When stations cannot hear each other

Listening first needs everyone to hear

Wi-Fi shares the air by listening before it sends: a station that senses another transmission holds off. That works only between stations that can sense each other. Two stations can each reach the access point and still be out of each other's range, because of distance, a wall or a floor between them, the way their antennas point, or noise near one of them. Each senses an idle channel, each sends, and the frames meet at the access point. Fouad Tobagi and Leonard Kleinrock, at UCLA, described this hidden terminal in December 1975.

On the bench the island does the hiding. Alpha and Bravo are an estimated -108 dBm apart, under the line where a float senses anything, while each is about -70 dBm at the access point.

What an overlap costs

The access point senses energy for as long as both frames last and decodes neither. It sends no acknowledgement, and the missing acknowledgement is the only way either sender learns of the loss. Each widens its random backoff and sends the whole frame again. Usually both are lost; sometimes the much stronger one still gets through, as Bravo's does once it moves closer. A hidden pair shows up as retries and a slow link while each station reports a good signal from the access point.

Asking first

RTS and CTS move the listening to the one place both stations can hear: the access point. The sender sends a short request to send, and the access point answers with clear to send. Both carry a Duration, how long the rest of the exchange will take, and every station that decodes either sets its NAV, a timer of virtual carrier sense, and stays quiet that long. Bravo cannot sense Alpha's request, but it decodes the access point's answer, so it waits. The radio amateur Phil Karn proposed asking first this way in September 1990, and the idea fed into later wireless LAN work.

Why it is not always on

The request and the answer are extra frames with their own gaps, paid for on every exchange they guard. So RTS/CTS is optional, and usually used only for frames above a size threshold, where losing the frame would cost more than asking. It is a hint, not ownership of the air: a station that does not decode the clear to send does not wait, and it does not fix every case.

Common mistakes

• Judging a link by the signal at the access point. Two stations can both show a strong signal there and still lose frames to each other: the trouble is the path between them, which neither signal reading shows.
• Blaming only walls. Distance, the way an antenna points and noise near one station hide stations as well as a wall does.
• Expecting RTS/CTS to end every collision. It costs airtime on every frame it guards, and only the stations that decode the clear to send wait.