I want to put together (1) a cheap home breadboard reference current supply to calibrate and test 3 1/2 digit DMMs and (2) a low noise power supply to replace batteries in battery-powered audio devices. This is mainly for me to use at home, so I do not need to worry about working over a wide range of temperatures or startup time, but I do need to worry about overheating.

I have an AD584LH precision voltage source that puts out 7.5 V with 0.05% accuracy, but only supplies 10 mA, and a 100 Ω 0.01% precision resistor (2ppm/ºC, 600mW). I also have a couple of ordinary 10 kΩ potentiometers and a 3-24 V variable switching power supply.

For (1) I figure if I can boost the current output to 100 mA with less than a 3 mV voltage drop from the AD584 output, I can safely put the 7.5 V through the resistor and get a precise enough 75 mA current output for calibrating the DMMs, although I would really like 500 mA to get closer to full range (600 mA) without pushing the fuse limit. And for (2) if I can boost the output to something like 2-3 A then I can replace any batteries I need.

UPDATE #1 on (1):

Thanks to Tim Williams, I'm thinking the following will suffice for calibrating my ammeter. The AD584LH also supplies a precision 5.0 V reference. I have a bunch of NiMH AA batteries to use as V1_batt and a 10 Ω trim pot to use as R1_trim. I also have a second DMM which, while not great, should be good enough to use as AM_null, and which I think is more sensitive and accurate in trying to measure zero current than trying to measure zero voltage. I can buy a precision 10 Ω 4 W resistor. Then I adjust the trim pot until AM_null reads zero, and I have a 500 mA reference current. The result should be accurate to within the tolerances of the voltage reference, resistor references, and ammeter sensitivity, all of which should be good enough for my purposes of calibrating a 3 1/2 digit DMM. The fact that it requires manual, rather than automatic, trim, and is not directly traceable to NIST standards, is in keeping with my initial statement that I'm building a home breadboard device, not a commercial lab standard.


simulate this circuit – Schematic created using CircuitLab

UPDATE #1 on (2)

I expected (based on an application note in the AD584 datasheet) that the solution to problem (1) would be some sort of linear amplifier. As it turns out, the solution to (1) in update #1 above does not amplify the output of the voltage reference and does require a low noise power supply, so it is not a basis for a solution to (2): a high power, low noise power supply to replace batteries.

However, imagine if someone did present a power amplification solution to problem (1). Apparently, people with pre-conceived notions of how to build a power supply could not fathom my idea, which to me remains pretty straightforward in theory, whether or not practical: If there were a linear amplifier design that boosted the current output of the voltage reference while maintaining the voltage via some form of feedback, then as long as the responsiveness of the feedback loop was fast and accurate enough, it would be sufficient to filter out audio frequency (i.e. < 20 kHz) noise from a noisy power source like an SMPS, and allow me to use it to replace batteries in battery-powered audio devices, which, because they expect clean power from batteries, lack any such input voltage filtering. Expecting responsiveness in the 100 kHz range did not and still does not seem unreasonable to me.

Maybe such a setup would not be ideal for mass production, but since I would have already had all or almost all of the components, it might make sense for me to use it rather than build something entirely different to achieve the same ends. I thought people here might be interested in thinking creatively about how to build an unconventional circuit, or educate me on why such a circuit would be impractical if not impossible.


Side note on batteries: The data sheet gives a schematic for varying the voltage output of the AD584, I just need a couple of potentiometers, and since in this application it is replacing batteries, precise voltage is not required, so I am covered on voltage. I have tried various commercial adjustable (switching) power supplies but they all have way too much noise for a battery-powered audio device that expects very clean power. The AD584 provides very clean power, just not enough of it.

The AD584 datasheet has a schematic for boosting the power output of the voltage reference, but it uses a Darlington pair that is obsolete, and even if it were available, would dissipate 18 watts (converting 15 V to 6 V at 2 A) which is way more heat than I want to deal with.

I'm thinking there should be a way to do something like the Darlington power amplifier, where the reference voltage regulates the output voltage, including filtering audio-frequency noise, but maybe using dual power supplies so the amplifier does not have to get so hot. Maybe I can power the AD584 with a battery and use the switching power supply, dialed to an appropriate level, to provide the necessary current.

Is this possible? Reasonable to do without spending more than $40 on additional parts, or requiring a massive heat sink?

