Digital adjustment of voltage reference output

[cellularmitosis]

Modern DMM's all support digital calibration, but I've noticed most of the voltage reference designs I've seen on the forum don't include any method of digital adjustment.  I thought perhaps I'd play with that idea!

Here's what I've come up with so far.  Discussion welcome!

This circuit is a typical 7V to 10V step up for e.g. an LTZ1000 reference.  A DAC is added via a large resistor to allow digital adjustment of the 7V to 10V amplifier.

I haven't yet looked at how to analyze the impact of tempco on both the resistor and DAC, and find the correct balance between those two components.

[cellularmitosis]

Here's an attempt at figuring out the impact of resistor tempco.

When the 7.5meg resistor drifts by 1000ppm, the output shifts by 1.16uV.

[cellularmitosis]

Hmm, apparently a 7.5meg resistor at 23C creates 11uV RMS of noise over 1000Hz of bandwidth.  That's already more than the LTZ1000 itself.

http://www.radio-electronics.com/info/rf-technology-design/noise/thermal-calculations.php

[Andreas]

But it is in parallel with a 3K and a 7K resistor (and thus divided) -> much lower noise

Before Lars mailed his SVR-T (Version 2) schematics I planned a 3rd order temperature correction
for a AD587 with a termistor and a PIC.
A 12 Bit DAC (LTC1257) with reference from the 10V output should be sufficient to trim for a selected +/- 1 ppm/K reference.

with best regards

Andreas

[chuckb]

That's a good technique. I have used something similar before to calibrate the gain and offset of a pressure sensor with +-30% initial tolerance to 0.001% resolution. I used an AD7846.

With a LTC1657 (parallel) you could control the voltage with HEX switches and skip the microprocessor.

Of course with the LTC1655 (serial) and a micro you could correct residual drift with temperature and time...

The voltage noise of the 7.5 meg resistor can be thought of as a noisy current source with 1.5 pa of noise (11uV noise / 7.5 Meg). That current noise when injected into a 3k node impedance would develop a 4.4 nV noise. This is 10% of the LTZ1000 noise.

[hwj-d]

Special thanks for opening this thread. 
(hereby subscribing for 'new replies')

[Andreas]

Hello,

similar idea for a 7 to 10V transfer:

the 50K and 20K resistors are made from a matched ratio-divider LT5400-3 with 0.2 ppm/K typ.
(2*10k in series and 2*100k in parallel)

The other resistor stages are arranged in a way that the LTC1257 output has a symmetrical +/- change.
And no resistor above 1 Meg. So 0.1% metal film resistors with <= 15ppm/K can be used.
The LTC1257 is then fed from the 10V output voltage. with 3rd order fine trimming for the T.C. by a PIC from a NTC.

with best regards

Andreas

[TiN]

I would first start replying the question of the trim goal. Is it to battle for LTZ output voltage variation? Or is it to make "perfect" 10V output? Or is it to twist the output tempco?
If it's yes for first question, that you would much wider range than few ppms. If it's to make perfect 10V, then one should worry more about long-term stability and not the tempco.

There is a reason why most of the commercial designs do not have digital trim, is to reduce possibility for added instability and noise in reference voltage. Also precise trim often not required because reference recipient (DAC, ADC, PA output stage) can be also trimmed to desired output result, rather than reference itself. Having DAC at the reference side however have good chance for ground loops, digital noise injection and gain instability. Long-term stability is not typically specified for these components. Sure, one can design in digital isolation (including the low-noise PS), careful current path control, etc, but at best you just reach stability of the output resistor network. So it all sounds rather expensive feat without visible gains.

If doing trim, maybe better option to make 7V-10V stage fixed and stable as possible, but trim actual LTZ output in first place. You can do so by injecting offset current at bias resistors (70K, etc) or minor adjustments to zener current (120R, etc). So you trim LTZ input for stable output stage match, rather than trying the opposite. It sound's easier to me, as you have more control points over output voltage and tempco trim on LTZ circuit itself. I recall our LTZ veteran Dr.Frank did something similar on his original reference design, bringing LTZ output to 7.00000V and then boosting that trimmed level to desired 10V.

