Accurate low voltage diff measurement

[Oznog]

I would like to measure my RV's secondary battery accurately through columetry, amp-hrs in versus amp-hrs out. I also want to have a bypass around one of the diodes in the inverter, so it will boost the voltage (and therefore the charging current) by at least 0.7V to the deep cycle battery when it needs bulk charging.

It will all be managed by a microcontroller system, one under the hood, and another managing an LCD & user interface inside.

I have a 50 mV @ 200 amp shunt that will of course go on the high side. I was interested in Hall-effect devices, but they can't detect the low current loads accurately, and the current surges from the power inverted would likely cause permanent magnetization inaccuracies.

So here's my problems. I'd like a resolution of 25nV, easy to get with the right ADC. But no ADC will accept a common mode voltage of 15+ volts, generally it's all 5V. A resistive voltage divider is completely out of the question due to the low amplitude of the differential signal. i.e. a 1% error between the two sides would mean an error of 50mV, or 200 amps in the reading!

The other option was a differential low-offset chopper op-amp, but I've got similar problems with resistor inaccuracies, and the "absolute maximum voltage" of the choppers I know is only 16V, generally too close for comfort.

The only thing I could see was to ground the adc and/or amp at one of the shunt terminals and charge pump a potential +5v above the positive terminal voltage. This is, however, incompatible with the negative ground on the interface inside the cabin.

How can I pull this off? Mainly I think it's a level conversion issue for the digital signals, probably a SPI which I'd like to run at a high speed. Is there a chip which can pull that off within these voltage differences? The positive terminal is also not a constant voltage, it could drop to 9V when the inverter loads in heavily, and this can't be allowed to corrupt the data, so it'd have to be differential. Is this a case for an optoisolator? Can I get an optoisolator to run at high speeds, like 2MHz?

 

[Optikon]

I would like to measure my RV's secondary battery accurately through columetry, amp-hrs in versus amp-hrs out. I also want to have a bypass around one of the diodes in the inverter, so it will boost the voltage (and therefore the charging current) by at least 0.7V to the deep cycle battery when it needs bulk charging.

It will all be managed by a microcontroller system, one under the hood, and another managing an LCD & user interface inside.

I have a 50 mV @ 200 amp shunt that will of course go on the high side. I was interested in Hall-effect devices, but they can't detect the low current loads accurately, and the current surges from the power inverted would likely cause permanent magnetization inaccuracies.

So here's my problems. I'd like a resolution of 25nV, easy to get with the right ADC. But no ADC will accept a common mode voltage of 15+ volts, generally it's all 5V. A resistive voltage divider is completely out of the question due to the low amplitude of the differential signal. i.e. a 1% error between the two sides would mean an error of 50mV, or 200 amps in the reading!

The other option was a differential low-offset chopper op-amp, but I've got similar problems with resistor inaccuracies, and the "absolute maximum voltage" of the choppers I know is only 16V, generally too close for comfort.

The only thing I could see was to ground the adc and/or amp at one of the shunt terminals and charge pump a potential +5v above the positive terminal voltage. This is, however, incompatible with the negative ground on the interface inside the cabin.

How can I pull this off? Mainly I think it's a level conversion issue for the digital signals, probably a SPI which I'd like to run at a high speed. Is there a chip which can pull that off within these voltage differences? The positive terminal is also not a constant voltage, it could drop to 9V when the inverter loads in heavily, and this can't be allowed to corrupt the data, so it'd have to be differential. Is this a case for an optoisolator? Can I get an optoisolator to run at high speeds, like 2MHz?

Question: Why do you need 25nV of resolution for measuring battery voltage? Sounds like major overkill to me - making the problem unnecessarily complicated.

You can get opto isolators to run at 2MHz (much more if needed - like 10MHz)

 

[Roff]

You basically want 200A full scale with a resolution of 100ua - right? That sounds like overkill to me too, but who am I to judge?
Although I have serious reservations about noise, how about amplifying the high side voltage by, say, 100 first, leaving it referenced to the high side. Then use the "flying capacitor" transfer trick to reference it to the input of the A/D (differential?). You basically charge a high-quality capacitor to the input voltage, disconnect both ends of it using relays, then connect it to the A/D with another set of relay contacts (DPDT relay). This has to be break-before-make. If you need high speed, this obviously won't work, but your 25nv A/D wouldn't be very fast, so I'm guessing you don't need speed. I've never done this - I've only read about it. I think some A/Ds actually do this internally, with CMOS switches instead of relays, but for somewhat different reasons.

 

[Oznog]

Small loads (around or under 100mA) over long periods can represent a lot of the deep cycle battery's drain, so it's important. Now that I think about it, 10mA might be a lot better. Noise shouldn't be an issue if I do a lot of oversampling and/or add a lowpass filter. The filter's impedance will be require careful consideration. The killer would be offset, but something like the LTC1053 is totally remarkable (the downside being cost and a huge recovery time from + or - output saturation overloads). Thermocouple effects where dissimilar metals form a junction and create a tiny DC voltage might be a serious problem though.

I'm starting to think amplifying the signal as you suggest, then passing it to a capacitive charge pump, could be the way to go. The cap would need to be nonpolarized though, which rules out tantalum, and I wouldn't put an NPO electrolytic under the hood. I'd probably be stuck with a mylar film which isn't great for capacitance, but we'll see. I think it's a good idea. I might be able to drop the external ADC in favor of my PIC's 10-bit ADC by having multiple ranges.

But how to do it? I would normally use a 4053 analog switch, but they're only rated for an absolute max of 20V. If I'm referencing +5v above the positive rail, that's too close. I see a MAX379 has a voltage up to +-65V! Lousy rdsOn of 2K ohms though. I guess that's my lowpass filter right there. Also, the PIC can't read a negative voltage, at best I might need a positive and negative scale, or do the charge transport to a biased 2.5V midpoint.

I tried to draw out what it would take with discrete MOSFETs, I sure don't see anything practical.

BTW, noise can be a GOOD thing at times! I've oversampled and averaged PIC's 10-bit ADC off of a barometric pressure sensor to make an altimeter that should have taken around 16 bits to get the 1 ft resolution I desired. If the noise distribution is OK, it does wonders.