VoltCard design with LM399: does this look sane?

[technix]

Another revision of the VoltCard. It is no longer standing off the main board (to accommodate the heater power without running that track all across the board) and I trimmed a bit of size off it. Actually if I get rid of the digital portion this module may be made even smaller, thanks to the SMT parts (the entire constant current source is two 0603 and two SOT-23 packages)

[Kleinstein]

If you have the amplifier to 10 V (or similar) you can just use a resistor from the 10 V as a current source. Depending on the OP you may need a startup circuit (e.g. a resistor from the OP output to supply).

The Trimmer position is better, but still not good enough - the pot may have a TL of something like 200 ppm/K and even the voltage fraction at the pot may change by some 10 ppm. So you likely want the pot to only go from something like 9.9 -10.1 V or even less. This may mean you need to use individually selected resistors as well.
Also the up to 250 K input impedance at the OP can give extra trouble. So better have the full 10 V at the trimmer, and use a resistor at the wiper to reduce the influence of the trimmer. This also only needs 2 precision resistors.

[GK]

The 500k trimpot ads up to 125k in series with the (-) input, not 250k (with the wiper at center position there is effectively two 250k resistors in parallel (disregarding the much smaller 200R). In any case 500k is much larger than necessary. 20k (in parallel with the 200R) would be large enough, effectively reducing the coefficient by a factor of 100. An adjustment range of 9.9 to 10.1V isn't enough to cover the LM339's initial tolerance of +/-2%.

[technix]

If you have the amplifier to 10 V (or similar) you can just use a resistor from the 10 V as a current source. Depending on the OP you may need a startup circuit (e.g. a resistor from the OP output to supply).

The Trimmer position is better, but still not good enough - the pot may have a TL of something like 200 ppm/K and even the voltage fraction at the pot may change by some 10 ppm. So you likely want the pot to only go from something like 9.9 -10.1 V or even less. This may mean you need to use individually selected resistors as well.
Also the up to 250 K input impedance at the OP can give extra trouble. So better have the full 10 V at the trimmer, and use a resistor at the wiper to reduce the influence of the trimmer. This also only needs 2 precision resistors.

I don't want all that bootstrapping trouble so current source is intentionally made independent to the voltage reference.

I am okay with the accuracy of this version of the circuit already. OP07C is a good OPA with low input offset current so that 250K will not give me too much trouble. I spun a new board revision, double side loading components to shave off even more board space, and go back to the breadboard-friendly single-in-line form factor. This revision also makes the module almost the same length ang width of a 40-pin DIP chip

[technix]

Final revision before sending off to the board house.oes this seem to work?

Ended up in a 0.1in SIP form factor with pins:

1. Heater V+
2. Heater V-
12. Reference Out
13. Reference Ground
14. Reference +12V
17. EEPROM GND
18. EEPROM SDA
19. EEPROM SCL
20. EEPROM Vcc

The digital section and its pins 17-20 can be omitted.

[technix]

It is not only the inrush current that matters for the LM399. Even under normal operation the current depends on the external temperature, the wire resistance usually has a TK of some 3000 ppm/K and the contact resistance is not that stable and predictable. So to get a reliable reference there is not way around having a separate ground connection for the voltage signal and the power. You might get away with the 1 mA current for the Ref. over the voltage contact, but even this is better avoided.

The negative heater terminal of the LM399 can be tied negative with respect to the anode of the zener by a substantial number of volts and the heater can accommodate a supply voltage as high as 40V. So if you can furnish split (+/-) supply rails, the heater can be powered directly between them, keeping the heater current out of the ground completely.

The ADR... Refs. in SMD case are not a real alternative they have significant more long term drift and hysteresis. Also humidity is likely a problem.

The ARD01 will deliver better short and long term performance than an LM399-based reference followed up by a simple and cheap voltage-scaling amplifier.

Actually I found someone offering ADR01B at US$1.5/each... Volt nuts, if you need those, call me.

I have already bought the LM399's and the percision resistors so this LM399 VoltCard ("Original") will be spunned and built. If I got interested in another voltage reference again I will spin an ADR01-based VoltCard with a compatible pinout.

