A link to the research paper referenced (http://www.nds.rub.de/media/nds/veroeffentlichungen/2012/07/...) were they explain the methodology including how they had to capture electricity consumption at a higher resolution than the smart meter broadcast and also that the TV had dynamic backlighting. The actual method of predicting usage based on brightness of screen and some of the false positive problems they had is interesting and also the inability to identify some kinds of broadcast (such as tv news) because the picture didn't change enough.
This is an interesting research investigation, but it hardly fits into (current) real-world applications of smart metering, at least in the US. In the US, distribution utilities are capturing usage data in 15-30 minute intervals. Some may go as far as 5-minute interval data. But, even at that level, I would think that the granularity of consumption aggregates way too much load activity to pinpoint individual devices or specific activity on those devices.
Maybe someday we'll see metering at 5 second, or smaller, intervals. But, at that point, the only advantage would be specifically to monitor household behavior. Utilities already have more data then they know what to do with. And, if the goal is to identify a household's TV viewing, well, there are better ways to get that data, aren't there?
Then, aren't there technical challenges for capturing 5 second intervals? From what I understand (not being an EE or hardware person), the smart grid RF mesh networks are on the 900MHz band, which have limited spectrum and lots of other usage. In my dealings with mesh network vendors, I have gotten the sense that bandwidth is pretty limited.
The research paper referenced above indicates a 0.5/sec sampling rate. Would a battery-backed inverter be able to sufficiently attenuate the power signature on this time-scale?
Usually the optical port is an easier interface. There are python libraries around somewhere I believe. Ping me on email and I'll follow up with my contacts.
I'm sure you can make marvels in a controlled environment ... but how you're going to know what someone watch on TV, while someone else in the house is using a hairdryer, the washing machine is running and a second TV is used in conjunction with a console.
Signal processing is a centuries-old field of engineering that deals with exactly that problem. We already have theoretical framework and practical ways of dealing with signal identification and separation problems.
Well they've got the hair dryer signature, the washing machine signature, etc., as isolated signals - I can't imagine that it's that hard of a problem in signal processing to isolate one from another up to a third or fourth level of varying signals. Like extracting signals from non-perfect carrier waves.
It's a shame that anyone would feel that this is something that a business should want to do against the will of their customers however.
