LM399 based 10 V reference

[miro123]

I decided to build second PCB version of my LM399/ADR1399 reference board.
One improvement point was - buffering the opamp output /as proposed many times in this thread with two npn transistors/
But I sow major drawback of proposed circuit - It is excelen voltage source and almost zero current sink. Zoutput source ~ 0 ohm    while Sink capability is limited to pulldown resistor ~22..100KOhms. see attachment.  I come to conclusion that this solution introduce more problems that it solves. Am I correct? Do I miss something?
I tend to use - combined Opamps Outer loop opa2182 for DC accuracy  inner loop//AD706.707 have many in my drawer / - for lower noise, filter/remove chopper artifact and reduce thermal stress on OPA2182

 

[iMo]

People sometimes use the opamp+LT1010 as well..

[David Hess]

People sometimes use the opamp+LT1010 as well..

I am not sure what that was a response to.

Buffering a precision operational amplifier preserves precision and there are lots of ways to do it.  A simple class-a emitter follower may be enough if there is a significant load.  A diamond buffer or current feedback buffer also works well and is likely less expensive than an LT1010.

[Kleinstein]

If one wants a seprate current buffer inside the loop, that extra buffer should be fast (faster than the outer loop, if there is no extra AC feedback), but it does not have to be accurate. So the AD706/7ß7 are not a good choice. It would be more like TL071 or NE5532.

The version with just the emitter follower is asymmetric in the source / sink capability, but one could add more resistive load (e.g. 5 K instead of the 22 K) if needed. The output impedance would still be just 1 value, unless the output needs to sink a much of the current from the resistor.

[iMo]

People sometimes use the opamp+LT1010 as well..

I am not sure what that was a response to.

What is wrong w/ it? LT1010 is 20MHz BW, 75V/us, +/-150mA.
Within the loop with the opamp ("opamp+LT1010") it could provide source or sink capability easily.

[David Hess]

What is wrong w/ it? LT1010 is 20MHz BW, 75V/us, +/-150mA.
Within the loop with the opamp ("opamp+LT1010") it could provide source or sink capability easily.

There is nothing wrong with using the LT1010; it is just more expensive than the alternatives.

[miro123]

Thanks, Kleinstein, David and Imo. I've got your point.
My point was that emitter follower promoted many times in this thread makes more harm than good.  Better choice is without it or alternative solution with composed amplifier. I agree that current feedback is the most suitable. Unfortunately I don't have experience with single ended current feedback amplifiers. Can somebody propose an suitable component. Requirement are low power, at least +15V Vcc, low noise and THD.

PS: I use such combo regularly to drive my ADCs input and Vref- OPA189/182 + Opa2625/AD8065. I find A8065 over the top for this application

[iMo]

..
My point was that emitter follower promoted many times in this thread makes more harm than good..

Could you elaborate a little bit what kind of harm it may cause?

[miro123]

..
My point was that emitter follower promoted many times in this thread makes more harm than good..

Could you elaborate a little bit what kind of harm it may cause?

Output impedance of 10..100K is no go for.
questions are
- how this circuit handle the chopper input of any DMM.
- Can I connect this output to any switched capacitor input? Sayying in simple enginering  jartgon - any CMOS  DAC ADC input or Vref
- Is there an use case where this circuit works?

My conclusion - this circuit is is only sutable to heat 10MOhm resistor. The switching behavour of DMM frontend and modern CMOS circuit requires low impedance sink and source output

[iMo]

That circuits people use with their 10Mohm (and higher) loads because it de-loads the 7->10V opamp and it has a current limiting function. For coping with xx nanoseconds wide fluctuations at the inputs of modern choppers or modern adcs you would need something better, indeed.

[Kleinstein]

The emitter follower has a relatively low output impedance. Typical a few ohms from the parasitic emitter resistance and than some 26 Ohm divided by the current in mA. So with a 20 K load (0.5 mA of current for the transistor) this would be a little more than 50 ohm - about comparable to the open loop output resistance of OP-amps.  This is the resistance without the OP-amp for the control loop - with the feedback one would have an output restance similar to the case without the emitter follower, just without the thermal error and no output stage cross over as the output stage is class A.  If there is no problem with oscillation due to excessive capacitive load the output should be able to also drive larger loads.