  • 1
    \$\begingroup\$ Please add actual numbers: what output impedance do you demand at which frequency range. Formulating this spec might lead you to acknowledge that you don't need both low output impedance and high accuracy simultaneously for your applications. \$\endgroup\$
    – tobalt
    Jan 3 at 5:04
  • \$\begingroup\$ @tobalt I don't fully understand your question. I have 2 applications, both seeking DC output, one to drive a reference current into a DMM reading amps, given a reference voltage and reference resistor, one to send say 3 A with no appreciable noise under 20 kHz with low enough power dissipation from the amp that I don't need a heat sink. I don't need low output impedance and high accuracy simultaneously and didn't ask for it. I asked for 75 mA at high accuracy and 2-3 A at low noise in the audio range (< 20kHz). I'm open to that being provided by one solution or two. \$\endgroup\$
    – Sara
    Jan 4 at 1:50
  • \$\begingroup\$ a) as noted the bare reference is sufficient to calibrate a voltmeter. b) if you have a called voltmeter, you don't need an accurate current anymore to calib your ammeter. you pass any current through your accurate resistor and measure an accurate voltage over the resistor with the called voltmeter c) high current, low noise, for audio you achieve by SMPS + copious filtering and linear post regulation. If this doesn't work, the issue is elsewhere, e.g. using inadequate single-ended audio, but this latter issue is best treated in a separate question as it has no relation to the reference. \$\endgroup\$
    – tobalt
    Jan 4 at 4:40

2 Answers 2


The underlying problem here, I think, is unfortunately hard to capture in an SE answer, so I'll instead give a rundown of what's missing, and by exploring that, to hint at where to look further.

You do have a pair of very nice components there -- but unfortunately it is far from trivial to make use of them.

This is in much the same way that novices might tear down some industrial equipment and, say, extract a couple juicy power transistors from it; the ratings might be quite provocative (say 1200V 600A, that's a lot of power! cool!), but not know how to actually put them to use, say to make an amusement like a Tesla coil. In fact, the power transistors are a fairly minor cost in a complete coil build, and it makes little difference what transistors you've salvaged, when you also need all the capacitors, raw materials (wire, tubing, metal strip, heatsinks...), etc. to complete a build. Plus the time to design (if applicable... or debug, if using random un-vetted schematics off the internet!) and build the controller, whether it be a simple (but failure-prone) fixed-frequency oscillator, or a full current-mode, frequency-tracking design plus musical modulator. (And now the simplicity of buying a kit starts to sound really appealing..!)

Not that the particulars of that kind of project are relevant here (or, if you know anything about it, if this example has even been meaningful to you?--), but the general problem here works the same way.

To cut to the chase, the domain you're exploring is metrology, the study of measurement.

Statistics is a prerequisite, as statistical methods are used to understand process variation, signals and noise, and how errors in parameters propagate through a given system of measurements.

The fundamental motivation behind electrical metrology is: everything has resistance. I'll assume you're already familiar with basic circuits and DC analysis (voltage and current flows, resistors and dividers).

Sometimes those resistances are nice and stable (like your precision specimen), but more often, they are nuisance resistances, large or small, with values not quite small or large enough to ignore, and with values variable enough (depending on temperature, mechanical fit, etc.) that you can't simply calibrate them out, either. (Not that calibration is necessarily a step available to you, with only two precision components to hand.)

The first step is, we draw all the wiring and contact resistances in the circuit, find the dominant and parasitic current paths, and find the voltage drops for each branch.

Suppose we make a 75mA current source by wiring the reference to the resistor. By definition, we have 75mA flowing through the 100Ω resistor with 7.5V across it:


simulate this circuit – Schematic created using CircuitLab

But if we put a real ammeter in series with it (notice this is just a simulation element in the schematic! it has zero resistance!), we measure an excessive error -- far greater than the tolerance of the components we're starting with. We've screwed up bad:


simulate this circuit

To wit, my BM235 DMMs -- I have a pair here, so can use one to measure the other -- measure 1.7Ω on the mA scale, 101.1Ω on the µA scale (evidently, the ohmmeter delivers 161.3µA at these voltages), and ≤0.1Ω on the A scale (<1mA test current measured). So the above circuit is a very reasonable error to expect in such a test.

Also, from experience, the contact resistance of these meters (and the cables I use with them) is in the 0.1Ω range, so consider that to be 1.7 ± 0.1 Ω, give or take. The variation of 0.1Ω, in relation to the 100Ω reference, is still larger than the given tolerances (0.1% vs. 0.05% or 0.01%), and we need it much smaller instead (so that we have a high statistical likelihood of measuring what we think we are). So this is totally unworkable.

