A CART based mechanism for collision detection in IEEE 802.11

The ability to detect and distinguish packet errors due to collisions from those caused by channel errors can significantly impact the performance of medium access control (MAC) protocols such as IEEE 802.11. In particular, such mechanisms affect the backoff mechanism as well as rate adaptation algorithms. This paper presents a real-time algorithm based on classification and regression trees (CART) for distinguishing packet corruption and losses due to channel errors from those caused by collisions with other simultaneous transmissions. Using a set of four metrics, we propose a classifier tree that reduces the classification errors by considering the impact of channel variations and collisions on bit errors from multiple, disparate perspectives. Extensive simulation results are used to verify the superior performance of the proposed technique over existing mechanisms.