0
\$\begingroup\$

Highly accurate instruments with a calibration are quite expensive, and anything electronic will tend to change over time. The effects of heat and oxidation, and perhaps changes in the material due to current directly. Perhaps there are other reasons circuits change over time.

So I was wondering about some kind of reference that could be built into equipment so it could recalibrate itself. Suppose you wanted to have a fixed power supply that would stay stable at a precise voltage or at least be capable of re-calibration without sending it for expensive recalibration at the factory.

Thinking about what kind of components could be stable, I think a wire-wound resistor with thick connectors and soldered connections. Zener diodes would seem less stable.

So can anyone suggest a way to build a circuit that would provide an extremely stable voltage reference? Given such a voltage reference, I think it would be possible to build an extremely stable power supply.

This question is mostly theoretical out of interest. I'm not asking for a practical design at this point though I wouldn't say no!

\$\endgroup\$
5
  • \$\begingroup\$ The issue then would be, how do you keep the thing you are calibrating from calibrated over time? \$\endgroup\$ – James T May 17 '16 at 13:24
  • \$\begingroup\$ This question is probably too broad and/or opinion based. Every material can possibly change, so the only resort you have is like a hydrogen maser connected to a josephson junction \$\endgroup\$ – PlasmaHH May 17 '16 at 13:25
  • \$\begingroup\$ The question is, can anyone suggest a voltage reference that is extremely stable? I think I could create a resistor that would be accurate and extremely stable. Trying to envision a circuit that would stay stable over time. \$\endgroup\$ – Dov May 17 '16 at 13:32
  • 1
    \$\begingroup\$ Think about why this isn't done already. Calibration is not a manufacturer scam, it's because accurate references are expensive. It's like asking for a caesium reference clock in every home. \$\endgroup\$ – pjc50 May 17 '16 at 13:53
  • 2
    \$\begingroup\$ (The usual implementation is a silicon 'bandgap' reference) \$\endgroup\$ – pjc50 May 17 '16 at 13:58
2
\$\begingroup\$

How far you have to go depends on how accurate you need. For many applications, a zener diode is sufficient. The next level might be a zener at a specific temperature, held there by a heater. There are also "reference" cells, which are really batteries that as little as possible current is drawn from. These need to be at the right temperature too for the best accuracy.

If you keep going, you have to worry about the temperature sensor drift used to keep whatever you are using as a reference at the right temperature. Then you have to worry about how exactly you are comparing the reference voltage to something else. Every means of comparison will have some error, which will drift over time, be dependent on temperature, etc. Even more extreme, you have to think about thermocouple offsets due to different parts of the circuit being of different temperatures and made of different materials.

Without a number of how accurate "good enough" is, it's impossible to say how far you have to go, how much you have to spend to get there, and whether it's even possible with current technology.

Even if you have a perfect power supply, what about the voltage drop in the wires between it and the load? You have to look at the whole picture and think carefully about what problem you are really trying to solve.

\$\endgroup\$
1
\$\begingroup\$

Probably the most stable voltage reference commonly available is a buried zener diode. It requires a lot of voltage (more than 7V) and significant power (especially if ovenized). There are more precise references such as the Josephson junction standard, but that's out of the budget of even most calibration labs, as well as being somewhat inconvenient, what with the cryostat and liquid helium.

There are other technologies with various advantages and disadvantages- you can read the datasheets- if it's a good reference they will give drift per 1000 hours or similar, typically the drift rate decreases over time as in the classic 1/f noise spectrum. Also the stability over time will be at a certain temperature and operating conditions. If you ovenize the reference, for example, if you allow the oven to go out of control there will be more drift in the reference, and there will typically be small changes from thermal hysteresis that are not time related.

Stable resistors are all well and good (as well as necessary) but they are required in addition to the voltage reference, not instead of, if it's a power supply. A stable reference resistor might be useful for an ohmmeter.

I question your premise- if you have an extremely stable reference built-in then why would you re-calibrate using the reference? Would it not be better use the reference directly to determine the output voltage rather than having a worse reference that you would calibrate to the good one... unless there are power constraints etc.

There are many circuits that continuously 'recalibrate' such as ratiometric measurements to a reference resistor or voltage, or chopper amplifiers. Consider the auto-zero dual slope ADC- it removes errors due to long term drift in the integrator capacitor, integrator resistor, amplifier offset voltage and clock frequency- each of those parameters only needs to be stable enough not to change much during each measurement cycle (milliseconds). As a huge advantage, this also removes errors due to initial tolerances and due to temperature change (provided the temperature does not change much during the measurement cycle).

\$\endgroup\$
0
\$\begingroup\$

There are stable references such as MAX6194AESA+ that drift by as little as 50ppm per 1000 hours of operation.

But no matter how stable your reference, you can always beat that accuracy by putting in an adjustment to periodically calibrate the supply from some external NIST traceable reference.

\$\endgroup\$
1
  • \$\begingroup\$ But calibrate from what? \$\endgroup\$ – Dov May 17 '16 at 15:45

Not the answer you're looking for? Browse other questions tagged or ask your own question.