Incidentally, note that 75mA is only 1/750 of the full reading on these instruments, so calibrating to this level would only capture a fraction of their capability -- and your components's. Preferably, the test current would be near or at the full-scale value (around 4000 or 6000 counts for these, I think?), and even more preferably, at multiple points between zero and max, so that we can calculate a regression curve, also measuring linearity and repeatability.

Which, on that note: we do at least get the zero point for free, and these meters do indeed read zero (± <1 count) when left open-circuit.

So, we need to somehow rearrange the circuit so that we can get the voltage here, but the ammeter there:


simulate this circuit

...but yet still somehow carrying the same current from the loop. Awkward, right?

Now we need some manner of control scheme. Which means we must bolt on control theory / dynamics to our list of prerequisites -- fortunately not the full scope, but the fact is, we're using systems with negative feedback, and we need to ensure that feedback stays negative, lest we construct an oscillator by accident!

The simplest such loop is completely manual. Suppose we have a voltmeter that doesn't read accurately, but does read zero offset:


simulate this circuit

Note that VM1 is a simulated component with infinite resistance, but as long as we can measure arbitrarily close to zero with it, its current flow will be similarly small, even if its resistance is finite. Therefore the current loop between V1 and AM1 is untouched, and the current is known.

We can simply adjust V1, or R1, until the error goes to zero. We need a variable source (or resistor), but it doesn't need to be accurate; just stable (stable enough that we can take the measurement without it drifting away while we're watching).

The automated equivalent, is to use an op-amp to measure the difference, and servo V1 such that this difference trends towards zero. (Tack on another prerequisite: operation of semiconductor devices, particularly op-amps.)

Note that the position of the ammeter doesn't matter, it can be anywhere in the loop; thus we can ground the 7.5V reference, or the voltmeter, in case that is advantageous to our measurement (op-amps need their inputs within a given range to function, for example). Or maybe V1 should be grounded as it will be an amplifier or something.

Real op-amps aren't perfect. There is inevitably a small error voltage in one's measurement: we specify it as the input offset voltage. This manifests as a voltage source in series with one input pin, typically a few mV, for example of the jellybean LM358. We can use such an amp to adjust this circuit towards zero difference, but that "zero" as read by the op-amp won't be true zero. To approach true zero, we need a much better performing device, like an OP07, OPA2182, etc. With an error of mere µV (or stability good enough to trim down to zero with a calibration step), we can get accuracy better than the instruments in question (and precision better than the reference components in question), and therefore achieve meaningful calibration.

That is, at least calibration to these components as standards. There is still the matter of where their accuracy comes from, which ideally should be traceable back to an international standard and the accepted definition of the volt, ampere and ohm. These units, themselves, are a matter of definition: arbitrary, not somehow extracted intrinsically from the universe itself. Thus, metrology also has a relational -- or if you like, social or political -- aspect to it: at the highest level, what these numbers actually represent, is a matter of agreement between parties (namely, the SI and its stakeholders).

As for other currents, 2-3A for example, you must have at least one other precision resistor. You cannot put this much current through the 100Ω (at least for long enough to obtain a measurement) without exceeding its rating, and its voltage drop would then far exceed your single reference anyway. Ultimately you will be doing a ratiometric or transfer measurement, perhaps using another resistor -- whose value is less well defined, but stable over time and temperature -- and this current source to calibrate it, and then to use a circuit like above to transfer its calibrated value to the instrument in question. Note that transfer errors compound, so the components/instruments used must be even higher precision to obtain the same confidence level in the final result.

Likewise, some calibrated-ratio voltage divider resistors would be desirable, say to reduce your 7.5V reference to a more manageable 1 or 0.1V or whatever, which enable the use of smaller resistances (and thus lower power dissipation) for higher-current testing.

Again, I won't go into detail on these techniques, this post is already long enough as it is -- but this should provide clues for where you need to look to find your true answers. Ultimately, metrology is a simple study, but a clever one, with many measurement setups and techniques that may not be obvious from basic intuition, but many of them share that quality of "profound obviousness once you see it".

Good luck!

