Hardware description:
Our hardware design utilizes a 0.5mili ohm surface mount sense resistor on a heavy (a total of 6 x 2oz) multi-layer copper PCB laminate for managing peak currents up to 150A inrush for a single-phase device. The kelvin nodes for sensing the voltage drop across this initial device are not ideal, which resulted in some unit-to-unit variance when solder quantity was not sufficient. Soldering corrections were specified to be convex with silver solder to reduce the error introduced by standard paste low solder quantities. This assembly change mitigated some variance across board assemblies and has tested well, but this has been difficult to quantify for this different issue.
The process that exposes this issue:
- Unit is tested fully for performance. Once completed, it will be sent on to calibration:
- Initial calibration consists of running a 10A calibrated constant-current load through the unit, reading the RAW Current RMS value from the chipset as “#.#####”.
- Calculate and apply the calibration value multiplier found during this process to the raw value to produce the final value of “10.0A” We save that value in EEPROM. (Calculated by comparing the RAW reading to the expected reading for a 10A load)
- If the power is not cycled on the unit, a variety of different current loads all provide accurate results over the normal operating current range. The readings behave normally and as expected for long periods of time under a variety of use cases and thermal conditions.
- Power-cycling the unit (and thus resetting the chipset) will sometimes result in a different RMS Current RAW value. The same calibration value we entered previously is used, resulting in significant error. Sometimes as much as 20%. This then stays consistent as long as the power is not cycled.
- Power-cycling the unit a few more times can result in returning to the state of the initial calibration, and the value is correct again. This suggests the problem is not in the hardware.
- Power-cycling some of the units more results in the error manifesting in 3 different states. Error above, on calibration, and error below the initial calibration. The RAW rms value varying by the same degree before calibration values are applied.
- Few units do not exhibit this issue at all across power cycles.
It is difficult to blame the hardware design when the readings remain consistent as long as the power does. Also, if the power is cycled enough times, it can return to the calibrated state. What we don’t know is which states could be the “error” states, so we could be performing the initial calibration on some units in the “error” state, and others in the “normal” state and wind up with a variety of outcomes.
During initial bring-up we did experience some odd behavior with the isolated AD converters when we were attempting to mitigate EMI by inserting ferrite beads into the input and output power paths to control noise. We found that was a Bad Idea for performance as readings would become inconsistent (large errors). The attached circuit is what we have in-place now, no ferrite beads. We do wonder if that could be related to something in the isolated power path we do not know to look for.
We have seen a similar thread on Engineering Zone:
ADE7933/ADE7932 input circuit - Q&A - Energy Monitoring and Metering - EngineerZone (analog.com)
We would appreciate any insight into what may be causing this, where and how to look for the root cause of the issue. Is the problem in initialization of the device, a sequence of events that may be triggering a bad setup of the device? Right now, we are open to any ideas how to track this down and resolve it.
Thank you.
