John Markoff at The New York Times is reporting onEtherlinx, a company that claims to have significantly extended the range of wireless broadband by hacking the firmware of current 802.11b cards and running CPEs (customer premises equipment) with 2 cards inside - one card that runs their software radio (non-standard, not 802.11b protocols) and one that retransmits signals into a house or other local broadband endpoint. They claim 2MBps speeds in field trials they've been conducting in Oakland.
Certainly, this is somthing that needs further investigation. The company claims all sorts of neat stuff, including security, QoS, and other features. This can be performed in the CPE, probably not at the radio layer. The CPE can also be built very cheaply, and sold at about a $100 price point. A number of questions remain - are they using FHSS (old-fashioned 802.11 signals maxed out at 2Mbps and were FHSS) or DSSS? How do the CPEs react to multipath loss, reflections, and loss of line-of-sight to the brodcast tower? How well does the technology scale? Can it be used in a mesh configuration or is it point-to-multipoint? They claim that their low-cost CPE can be deployed without the need for an installer, which means it must be robust indeed.
This can be compared to Navini networks, which has developed a base station that uses phased antenna arrays (essentially smart antennas) to direct power at the CPEs, and simple CPEs that can be installed indoors.
Phased array antenna approaches allow for better penetration and even a notional non-line of sight capability, but they require a managed base station and can't be used for organic mesh networks, like the meshes that are created as a part of Nokia Rooftop's or Sky Pilot's solutions.