LM399 based 10 V reference

[iMo]

While playing with the new LTSpice24 - here are couple of sims with the traditional 7V to 10V buffer made of a single n-jfet and OP07.
The current limit value depends on the Idss, in the sim it is about 48mA.
Mind in case of a short the current flows from the OP07's output via n-jfet's gate, therefore at least 3k3 in the gate would be nice to have.
The low value resistors in series with capacitors are their esrs.

[ivo]

If one wants a simple buffer in-front of the opamp for the slight bit extra power/safety over its bare output stage, isn't this a classic push pull design? the 1k resistor allows the opamp to fine control the level when it's settled on the exact value, but probably only has to contribute single digit mA if the output is yanked on by a bigger load. I remember reading it from some app note.

[Kleinstein]

The transistors in the simple push pull would only work one at a time and not at all a light loads. So this is more like a class B output stage.  One would like more like a class AB output stage, so that both transistors would be active at the same time.  A simple for may work with an NJFET and PNP transistor. To limit the current one would likely need a resistor at the source side and somewhat selected threshold and / or extra diode in series.
If one takes into account current limiting an OP-amps as driver does not look that complicated.

For the response wanted, this may be different dependent on the meter / sensing amplifier in mind. With an AZ amplifier at the meter input one would wont a well behaved response with not much ringing or resonance. Quite some capacitance at the output could be a brute fore way to get there, maybe combined with an RC combination.

For an AZ switching DMM (like more HP meters) this may not be enough, as a large capacitance also slows down the pulse response. So the ideal would be more fast settling after a pulse. There should be at least some 10 µs before the meater actually reads the singal, and for the first few µs one does not need to be settled to sub ppm, just that one average the error is not to large in error. Setting in some 100 µs may still be fast enough if not starting to large in excursion.

[iMo]

The settling time sim - after the same NEG pulse as above in my ADR1001#1 buffer.
With the POS pulse it is 46us. The CLC filter adds the damped resonance, of course. Not counting capacitance and inductance of the leads.
Without knowing how the typical pulses look like it would be just wild guessing, imho (after 11 years of this thread) 

[Kleinstein]

The settling time should not depend much on the pulse. The pulse can effect how much the settling part is excited, but the time constant should not change.

[iMo]

Here is a doc with chopper's pulses at their inputs.
10-15ns wide and 4-15uApp of amplitude..
https://www.eevblog.com/forum/metrology/zero-drift-amplifier-input-bias-current/msg4991131/#msg4991131
The pulses from DMMs may differ significanlty, of course..

PS: Not related to the DMMs inputs directly - a 15ns pulse (2ns edges) -10uA amplitude will create aprox 2uVpp response with 26us settling in above sim with OP07 and jfet transistor. Also mind the modern opamps have the pulses X usecs apart..

[miro123]

The settling time sim - after the same NEG pulse as above in my ADR1001#1 buffer.
With the POS pulse it is 46us. The CLC filter adds the damped resonance, of course. Not counting capacitance and inductance of the leads.
Without knowing how the typical pulses look like it would be just wild guessing, imho (after 11 years of this thread) 

Hi Imo,
Few remarks on your simulation
Common mode chokes are not ideal inductance. their core is optimazed for max loss instead of efficiency. It is nice to look at IR images of such chokes.
Common mode chokes also behaves as diferential.
Common mode has parasitic capacitance. Capacitance depend on inductance - number of turns.
Common mode chokes are in many flavors. Depends on application that you want to use it.
More info at manuafascture sites datasheets and ANs. Top manufacture Wurth, TDK - few links https://www.we-online.com/components/products/datasheet/744282100.pdf
I tend to use them sparely and only If I know which frequency range I want to attenuate.  If you dont follow this rule you end up with worse solution. Having an spectrum analyzer with all accesories arround it help a lot.
The spectrum also depends on type/coax or twisted pair or random wires / and length of used cables

[miro123]

Hello here is the draft schematics revision of 4,66 , 7V and 10V
Few  requirements
Long term stable 6,9...7.1 Vout
4,6V short term stability and lower noise
10V - the most lose specified - I dont know do I really need. It was just available by the  bootstrap circuit.
Opamps are still not specified. I will build at least two  boards + one for test. I will try different opamps.
The board wil be in separated case - Bigger case will contain   Vref + R reference cases. Resistors are almost ready. I still waiting for final cure of the sealant. I wil post some pictures later on. Four selected Vishay foil resistors + NTC+PT100 are submerged in silicone oil  bath.