When driving directly a switched capacitor type ADC (most SD and SAR ADC chips) it would need extra care / filtering close to the ADC and ideally a ADC driver at the ADC. These ADCs just need a defined drive characteristic and generally don't work well with a variable impedance - the cable length alone could make a difference. So directly to the ADC is not a good input design.
Also the ADC / DAC ref. inputs may need a dedicated driver close to the chip and some want quite some capacitance there.

Chopped inputs at a DMM should include the filtering needed, so that the source impedance is not as critical. This is one difficulty in using AZ OP amps at a DMM input. If the input is sensitive to the source impedance this is more a problem with the DMM not the signal source.  The HP like AZ switching produces fewer (e.g. 5-50 Hz compared to some 50 kHz) current spikes than AZ amplifiers, but the spikes tend to be larger and longer. Some settling time is usually alowed, but this may not be enough for a slow driver.   The current spike from AZ switching could be an issue with the output, if build to tolerate a large capacitance, as this slow down the settling.

There is nothing bad with the LT1010. Similar to the AD8065 it may be a bit over the top. The in loop driver does not need to be low drift, low noise and low THD. One would like good speed and good tolerance to capacitive loading, whicht often comes with a bit higher power consumption. Quite some of the faster OP-amps are made with audio in mind and are thus still low noise and low THD, but this is not really needed. So the NE5532 was suggested for it's speed and availabilty, not the noise or THD. It is however quite power hungry. For a 10 V output is usually OK to power the amplifier from one the +15 ond GND, with no need to also have -15 V.

[iMo]

The output impedance in Ohm..

[r6502]

@MIRO123:
Does your schematic contain an error?


It helps, when you generate a part with multiple units. This units you can than pace like single parts on the schematic, see sample below, just a sample, but much clearer schematic and helps to prevent errors.


Guido

[iMo]

Despite the fact the output impedance of the emitter follower buffer is low, it handles the pulses coming from DMM's side poorly. I've made sims where I can see the output voltage changes based on pulses parameters.

If somebody knows how the DMM pulses look like we may make the sim (I think there was a post in past on the DMM's output transients comparison, but I cannot find the post).

The buffer with ie AD8065 works much better in the sim, of course.

Edit: Nope, the buffer with the emitter follower (with the current limit) behaves the same as with the AD8605. My error in the simulation..
I deleted the next post on the droop as well..

[Wolfgang]

The regulating loop is slow, and the output cap is a bit meager. So pulse loads will not be regulated out in time.

[miro123]

If somebody knows how the DMM pulses look like we may make the sim

See attached screenshot from Keysight 3446x data-shit. "Noise and injected current: Keysight Truevolt DMMs contribute less than 30% of the injected current than alternatives. Compared to some lower cost alternatives, Truevolt DMMs offer almost 100% less noise"

My trial and error experience is that my PICO M3500 is the worse of all my own DMMs - If I have time I can measure it with my Keithley DMM6500 digitizer. I have bunch of HP3456, HP34970 and Pico/Array M3500.
Back on topic
  - My goal is to make reliable voltage reference. Reliable translates to lower uncertainty
  - It must work in forward and reverse polarity on battery as well as transformer powered.
  - I want to use the voltage reference in real modern world. I dont blame for fout the DMMs auto-zero implementation or the modern zero-drift OpAmps. I assuem that they exist.
  - cope with EMI challenges from modern world - from solar Inverters, to EV chargers. from LED bulbs to BLDC air-conditioning /refrigerator.

[David Hess]

In the same sim schematics and the same pulses: with the transistors the voltage droop is say 12ppm, when transistors are replaced by a faster opamp like the AD8065 there are just 0.1ppm transients left.

An class-ab stage doubles the transconductance so should have half of the voltage droop.

I suspect this explains why Sziklai pairs are sometimes used at low current, especially when the first transistor is an FET.

[guenthert]

The motivation for the transistor behind the OpAmp were
      i) boost output current
      ii) cheap device to be replaced (there was a report regarding failed OpAmp potentially due to ESD) and
      iii) avoid uneven warming of the OpAmp due to excessive load, which might change input offset voltage

The pulses of DVMs are intense (in the order of a few nA -- observable using an oscilloscope), but brief.  Unless one uses very short integration times, the effect ought to be small.  It would be interesting to compare implementations with and w/o output transistor.  I recall having read the advice to observe the deviation when connecting a DVM using a passive null detector (I suppose a galvanometer was meant, but I don't think the passive ones were sensitive and certainly not fast enough)

To put things into perspective, the Fluke 515A portable standard has an output resistance (DC 1V, 10V ranges) of 300Ohm, the HP735A transfer standard even 1kOhm (comparable to Weston standard cells), but yeah, they can hardly be called 'modern'.

