LEAD: Leveraging Protocol Signatures for Improving 802.11 Decoding Performance

Error correction is a fundamental problem in wireless system design as wireless links often suffer high bit error rate (BER) due to signal attenuation and interference. This paper presents a new cross-layer solution called LEAD to improve the performance of existing PHY layer error correcting codes. While the traditional wisdom of cross-layer wireless design is to expose and exploit PHY layer information at upper-layers, LEAD represents a paradigm shift in that it leverages upper-layer protocol signatures to improve PHY layer decoding performance. The design of LEAD is motivated by two key insights. First, error correcting code can correct more bit errors when the values of some bits, which we refer to as pilots, are known before decoding. Second, some header fields of upper-layer (MAC, logical link control, network, and transport) protocols are often fixed or highly biased toward certain values. These distinctive bit pattern signatures can thus be exploited as pilots to assist decoding. To realize this idea, we first characterize bit bias in reallife network traffic, and develop an efficient algorithm to extract pilot bits whose values have high predictability. We then augment three decoders, including the BCJR decoder of convolutional code, the belief propagation decoder of LDPC code, and the bubble decoder of rateless Spinal code, to effectively exploit extracted pilots. We implement LEAD on the USRP platform and evaluate its performance by replaying real-life 802.11 packet traces on a testbed of 12 USRP links. Our results show that LEAD reduces BER by more than 90% for half of the error packets, while incurring very low overhead.