Comments and remarks are welcome.  I'm sure there are many mistake at this stage. Keep in mind that I'm beginner in CAD systems. Last time I did it ~30 years ago when I was junior engineer
BR,
Miro

[iMo]

..Common mode chokes are not ideal inductance. their core is optimazed for max loss instead of efficiency. It is nice to look at IR images of such chokes.
Common mode chokes also behaves as diferential.
Common mode has parasitic capacitance. Capacitance depend on inductance - number of turns.
Common mode chokes are in many flavors. Depends on application that you want to use it..

My common mode choke is made of a small ferrite toroid core with two or three (I forgot) turns of twisted enameled copper wire coming from the Vref board and going to the terminal posts. Inductance around 10-20uH I guess. The CM choke is intended to suppress the higher frequency bands (like XX MHz up)..

[Kleinstein]

The OPA2209 is a rather expensive choice for the output buffers inside the loop. A NE5532 should be about as good though maybe with a little  more power consumption.
The filters before the output buffers may effect the loop stability. One may want an extra capacitor in direct feedback to get the fast feedback part for U3B and U1B, at least as an option in case it oscillates.

The capacitor C5 at the 10 V output should not come from the ouput, but directly the OP-amp output or maybe from the the other side of R6. The idea is to get the higher frequency feedback this way and this is needed to get the tolerace for large capacitance at the output.

The filtering for the reference voltage looks rather modest. If there is space large capacitors (possibly a few µF) for C1,C6 and C12 can make sense

As a simplification the 4.6 V part could get away be using the 7 V output as a buffer to start with. So one could get away without U3A.

The 10 V output would be about 10 V, not an accurate 10 V, with no trim.

The LM399 heater voltage should ideally also be regulated. There is a slight effect of the voltage on the temperature and this way the ouput voltage.

[Andreas]

Just some remarks:

U4 seems to have no power supply.

I would use the same protection zener for all outputs.
(All Op-Amps are supplied from the same 15V. So the zener only needs to limit to ~15V with (current limited) voltage at the output).
I prefer the PTVS11VP1UP,115 Diode for my references up to 10V.

Is there any reason to  put R13, R16, R17 outside the feedback loop (R14, R22).
With 10 Meg input resistance of the DMM 20 Ohms give already 2 ppm error.

How do you dimension the output capacitors? (for which frequency coupled into the output)
C4 with 100 pF is rather low compared to ~1 nF of the transient zener.

I would wire C8, C10 not to GND but to the negative Input (PIN 6) of U1B / U3B to get stability against capacitive loads at the output (Coax cables etc.)
(See also AN86 of LT for the combo outputs).

with best regards

Andreas

[miro123]

The OPA2209 is a rather expensive choice for the output buffers inside the loop. A NE5532 should be about as good though maybe with a little  more power consumption.
The filters before the output buffers may effect the loop stability. One may want an extra capacitor in direct feedback to get the fast feedback part for U3B and U1B, at least as an option in case it oscillates.

Many thanks for quick and relevant review Kleinstein.
I will evaluate bunch of opamps in first one or two prototype boards. OPA2209 is know component for me but expensive. NE5532 is in my order list for evaluation. I have doubt about 22186 too. It is new component for me. I don't know how bad is compare to my well known OPA2198 opa2182. According to datasheet is more noisy an slower I have no idea about switching frequency.
All other remarks are relevant too.
Thanks
Miro

[Kleinstein]

The OPA186 seems to be a lower power version of the OPA187. The noise curves give a crude hint where the chopper frequency should be.
Similar the OPA182 is a slightly lower power / lower speed variant of the OPA189. Here the lower speed can be a good thing as the OPA189 is quite fast, when used with low gain.

