV inp offset, with 0.5uV/degC. The input current noise density is 30% lower, voltage inp noise similar, and the input voltage offset's TC is 10x higher compared to the OPA189, no chopping artifacts.

You may experiment with this (example only, R5 based on the zener used, etc.)..

[bastl_r]

Quality Foil = Siemens MKL? I have a couple of them with 6.8µF and 10µF.

[dietert1]

In our ADR1399 reference the filter has been made a Sallen-Key 2nd order low pass with 2x Wima MKP4 10uF 250V. A very similar circuit, just add one more capacitor.
At time constants of several seconds or more it makes sense to have the capacitors in an oven as even the best film caps have some capacitance variation with temperature and that will show up as voltage variation at the output of the filter.

Regards, Dieter

[Kleinstein]

It very much depends on the use case how much filtering is useful. A filter may improbe the 0.1 to 10 Hz band noise that is kind of the standard band. However for DC use the relevant frequencies are lower and there a filter gets very tricky and less practical.

When used with a DMM the DMMs are quite good in limiting the bandwidth and reduce the effect of higher frequency noise.
Many DMMs us AZ switching that is sampling the input only half the time. This results in some response to noise at frequencies close to the reading rate and the odd harmonics. Still this effect is quite limited, as the low frequency noise is higher from the 1/f noise. The 2.5 Hz for 10 PLC may still be accessible to filtering. After that there may well be a gap with essentially no effect when used with a DMM to read the reference. So there is very limited use for a cross over below some 1 Hz.


Some filtering makes sense, if there is not much extra effort, like if there is an ampifier anyway. There is a balance of effort for a filter and starting with a better reference (ADR1399) or 2 in parallel.

[iMo]

FYI

[David Hess]

It very much depends on the use case how much filtering is useful. A filter may improbe the 0.1 to 10 Hz band noise that is kind of the standard band. However for DC use the relevant frequencies are lower and there a filter gets very tricky and less practical.

Even for just 0.1 to 10 Hz, it is very difficult to make a filter which does not make things worse.  It is almost always better to reduce the flicker noise at the source, like with parallel or series references, or use a design which inherently rejects flicker noise like correlated double sampling (chopping), or ratiometric operation.

Also it is difficult to distinguish low frequency noise from drift, including the effects of inadvertent thermocouples, so steps will be needed to handle these as well.

[dietert1]

There is another thread about making an analog low pass filter that shall reduce LTZ1000 reference noise to 10 nVrms or so by averaging at time scales of hours or days. The last state of discussion was that one might use LiFePo4 batteries. Definitely not something you make just out of curiosity. People appear to be very busy...

Regards, Dieter

[iMo]

I made an experiment with the RC filter inside my own 34401A.
Cut the trace near the AD706 (the 7->10V opamp) and extremely carefully soldered the RC there somehow.
Tried with 10k/47k/100k and 3.3uF wima foil.
Measured against my ADR1001. Frankly, I had not seen any tangible decrease in the noise (evaluated the stddev).

I saw a nice change (like a 20-30% drop) when I replaced the original HP399 with one of mine best LM399 (with the lowest noise when not shooting, I never saw a shot during the noise measurement with my noise tester, so I thought it has got none), I published here the result, afaik. The problem with it was the LM399 made a couple of times per day a single shot in one direction by the classic ~5uV @10V range (the other problem was - the readjustment of the entire meter then). The HP399 is noisier but the popcorning is frequent and it is distributed evenly (I think they are selecting the zeners that way as the chance to average the shots during the measurement process is high).

Edit: added the shot..

[Birb]

one question about grounding: Should the metal case the voltage reference is housed in be grounded to earth or circuit ground? (I’d imagine circuit ground to ensure the potentials are absolutely equal, but I’d like to check.)
Thanks!

[Korchahin]

Hello guys. Please tell me why stable current generators are used on op-amps (OP07, OPAx189) and not on discrete circuits? For example, a simple circuit can provide a fairly stable current. What is the superiority of op-amps?

[iMo]

You may run the 399 off a simple resistor (afaik you may find it in DS). The opamp's are used because people do 7->10V and that allows to create the stable current (thanks to the ref itself, a bonus)..

[David Hess]

Hello guys. Please tell me why stable current generators are used on op-amps (OP07, OPAx189) and not on discrete circuits? For example, a simple circuit can provide a fairly stable current. What is the superiority of op-amps?

Using an operational amplifier removes or minimizes temperature dependent terms like Vbe.  This also makes it easier to apply very small amounts temperature correction.

[ivo]

Hello guys. Please tell me why stable current generators are used on op-amps (OP07, OPAx189) and not on discrete circuits? For example, a simple circuit can provide a fairly stable current. What is the superiority of op-amps?

Your circuit presented, you have put in a ultra low tempco 1N829A... but then you have servoed it with the Vbe drop of the BC556! So the expensive zener contributes 'no particular purpose', you are still feeding a temperature-dependant current (because of the BC556's Vbe junction tempco) to the ADR1399. An even more stable design might be using a zener with an intentional +2mV tempco, usually they are 5.6V or 6.2V, that would counteract the rough -2mV tempco of the transistor.

Using an opamp feedback circuit to drive the current, it means that the current driving the reference, itself derives its flat tempco of the reference. So it's a self-stable referential loop. Only the tempcos of the voltage & current setting resistors, contribute to any tempco of the reference current.

If one is appreciative of such things, you might use the phrase "belt and braces" for such design; alternative if you were sceptical you might call it "gilding the lily". If I'm reading the ADR1399 datasheet correctly, LM399 changes its reference value by roughly 0.5uV per 1uA change in current, while ADR1399 is capable of only 0.04uV per 1uA change. In any case, a good modern opamp + decent resistor is cheaper than an as-new 1N829A.

[dietert1]

Don't know whether it has been mentioned before: A LM1117-adj with a metal film load resistor can be a simple and quite good current source, with built-in reference and regulator. It has low noise and low temperature coefficient within a voltage range of 1.5 to 15 V. Minimum current is about 2 or 3 mA depending on required temperature range, so a 390R resistor should work. The circuit requires an output capacitor to avoid oscillation. That's no problem when driving a ADR1399. Maybe the recommended ADR1399 snubber is already enough.

Regards, Dieter

[Kleinstein]

From the stability and simplicity it is hard to beat the resistor from the 10 V output. Even when build with separate drive and sense it is still the preferred solution.