Improving absolute precision of ADC to balance battery voltage

Question

I plan to use a relatively high-end microcontroller (Piccolo TMS320F28035, 12-bit resolution, +/- 4LSB offset, +/- 60 LSB gain) to measure voltage across stacked battery cells and control associated analog electronics to equalize their charge. The microcontroller will also store data in an eeprom memory (blackbox).

The current plan is to read up to 10 cell voltages. The problem is the large common mode voltage (each cell can go up to 3.5 V) - I cannot use isolated amplifiers such as INA124 or non-isolated high-precision INA117 due to high cost.

The current plan is to use voltage dividers (0.1%) at each tap and calculate cell voltages relatively to each other:

1. V1 is measured directly
2. V2 is calculated as measured V2 value less calculated V1
3. and so forth

The problem with this setup is that the tenth measured voltage will suffer from high voltage-divider ratio and thus could be off by too much.

Another approach is to use a battery monitoring chip and use is as an analog front end (BQ77PL900) but the cost is quite high as well.

Are there better ways to precisely read battery cell voltages?

Thanks,

SBNY

Answer 1

I was thinking about a circuit like this one to 'extract' the single cell's voltage:
(My first shot at circuitlab, please be lenient :))

^{simulate this circuit – Schematic created using CircuitLab}

This scheme can be applied identically to every cell to be measured.

I'm not sure if this works when "self-powered" from the battery pack as sketched. You should definitely go for a rail-to-rail op-amp, which kind of excludes bi-polar devices.

Are you actually charging batteries or rather "supercapacitors"? - Maybe you can find some ideas for either case by searching for "supercap balancing", resulting for example in this paper on the topic.

Edit:

Maybe it's worth mentioning that with this op-amp-based circuit you can of course scale the output value to your exact needs by varying the resistor values. For example, one might want to scale some "3V max." of the cells (= range from 0V to 3V) to an output of "5V max." (range 0V to 5V) to feed into a 5V-ADC. Or one may scale the voltage down, for example to measure "4.2V max." LiPos with a 3.3V µC/ADC.

Edit #2:

With one more op-amp per cell it's also possible to remove some constant offset voltage and increase the resolution of the measurement. If, for example, a single cell needs only be measured between 2.5V and 3V, the 2.5V constant voltage can be subtracted by the 2nd op-amp, and the resulting, limited voltage range of only 0V - 0.5V can then be scaled up to, e.g., 0V - 3.3V. With a 10-bit ADC this would yield a (theoretical) resolution of 0.5V/1024 ~ 0.5mV/LSB.

Answer 2

A radical approach would be to put a separate microcontroller at each battery. A uC with a 10-bit A/D costs almost nothing these days. No need for a resistor divider, simple use a reference (TL431 or similar) and have the uC measure the reference's voltage with respect to its supply voltage. The uC's can send their measurements to the central uC via optocouplers or and IR LED (one per slacve uC) + a TSOP receiver (one).

Answer 3

Use an analogue multiplexer followed by a single high-precision resistor divider. The multiplexer will need to be powered from the highest point in the battery stack for it to work correctly. Control the multiplexer from IO pins on your MCU - 4x IO pins will address up to 16 multiplexed inputs - in this way you also save 9x analogue input pins. Net gain is 5 pins.

EDIT - I would recommend using an external ADC and one that is 16-bit. It can sit on the output of the precision attenuator. You can find decent ones that are much more linear than internal MCU ADCs.

I would also suggest powering the MCU at midrail on the battery stack so that a cheaper lower voltage range multiplexer can be used. This will mean that the ADC has to measure negative voltages but there are some that can do this.