Precision zener-based reference - constant current supply schematic?

[PCB.Wiz]

Regarding my sziklai pair contraption, I have put there a collector load for the PNP Q3 (R1 in my schematic, see post #21). Shall this stay and if yes, what would be determined optimal value there and how? And if not, would it be better to switch to just a resistor across BE of Q4? And again, if that, what value would be considered optimal and why?
I think R1 against GND gives the most speed to the output NPN, as it can remove charge from the base faster, than R across b-e. Or maybe not? ...not that speed would be any concern here.

That value is quite broad.
It does not need to be low, something that reduces the Ib effect and loads Q3 more constantly likely helps, but the feedback is very forgiving.

When using the PNP + EF  NPN, wouldn't it be beneficial to lower the bias through the diff pair too? (noise? lower heating -> more thermally stable, ??)
I see you have cranked the bias much higher  (x2) compared to original schematic.

Numbers sub 1mA do not give much heating, but that does raise an interesting point.
I noticed the corrected parabola of the two choices, are flipped. Maybe there is a choice somewhere that can null better ?

Seems there is: here is a bias point of 1.07mA for Q1,Q2 which nulls better to two flats.

 

[Yansi]

Hello,
sorry for not responding any sooner, was on vacation and trying to rest.   

Those tempco nulls I guess are very component dependent and will be wildly different when using different physical parts. Practical realization would mean I'd need trim pots at those places. Setting them would be another wild challenge I guess.

Does it even make any sense to place those trims there, will they have any practical use, considering practical tempcos of resistors?  I have looked in my stash of retro period correct resistors and what I have is usually in the 50-100 ppm range and yes, that was considered precision metal film at that time around here ("normal" metal film resistors were in the range of 200-500 ppm/K  :'( ).

For the 68R resistor, I think I have found a 68R1 100ppm/K 1% 1W resistor and for the 1:2 feedback divider I think I should have something at 50ppm/K or even 25 ppm/K. Would be interesting to verify the tempco of at least the 68R resistor, which I think is most critical of the whole bunch.

Any ideas how to practically verify/measure tempco of such a resistor? The best I could think of would be to get a peltier cooler/heater with an oil bath cup on it and place the resistor in there, together with a temperature probe and just log the resistance. I hope my 34401 will be precise enough to capture any meaningful data. Rest of the setup I would really need to prepare, including the metal container for the oil bath (likely I will mill it out of aluminium).

Yes, a lot of work with a "simple" project, but I am really interested in learning something here, not to be finished in 15 minutes.

Maybe I should also start drawing some test PCB for it all.

The only area I need to investigate, is the PSSR the circuit will in reality have. Is it wise tu supply it directly from a rectified AC voltage, without any pre-filtering? Likely not, hmm...  Even though at the sim it loooked very good already by itself. 

Y.

[jaromir]

For those curious, each of these bricks were specified individually for a current of 15 to 100mA, together with voltage stamped on it. Mine is 8.19V at 60mA. The tempco is specified <10E-6   1/°C  (= 1 ppm/K?), dynamic resistance of < 15 ohm.

That sounds very much like KZZ82. I measured more than dozen of them, coming from different sources and none of them behaved according to the specs. Mostly being orders of magnitude off.
See https://www.eevblog.com/forum/metrology/kzz82-voltage-reference/msg4253485/#msg4253485 for sample of measured values.

[PCB.Wiz]

That sounds very much like KZZ82. I measured more than dozen of them, coming from different sources and none of them behaved according to the specs. Mostly being orders of magnitude off.
See https://www.eevblog.com/forum/metrology/kzz82-voltage-reference/msg4253485/#msg4253485 for sample of measured values.

Good point, a first step should be to first establish this part really (still?) does have a temperature null operating point

[PCB.Wiz]

Those tempco nulls I guess are very component dependent and will be wildly different when using different physical parts. Practical realization would mean I'd need trim pots at those places. Setting them would be another wild challenge I guess.

Yes, that would need adjustment with real parts, and with good temperature tracking.
The final result was academic, it is far better than your target spec.

