Those 23 broadcasters are all on different channels. We're talking about devices that could potentially be on the same channel as an exiting TV station. This is not a trivial problem to solve, because you need very sensitive detectors on the "smart" device to ensure that it correctly detects when there is an active station on a given channel: http://www.sharedspectrum.com/wp-content/uploads/2008-10_SSC....

One of the most complex aspects of doing this sort of thing is various permutations of the "hidden node" problem. Essentially, the problem is that a "smart" transmitter may not hear a dumb one, and use an in-use channel, interfering with a dumb receiver, which otherwise could hear the dumb transmitter. This tends to happen because geographic obstacles can cause individual nodes to have a different view of the spectrum environment.

Good point, but devices don't have to be able to sense broadcasts on a channel in order to avoid using that channel. Alternatively, the WISP WAP could just continuously broadcast (on known safe channels, as a part of the normal SSID etc. beacon) the list of currently safe channels, as configured and maintained by WISP personnel, and consumer devices could simply not transmit until they received the list.

Even if we were limited to listen-before-talk, the linked study recommended 10 W rather than 4 W, and that was in the urban context of Baltimore-DC, rather than out here in the hills where we rural people would like a choice in ISPs.