5 decade precision current sense application

I'm concepting a current sense application covering about 5 decades from 500u to 5A where high precision <0.1% AC/DC and high bandwidth ~1Mhz are required (high precision applies to <<1mhz). Ultimately the signal is fed to an ADC (>5MSPS)

So first I think it's obvious this requires multiple shunts and my thought is to space them every decade -> 0.01/0.1/1/10/100. These have to be low PPM 4 terminal type shunts with kelvin sensing.

Next I'm wondering about a few options for muxing these shunts and given the bandwidth and precision requirements I'm worried about the challenge of switching such low level signals. The shunts themselves have to be switched with fets or relays.

1) Use analog muxes to switch the shunt sense leads into a single differential amplifier stage
Pros: Shares one high quality diff-amp stage.
Cons: Has to switch 50mV type signals and lengthens their traces

2) Pair a differential amplifier with each shunt and mux their outputs
Pros: switching happens on higher amplitude signals. Can optimize layout around the shunts
Cons: Differential stage (with matched resistors) must be repeated

2A) Use opamps with enable/disable and wire OR their outputs to implement the mux. For some reason I've been studying this lately and like it
Pros: Eliminates analog switches with their associated parasitics/cost/size
Cons: Limits opamp choice (needs enable feature and 'mux friendly' input).


Thoughts about these options or the problem in general?

Hi,

500uA to 5A is 4 decades, not 5.

It could be easily converted with a 16 bit ADC without shunt switching.

So either I don't understand your requirement or you over complicate things.

Klaus

Ok 4 in this case.

If I have a single +/-5A range and put 500uA into it that's 0.01% of full scale or like +/-3 counts of a 16-bit ADC...now I need to measure that ~500uA signal with 0.1% accuracy (+/- 500nA).

So it's obvious ranges are required. There is flexibility on the shunt choice and spacing. It doesn't have to be every decade, but the demands of the amplitude and bandwidth requirements make me think that's the right tradeoff.

Multiple shunts may be reasonable under circumstances. But if you use it, you most likely need (a) short switch(es) for the low current shunt(s) which is the actually demanding part of your design.

And exactly what do you mean by 'short switches'? Short traces from the switches? To be specific it's either going to be Analog ADG series switches or reed relays (but probably ADG).

That's why I suggested just dedicating an amplifier to each shunt which allows the layout to be maximally optimized in that area.

You are specifying a current measurement device with 4 or 5 decades range. But the low current shunt can't withstand 5A, so it must be shorted in the high current range.

Oh yes of course. I’m picturing parallel shunts where the larger ones can be opened for low current measurements using fet switches probably.

I prefer option 2).
Use a differential instrumentation (or difference) amp IC for each shunt and you don't have to worry about matching resistors.
They are already matched inside the IC.

The muxing can be done with one IC analog multiplexer.

Would a log amp be a workable approach? If you had a set of
matched diodes (like maybe a NPN transistor array) you might
be able to take out temperature skew and/or cal-map the data
based on sensed temp.

A multichannel ADC, each reading from one of 4 divider ladders
(overrange protected) or 4 taps on one, could be near real time
without muxing (but you'd have to deal with 4 data streams).

If you used instrumentation amp per tap, and some comparators
you might be able to autorange with scale-bits plus the ADC data.

The requested accuracy of 0.1% is hard to achieve with logarithmic amplifiers. Without shunt switching, the overall dynamic of 1e7 causes problems of DC stability, even with chopper stabilized amplifier, and also wideband signal-to-noise ratio is critical.

crutschow said:

I prefer option 2).
Use a differential instrumentation (or difference) amp IC for each shunt and you don't have to worry about matching resistors.
They are already matched inside the IC.

The muxing can be done with one IC analog multiplexer.

Click to expand...

The bandwidth and accuracy requirements rule out many parts for one reason or the other. Most shunt amplifiers or difference amplifiers are ruled out on bandwidth.

Instrumentation amplifiers are abundant with well matched resistors but their absolute tempco is usually mediocre meaning you can't use an external gain setting resistor without throwing much of the accuracy out the window - and I think the shut amplifiers need gain

So while I'd keep looking for either of those types of parts the fallback is one of the handful of matched quad resistor packs with gain setting ratios (LT5400, ACAS) and a precision op amp.


Thanks for all the replies.

One more question:

Would you agree that a transimpedance topology is usually superior at low currents (<~10mA)?

Once you're below the the output current capabilities of typical opamps I see few downsides (and assuming you already have a sufficient negative supply) and lots of upsides. Particularly that by returning to a virtual ground the measured source doesn't 'see' the sense resistor and thus a much larger sense resistor can be chosen.

Hi,

I agree, I also prefer the TIA with low currents.
Especially for very low currents. Nanoamperes, picoamperes.

The problem of non TIA systems is, that they work with voltage drop. But voltage drop at low currents means high ohmic resistors.
High ohmic resistors are prone for capacitive errors, even for frequencies in the high kHz range.

Klaus