MAX6226 voltage reference

[Kleinstein]

The readings are unstable and there is thus something wrong. Oscillation of the reference or maybe at the supply is one possibilty, though not the only one.

Some additional RC loading (possibly just a 10 µF electrolytic cap) could be a try to fight osicillation.

[dietert1]

Yes, damping at the output should help.
Something else one can do without a scope is insert a resistor (maybe 10 or 100 Ohm) into the reference supply and measure current consumption. I would expect much higher consumption when the device is oscillationg, let's say more than 10 mA instead of the 380 to 550 uA spec.

Regards, Dieter

[iMo]

Another point is the thermal stability there - you have to put the pcb into a box, thermally isolated from the ambient air flow. The large part of the p-p voltage may come from the open air situation. The 30uVpp looks pretty high, indeed, I would expect something like 5-10x smaller value with no air drafts.
Would be interesting to observe the chip current Icc - whether it is higher than those 380uA, you may try to load the output with different capacitancies and the Icc should be changing then when oscillating, imho.

[Svgeesus]

I added a 10R (Dale CMF60) and 100µF (Nichicon electrolytic) filter between the power supply and the voltage reference board (the second one, with no op-amp fitted).

Initial results indicate much better stability.

I had measured the exact resistor value (4 wire measurement) to calculate the current draw. But since the oscillation is removed by the resistor, happily that is no longer needed.

I will leave this running for a while, then try the first board (with op-amp) with a pair of RC filters on the positive and negative rails (and soldered test wires). If that goes well I will put it in an insulated box and leave it running for a while.

Thanks to those who suggested improvements and experiments; much appreciated!

[thermistor-guy]

Another point is the thermal stability there - you have to put the pcb into a box, thermally isolated from the ambient air flow. ...

For maximum benefit, also consider suppressing convection currents inside the box, if it's easy to do:

https://www.eevblog.com/forum/metrology/thermal-fluctuations-on-pcb/msg4705313/#msg4705313

[Svgeesus]

For maximum benefit, also consider suppressing convection currents inside the box, if it's easy to do:

https://www.eevblog.com/forum/metrology/thermal-fluctuations-on-pcb/msg4705313/#msg4705313

Thanks for the advice. My current setup sandwiches the (small) PCB in a folded hand towel - four thicknesses below, four above. So air convection should be minimal.

[Svgeesus]

My main problem is the AZ (chopper-stabilized) OPA2186 op-amp. On one board, without it, and with 10R 100µF RC filtering between the power supply and the board, everything is nice and stable. On my other board, with the same RC filtering and the op-amp fitted, it can be stable for some hours then jump into unstable mode for some more hours. A 22R output resistor after the op-amp and before the DMM did not improve things at all.

[dietert1]

The OPAx186 is a very low power device and from its datasheet Figs. 6.25 and 6.26 you can see that a 50 Ohm isolation resistor doesn't help with a capacitive load. You could try an OPA2189 instead.
Another mod to try would be a low leakage capacitor parallel to R2, e.g. 1 or 10 nF.

Regards, Dieter

[iMo]

Btw showing us your actual schematics would help..

[Svgeesus]

Here is the schematic.

[GigaJoe]

Just wondering, why do it need C4 , C5 ?

[RandallMcRee]

Just wondering, why do it need C4 , C5 ?

It is specifically mentioned in the data sheet.  As with most refs!

[Svgeesus]

Just wondering, why do it need C4 , C5 ?

From the MAX6226 data sheet:

Output Bypassing
The MAX6226 requires an output capacitor between
0.1μF and 10μF. Locate the output capacitor as close
to OUTF as possible. For applications driving switching
capacitive loads or rapidly changing load currents, it is
advantageous to use a 10μF capacitor in parallel with a
0.1μF capacitor. Larger capacitor values reduce transients
on the reference output.

However, I do wonder if that capacitive load on the input to the voltage follower op-amp upsets it.

[iMo]

That has been a topic here recently. I put an 1k resistor in front of my OPA189 therefore (as recommended with the choppers)..

..However, I do wonder if that capacitive load on the input to the voltage follower op-amp upsets it.

[exe]

That has been a topic here recently. I put an 1k resistor in front of my OPA189 therefore (as recommended with the choppers)..

Why is this recommended? I think it increases noise as choppers have current noise due to switching inputs. I'd suggest an RC-filter instead of just a resistor if bandwidth is not a concern.

[iMo]

https://www.eevblog.com/forum/metrology/zero-drift-amplifier-input-bias-current/

FYI - the voltage peaks at the inputs of the OPA188 in the TI's presentation (in the above link) were 35mVpp and 13ns width.

[Svgeesus]

While a series 10k resistor at the op-amp input makes sense in the typical 7V-to-10V reference buffer, I don't see how to do that with a Vref that has force and sense connections. A resistor in the force line (which is supposed to be broad and low resistance) seems odd and would just drop the voltage a little, although the feedback via the sense line would presumably correct that. But would you also need a resistor in the sense line as well.

I'm thinking that the next iteration of this board will use a low-drift, e-trim opamp such as OPA2192. The input bias is way higher (4.8nA instead of 5pA) but comparable to the 4.5nA of the AD8676B that I saw recommended by Egan. The OPA2192 is fine driving modest capacitative load and doesn't have the charge injection issue which this MAX Vref seems sensitive to. I'm also considering using the in-the-loop isolation resistor configuration (two resistors and a capacitor) to avoid any ringing or instability there.

Egan, M. The 20-Bit DAC Is the Easiest Part of a 1-ppm-Accurate Precision Voltage Source

[ivo]

A resistor in the force line (which is supposed to be broad and low resistance) seems odd and would just drop the voltage a little, although the feedback via the sense line would presumably correct that. But would you also need a resistor in the sense line as well.

This is true while the REF is trying to drive a tricky/heavy load, but falls apart if that's not the case. The problem is the chopper's charge switching will enter in tiny amounts on the inputs, and you don't want the REF's own error amp reacting resonantly with that to create instability. In many cases excessive direct capacitance going into the chopper's inputs can also mess with its chopping.

When the only draw for the REF is less than nA of current from an opamp, loading down the force line with resistance shouldn't be much of a problem because it's not like it has to provide current anyway, and the actual error is correct through the sense line. The poster child for the opposite case is of course an ADC which is swinging around Vref trying to charge capacitors and switching currents.

The primary critical thing for DC accuracy is the sense line can accurately sense the the voltage where you want the REF's output. So I would put a larger resistance on the force and a much smaller resistance on the sense.

Compromise of the OPA192 is much larger voltage noise (186: ~600nV, 192: ~1uV, 189: 100nV, all peak-peak) although there's trade-offs in all of them, depends which thing you want don't mind sacrificing

I've played with a 186 as well (was most attracted to 1uV offset and <10pA bias) and yeah it doesn't like capacitance at all, I found on its inputs or outputs. Just connecting it to a DVM I found it needed more loading to stabilise (practical workaround), I'm presuming cables and only 10M made it wobble.