There are also alternative ways, for which DAC trim wouldn't be much help. E.g. multiply LTZ-7V by LTC1043 or similar flying cap block, and then divide output to the ratio 10Vish back. That would be also more noisy, but still acceptable if injected noise much under own LTZ output noise. 

[branadic]

Hello,

similar idea for a 7 to 10V transfer:

the 50K and 20K resistors are made from a matched ratio-divider LT5400-3 with 0.2 ppm/K typ.
(2*10k in series and 2*100k in parallel)

The other resistor stages are arranged in a way that the LTC1257 output has a symmetrical +/- change.
And no resistor above 1 Meg. So 0.1% metal film resistors with <= 15ppm/K can be used.
The LTC1257 is then fed from the 10V output voltage. with 3rd order fine trimming for the T.C. by a PIC from a NTC.

with best regards

Andreas

I do like the idea of this gain stage. Have you tried building it on a breadboard to verify performance? Have you simulated the TC influence of the resistors? Would you mind sharing the .asc file? I don't get the same results as you do when reconstructing the schematic in LTspice. Lot's of questions

-branadic-

[Andreas]

Hello branadic,

all still on my (never ending) todo-list. I fear that the LT5400 is not suitable for a perf-board.

Be aware that the simulation is not from the same LTZ than the cirquit diagram.
ASCII-files attached.
The T.C. influence of the resistors (and some ageing of the LTZ)
is that what shall be corrected with the DAC.
(I expect a decreasing LTZ-voltage over time).

with best regards

Andreas

[branadic]

Ah, schematic is LTZ1 and diagram is LTZ6... I was wondering exacly about that.
How would you modify this circuit for LM399?

-branadic-

[Andreas]

Hello branadic,

in principle the cirquit is always the same.
Only the values of the resistors (mainly R4-R6 and eventually the T.C. of R4) have to be adapted.
With 1 Meg for R4 the T.C. of R4 can be about factor 70 of  the LT5400 T.C. (0.2ppm).

with best regards

Andreas

[d-smes]

I fear that the LT5400 is not suitable for a perf-board.

I used the Vishay SMNZ network with two 1K and two 10K resistors connected series / parallel to create 2K over 5K divider like what you describe.  Unfortunately, DigiKey no longer sells these in single-piece bulk form: https://www.digikey.com/product-detail/en/vishay-foil-resistors-division-of-vishay-precision-group/Y1747V0205QT9W/Y1747V0205QT9W-ND/4232601   Maybe some other distributor still carries them...

If doing trim, maybe better option to make 7V-10V stage fixed and stable as possible, but trim actual LTZ output in first place. You can do so by injecting offset current at bias resistors (70K, etc) or minor adjustments to zener current (120R, etc). So you trim LTZ input for stable output stage match, rather than trying the opposite. It sound's easier to me, as you have more control points over output voltage and tempco trim on LTZ circuit itself. I recall our LTZ veteran Dr.Frank did something similar on his original reference design, bringing LTZ output to 7.00000V and then boosting that trimmed level to desired 10V.

The 2.5:1 ratio seems to be common or easily configurable from a 4-resistor network.  If this is ratio used in a fixed 7V - 10V gain stage, then trimming a LTZ1000 to 7.14286V might be the way to go because 7.14xxx seems to be closer to the "typical" LTZ voltage.  The question is-  How far can the LTZ be trimmed from it's native voltage before temperature coefficient is compromised?

[Kleinstein]

There is not much room for trimming the LTZ1000 voltage. Maybe some 20 mV up and down with a higher/lower transistor current, with little effect on the sensitivity to the resistors. Changing the Temerature would work too, but the temperature is usually set by other requirements. The other option is a good quality resistor in series to the zener. Up to a certain point (e.g. about 60 mV more), there is even a chance that the TC could improve and some of the other resistors get less important. However that extra resistor and the 120 Ohms one to set the current get more important.

Even if one starts with a resister network for a good ration like 1:2.5 , one can still do the fine trim in that stage.