Piccett: Protocol-independent classification of corrupted error-tolerant traffic

Bit errors regularly occur in wireless communications. While many media streaming codecs in principle provide bit error tolerance and resilience, packet-based communication typically drops packets that are not transmitted perfectly. We present PICCETT, a method to heuristically identify which connections corrupted packets belong to, and to assign them to the correct applications instead of dropping them. PICCETT is a receiver-side classifier that requires no support from the sender or network, and no information which communication protocols are used. We show that PICCETT can assign virtually all packets to the correct connections at bit error rates up to 7-10%, and prevents misassignments even during error bursts. PICCETT's classification algorithm needs no prior offline training and both trains and classifies fast enough to easily keep up with IEEE 802.11 communication speeds.