improving a current sense circuit

[wergor]

for my hobby project I'm designing a current sense circuit. the requirements are:

• high accuracy (<100ppm after calibration, <50ppm would be great)
• selectable ranges, 2A to 2uA
• bidirectional

i have come up with a simple circuit:

The MAX4209 is a 10pA Ib max, a = 100 instrumentation amplifier. The MAX4821 closes one of the relays and the ADG708 connects the high side of the active shunt resistor to the MAX4209. I have chosen this part over a 74HC238 and ULN2003 because with it, I can avoid breaking the circuit when switching ranges. Not shown is the voltage buffer at FORCE_LOW that I use to determine if the voltage is safe for the MAX4209.
I am now trying to find ways to improve the design.

• While the error budged caused by the MAX4209's Ib is very low (~5ppm at the lowest range) the error budget caused by the ADG708's on leakage current alone is ~3000ppm. unfortunately, the ADG708 is already almost as low in terms of on leakage current as these muxes go (as far as I know). apart from using relays instead of the ADG708 and using one MAX4209 for each range, is there a solution I have overlooked?
• the ADG708 operates off a +-2.5V supply, but is controlled by a 3.3V uC, so I'd have to use a logic level converter for it. I was thinking about using a ADG658 instead, which has similar characteristics to the ADG708 and powering it with +3.3V / -2.5V. Is there a better solution?
• any other traps I have fallen into?

[Kleinstein]

To simplify the circuit and essentially avoid the MUX, one can have the shunt resistors in a single string in series. So when sensing the higher currents, the higher shunts would be in series with the voltage sensing. Except for the very high value shunts this is not a problem.

The ADG708 leakage can be relatively high - there are better ones, though with higher on resistance. However one would likely not need a full MUX anyway.

Often one side of the shunts can be ground - so one does not really need an instrumentation amplifier. So one is free to use a normal OP with separate resistors for the gain. The Max4209 is very low bias, but not low noise and not very linear / gain stable.

For the low currents there is also the option to use a trans-impedance amplifier. This may simpifiy things as the dynamic range my be higher and one may be able to skip a range or 2. So one may end up with an TIA for the 2 lowest ranges and the chain of shunts for the higher currents.  Only the TIA needs a low bias (or more exact low current noise -  a stable bias could be compensated). With a higher voltage at the "shunt"/FB resistor voltage noise and drift are less important. The amplifier for the shunts can be lower noise, as the current noise is no longer as critical. So one could use a lower voltage drop and this can help the high current range shunts as self heating gets smaller. It still needs good / stable shunts.
Protection could be demanding to avoid leakage currents to effect the low current ranges. It is possible but not that simple.

The calibration can be very demanding - it needs a suitable source, which can be hard at the upper and lower end.

[wergor]

To simplify the circuit and essentially avoid the MUX, one can have the shunt resistors in a single string in series. So when sensing the higher currents, the higher shunts would be in series with the voltage sensing. Except for the very high value shunts this is not a problem.

could you sketch that for me, please?

Often one side of the shunts can be ground - so one does not really need an instrumentation amplifier. So one is free to use a normal OP with separate resistors for the gain. The Max4209 is very low bias, but not low noise and not very linear / gain stable.

I hoped to be able to use a fixed amplification instrumentation amplifier to avoid having to use matched resistor networks. But thank you for pointing that out, I was about to use it in the voltage sense circuit as well, but it looks like I have to go find another amplifier once again

Protection could be demanding to avoid leakage currents to effect the low current ranges. It is possible but not that simple.

the protection scheme I have in mind is to monitor the voltage with a low input bias voltage (~1 pA) buffer and switch to higher current ranges or disconnect the load entirely if the voltage goes too high. that voltage buffer wouldn't need to be terribly accurate either because the MAX4209 would saturate at about 25mV differential input voltage but only be in danger at differental voltages > 2.5V.

The calibration can be very demanding - it needs a suitable source, which can be hard at the upper and lower end.

of course, but I'll cross that bridge when I get there

[Kleinstein]

I have sketched a possible current input section, including some protection.
The exact resistor values and OPs to use are still free to adjust. The switches can be relays and in some places also FET switches.
The outputs are still separate for low and higher currents.

For protection opening the circuit could be tricky if used with a highly inductive source (e.g. electromagnet). It is still the worst case when the fuse blows. One may want some extra protection (e.g. spark-gap/MOV + cap) in such a case.

[wergor]

thank you for the sketch!
The reason I have a mux in my circuit is that I want to avoid any influences the relays could have on the measurement. The relays I intend to use ( MS05-1A87-75D ) have contact resistances of up to 150mOhm, and the datasheet does not contain information about repeatability or temperature / age dependent drifts. Am I worrying too much?

[Kleinstein]

The switch resistance can be significant. So it should not be directly part of the shunt. The series connection of the shunts is a way to avoid this without using an extra MUX. At least for the lower resistance (e.g. up to some 10 K) the other shunts in series to the amplifier is not a problem.
 
For the highest shunt (e.g. > 10 K) the contract resistance may be acceptable.

[BobSacamano]

Often one side of the shunts can be ground - so one does not really need an instrumentation amplifier.

Can you explain how one side of the shunts can be ground for a high side current sensing topology?

[Marco]

As I've said before, I think biased diodes can isolate the paths in a multirange TIA.

I think something like this would work to isolate the paths, when the switch is off the guard voltage (from a LMC662) will bias the diodes at a couple mV. Even when the higher current ranges will need a couple diodes in parallel, something like BAS116 should leak bugger all with a couple mV bias. When the switch is on the main TIA opamp needs to pull a little more current through the guard voltage resistor, but that doesn't affect the voltage at VOUT. Each VOUT needs it's own buffer (more LMC662s).

PS. the TIA would need a current boost stage for higher currents.

[moffy]

Marco, I think you meant to connect the inverting input of X1 to its output rather than to ground as shown.

[Marco]

Oops, yeah. But it's just meant to explain the concept of the guard voltage biased diodes. Instead of the diodes, you could also use optofets and bias the bottom of the optofets. It would have far more leakage than the diodes, but <10 pA.

For the original design, using optofets and relays should also reduce leakage currents to pA range, no guard necessary since burden voltage is only a couple mV to begin with. Would only work on the upper couple ranges though, so might as well go all relay at that point just for simplicity.

[Doctorandus_P]

It's quite common make the sense amplifier configurable. If yo can switch it for example between 1:10 and 1:100, then you can halve the number of sense resistors.

When you get into the lower ppm's then temperature coefficients of your sense resistors (and price etc) becomes a factor and a single low power relay can then be used instead of three higher current relays and sense resistors.