This seems like it fits in with a lot of the interests here. I know there’s quite a few software engineers amongst you.
Very cool and informative. Nice interview.
… but where’s the part where Jonah reveals the Easter Eggs? My theory is if Sense detects an ELP it’s going to rock out on the signature.
Great blog article. Nice job @JonahAtSense in making all the functionality of the monitor work within the memory, DMA speed, practical communication bandwidth and CPU capabilities. Thanks for the details on the operation of monitor, plus the special challenges. I’m guessing that some of the 2017 Sense dropouts might have been connected with that SPI issue. Congrats on solving.
I really like this statement. I’ve always believed that Sense was taking measurements on the CT’ AND THE LOAD LINES. This confirms that.
I like the part below that about probabilistic state. That hints that the monitor is using some form of Hidden Markov Models to discover and classify transitions of interest. That’s what MIT was experimenting with a while back.
The part about feature extraction is also of interest. That’s the magic that converts the raw voltage and current microsecond measurements into the features that machine learning uses to make decisions.
I don’t know the date of that research study but there are references to other studies from 2006. That’s not very long ago and a lot of what they have there looks antiquated today. It’s amazing how quickly things move.
That paper was actually from 2011… You can still buy the V2 version of their funky WiFi connected PowerPorts for only 125$ plus 12$/yr monitoring fee.
Makes the HS300 look like a good deal. More outlets, lower price, no monitoring fee (with Sense)