[Kleinstein]

The better way to combine trimmer with fixed resistors is shown here:
https://www.eevblog.com/forum/projects/bought-several-used-lm399-how-to-use-it/msg669177/#msg669177
This only needs 2 of the stable resistors, not 3. The other advantage is that its easy to adjust the range covered by the trimmer. If needed a 3rd. stage with 2 more fixed resistors could be used to do a coarse trim.

If you keep the high impedance feedback, the circuit should have a capacitor in feedback, to improve stability against oscillation.

The divider with only 1.3 K and 7 Ohms uses 5 mA of current, that's quite a lot for the ref. ground pin. The strange current source also adds about 2 mA, that can be avoided.

The Bootstraping current source is far from complicated. With the OP07 to already delivering 10 V, this is just 2 resistors: One to set the current and one to the supply to ensure startup.

The layout has a asymmetric connection to the LM399: one contact is just on top, the other has two lines going from the pin. This can result in avoidable temperature differences.

[technix]

The better way to combine trimmer with fixed resistors is shown here:
https://www.eevblog.com/forum/projects/bought-several-used-lm399-how-to-use-it/msg669177/#msg669177
This only needs 2 of the stable resistors, not 3. The other advantage is that its easy to adjust the range covered by the trimmer. If needed a 3rd. stage with 2 more fixed resistors could be used to do a coarse trim.

If you keep the high impedance feedback, the circuit should have a capacitor in feedback, to improve stability against oscillation.

The divider with only 1.3 K and 7 Ohms uses 5 mA of current, that's quite a lot for the ref. ground pin. The strange current source also adds about 2 mA, that can be avoided.

The Bootstraping current source is far from complicated. With the OP07 to already delivering 10 V, this is just 2 resistors: One to set the current and one to the supply to ensure startup.

The layout has a asymmetric connection to the LM399: one contact is just on top, the other has two lines going from the pin. This can result in avoidable temperature differences.

1) The current resistor set consists of only E24 and E96 values: 1.3k, 200 ohms (both E24) and 499 ohms (E96) so they are way cheaper. And I never bothered to sit down and calculate the resistors.

2) Thanks for the info, and the current layout allow me to tuck that cap right across R6 on the other side of the board (I am double side loading anyway) but what is the optimal capacitance? And will a cheap MLCC affect the circuit's accuracy?

3) I am not bothering that, and the TL431 current mirror is good enough for me.

4) This asymmetric-looking connection have its roots at four-wire measurement. The trace on the back is a constant current source, and it merges with the trace on the front (sense pin to the op amp) right at the pad to the Zener.

[Kleinstein]

The capacitor in feedback is not critical. It just sets the upper bandwidth limit for the amplified part. Anything from 100 pF to 1 µF would be OK - possibly less in a high impedance setup. A normal cap. should be OK, as long as leakage is not extremely high. The position across R6 is not helping much, the critical part is the high impedance from the trimmer. The right place is from the inv. Input to the output. 

As for the layout at the reference, symmetric would be preferred, even if this would mean having an extra blind connection at the other pin. It's also better to have the 2 lines close together, not separated as in the current layout. Another option would be to make the bottom a complete ring around the 4 Pins to reduce thermal gradients.

If there is place is might also help to have something to fix a cover to the board, as to prevent turbulent airflow around the OP.

[technix]

The capacitor in feedback is not critical. It just sets the upper bandwidth limit for the amplified part. Anything from 100 pF to 1 µF would be OK - possibly less in a high impedance setup. A normal cap. should be OK, as long as leakage is not extremely high. The position across R6 is not helping much, the critical part is the high impedance from the trimmer. The right place is from the inv. Input to the output. 

As for the layout at the reference, symmetric would be preferred, even if this would mean having an extra blind connection at the other pin. It's also better to have the 2 lines close together, not separated as in the current layout. Another option would be to make the bottom a complete ring around the 4 Pins to reduce thermal gradients.

If there is place is might also help to have something to fix a cover to the board, as to prevent turbulent airflow around the OP.

