Isolated HV measurement circuit using opto & pulse transformer?

[max_torque]

Hi all,

I have this little HV DC measurement pcb from a nissan leaf, used in the HVPDU to measure the voltage on the DC bus. Of course, it includes a suitable galvanic isolation architecture, but what i found interesting was how it might actually sense the analogue HV across that isolation??


I have attached a couple of pics of the interesting bit of the circuit

You can see a standard arrangement of series resistors to drop the HV into a suitable low voltage range, but then that signal looks to pass across an optocoupler, however there is a pulse transformer doing something clever, i guess used to some how modultate or calibrate the optocouplers CTR co-efficient?  Anyone got any insight into what technique is being used here?

[schmitt trigger]

Can you read the optocoupler’s part number?
That would provide another puzzle piece.

[max_torque]

Optocoupler is labelled N2HB2  with a sort of curly "M" symbol

[Ian.M]

Hmmm.  One approach would be to wire the pulse transformer in series with the opto LED with the voltage being monitored applying reverse bias to the LED, and apply increasing amplitude pulses till you get a signal on the opto output, then decrease till it stops.  Repeat to track the input voltage.  You'll probably need to compensate for LED Vf variation vs temperature, and may need initial calibration for opto CTR, (and some means to recalibrate for aging).

[D Straney]

For what it's worth, if that optocoupler has one of the standard pinouts (pins 1 and 2 for LED, 3 and 4 for whatever output device) then the LED is on the "comms" side instead of the "HV" side.  (Actually looking at it again, you can see the "3300" series resistor and transistor used to drive the optocoupler's LED, on the "comms" side)

If on the HV side you used a phototransistor or photo-MOSFET to quickly connect then disconnect the sense voltage from the transformer, then you'd get a nice little square pulse with the sense voltage amplitude on the other side of the transformer, which you could sample in the middle of the pulse and then convert to digital at your leisure.  Of course things like the magnetizing inductance of the transformer and the source impedance of the divider would kill the accuracy of that pulsed voltage if not done right, but it could work if you can get the right combination of...

• a high-enough inductance transformer
• a short enough pulse for the magnetizing current to stay sufficiently low over the pulse time
• a long enough pulse that transients related to the leakage inductance etc. have enough time to reliably settle out before sampling the voltage
• a low-enough source impedance to not be bothered by sourcing the transformer's rising magnetizing current over the pulse duration (this might be what the ceramic cap on the HV side is for, providing some buffering for the sense voltage during the pulse)

[D Straney]

Was thinking a bit more about this, as I might have a use for isolated voltage measurements myself in the future, so here's some sample numbers:
If you set up a circuit like I described before and can get the equivalent-RL-circuit of transformer's leakage inductance and winding resistance (as these limit the ramp-up time for the voltage on the other side) to settle in, let's say, 50 µs, then it can actually work with some optimistic but not unreasonable values:

• 10V (ballpark) signal across a 10 mH primary inductance for 50 µs = 50 mA peak magnetizing current
• 0-50 mA triangular current out of a 10 µF ceramic cap for 50 µs = 125 mV sag in cap voltage

So overall that's only a -1.3% error in the reading created by the transformer's magnetizing current pulling charge out of the filtering cap on the voltage divider's output.  Seems pretty decent!  If you wanted to use a <10µF cap / smaller transformer / longer pulse settling time / etc. and try to compensate for the larger error in software, it would be a little more difficult than just a fixed multiplier or offset considering that the error depends directly on the original reading, but all the values that go into it are pretty predictable so you could definitely create a correction curve for it.  And even if the transformer's magnetizing inductance (the most variable part here) changes by 20%, but that's out of a 10% error for example, then that change only creates a 2% difference in the reading.

As for that 50 µs settling time, a 500 µH leakage inductance (~0.95 coupling coefficient for 10 mH transformer) and 33Ω series resistance would create a 15 µs time constant, good enough for settling to reasonable accuracy (couple %?) within 50 µs.  Some of that 33Ω series resistance could be added externally to lower this settling time: the 33Ω * 50 mA peak primary current will produce a 1.65V "sag" in the reading, much larger than the capacitor-voltage-sag-induced error, but this is a predictable value in the same way as the one described above, and can be compensated for.

Finally, looking at that last close-up photo gives a pretty good idea of the schematic:

[max_torque]

wow, thanks for those replies, really interesting approach!

It makes sense that this can be a relatively low accuracy measuremen, primarily to determine the prescence or otherwise of HV DC on the vehicle systems, so a cheap, simple architecture that uses some more complex software and dev to make it work seems to make sense to me.  Anything Automotive is driven by cost due to the volumes being manufactured so here a larger upfront cost can be easily offset by the lower BOM costs etc

This could be quite a useful circuit architecture i think too :-)