  • \$\begingroup\$ Thank you very much, Tim. I modified your ammeter calibration solution slightly and added it to my question as a proposed solution. Please LMK if you think it need further tweaks. I also tried to answer some of your questions, and better explain my thinking about why a solution to (1) would be a solution to (2), and why it would be appealing to me. Regarding (2), the high power output, what I was hoping was that the linear amp I expected to be the solution to (1) could be repurposed as a kind of noise filter, without additional parts, for use powering audio circuits expecting batteries. \$\endgroup\$
    – Sara
    Jan 5 at 1:02
  • \$\begingroup\$ That sounds good. Regarding (2), an op-amp can servo to the ref (and its low noise level) over a reasonable bandwidth, and yes it can be made to filter arbitrarily low (leaving LF noise as ref error) while keeping a low Zout. The manual servo process described above is functionally a slow voltage follower, and general op-amp, and power supply and audio amp, designs and concepts can take over from here. \$\endgroup\$ Jan 5 at 3:24
  • \$\begingroup\$ Regarding "snark": try to keep a dispassionate tone. Reacting negatively to help given (even if it's, in your opinion, poor quality, or not what you expected) reflects poorly on you, and makes others less likely to help. For my part, I tend to get... overly frank? when I'm writing an answer to a poorly worded or motivated question; it's hard to realize the proper tone of a post while writing in the moment. As such, I try to give generous leeway to others who may have similar, eh, issues? habits? styles?, and try to give constructive criticism where the tone is achingly off. \$\endgroup\$ Jan 5 at 3:32
  • \$\begingroup\$ Hey Tim, just chiming to upvote this nice and extensive answer that highlights very nicely what I only tried to briefly stress in my answer! Thanks! \$\endgroup\$ Jan 5 at 6:45

Sounds Like a Job for a Voltage Follower

-- numerous people needing to increase the drive strength of a voltage signal.

It's the simplest of all opamp circuits; you'll manage. It would make sense to use an opamp with a specifically low offset voltage, or even a chopping/self zeroing one, if the choice of construction wouldn't already imply relatively large noise and parasitic crosstalk. Depending on how good you still want this to be, additional costs between 0.25 and maybe 4 AUD.

Certainly not a use case for a simple Darlington pair, since that tends to have around 1.4 V difference, heavily depending on temperature.

Regarding your use cases: I doubt you'll build something much better than your DMM. Calibration-grade voltage sources might neither be a good breadboard nor a good beginner's project. Also, to calibrate a voltmeter, you don't need 100 mA, you don't even need 1 mA. To calibrate an amperemeter, you need a precision current source, not a precision voltage source, because you don't want to have the internal resistance of your DMM affect the calibration. So, really not sure where you're going with this project, altogether!

Furthermore, the precision of your voltage source is in very bad relation to the voltage drops you'll see for breadboard contact resistance when carrying currents in the 10 mA range or higher!

You don't at all need a precision voltage source for a low-noise power source for an audio device. Actually, that feels quite opposed to what you actually need. (Which is probably just a modern switch-mode power supply with sufficient output filtering, seeing that any portable audio service internally definitely uses such a supply already, and being lower in noise than that has no advantage whatsoever).

Think about it: fresh batteries might have 1.5 V, batteries that still work beautifully night have 1.25 V. There's absolutely nothing precise about the voltage coming out of batteries!

  • \$\begingroup\$ good bit of mythbusting \$\endgroup\$
    – tobalt
    Jan 3 at 5:04
  • \$\begingroup\$ @tobalt thanks! Though in a previous question I just read, Sara shared a circuit where there was a Darlington in what I'd call supply-current-controlled configuration. On one hand, that suggests I haven't thought of all ways a Darlington pair might be used here, on the other hand, that circuit did look very precarious. Don't have the bandwidth (not a pun) to simulate that right now, but something tells me it's not going to be low in potential to oscillate... \$\endgroup\$ Jan 3 at 5:09
  • \$\begingroup\$ A usual way how Darlingtons come into play is by strapping them to the output of an opamp that does the precision control. But I agree with your general stance, that this is not required for the tasks mentioned by the OP \$\endgroup\$
    – tobalt
    Jan 3 at 5:14
  • 1
    \$\begingroup\$ @tobalt yep, connection to the controlled output is what I expected, but here, the Darlington is basically connected to the positive supply of the opamp. Slightly confusing. electronics.stackexchange.com/q/695841/359402 \$\endgroup\$ Jan 3 at 5:20
  • 1
    \$\begingroup\$ That's a nice circuit there 😁 I think designers in the dawn era of analog IC were much more able to use the supply terminals of these IC for useful functionality, because they had better recognition of the inner workings. \$\endgroup\$
    – tobalt
    Jan 3 at 5:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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