The LM399 reference is still relatively noisy (e.g. 100 nV/sqrt(Hz) of white noise part and quite some 1/f noise on top) . So the op-amps for the buffer should not be that critical. The question is more if one needs AZ op-amps with there possible spikes or could use a classic precision OP-amp like OP07, OP177 or newer OPA207, OPA205,  ADA4077.

[miro123]

U4 seems to have no power supply.

Thanks you save two patches in PCB

I prefer the PTVS11VP1UP,115 Diode for my references up to 10V.

I will order a bunch of TVS and select the one with lower leakage across T range

Is there any reason to  put R13, R16, R17 outside the feedback loop (R14, R22).

I have doubt too. I still did not make a choise between OpAmp stability vs Performance/LowZ

How do you dimension the output capacitors? (for which frequency coupled into the output)
C4 with 100 pF is rather low compared to ~1 nF of the transient zener.

Still work ToDo - I put some random numbers :-) Thanks for the tip I will on LT AN86

I would wire C8, C10 not to GND but to the negative Input (PIN 6) of U1B / U3B

I dont like placing this capacitors to GND too due to worseing of stability  but I was also aware of choper injection back input. I will place them both and decide which to use later on
Thanks
Miro

[miro123]

The LM399 reference is still relatively noisy (e.g. 100 nV/sqrt(Hz) of white noise part and quite some 1/f noise on top) . So the op-amps for the buffer should not be that critical. The question is more if one needs AZ op-amps with there possible spikes or could use a classic precision OP-amp like OP07, OP177 or newer OPA207, OPA205,  ADA4077.

OPA2186 took my atention due to extremely low price. I'm reluctant to use it blindly since it seems different from other TI modern Az opamps. The max voltage is 24V. All other TI AZ OpAmp members have 36V. Is it different architecture or different Fab process? - I will order few and evaluate them. I like OPA207, OPA205,  ADA4077 and especially OPA2205 but I don't like their price.
OPA2140 is another chiose - since one opamp can serve as both - High accuracy, low Ib, and high BW - no need of dual stage opamps. So one expensive instead of two cheap

[iMo]

The NE5532 needs a small resistor at the output (within the main loop) when wired as the unity gain final stage in the buffer, otherwise it oscillates with capacitive loads (at least in my sims).

PS: below with 1uF ceramic multilayer at the output. The output impedance is max 1.9ohm, peaks at some 100KHz.
Note: sim only..

PS1: with 180pF FB there is still 47deg PM at 82kHz, but it starts ring..

PS2: replaced the 1uA/500ns pulses with 10uA/1us pulses in order to see the responses better. Output settles in 11/15usecs..

[iMo]

.. and yes, the LTspice can draw nice smooth graphs in sub uV range too.. 

[Kleinstein]

The LM399 reference is still relatively noisy (e.g. 100 nV/sqrt(Hz) of white noise part and quite some 1/f noise on top) . So the op-amps for the buffer should not be that critical. The question is more if one needs AZ op-amps with there possible spikes or could use a classic precision OP-amp like OP07, OP177 or newer OPA207, OPA205,  ADA4077.

OPA2186 took my atention due to extremely low price. I'm reluctant to use it blindly since it seems different from other TI modern Az opamps. The max voltage is 24V. All other TI AZ OpAmp members have 36V. Is it different architecture or different Fab process? - I will order few and evaluate them. I like OPA207, OPA205,  ADA4077 and especially OPA2205 but I don't like their price.
OPA2140 is another chiose - since one opamp can serve as both - High accuracy, low Ib, and high BW - no need of dual stage opamps. So one expensive instead of two cheap

If you look for a cheap precision OP-amp, the OPA202 / OPA2202 are good candidates. They can kind of replace the OP07 with less power consumption and slightly better performance in most aspects.

The extra buffer is not just for high BW, but also to remove the thermal effect from the precision OP-amp. With the thermal effect the OPA2140 may be equally or even more sensitive than other precision amplifiers.

The simulated pulse response looks good enough. It should be well fast enough to be used with a DMM, likely even the R6581 with it's rather strong current pulse (but at least quite some waiting time).
A real world NE5532 and many other OP-amps may want more than 1 ohm in series to avoid oscillation from capacitive loading. This would slow down the response a little.

