We define wireless links using packet loss rate as in [4]. Packet loss rates are easy to measure and reflect the link quality experienced by higher layers. For a pair of nodes that communicates using the 802.11 protocol, packet loss rate in both directions matters, since a unicast packet transmission is considered successful only if the sender successfully receives the ACK sent by the receiver. To discover links that have reasonably low packet loss rate in both directions, we carried out the following experiment.
We had each node in our testbed broadcast 1000 byte packets for 30 seconds. Only one node was active at a time. We measured the packet reception rate at all other nodes in the network. The entire test was repeated 50 times. This data gives us the average packet loss rate between every ordered pair of nodes in our testbed. For two nodes and , let be the packet loss rate from to , and let be the loss rate from to . We say that links and exist if: , where is some threshold value. This definition was proposed in [4], where the ratio is called the ETX (expected transmissions) value of the link. For the purposes of our paper it is sufficient to note that a high ETX value implies that the link is lossy in either one or both directions.
We use the threshold of in the rest of the paper, to weed out highly lossy links. Any reasonable routing protocol will avoid such poor quality links. Of the possible links in our testbed, 152 links have . The average loss rate of these 152 links is 2.9%. We have experimented with lower values of ; lower values reduce the number of links, but our interference results remain similar.
Note that we measure packet loss rate in both directions with 1000 byte packets, even though ACK packets are much smaller than data packets. It is not our aim to accurately characterize the loss rate of a link - we only want to eliminate highly lossy links that a routing protocol will avoid. Similar approach has been used in [4,5].