The most important differences between the wireless LAN and the MAC protocol of most wired networking applications is the impossibility to detect collissions. With the receiving and sending antennas immediately next to each other, a station is unable to see any signal but its own. As a result, the complete packet will be sent before the incorrect checksum reveals that a collission has happened. It is therefore of utmost importance that the number of collissions be limited to the absolute minimum.
This is achieved by a protocol called Carrier Sense Multiple Access with Collission Avoidance. The idea is to prevent collissions at the moment they are most likely to occur , i.e. when the bus is released. All clients are forced to wait for a random number of timeslots and then sense the medium again, before starting a transmission. If the medium is sensed to be busy, the client freezes its timer until it becomes free again. Thus, the chance of two clients starting to send simultaneously is reduced.
Of course, the overhead introduced by the Collission Avoidance delays should be as small as possible. On the other hand, the protocol should keep the number of collissions to a minimum, even under the highest possible load. To this end, the range of the random delay, or the contention window, is set to vary with the load. In the case of a collission, the delay is doubled progressively: 15, 31, 63,...1023, until a succesfull transmission occurs and the delay is reset to the minimal value. The 802.11 standard does not fix the minimum and maximum values of the contention window. However, it does advise a minimum of 15 or 31 and a maximum of 1023. The figure below follows the Random Delay (red) and the timer (green) through a collission and a successfull transmission. During the whole period the timer can be seen to reset and restart due to the transmissions of other clients.