Many papers have presented strategies to improve the performance of reliable transports, such as TCP, over wireless networks. In [1], Balakrishnan et al. compare and analyze several techniques for improving TCP throughput over a precursor to 802.11, including modifications to TCP endpoints such as selective acknowledgements, and making the link-layer protocol reliable and ``TCP-aware,'' with immediate retransmission of lost packets. The authors were not able to obtain the contents of errored frames4and did not analyze forward error correction in this context.
Khayam at al. ([2]) measured the error characteristics of two indoor one-way 802.11 links across a hallway at 2, 5.5, and 11 Mbits/sec. Because their links were one-way (for multicast video broadcast), they used forward error correction to recover from dropped packets as well as bit errors.5 The authors found that, for these two links operating at 5.5 Mbits/sec, giving the application layer access to frames received with bit errors reduced by about 40 percent the amount of redundant information necessary to achieve perfect transmission. At 11 Mbits/sec, there was no advantage to looking at frames received with bit errors, because the vast majority of errors were long strings of dropped packets, not frames received with bit errors. Their two links had almost no dropped packets or errored octets at 2 Mbits/sec. The Khayam authors did not try to extend 802.11's range or discuss two-way connections.