A Framework and Experimental Study for Discrimination of Collision and Channel Errors in Wireless LANs

A fundamental unresolved problem in wireless networks is that of distinguishing packet errors that are caused by deteriorated link conditions and noise, from errors that occur due to packet collisions. In this paper, we develop advanced algorithms based on Cyclic Redundancy Check (CRC) [10] that solve this problem. Specifically, our innovation is that we form multiple CRCs, each of which is responsible for a different segment in a packet. The CRCs are appended after each segment. In this way, we can essentially visualize the pattern of errors across the packet. If the number of successive erroneous segments exceeds a threshold, we decide in favor of a collision. We integrate our approach with SampleRate. Our approach is implemented in MadWiFi [7] and is validated through realistic test-bed experiments. Our technique is shown to significantly outperform current error identification techniques, while having low complexity, and it constitutes an approach that can be readily incorporated in existing wireless protocols.