Very low ref voltage

[Marco Trapanese]

Hello,

to test and calibrate several acquisition boards I'd need a very low
reference voltages.

I'd prefer a good thermal and short-term stability rather than a very
accurate output voltage.

My goal is to build a circuit with these outputs:

- 100 uV
- 1 mV
- 10 mV
- 100 mV

also a 10 uV output will be very appreciated.
The output current could be quite small, less than 1 mA.

Obviously, I'm looking for the lowest noise and good short-term
stability - I want the output doesn't change more than 1/1000 in a
couple of hours.

Do you think it's possible to make such a circuit? Any idea how to get
those low voltages?

Thanks
Marco

 

[[email protected]]

Hello,

to test and calibrate several acquisition boards I'd need a very low
reference voltages.

I'd prefer a good thermal and short-term stability rather than a very
accurate output voltage.

My goal is to build a circuit with these outputs:

- 100 uV
- 1 mV
- 10 mV
- 100 mV

also a 10 uV output will be very appreciated.
The output current could be quite small, less than 1 mA.

Obviously, I'm looking for the lowest noise and good short-term
stability - I want the output doesn't change more than 1/1000 in a
couple of hours.

Do you think it's possible to make such a circuit? Any idea how to get
those low voltages?

Thanks
Marco

What kind of circuit topology are you thinking about ?

At least in any unbalanced circuits, the ground noise could swamp the
reference voltage.

 

[Spehro Pefhany]

Hello,

to test and calibrate several acquisition boards I'd need a very low
reference voltages.

I'd prefer a good thermal and short-term stability rather than a very
accurate output voltage.

My goal is to build a circuit with these outputs:

- 100 uV
- 1 mV
- 10 mV
- 100 mV

also a 10 uV output will be very appreciated.
The output current could be quite small, less than 1 mA.

1mA is actually quite a bit of current at 10uV, that implies a load
resistance of 10 milliohm.

Most DC-accurate data acquisition boards have relatively high
impedance inputs- gigohms, megohms.

Obviously, I'm looking for the lowest noise and good short-term
stability - I want the output doesn't change more than 1/1000 in a
couple of hours.

1/1000 of 10uV is 10nV. That will be extremely difficult because of
thermal EMFs. 0.1% + (say) 5uV might not be very hard.

Do you think it's possible to make such a circuit? Any idea how to get
those low voltages?

Thanks
Marco

Can't you just use a voltage divider off of an ordinary reference?

 

[Marco Trapanese]

Il 21/06/2012 15:35, Spehro Pefhany ha scritto:

1mA is actually quite a bit of current at 10uV, that implies a load
resistance of 10 milliohm.

1/1000 of 10uV is 10nV. That will be extremely difficult because of
thermal EMFs. 0.1% + (say) 5uV might not be very hard.

I'm sorry, I mean at F.S. And of course the requested current will be
negligible in most applications.

Can't you just use a voltage divider off of an ordinary reference?

In this case what kind of resistors should I use?

Thanks
Marco

 

[Marco Trapanese]

Il 21/06/2012 16:48, George Herold ha scritto:

Ahh 1% metal film? Or the 0.1%ers if you want to be fancy. Getting
low values (below 50 ohms) is expensive though.

The input stages are op-amps and INA as already pointed out by the other
friend. Sensors have an impedance about 350 ohm typically (e.g. strain
gauges).

If I use a 1.25V as standard ref, to have 100 uV I may use 1.5M over 120
ohm. Not bad.

Do you really need a
milli amp?

Well, I think I over-estimated that

Marco

 

[Spehro Pefhany]

In this case what kind of resistors should I use?

Thanks
Marco

For 0.1%-ish lab stability, just about any precision metal film
resistor will do.

Keep the lower end resistor to something like <10 ohms and you should
be okay.

If you want to get fancy you can use low-value resistors with 4
terminations, but that gets expensive.

Here's a 1206 0.1% 10 ohm resistor +/-25ppm/°C for 44 cents.

http://www.digikey.ca/product-detail/en/CRT1206-BY-10R0ELF

 

[Marco Trapanese]

Il 23/06/2012 22:54, whit3rd ha scritto:

Best (for accuracy) is to buy a divider or trimmed divider
so the two resistors are the same material and batch.
Look at DigiKey's 'RM32A' series voltage dividers, for instance.

Thanks for the hint.
Marco

 

[Uwe Hercksen]

Best (for accuracy) is to buy a divider or trimmed divider
so the two resistors are the same material and batch.

Hello,

if the two resistors are on the same isolating base material, the
thermal match is even better.

Bye

 

[[email protected]]

Best (for accuracy) is to buy a divider or trimmed divider
so the two resistors are the same material and batch.
Look at DigiKey's 'RM32A' series voltage dividers, for instance.

