Power lines make terrible transmission lines:
- The insulators are lossy, typically PVC. Hopeless above, say, 100MHz, so don't expect much more than comparable baud rate. (If the noise level is low, you can still throw in lots of bits per symbol, but I wouldn't count on more than a few.)
- No impedances are controlled. Romex will be around 100 ohms, but unbalanced. Plastic junction boxes will radiate randomly. BX, conduit and metal boxes will have reasonable shielding, for the most part, but random impedances in the vicinity of 100 ohms (differential). And mismatched propagation delays for each conductor versus ground.
- Random networks. Circuits are typically chained straight from the panel, but frequently, they branch to reach multiple locations. Certainly, it's uncommon that a single outlet (where you've plugged in the modem) will be at the end of a straight circuit.
- Dispersive and generally nasty loads. Transformers and motors will absorb RF unevenly (add phase shifts and attenuation to random length reflections), SMPSs will generally reflect as shorts (due to the X type capacitors in the filter), and unfiltered rectifiers (CFLs for example) will alternately reflect (open vs. short) RF, leading to huge backscatter modulation!
And that's just for a circuit. Where the circuits meet up at the panel, half and half go to either side of the power company's transformer (US mains). Even if the reflection and attenuation from 5-10 circuits (per side) branching at the panel isn't bad enough, you've got a split-brained house where half the outlets randomly won't communicate as well with the other half.
Let alone between houses. The wires on the power poles are usually widely spaced, so that the impedance will be more like 200-300 ohms for a considerable distance (blocks?). The transformer itself will probably be a reasonable short to RF (being made of heavy copper or aluminum strap), adding yet another reflection to the mix.
But all that still isn't insurmountable. You can do what DSL does, and use a massive array of relatively narrow channels. By testing each channel for performance, you can figure out the available SNR and therefore bitrate for a given link. But you're still left with two problems:
1. You aren't guaranteed to find any single channel that's good enough to use, even at low bitrate;
2. You can't radiate higher than Part 15 limits on conducted and radiated EMI. Which is, coincidentally, also the (hopefully) worst case noise floor you have to work against.
So, most of the time, on most channels, your SNR will be negative. You'll get lucky breaks between bands of EMI, and where reflections manage to reinforce rather than null your signal along a given path. But I can't imagine these are all that common, and still, we're talking about between neighbors at best.
It's way more practical to just run CAT5 into everyone's house; or better yet, use an existing twisted pair, like DSL, or better still, coax cable!
It's a horrible premise, with an awful solution, in search of a problem that doesn't exist!
Tim