Of course, if the zener-block is 'not great' anymore, a better precision current can give the illusion of a better reference.
In this case, you may want to create a deliberate uA/K temperature slope to fix the zener's best operating point.

For the 68R resistor, I think I have found a 68R1 100ppm/K 1% 1W resistor and for the 1:2 feedback divider I think I should have something at 50ppm/K or even 25 ppm/K. Would be interesting to verify the tempco of at least the 68R resistor, which I think is most critical of the whole bunch.

Note that is ppm/K, so you can help by using a larger resistor or even 4 of them, to reduce the deviation from the average assembly temperature.
The numbers they give are worst case, and conservative.
The tempco trim here can also fix a constant ppm/K, provided the temperatures can track.
Practical Temp tracking is likely to be a challenge here, which is why I changed to Power-package BJT sims.

Any ideas how to practically verify/measure tempco of such a resistor? The best I could think of would be to get a peltier cooler/heater with an oil bath cup on it and place the resistor in there, together with a temperature probe and just log the resistance. I hope my 34401 will be precise enough to capture any meaningful data. Rest of the setup I would really need to prepare, including the metal container for the oil bath (likely I will mill it out of aluminium).
Yes, a lot of work with a "simple" project, but I am really interested in learning something here, not to be finished in 15 minutes.

I would sanity check your Zener-block first, with currents 15-100mA to see if it really does have a null at 60mA (see #27)

[PCB.Wiz]

For those curious, each of these bricks were specified individually for a current of 15 to 100mA, together with voltage stamped on it. Mine is 8.19V at 60mA. The tempco is specified <10E-6   1/°C  (= 1 ppm/K?), dynamic resistance of < 15 ohm.

Here is a sanity-check of a Zener + 2 diodes, with some choices of diodes and Zeners. I pushed to the higher current end, given these specs.
You can get a parabola in all cases, with a correct choice of current.  Real-components tests would be next.
The quality of the parabola also varies a bit, 1ppm/K is possible within a narrower temperature range.
The parabola shape does vary with current, but the finite resistance of the whole setup means good current regulation certainly helps.

The final operating current point would be chosen for the whole circuit + tight thermal coupling assembly.

[EE_Audio]

Looks like you have some good options using transistors. In case you are interested, there's a good single op amp solution as well. See Attached.

[Yansi]

Sorry guys, I was not able to respond, overloaded with important work as usual and not time for toying with things.

For the 68R resistor, I think I have found a 68R1 100ppm/K 1% 1W resistor and for the 1:2 feedback divider I think I should have something at 50ppm/K or even 25 ppm/K. Would be interesting to verify the tempco of at least the 68R resistor, which I think is most critical of the whole bunch.

Note that is ppm/K, so you can help by using a larger resistor or even 4 of them, to reduce the deviation from the average assembly temperature.
The numbers they give are worst case, and conservative.
The tempco trim here can also fix a constant ppm/K, provided the temperatures can track.
Practical Temp tracking is likely to be a challenge here, which is why I changed to Power-package BJT sims.

Any ideas how to practically verify/measure tempco of such a resistor? The best I could think of would be to get a peltier cooler/heater with an oil bath cup on it and place the resistor in there, together with a temperature probe and just log the resistance. I hope my 34401 will be precise enough to capture any meaningful data. Rest of the setup I would really need to prepare, including the metal container for the oil bath (likely I will mill it out of aluminium).
Yes, a lot of work with a "simple" project, but I am really interested in learning something here, not to be finished in 15 minutes.

I would sanity check your Zener-block first, with currents 15-100mA to see if it really does have a null at 60mA (see #27)

Not sure if I understand how a combination of multiple resistors will help with the tempco? I can sure replace 68R with three parallel 205 ohm (I think I have 205 ohm, but alsi in 100ppm/K grade) or four parallel 270 ohm (also 100ppm/K). But shouldn't the tempco of the triplet or quad be still the same? Especially taking resistors from the same batch, I would expect even the tempco to quite well track between different pieces and the resulting tempco to be about the same.

Regarding resistor tempco verification, I have meanwhile manage to mill an aluminum "tub" that can be mounted on a peltier cooler/heater, resistor placed inside the tub and filled with mineral oil / demi-water / isopropyl or something... That should not be difficult to make the tempco verified.