Added another blind ground connection to the reference (and I happen to have a ground via to connect it to.) The drive/sense side is a bit sad as the board layout comes into play and I cannot just run the traces parallel across the board itself to each other for a small part of the circuit.

This board will be small enough to apply a heat shrink tube across the mid-section (about the same physical size as a PDIP-40 chip body not counting the reference chip, headers and the pot,) covering up the OP and constant current source (before the reference chip or the pot is soldered of course) so the cover is no issue.

[GK]

I already showed you that alternative pot/divider resistor arrangement in reply#8, but you didn't seem to like it. The advantage of the second arrangement, however, is that the adjustment range can be set symmetrically +/- about the target voltage if you choose the resistor values correctly. This is potentially an advantage if you desire to minimize the influence of the trim-pot. with the first arrangement this is only possible if the targeted output voltage is 2 times the reference voltage (or half the reference voltage when used as a passive divider).
 

[technix]

I already showed you that alternative pot/divider resistor arrangement in reply#8, but you didn't seem to like it. The advantage of the second arrangement, however, is that the adjustment range can be set symmetrically +/- about the target voltage if you choose the resistor values correctly. This is potentially an advantage if you desire to minimize the influence of the trim-pot. with the first arrangement this is only possible if the targeted output voltage is 2 times the reference voltage (or half the reference voltage when used as a passive divider).

This design already have the pot's tempco effect within 0.5ppm and that is already good enough for me (or anybody, since 0.5ppm is within LM399's own tempco and noise threshold.) No need to guild the lily here.

[Andreas]

Hello,

the 0.5-1ppm/K for the LM399 is the average spec over the whole temperature range.

I measured around 7-8uV output voltage change over 10-40 deg C (>30 K range) for my LM399#2 reference.
(see picture: X-axis temperature, the right y-scale is the voltage difference in mV against a second LM399 kept at room temperature).

So around room temperature you might be much better than spec for the LM399 alone.

With best regards

Andreas

[Kleinstein]

The current trimmer setup has two disadvantages:
1) It needs a 3rd highly stable resistor
2) the output impedance is relatively high. This gets better if a slightly lower value trimmer is used, but the range is limited. This gives extra noise (especially from current noise of the OP) and also drift due to the bias current.

Depending on the quality of the trimmer, the extra instability may not be more than the LM399's intrinsic stability, but it still adds up and prediction how stable the trimmer setting is, is difficult. It is anyway difficult to do the adjustment over the full 2% uncertainty range of the LM399 with one analog trimmer alone. It's not only the TK, but also the mechanical stability of the trimmer. The classical solution is using a 2 step trimming: coarse trimming with a individually chosen resistor and only fine tuning with the trimmer.

[GK]

I'm sorry, but the noise is still negligible in comparison to the LM399's noise. Figure 6, page 4 graphs ~90nV/sqrt-hz above the flicker (1/f) corner at ~ 100Hz:

http://www.ti.com/lit/ds/snvs751c/snvs751c.pdf

90nV is equivalent to the thermal noise generated at 25 deg.C by a 500k resistor.

And don't forget that uncorrelated noise sources sum as the square root of the sum of the squares. E.g. two 90nV noise sources sum to 127nV not 180nV. So even if the OP's amplifier circuit was as noisy as a 500k resistor it would only worsen the overall noise performance by 3dB.

In any case the LM399 is a terribly noisy reference and any truly serious application requires filtering. So far the OP hasn't mentioned anything about his noise performance requirements. A no holds barred approach would use auto-zeroing op-amps in differential/instrumentation amplifier configuration to allow low feedback impedances while rejecting ground currents and attendant I*R voltage drops and incorporate very low-freq. filtering and multiple trim-pots for the scale calibration. AD make some truly low noise auto-zeroing op-amps. How far do you want to go?
 

[technix]

I'm sorry, but the noise is still negligible in comparison to the LM399's noise. Figure 6, page 4 graphs ~90nV/sqrt-hz above the flicker (1/f) corner at ~ 100Hz:

http://www.ti.com/lit/ds/snvs751c/snvs751c.pdf

90nV is equivalent to the thermal noise generated at 25 deg.C by a 500k resistor.

