... specification¹

ANSI/IEEE Std 802.11, 1999 edition.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Ethernet²

Formally, ISO/IEC 8802-3:2000(E) or IEEE Std 802.3, 2000 edition.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... project³

http://www.pdos.lcs.mit.edu/roofnet/

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... frames⁴

Balakrishnan, personal communication, November 2003.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... errors.⁵

When a two-way connection is available and latency is not important, forward error correction is not useful for dealing with dropped packets. The ideal strategy is acknowledgement and retransmission of the dropped packet, because the redundantly-transmitted information is exactly what was lost.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Intersil⁶

Now sold by GlobespanVirata, Inc.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... driver⁷

By Jouni Malinen. http://hostap.epitest.fi/.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... link?⁸

We acknowledge that these definitions are subjective, and should not substitute for rigorous link-quality statistics. How many successful transmissions must a link receive to be ``functional''? What percentage of packets must be received before we can say our link does so ``consistently''?

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... miles.⁹

More rigorous measurements are needed here, along with signal strength. The zone of perfect reception dropped out conclusively at 0.50 miles, but seemed to reappear briefly at one spot 0.70 miles away. Was this the result of a coherence of lucky bounces, a low-noise spot, or what? Unfortunately, the Prism 2 card does not produce any measurement of signal strength at these long distances.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... percent.¹⁰

Note that most 802.11 implementations retransmit four times, so for non-broadcast traffic the loss rate that normal 802.11 users would experience in practice is closer to $0.68^4 = 0.21$.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... percent.¹¹

Why all the talk of incorrect octets, instead of bit errors? 802.11 at 1 Mbit/sec uses differential binary phase shift keying (DBPSK), where each bit gets its own modulated symbol, so there does not seem to be a theoretical reason to expect errors to come in octets. Nevertheless, the errors we observe in practice generally corrupt a whole octet or a string of octets, not just scattered bits. See Figure 2.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... packet.¹²

We have to increase the MTU of the wireless card to this number, from the default of 1,500 bytes.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... manner.¹³

We leave the Ethernet header unscrambled so the card knows we are sending to the broadcast address and turns off link-layer retransmissions.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... errors.¹⁴

Most transmission problems were in one direction, because one card, the ``access point,'' was operating at 200 mW and the other at 32 mW.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.