One way to get a stable and accurate voltage divider is to get a large
number of XX kohm resistors, hopefully from the same batch.

To get a 10:1 voltage divider, put nine of these resistors in series
and one to ground.

Likewise for current measurement, put multiple equal values in
parallel and a single one through the actual current sensor.

This will compensate for the temperature coefficient as well as
inaccurate XX values.

 

[[email protected]]

That's the way we do it in the integrated circuit world. Compensates
out end-effects and TC.

A better way, in CMOS is to use gate leakage (tunneling current). The gate
oxide thickness is held to very tight tolerances as are the gates themselves.
It turns out to be a very good high value divider. Ratios other than 1/2 can
be had by varying the gate size but it's not as good as a 1:1 divider. Of
course you can't load it much, either. ;-)

 

[Jamie]

A better way, in CMOS is to use gate leakage (tunneling current). The gate
oxide thickness is held to very tight tolerances as are the gates themselves.
It turns out to be a very good high value divider. Ratios other than 1/2 can
be had by varying the gate size but it's not as good as a 1:1 divider. Of
course you can't load it much, either. ;-)

Have used Mosfets for voltage references due to their T coefficiency
being +, and - on the Vg(th).

For the better part, it seems to work. But, they don't seem to be
very fast and you have to keep them with in a window. At least the Mfets
I used were like that and they do operate in a cold environment.

I didn't use an RF mosfet and there was a reason for that, I just
can't remember why?

Jamie

 

[SoothSayer]

That's the way we do it in the integrated circuit world. Compensates
out end-effects and TC.

I was able to make cheap 8-bit DAC's 50 years ago by applying that
method.


...Jim Thompson

Creating too low a voltage reference introduces problems one must keep
an eye out for.

Can't be considered 'stable' or even 'referenceable' once it gets below
a certain point.

Pretty much why the industry has them where they are. They have
already found the most stable, best value to provide.

 

[SoothSayer]

We had a similar problem and used the Nat Semi LM4140 as the ref, and the
AD706 op-amp as the buffer. We have to be stable to better than 5V +/- 100
uV for hours, so the thermal drifts and 1/f noise didn't contaminate
14-bit measurements.

Jon

Abating the 1/f drops the noise floor and get you a lot of benefit.

It is important to excise as much as possible.

 

[Marco Trapanese]

Il 26/06/2012 05:16, SoothSayer ha scritto:

Creating too low a voltage reference introduces problems one must keep
an eye out for.

Can't be considered 'stable' or even 'referenceable' once it gets below
a certain point.

What is this "certain point", approximately?

Marco

 

[Marco Trapanese]

Il 26/06/2012 15:16, George Herold ha scritto:

"> What is this "certain point", approximately?"

1 volt? (WAG)

So you're saying I can't get my low refs...
What about the ideas proposed by other friends?

Marco

 

[Spehro Pefhany]

"> What is this "certain point", approximately?"

1 volt? (WAG)

George H.

A 1-volt reference is pretty useless for calibrating thermocouple
instrumentation.

The most accurate and stable voltage reference that I know of is only
about 500-700uV, which is inconvenient in some ways, but worth working
with in some cases.

Hey, I can now generate ratios of AC voltages on my bench with a
digitally selected resolution of 0.01ppm. Happy happy, joy joy.

 

[Spehro Pefhany]

Do tell. I guess I was just thinking about the old Weston cells. I
really don't know what is used for voltage references today.  (I
tend just to believe my bench DMM.)

George H.

The best reference that I know of that is commercially available is
thousands of Josephson junctions in series that are pumped (?) by
~70GHz microwaves. The guts needs to be cold (4K) to work.

 

[tm]

The best reference that I know of that is commercially available is
thousands of Josephson junctions in series that are pumped (?) by
~70GHz microwaves. The guts needs to be cold (4K) to work.

But you can't find one on ebay.

 

[Marco Trapanese]

Il 26/06/2012 16:51, George Herold ha scritto:

Hi Marco, Sorry I should have kept my mouth shut.
I see no problem with dividing down the 1 volt reference.

Don't worry, thank you anyway.

Marco

 

[Spehro Pefhany]

OK, I did that as a lab back in grad school. PITA to get a sample to
work. I remember staying up all night trying sample after sample.
(samples were just pieces of niobium wire twisted together.) I fianly
made a probe that would hold 4 samples at a time and got some OK
data.

From what I've read, it sounds like the JJs nowadays are a lot better.

Say are there any high Tc super Josephsom junction? I know there are
HTc squids.

George H.

Apparently (http://arxiv.org/abs/0708.0904) but all the stuff I'm
doing is LTc. There's supply issues with He these days too.