ADR1399 Reference & Precision Buffer
An always-on DC voltage reference box built around an ADR1399 module, with extra outputs from a surplus HP precision resistor network and my own bipolar 4-wire buffer. It gives my bench a stable 10 V, 5 V, 1 V and 0.1 V to compare my meters against.
Goal
DC volts is the backbone of my bench. My HP 3456A, Keithley 2000 and 34410A agree with each other on the 10 V and 1 V ranges, but three meters checking each other is a closed loop. I wanted a reference that stays powered all the time, sits on the bench and gives me a few stable voltages to log the meters against over weeks and months.
I also wanted a buffer that can drive a load without the reference noticing it, with 4-wire (force/sense) outputs so lead resistance doesn't matter. Its input is exposed on the front panel, so the same buffer can add remote sense to my HP 3245A.
Approach
The box has three parts:
- The reference. A ready-made ADR1399 module (a "1399KEZ" board) with outputs at 10 V, 7 V, 1 V and 0.1 V. It came with a calibration label taken at 24 °C.
- Extra outputs. A 5 V output from a surplus HP 3456A Fineline thin-film network (U200). It uses the network's matched 40 kΩ pair as a ½ divider, so the ratio depends on two resistors on the same substrate rather than on two separate parts. U200 can also provide 1 V and 0.1 V. For now those come from the module's own taps.
- The buffer. A bipolar 4-wire buffer of my own design: two identical channels, one for each input terminal.
The box has no power switch. It is meant to stay on, because the reference is most stable when it is never cycled.
The buffer
Each channel is a composite amplifier:
- An ADA4523 chopper op amp sets the DC accuracy (very low offset and drift).
- A BUF634A inside its loop supplies the output current.
- The loop is closed twice. At DC, feedback comes from the sense terminal, so the voltage at the load is correct whatever the lead resistance. At higher frequencies, a capacitor takes feedback from the BUF634A output instead, which keeps the loop stable.
- A small resistor separates the BUF634A from the force terminal. It isolates the loop from capacitive loads and limits the current into a short.
- A 200 kΩ resistor from force to sense keeps the output defined if the sense leads are left unconnected.
- An RC filter at each input limits what can reach the op amp from outside.
The two channels follow the HI and LO input terminals. The output is the difference between them, so the input can float relative to the buffer's supply.
PCB layout
At the microvolt level, the board's enemies are thermoelectric EMF (every junction of two metals at slightly different temperatures is a small thermocouple) and leakage across the board. Most layout decisions follow from that:
- Identical channels. The second channel is a translated copy of the first. Any thermal effect then hits both channels the same way and largely cancels in the differential output.
- Matched pairs placed side by side, so the resistors in each pair see the same temperature.
- High-impedance nodes only on the top layer, with no vias, and kept well away from the supply rails.
- Slots and a split ground pour to slow heat flowing from the warm parts (the BUF634As and the supply) towards the inputs. Copper dominates heat flow on a PCB, so cutting it matters more than the slots themselves.
- Bare copper for the terminals. The board uses an OSP finish (bare copper, no nickel or tin), with the solder mask removed around the binding-post pads. Crimped copper ring lugs then make copper-to-copper contact all the way from the wire to the PCB.
Interactive view of the v6 buffer board: drag to pan, scroll to zoom, click a part to select it.
The box
- Supply: a linear supply with a split-bobbin mains transformer and LM317/LM337 pre-regulators. A split bobbin keeps the capacitance between primary and secondary low. With the circuit floating, 1.5 V AC appeared across a 10 MΩ resistor to earth, about 150 nA of leakage current.
- Grounding: the −REF binding post on the front panel is the single star point. The reference, the buffer's LO input and the U200 divider all meet there. A 10 MΩ ∥ 10 nF bleeder from the circuit to the chassis is currently disconnected. I plan to put it on a switch at the back, so the box can either be referred to earth or float.
- Thermal: U200 sits in a foam box to stop air movement around it. I'm considering a thin metal shell inside the foam to keep the network and its wire junctions isothermal.
Status
The reference and the buffer are assembled and running in the enclosure.
First readings. On my bench, after warm-up, the 10 V output read 10.00076 V. That's about 40 ppm above the value on the label (10.0003518 V). Two days later it read 10.00084 V. A new ADR1399, especially in the ceramic package, is known to take a while to settle, and I don't yet know how much of the difference from the label comes from the module and how much from my meter. Other outputs:
- 1 V tap: 0.999783 V.
- 5 V from U200: 5.00005 V.
- 0.1 V tap: 100.54 mV. It is not trimmed on this module, about 0.5 % high, so U200 is the better source for 0.1 V.
Buffer characterisation (in progress).
- With an HP 3400A RMS voltmeter (10 Hz–10 MHz) across the outputs I measured 1–2 mV RMS. That's far more than the parts themselves should produce.
- The reading depends on whether the box is connected to mains earth.
- Filtering the buffer's supply rails with series resistors made no difference.
- My current explanation is that common-mode current from the earthed meter flows through the output resistors and the measuring cable, rather than noise coming from the buffer. Next, I'll measure with a floating meter to separate the two.
- One output resistor had overheated at some point. I suspect an accidental short of the output.
Next steps:
- Replace the output resistors with 10 Ω high-power parts.
- Add film capacitors from each force terminal to −REF.
- Check stability with load steps into capacitive loads.
- Measure the buffer's low-frequency noise and offset directly, as the difference between its input and its output, with a DMM.
- Log the 10 V, the 7 V and the ratios between outputs for several weeks, together with the temperature inside the box.
Links
- 4-Wire Precision Buffer on GitHub: KiCad project and Gerber files for the buffer (MIT licence)
- Buffer schematic in KiCanvas
- Buffer PCB layout in KiCanvas