And don't forget that uncorrelated noise sources sum as the square root of the sum of the squares. E.g. two 90nV noise sources sum to 127nV not 180nV. So even if the OP's amplifier circuit was as noisy as a 500k resistor it would only worsen the overall noise performance by 3dB.

In any case the LM399 is a terribly noisy reference and any truly serious application requires filtering. So far the OP hasn't mentioned anything about his noise performance requirements. A no holds barred approach would use auto-zeroing op-amps in differential/instrumentation amplifier configuration to allow low feedback impedances while rejecting ground currents and attendant I*R voltage drops and incorporate very low-freq. filtering and multiple trim-pots for the scale calibration. AD make some truly low noise auto-zeroing op-amps. How far do you want to go?
 

1uF 0603 MLCC to ground at the input of the OPA will be added and that is it, none if board space is not adequate. The OPA also have a cap on its feedback path.

[GK]

1uF directly in parallel with the LM399 won't do much due to the LM399's low dynamic impedance. I'm also not sure that the LM399 is stable with a large capacitive load. A series resistor will also be required to make a worthwhile filter. Make sure the cap is a low leakage type.

Having substantial noise filtering prior to the scaling amplifier will make the scaling amplifiers noise dominant, so then you might want to reconsider its design.

A capacitor in parallel with the feedback resistor will not do a lot either, as the high frequency noise attenuation in this case will plateau at only -3.2dB.
 
If you're adverse to implementing filtering you may also want to test your individual LM399 for noise performance as the specified parameter spread is pretty large. The datasheet specifies 50uV worse case rms noise in a 10Hz-10kHz bandwidth. What?? That's 500nV/sqrt-Hz! 

[cellularmitosis]

I'm sorry, but the noise is still negligible in comparison to the LM399's noise. Figure 6, page 4 graphs ~90nV/sqrt-hz above the flicker (1/f) corner at ~ 100Hz:

GK, I wanted to say thank you for actually doing the math and keeping us honest.  Some of the results have surprised me (e.g., the LM317 drift being more than adequate).

[technix]

1uF directly in parallel with the LM399 won't do much due to the LM399's low dynamic impedance. I'm also not sure that the LM399 is stable with a large capacitive load. A series resistor will also be required to make a worthwhile filter. Make sure the cap is a low leakage type.

Having substantial noise filtering prior to the scaling amplifier will make the scaling amplifiers noise dominant, so then you might want to reconsider its design.

A capacitor in parallel with the feedback resistor will not do a lot either, as the high frequency noise attenuation in this case will plateau at only -3.2dB.
 
If you're adverse to implementing filtering you may also want to test your individual LM399 for noise performance as the specified parameter spread is pretty large. The datasheet specifies 50uV worse case rms noise in a 10Hz-10kHz bandwidth. What?? That's 500nV/sqrt-Hz! 

If that is the case, I will leave the filtering to the main board. I am running out of space here.

[Kleinstein]

If you already have amplification, it would be easy to add the filter in front, it's just a resistor and a low leakage capacitor. The main board might need an extra amplifier.

[technix]

If you already have amplification, it would be easy to add the filter in front, it's just a resistor and a low leakage capacitor. The main board might need an extra amplifier.

The main board will have at least a voltage follower on it (for the VoltCard DMM, there will be a reducing amplifier that emits a percise 2.5V using two matched resistor divider stages, for the adjustable voltage standard there is the DAC chip) so noise filtering can be moved to the main board safely.

[Kleinstein]

Every extra amplifier adds drift. This is one reason why the ref. voltage without any modification, except possibly a buffer if needed (the LM399 does not absolutely need one - the LTZ1000 might like to have one).

Also with a higher resolution DAC, there is often no need to have an exact value of the ref. voltage. Trimming can be often done in software, choosing the right DAC values. So no troublesome mechanical trimmer is needed. Its only if something like a decimal KV divider is uses directly to set values when you need something like a 10.000 V reference. Otherwise 10.x is often ok as long as you know x.