Measurements

In order to design a reasonable error-correction strategy for 802.11 packets with bit errors, we sought to observe the error characteristics of several indoor and outdoor environments. Unfortunately, most consumer 802.11 implementations do not advertise an interface that allows the host to receive frames with failed CRC. To our knowledge, only cards based on the Intersil⁶ Prism 2/2.5/3 chipset have a publicly known method to access errored frames.

We modified the GNU/Linux Host AP driver⁷ for these cards to ignore the card's report of a failed CRC and put the card in ``HFA384X_RID_PROMISCUOUSMODE.'' These two changes were necessary to receive errored frames outside of the card's ``monitor mode.''

Because our goal is to improve 802.11's range, we conducted all measurements at 1 Mbit/sec, the slowest and most noise-resilient 802.11 modulation. We set up a laptop with a 32 mW Lucent Orinoco card on top of MIT's Green Building, about 310 feet above ground level. The laptop transmitted 1,400-byte ICMP ping packets to the Ethernet broadcast address. We wanted there to be no link-layer retransmissions, because they don't make sense when we are trying to salvage errored frames: the sender will keep trying to retransmit until the receiver gets the frame perfectly. This is not what we want.

We attempted to receive the pings at various distances with a Prism 2 card in another laptop, using a Global Positioning System receiver to mark our location. The links we tested were line-of-sight over the Charles River, which separates Boston and Cambridge.

We made three measurements: (1) How far is a line-of-sight outdoor link functional without receiving bit errors? (2) How far out can a line-of-sight outdoor link consistently receive packets, with or without bit errors? and (3) What is the distribution of bit errors per packet on an example long-distance link?⁸

The difference between (2) and (1) is essentially the ``marginal zone'': the increase in range we expect to receive by using forward error correction.

We made measurement (3) by taking just over three minutes of data on the Boston bank of the river, 0.55 miles away from the transmitter. See Figure 3.

Figure 3: The 0.55-mile marginal link across the Charles River. Courtesy MassGIS and the MIT Department of Urban Studies and Planning.