[iMo]

In the same sim schematics and the same pulses: with the transistors the voltage droop is say 12ppm, when transistors are replaced by a faster opamp like the AD8065 there are just 0.1ppm transients left.

An class-ab stage doubles the transconductance so should have half of the voltage droop.

I suspect this explains why Sziklai pairs are sometimes used at low current, especially when the first transistor is an FET.

See above I corrected the post on the sim -> no 12ppm droop, my mistake. There is a transient visible, like 1uV with 100nA/50ns pulse injected into the output.

Added 500ns long 1uA pulse, some 6uV response.

Added 500ns long -1uA pulse, some -6uV response.

Added 500ns long -1uA pulse with 2u2 capacitor, some -4uV response.

The 4.7ohm is not "esr" but "ser".

[iMo]

And the stability plots (FRA) with 47p/470p/4n7/47n FB capacitor and 2u2 ser 4R7 load.

[Kleinstein]

It absolutely makes sense to want an output that is no effected by a modern DMM. So good tolerance to current spikes and fast settling is really desirable. For this it help, if there is no large capacitance (e.g. 100 nF range) at the output. Getting fast settling with a large capacitor gets really hard. It could also help to have some actual resistance (e.g. 50 ohm range) in series, to reduce RF oddities from cable reflections.


The current spikes from AZ switching should be larger and with most meters also longer. The precharge part can compensate for this somewhat, but there is still the nonlinearity in the capacitance of the FETs.
Especially with a non zero voltage expect something like some 1-10 pQ of charge pulse, that would be some 1-10 µA for 1 µs or maybe a bit more stretched out.
The pulse from an AZ amplifier are shorter and smaller charge, but a lot more of them. There should be at least some filtering meter internal to smear out the pulse and make the peak current smaller.


I don't know what meters KS used in the comparison. A candidate for the really bad one is the ADT R6581.

[iMo]

.. a voltnut here made the measurements years back, for several meters afaik, the pulses repetition rates and amplitudes..

[Andreas]

Hello,

the question is: are these peak values, DC-values or RMS values? (with a large peak to RMS ratio?)

  - It must work in forward and reverse polarity on battery as well as transformer powered.
  - I want to use the voltage reference in real modern world. I dont blame for fout the DMMs auto-zero implementation or the modern zero-drift OpAmps. I assuem that they exist.
  - cope with EMI challenges from modern world - from solar Inverters, to EV chargers. from LED bulbs to BLDC air-conditioning /refrigerator.

So you will have a lot of work and additional components to spend like
- shielded transformer
- EMI filters on output (which increase the output impedance again)
- a good (double shielded) enclosure
...
And how do you check wether the requirements are really met?

with best regards

Andreas

[jorgemef]

Poor quality resistors not only have a problem with the TC but also with long term drift and noise.  For a reference source the long term drift is the more serious issue. Even with high grade parts it is hard to get specs on the long term drift and already the possibilty for drift is an issue, as one would not easily know if they drift.
The TC part is only the point that is easy to measure and compare.  A resistor array would really be the more sensible choice.

Finally found a local provider that can provide the NOMCT1603 5K at a reasonable price so will give it a try. Shipping costs from Mouser or Digikey are around 20 euros so as much as what the local provider will charge for four pieces plus shipping cost inside the country.

At least the tracking TCR should be more stable. I think will attach the part on top of my boards with thermal glue as dead bug configuration with the required connections between pins and just 3 connections to the board.

PS: would it be better to thermocouple to the board the resistor network on on the contrary to try to thermally isolate it like with some plastic spacer?

[iMo]

PS: would it be better to thermocouple to the board the resistor network on on the contrary to try to thermally isolate it like with some plastic spacer?

The temperature differences matter as they may create the EMF (in case of material differences), thus they mess with the voltages somewhere on the board. Also the temperature differences may create thermal waves bouncing in your box, creating "oscillations". So it is a rather difficult area - the modeling of thermal processes on the board and in the box.
The best solution, imho, would be to keep the board and the components in a thermal equilibrium, with constant thermal gradients (whatever they are).

Not an easy task, but the 399 is a cheap noisy reference, so one would not invest too much into it.. As Kleinstein wrote above the long term stability of the resistors is the key you have to focus on. The temperature could be measured (and you can make the TC compensation in math), but long term stability is something you cannot easily predict or adjust.