[David Hess]

If you look for a cheap precision OP-amp, the OPA202 / OPA2202 are good candidates. They can kind of replace the OP07 with less power consumption and slightly better performance in most aspects.

The price and availability are so good that I wonder what the catch is with the OPA202.  The drift is only average and there is no offset null to correct it.

This is the first time I have noticed a high voltage complementary bipolar process with super beta transistors.  Manufacturers used to publish articles describing their semiconductor processes, but that data all seems to be secret now.

I am not sure why, but it does not show up in the selection guide at Mouser, at least when I search for it.

[argintviu]

Hi all! The time has come to finish a reference project that has been on the back burner for a long time. The project is very simple, based on an older design by Mickle T.. I don't need anything fancy, just to be able to check some 4 1/2 digit meters for drift. I have a few MAB399, OP07CZ (ceramic) and a 5k NOMCA network. The power supply is based on a transformer and uA723 set for 15V output.

I have a few questions about the general design:

- Does the heater need bypass caps? I was thinking to add a 10uF tantalum + 100nF. The thing is, won't the tantalum cap die from the heat if it's mounted near the MAB399 pins? Would a 100nF suffice?

- Does the "zener" need a bypass cap (CZByp)?

- The bias compensation resistor R38 makes a convenient RC filter with CZFilt (as seen in AD587 datasheet). What kind of capacitor would be needed here? I was thinking to use 1uF 250V polypropylene (or polyester). C16 and C17 would then be of the same type.

- It would be nice to also have access to the raw voltage of the reference. Should it be buffered with a unity gain op amp?

My apologies if some of the questions have been answered. I have read this thread and many others ~1 year ago but I tend to forget.

[Kleinstein]

The LM399 gets hot inside but has already some insulation. So the power consuption in at about 200 mW (at bit more with short leads) and it does not heat up the PCB and surrounding parts very much.
The larger capacitor for the decoupling can be a bit away from the reference. 100 nF or maybe 1 µF at the reference heater would be enough. Some larger capacitor  for the 15 V supply in general could be good. C18 could already do that job.

The zener sides does not need a parallel capacitors. Just a capacitor could even cause more problems than good. The ADR1399 wants some 5 ohm and 1 µF as series element. A series RC would probably also better for the LM399.  So it makes sense to include the series RC in a PCB to have the option to also use an ADR1399. The capacitor could be a SMD ceramic (not too small form factor to no loose to much of the capacitance).

CZfilt should be a low leakage type. A polyester type should be good enough. PP tends to be bulky. There is no need for a low TC resistor for R38.

For the raw voltage it depends on the use if one wants / needs a buffer. A buffer offers some extra protection and may help with more lower resistance meters. A high Z DVM could be OK without.

The capacitor C16 to slow down the buffer is a bit on the large side. A smaller capacitor (1-10 nF range) would give faster settling, though lower maximum tolerated capacitance at the output.
For the trim resistor R52 the value depends on the individual reference voltage. Depending on the unit one may also need to trim up, e.g. with a resistor in parallel to R50.

[iMo]

I had several MAB399 and MAC199. It seems to me there is a design flaw with the heater, without a series protection resistor many of my chips died (the heater died). Also I would recommend the 1u ser 5ohm on the MAB399 as well for the stability.

I found out the soldering any stuff onto the 399 pins will increase the noise (ie by 20% in 34401A) (thermal flows). So use thin wires or tracks to the pins in some length.

Below my mods.

PS: you may go with higher voltage than the 15V, of course.

[macaba]

What are the practical advantages (no theory please!) of having these discrete transistor output buffers over using a buffer opamp?

Just wondering if I should change my design...

[dietert1]

I'd guess the 2N2222 with its 800 mA current spec is more robust than most opamps, so the idea is a decent short-circuit behaviour. The diode between amplifier and output stage together with a bigger R39 help, too.

Regards, Dieter

[bastl_r]

I'd guess the 2N2222 with its 800 mA current spec is more robust than most opamps

And you can easily create a very fast over current- and short- protection via Q8 and R40.
I think.
Correct me if i'm wrong.