Over the last couple of weeks I will have been assembling and testing the OPQBox2.6 power quality device This is actually the fourth PQ device I put together as my colleagues and I are trying to push forward our vision of the Hawaii island power quality monitoring network.
Arguably the first power quality device I put together was arguable the Power-is-Power(PIP) board. It consisted of an AVR MCU a couple of clamp-on current sense transformers. While nothing more then a class project, it was enough to get me interested in the field of sensor networks and power quality. A few semesters later I came back to the project and developed Opqbox 1. It was the first in the series of opq devices, and certainly the simplest one. It consisted of two wall-wart transformers, one for power and another for measurements. Measurements were performed with a MSP430 MCU with an on board 24bit ADC. Processing and communication was done by the Raspberry PI. Box1 was truly a minimum viable product, but with a few boxes we were able to make a real distributed measurement. It was the building block for the first public power quality monitoring sensor network to operate in Hawaii.
The next big thing.
A few things became very clear as we deployed Opqbox1:
- No two wall-warts are alike, and are quite painful to calibrate.
- No battery backup made us miss interesting event during the shutdown and startup of the grid.
- NTP is not great and the PI can't be trusted to maintain timing
- With no over the air updates for the MCU, bugs and updates requred each box to come back to the lab.
Hence OPBox2 was born. First task was to get rid of the wall-wart transformer. Instead an isolation amplifier (AMC1100) was used to bring the signal to the PI side. In the first revision of the OPQBox2 I used an AC capacitive dropper to power the hot side of AMC1100. This however presented a problem. Capacitive droppers have a questionable reputation when it comes to safety. Further more in an event of a power outage, even if the isolated side was battery powered, the hot side would be un-powered, and the device would be unusable until the power came back on. In order to get make OPQBox2 battery powered, I needed to include an isolated DC-DC converter to power the hot side of the AMC1100, this was accomplished via ADUM5010. A chip-scale isolated DC-DC fit the power budget and saved a lot of board space and design time.
Now that the power was centralized came the task of adding the battery circuitry. At first I tried to use a supercap on the power rail to allow the device circuitry to ride out a short power interruption. It quickly became apparent that a real chemical solution was requred. I am quite weary of LiPo batteries, and would prefer not to use them unless I have to. In this case, since the weight and volume requirements are rather relaxed I decided to go with a safer chemistry. LiFePo4 seemed like an appropriate choice. Unfortunately the only LiFePo4 single cell charger which did not blow up my BOM came in a 3x3mm package with a 3mm pitch. Given that I assemble all my PCBs by hand, soldering this became quite a challenge. None the less by the third OPQbox 2 prototype I had a working battery charger.
Finally I had to figure out timing. This was a pretty straight forward: Add a TCXO and a GPS option, and use the MCU to control sampling and realtime processing. This warranted an MCU change, since it became quickly apparent that the anemic MSP430 was not up to the task. Instead an STM32F373 was used as a DSP and the analog frontend. This 72Mhz monster includes a 16Bit 50ksps ADC, perfect for the measurement I am trying to make.
I believe that the next batch of boards will be the production run, and the 10PCBs that I have in hand, are enough for another pilot run. Stay tuned.