Microvolt offset difference amp?

[Ionforbes]

I've been working on a milli-ohm meter project that uses 4 wire measurement to get a 2 Ohm range with 0.1 milliohm resolution. The input is just a unity gain difference amp with the output going to an inverting x10 amp and then to an ADC. The everything is on a single-supply 5V rail and the 'Midref' virtual ground is a buffered 2.5V reference. The idea is that the voltage in volts between the virtual ground and x10 amp output = resistance in Ohms (the current source for measurement iss 100mA).

It all works fine in SPICE but building it in real life show that it works fine for large resistances but the difference amp has large errors (as in a few TIMES off) trying to measure something like 100mOhms. I'm assuming it's down to my 1% resistors not being good enough for the difference amp https://e2e.ti.com/blogs_/archives/b/thesignal/archive/2012/04/24/difference-amplifiers-the-need-for-well-matched-resistors

So I guess I need to use a dedicated difference amp with onboard trimmed resistors. My problem is that I need super low offset voltage to gett 100uOhm resolution, but I can't find any good enough. For example the best one on TI's site is 40uV but I can get op amps with just one or two microvolts of offset. Is there any other way with op amps I've been misssing? Any opinions greatly appreciated!

Thanks, Ion

[Kleinstein]

The difference amplifier is not very accurate.  Using 2 stages with x 1 difference and than x 10 inverting is rather complicated.  The difference stage can directly have gain. It can also make a difference which side sees which signal.
With gain in the difference stage the resistors are a little less critical. It also depends how stable the mid range voltage is.

The MCP6V64 is an AZ OP, but it can still have some offset. 120 K are already relatively high impedance, so that the bias current of the OP can produce additional offset. With AZ OPs the input currents are more like opposite sign for both inputs and thus they don't compensate with symmetric impedance, like with classical BJT based OPs.
One can usually compensate for the fixed offset with a separate zero measurement.

One can use 2 OPs to build the input part of the 3 OP instrumentation-amplifier. The differential input of the ADC replaces the difference stage - so no more need for the 3rd OP.  This version does not need matched resistors (10% would be good enough).

For best offset compensation one should still turn on and off the current thus way one can also compensate for thermal EMF at the terminals and the DUT itself.

[Marco]

Modern instrumentation amplifiers use an autozero current mode topology which doesn't rely on resistor matching at all for offset voltage, only for gain. Just use one of those.

MAX4208/MAX4209/AD8293/MCP6N16

[Zero999]

Use an instrumentation amplifier. The data sheet for the ISL28533, ISL28534, ISL28535, ISL28633, ISL28634, ISL28635 look good. The programmable gain range could give different resistance ranges if necessary.
https://cdn.datasheetspdf.com/pdf-down/I/S/L/ISL28533-Renesas.pdf

[Ionforbes]

Thanks for the help; I guess I just wasn't looking hard enough!

[jmelson]

First, R3, R4 R6 and R7 should be matched, or 1% tolerance, at least.  Second, WHY on earth use such high values?  That makes any bias current difference become a huge offset.  Maybe 1K would be a better choice for all 4 resistors.  Then, put a small cap to ground on each input pin (2 and 3) to remove ambient RF that can be rectified by the input stage.

Jon

[Kleinstein]

The circuit as shown would need good matching of R3,R4,R6,R7 more like 0.1 % range.

It is not a good idea to have additional capacitance to ground for the inverting input: this adds delay to the feedback and makes the circuit prone to oscillation. Some RF filtering is still a good idea. Anyway the initial circuit configuration is not that great. The instrumentation amplifier circuits (either as a special chip or with 2 OPs and using the differential input of the ADC) can have the capacitors at the inputs.
With AZ amplifier capacitors at the inputs also help with emissions from the amplifiers. Some show an offset, that depends on the capacitance / impedance seen at the input. Already quite some capacitance to start with reduces the change from external capacitance.

120 K make some sense for protection to limit the current in case of a fault. It is a little on the high side, but depends on the level of protection one needs / wants. The instrumentation amplifier version should also have resistors and capacitors at the inputs for protection and EMI filtering.

[moffy]

If you can afford a simple micro then you could  use the LTC2400 24bit ADC to read directly. It is stated to have an offset of 0.5ppm. Might need a buffer amp though.

[Ionforbes]

What sort of capacitor values would you consider at the input? I was also considering ferrite beads in series with the inputs. Would you recommend putting them before or after the caps? I'm completely new to this whole EMI thing. Also it would have ~5k overvoltage-current-limiting resistors before the BAV199 diodes which should have compeletely negligible offset whith thee MCP6N16's 100pA input bias.

Thanks, Ion

[Kleinstein]

The suitable capacitor value depends on the resistors and the needed speed / switching. To get reasonable fast response when the current is turned on and off, I would plan for RC somewhere in the range of 0.1 - 1 ms.  Some 1 nF should be enough to keep out the usual RF. The AZ amplifiers should be OK with this too, though 10 nF may be better for the very low noise ones. With already some series resistors in the 10-100 K range there is little need for a ferrite bead there. The bead may be suitable at the current source.

Edit: I don't see a need for the MCP6N16. Chances are a dual OP like MCP6V77 and 3 resistors to set the gain would be better suited with the differential input ADC.

[Ionforbes]

Thanks for the reply, I'll just use 1nf since I have it laying around. It's only going to be read by a primitive human, so a few ms of settling time is more than good enough for me! I'm not using the differential input of the ADC as I want to keep the negative input at the middle virtual ground; that way I can connect the test leads either way round, at the cost of 15bit resolution rather than 16.