Muxing of current sense amplifier inputs

[HwAoRrDk]

This is sort of a continuation from a previous thread asking about auto-ranging current sensing. Since then, I have reconsidered my design because I found a couple more current-sense amplifiers that can do negative common mode input voltages - the MAX9919F and TSC1101. These two have much better specifications than the INA197 (in particular input offset voltage, <200uV), so I think I can adequately handle a wide current range with a single amp, without resorting to ranging over separate sense resistors.

Anyway, I've been looking again at this project, and I can't get over how expensive these current sense amplifiers are - around £2 each. As things stand, I am going to need four of them! So I've been trying to think of a way to accomplish the same goal - current sensing of four different supply voltage rails (+5V, -5V, +12V, -12V) - in a cheaper way.

One thought I had was of multiplexing the sense inputs into a single current sense amplifier. I'm assuming that I would need to use an analogue switch that has dual supply and feed it from the most positive and negative supply rails it'll be connected to (i.e. +12V and -12V). Something like a DG409 or MUX509 that has a suitably wide supply voltage range. I don't think the fairly high on-resistance (100R) of the mux will hurt - maybe even take advantage of it for some filtering (when combined with a cap across the amp inputs).



But, I have no idea how the current sense amp will handle having input signals with widely varying common mode voltages switched to its inputs very frequently. Would this be a problem?

[David Hess]

That should work fine within the bandwidth limits of the amplifier.  The input offset current of 8 microamps through the 15 ohm switch mismatch will create up to 120 microamps of error between channels which might be unacceptable however depending on your requirements.  The MAX9919 is intended for lower values of source resistance.

[HwAoRrDk]

I must admit I only roughly understand the principle of input offset current with op-amps (mismatch in the currents flowing in/out of input pins, causing a slight additional input offset voltage, right?), so I'm having trouble understanding this.

That 15 ohm figure is the ΔR_DS(on) figure from the DG409 datasheet, right? And that's the max possible difference in on-resistance between each channel (a/b) of the mux? So the input offset current of the MAX9919 will mean there will be an additional voltage drop across one of the mux channels, of up to 8uA*15R=120uV? So, it will effectively increase the input offset voltage of the current sense amplifier by that much?

I think this would basically double my offset error at lowest current levels. I'll need to think some about whether this will be okay. I'm only shooting for +/-10mA accuracy, so my demands aren't too special.

But other than that, there's no problems with this approach?

[HwAoRrDk]

Would the TI MUX509 be a better choice? I see from the datasheet it has a max on-resistance delta of 6 ohms under the same conditions, which if my calculations are correct, would only contribute up to an additional 48uV offset voltage.

[fcb]

Page 12 of the MAX9919 d/s shows the "COMMON-MODE STEP RESPONSE", looks to take 10-20 us to recover from common-mode step (i.e. changing your MUX).

This will limit the speed at which you round-robin the inputs, if you can cope with this then you should be OK with your proposed design. If you don't need super accuracy, possibly can cope with a bit of calibration, and need a cheap solution - then you can do it with 5 opamps (2.5 TL072's and a 4052).

[TimFox]

Offset current:
The two inputs of an op amp each have a DC bias current, which can be substantial for a BJT device (80 nA typical, 800 nA max over temperature for a 741C).  The offset current is the difference between the two bias currents, which, due to matched input transistors, is roughly 4x smaller.  If the total DC resistances seen by each of the two inputs are well-matched, then the offset current is the important spec, otherwise the bias current is more relevant.  There is only an offset voltage, which is the difference between the two inputs required to keep the output within its useful range.
Since the noise of the two input currents is uncorrelated, the offset is not relevant for noise calculation.
When computing the DC errors and noise of the total circuit, it is convenient to convert the currents to voltages at the input by multiplying their value by the total resistance at that node.

[David Hess]

I must admit I only roughly understand the principle of input offset current with op-amps (mismatch in the currents flowing in/out of input pins, causing a slight additional input offset voltage, right?), so I'm having trouble understanding this.

That 15 ohm figure is the ΔR_DS(on) figure from the DG409 datasheet, right? And that's the max possible difference in on-resistance between each channel (a/b) of the mux? So the input offset current of the MAX9919 will mean there will be an additional voltage drop across one of the mux channels, of up to 8uA*15R=120uV? So, it will effectively increase the input offset voltage of the current sense amplifier by that much?

I think this would basically double my offset error at lowest current levels. I'll need to think some about whether this will be okay. I'm only shooting for +/-10mA accuracy, so my demands aren't too special.

But other than that, there's no problems with this approach?

You got it.  In practice it will be much better than 120 microvolts but the safe thing is to first consider the worst case values.

The reason I bought it up is that current sense amplifiers almost always have high input bias current because they are intended to be used with the very low source impedance of the current shunt, and this high input bias current results in a high input offset current which places restrictions on input multiplexing.

[HwAoRrDk]

Page 12 of the MAX9919 d/s shows the "COMMON-MODE STEP RESPONSE", looks to take 10-20 us to recover from common-mode step (i.e. changing your MUX).

This will limit the speed at which you round-robin the inputs, if you can cope with this then you should be OK with your proposed design. If you don't need super accuracy, possibly can cope with a bit of calibration, and need a cheap solution - then you can do it with 5 opamps (2.5 TL072's and a 4052).

Ah, thanks for pointing that out, I hadn't noticed that graph.

So, I will need to wait around 20us after switching the mux before sampling the amp output with the ADC. That should be no problem. I was planning on sampling each channel at 1kHz, so that gives me plenty of time (200us) to accommodate a settling delay.