With wireless mobile IEEE 802.11a/g networks, collisions are currently inevitable despite effective counter measures. This work proposes an approach to detect the MAC addresses of transmitting stations in case of a collision, and measures its practical feasibility. Recognizing senders using cross-correlation in the time domain worked surprisingly well in simulations using Additive White Gaussian Noise (AWGN) and standard Matlab channel models.

Real-world experiments using software-defined radios also showed promising results in spite of decreased accuracy due to channel effects. During the experiments, various Modulation and Coding Schemes (MCSs) and scrambler initialization values were compared. Knowledge about which senders were transmitting leading up to a collision could help develop new improvements to the 802.11 MAC coordination function, or serve as a feature for learning-based algorithms.


Collisions on wireless networks most likely lead to packet losses. Current network protocols typically recover from these situations by retransmissions. In doing so, the overall network capacity is reduced and the network delay increases with the amount and duration of collisions. However, collided frames may still reveal valuable information that might be suitable for advanced protocol designs.


  • Detect frame alignments of collided frames at the PHY.
  • Devise techniques to detect known data, such as MAC header fields.
  • Analyze real network scenarios with respect to collisions, classify observed events (e.g., pairs of hidden terminals) and generate statistics.