Before we all write the idea off as cranky, take a look at this POWIFI paper http://arxiv.org/pdf/1505.06815v1.pdf
It seems to have been covered in a number of popular news outlets, including several mainstream ones, over the past two or three days. I was on the tube this morning and noticed it in the Metro's pop sci section overlooking someone's shoulders. Brain said "bollocks", but I thought I'd look at the underlying details, from both what they're claiming it can do, and how they apparently achieve it.
Once you get beyond the media hyperbole, the actual claims are rather more sensible. Furthermore, I spent half an hour or so with the back of an envelope and, against my gut feel, it does seem workable.
Key to this is that the router needs modifying to transmit almost continuously at full power on multiple channels at times when the channels are not utilised for real traffic. The harvester end itself is actually quite simple but there have been some rather careful parts.
I've not yet read the paper, but surely you are always going to be limited by the max legally allowed ERP of the Tx, and the old bastard of the inverse square law...?
All of this is dealt with in the paper if you read it, it is a good read, and, as I said I was most sceptical until I'd read it and done a few back of envelope calculations.
The only thing glossed over in the paper as far as I can see is the energy harvesting matching network, although it appears to match reasonably well from their graphs. I can't see how it achieves this considering the rectifier after the matching network and the matching network itself.
To reiterate, I smelt bullshit, tons of it, but there's little I can question having read it and run the numbers.
Assume 3W* output (34.7dBm) aggregate across three non-overlapping channels and a 4dBi tx antenna and 0dBi receive antenna. Path loss over 17' is 54.5dB giving net -15.8dBm or 26.3uW at the receiver.
At 50 ohm that is 0.036V rms or 0.10Vpp, the rectifier doubler gets that to 0.20V at 131uA peak so you're now well within the territory for energy harvesting capabilities of the BQ25570 (100mV min, operating quiescent current <500nA).
So, in 30 minutes you can accumulate 26.3 x 60 x 30 = 47mJ.
Alternatively, that's 47mW for 1 second, that's quite a lot of MCU processing these days.
While my calculations assume 100% efficiency, they are there to show that this doesn't appear to be complete baloney.
* edit: I would question the legality of running 3W from a single unit as a use case under FCC regs which specify 1W max. That does not stop the end user from deploying three 1W units themselves.
Edit 2: To put this into perspective, this means it takes 10.6 hours to generate 1J. A typical phone battery is 5Wh, or 18kJ. It will take about 21 years to charge that battery assuming 100% efficiency and no self-discharge at a distance of 17'.
On the other hand, if we place the energy harvester 2" away, the path loss is only 14dB, giving a power of 290mW. This would take 17.2 hours to charge a 5Wh battery assuming 100% efficiency.
But then unlike other "products" the POWIFI is not claiming to be a phone charger.