The two design criteria are principally near or exceeding LAN speed throughput and link reliability. We'll take security as a given since it's pretty easy to configure.
There are, broadly speaking, just a couple of technologies on the market place which can deliver this sort of performance. Below is a summary of those on the market including the type, base bitrate and what commercial products they can be found in today.
- Airgo Gen 3 - Boosted 802.11g. 240mbps. Netgear RangeMax 240, Linksys SRX400
- Marvel Topdog - Draft 802.11n. 300mbps. Netgear RangeMax Next.
- Broadcom nFinity - Draft 802.11n. 300mbps. Linksys Wireless-N.
The latter two are basically implementations of the draft 802.11n standard which recently failed to be ratified. It should be noted that we don't care about interoperability because we're doing point to point with, presumably, the same equipment/brand.
The only real comparitive analysis of these technologies has been done by Tom's Networking where they basically conclude, somewhat surprisingly, that the Airgo Gen 3 stuff is far better at throughput-versus-range than the new draft-n gear. In fact so is plain old 802.11g which is a bit tragic. However this shouldn't discount those technologies from consideration for this application because we will not be experiencing line losses which will impact upon performance. This application should be able to achieve full speed with each and so the latter two have the advantage of a slightly higher bitrate.
The difficulty, so far, is in finding any equipment at all based on any of this wireless technology which features a so-called bridge mode. In other words. The product ranges are very heavily skewed to the average consumer needs and there isn't even a regular access point available. This is problematic because simply placing two routers at each end will not suffice, the routers will not establish an encrypted network between themselves. For that you need bridge functionality. Eg the product needs a place in the firmware interface to place the SSID of the network you wish to connect to and the WPA keys.
The only other way around this would be to place a computer at one end with a regular wireless client such as a PCI card and have it connect to an access point at the other. Unnecessarily complex but if that's what has to be done, then that's what has to be done.
The second stage of research here involves the construction of directional point to point antennas. Normally this is pretty simple and may remain simple but all of this technology is generally based upon multiple antennas. It's unclear if this is necessary for a direct point to point or if a multiple input antenna array is required. Normally MIMO technologies solve the issue of reflections so it is expected that a MIMO antenna wont actually be necessary since reflections can be ignored. This will probably be an issue of experimentation.
It's also worth noting that these multiple antenna products generally don't have detachable antennas. I anticipate that the products will need to be removed from their packaging and coax wired directly to the mainboards to external antenna(s).