However, verifying the zener block will be much more difficult, as it is much larger and complex thermal mass, that I do not think would survive dipping in water/oil whatever. Also, one would need to always wait quite long time for the temperatures inside the block to equalize. Best case, I could stick it in clima chamber at work, but would need someone that would allow me to add this to their setup if they will use acceptable temperature range/profile that I could use for the measurement. Also, such measurement would require repeated temperature sweeps for different supply currents to the block, so not really a quick measurement. (or vice versa, sweep the current at different temperatures? What do you suggest would be better?)

Meanwhile I was able to acquire another bunch of those zener blocks. Some are definitely bad (voltage ver much outside of the marking on the block) and some seem to sit precisely as indicated. But sure would be nice to verify the stability/tempco.

This weekend (i seriously hope) will be able to prepare some test PCBs for the transistor based amplifier/driver and start putting the resistor tempco measurements in practice.

Thank you for keeping with me 

[PCB.Wiz]

Sorry guys, I was not able to respond, overloaded with important work as usual and not time for toying with things.

For the 68R resistor, I think I have found a 68R1 100ppm/K 1% 1W resistor and for the 1:2 feedback divider I think I should have something at 50ppm/K or even 25 ppm/K. Would be interesting to verify the tempco of at least the 68R resistor, which I think is most critical of the whole bunch.

Note that is ppm/K, so you can help by using a larger resistor or even 4 of them, to reduce the deviation from the average assembly temperature.
The numbers they give are worst case, and conservative.
The tempco trim here can also fix a constant ppm/K, provided the temperatures can track.
Practical Temp tracking is likely to be a challenge here, which is why I changed to Power-package BJT sims.

Not sure if I understand how a combination of multiple resistors will help with the tempco? I can sure replace 68R with three parallel 205 ohm (I think I have 205 ohm, but alsi in 100ppm/K grade) or four parallel 270 ohm (also 100ppm/K). But shouldn't the tempco of the triplet or quad be still the same?

I was meaning the self heating effect - if you can reduce the temperature change, you reduce that added variation, and remove another time constant to worry about.
So spreading the power over 4 resistors drops the self-heating to 25%.

At the quite high 60mA levels mentioned here, there is significant self heating.

You do not actually need to dissipate ~250mW in a resistor here - I think the 68 ohms was chosen for simple connect to 12V ?
Once you change to active current regulation, the transistors take the heat.

[Yansi]

That why I have the 68R1 resistor in the size of a 4-5W one.

You do not actually need to dissipate ~250mW in a resistor here - I think the 68 ohms was chosen for simple connect to 12V ?
Once you change to active current regulation, the transistors take the heat.

What?  As the above designed transistor amplifier/driver multiplies voltage by 1.5x, from 8.19V nominal from the ref block, I get 12.285V.  That minus block voltage will be on the supply resistor: 12.285 - 8.19 = 4.095V. That divided by required 60mA, I get 68.25 ohm. (power dissipation over it is then about 1/4 W). This is the R10 in your latest schematic.

So what "simple connection to 12V" are we talking about? Sorry I got somehow lost...

[PCB.Wiz]

That why I have the 68R1 resistor in the size of a 4-5W one.

You do not actually need to dissipate ~250mW in a resistor here - I think the 68 ohms was chosen for simple connect to 12V ?
Once you change to active current regulation, the transistors take the heat.

What?  As the above designed transistor amplifier/driver multiplies voltage by 1.5x, from 8.19V nominal from the ref block, I get 12.285V.  That minus block voltage will be on the supply resistor: 12.285 - 8.19 = 4.095V. That divided by required 60mA, I get 68.25 ohm. (power dissipation over it is then about 1/4 W). This is the R10 in your latest schematic.

So what "simple connection to 12V" are we talking about? Sorry I got somehow lost...

It's a trade off between simplicity and performance.
Usually, if you want a gain of 1.5, you would use 3 equal resistors to give you that, independent of any current drive. The tempco tracking sets the errors in this case.
Using a single resistor to give both voltage gain, and current drive, is simple, but compromised.
It might be 'good enough' - time to try it?