“When developing a data acquisition system, I ran into a need of having fairly accurate current reference to compare against, 0.1% accuracy or better. This is not a particularly high standard, but unable to find a suitable device in my price range, I chose to design my own.
EBay has plenty of high-accuracy voltage references available, but not many current references. There is also DMMCheck, but it is a bit too expensive for me.
Three approaches to high-accuracy references
The gold standard when it comes to calibration references is tying them to some reproducible physical behavior. For current references, that would be a Kibble balance and could achieve accuracy on the level of 0.1 ppm (i.e. 0.00001%). Obviously that is not cheap or easy to build.
The next level down is building a very stable reference, and then measuring that against such a gold standard. This is the approach usually used in accredited calibration laboratories, and there are DIY projects for high-stability references also. But typically such references have a fairly poor initial accuracy, so they are of no use if you don’t have something to calibrate them against.
And then there is the cheapest way: just buying off-the-shelf parts and relying on their initial accuracy specifications from the manufacturers. Typically the manufacturers will laser trim the parts to be as close to the specification as they can, and then measure a large batch of devices to determine how good accuracy they can guarantee. It is hard to find parts with better than 0.01% accuracy, but that is good enough for me.
Part selection
To start with, I searched Digikey for the most accurate voltage reference that wasn’t ridicuously expensive. Currently that is ADR45xxBRZ, with ±0.02% accuracy and selling at 6.70 EUR. There used to be a more accurate part X60008 with ±0.01% accuracy, but that is no longer available.
For my purposes, the 2.048 V model was most suitable, while the series has options up to 5.0 V output voltage available. The B grade has ±0.02% initial accuracy, but on top of that there is up to ±0.02% shift due to soldering heat and further ±0.02% temperature drift over the -10°C to +80°C temperature range. One can try to minimize these by careful soldering technique and by using the device in room temperature.
To make this into a current reference, a precision shunt resistor is also needed. I chose a 100 ohm 0.01% model, which gives 20.48 mA reference current. This is large enough to get suitable accuracy, but low enough to avoid any excess heat that could affect the accuracy.
The current through the resistor is controlled through a transistor, which in turn is controlled by a low-offset operational amplifier. The goal is to make the voltage over the shunt resistor exactly equal to reference voltage, and any input offset would cause inaccuracy. I chose MCP6V51T, which has input offset of just ±2.4 µV, well below the ±410µV inaccuracy in the voltage reference.”