Updated with corrections from Bill McFarland 1/29/07
In both private discussions and during the Intel draft 11n chipset webcast, I have been hearing hints about 11n's "bad neighbor" problem having been addressed in draft 1.10. I was able to get Atheros' CTO, Bill McFarland on the phone to bring me up to speed on what actually got into the 1.10 draft. Note that some of these mechanisms were being debated back when Draft 1.0 was being finalized. But since consensus couldn't be reached, the 11n task force punted and didn't include any of them in 1.0.
Bill said that Draft 1.10 includes not one, but three mechanisms to prevent 802.11n products from using the channel bonding 40 MHz mode that causes interference with 802.11b/g WLANs. The first is Clear Channel Assessment (CCA). This mechanism is a check made before each transmission to see whether any legacy devices are using the channel. If any are, then only 20 MHz bandwidth (same as used by 802.11b/g gear) is allowed to be used. This check is now mandatory in both the 2.4 and 5 GHz bands in Draft 1.10. Note that a form of CCA actually is used in 802.11b and g as part of the Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) protocol.
The second mechanism is a little fuzzier, but involves actively scanning channels adjacent to an access point's primary channel to see if legacy devices are in use. This mechanism is mandatory in 2.4 and not used in 5 GHz. Once again, if any legacy devices are detected, 40 MHz mode is disabled for a period of time. Bill refers to this mechanism as "politeness", but it doesn't have a formal name yet.
McFarland said that Draft 1.10 actually contains specifics for the scanning algorithm, even including times. The only specific example he gave was that the 40 MHz disable time if legacy activity is detected is initially set to 30 minutes. But this is the least-baked of the three mechanisms and will most likely be tuned and tweaked over the next year or so.
The third protective mechanism is called the "intolerant bit" and used in the 2.4 GHz band only. You can think of this a 40 MHz mode kill switch that any STA (client) can throw if it decides that it doesn't want channel bonding to be used by the AP that it is associated with. It was initially described as included for use by devices that have both WLAN and Bluetooth radios, ostensibly to prevent channel-bonding from interfering with Bluetooth operation. But McFarland admitted that the mechanism isn't specific to Bluetooth and will need to be supported by all 2.4 GHz 11n gear.
Update 1/29/07: Further info indicates that the "intolerant" bit can be used in a way such that all in-range APs, not just the associated AP, would switch to 20 MHz mode if a STA set the bit.
I have to admit that three mechanisms seem almost like overkill for legacy WLAN protection. But given everything that's at stake, I'm glad that the industry (finally) erred on the side of overkill so as not to kill the goose that laid the golden wireless egg. My colleague George Ou may want to restrict 11n to using the 5 GHz band only. But I think that assuming that all three mechanisms remain in the spec and that they are implemented correctly, our 2.4 GHz airwaves may be safe for existing WLANs again.