Characterizing Interference in 802.11-based Wireless Mesh Networks Anand Kashyap 1 , Samrat Ganguly 2 , Samir R. Das 1 1 Computer Science Department, Stony Brook University, Stony Brook, NY, 11794 2 NEC Laboratories, Princeton, NJ ABSTRACT In a multihop wireless mesh network, interference among links has a significant impact on the link capacity. Thus, characterizing the interference is critical in understanding and improving the performance of the network. A common scenario in a mesh network is the existence of multiple in- terferers. Modeling the effect of multiple interferers is chal- lenging as the interference level not only depends upon the number of interferers but also on their relative positions. In this work, we present a measurement based interference model that captures the effect of multiple interferers. The interference model consists of metrics that determines the transmission capacity at the sender and collision probability at receiver in presence of interferers. The presented interfer- ence model can predict the capacity of any link in the mesh network using O(n) measurements. 1. INTRODUCTION The achievable capacity in a multihop wireless mesh network is interference limited as studied in [3, 4, 5, 6, 2]. Much of these works use a somewhat abstract and idealized model of interference, which describes a binary relation of inter- ference between two links, based on physical distances be- tween the transmitters and receivers, simplified radio prop- agation model, idealized transmitter and receiver character- istics, and so on. Interference is not binary in practice [7], and it is very difficult to develop an analytical model for capacity if a more realistic non-binary interference model is assumed. Thus, although these capacity models are suffi- cient for getting a high-level understanding of the behavior of various protocols in wireless networks, it has remained unclear how these models can be useful in practice for an operational network. Our work is motivated by the view that characterizing and modeling of interference in operational mesh networks would be quite useful in developing sophisticated protocol support. For example, routing and load balancing protocols can be improved by designing good metrics based on such inter- ference models. Scheduling algorithms can schedule trans- missions based on how links actually interfere rather than relying on graph abstractions. Admission control protocols supporting real-time/interactive media can also use this to develop capacity models and take admission control and routing decisions. In this work, our goal is to provide an empirical model to determine the capacity of a link in a 802.11 mesh network in the presence of interference. The capacity of a link is determined by two main factors - how good the link is, and what is the impact of interference on the link from traf- fic on neighboring links. The quality of the link is given by the delivery ratio, or the loss rate. Thus, in absence of any interference, the link capacity would be the product of the maximum sending rate of the sender and the delivery ratio of the link. Interference impacts the sender by reduc- ing its maximum sending rate, and it impacts the receiver by reducing the probability of receiving a packet success- fully by causing collisions. Our work models the sender side and receiver side interference independently to determine the capacity of a link. Our goal is to use commodity 802.11 radios and characterize the interference behavior such that throughput on given links can be predicted in the presence of multiple interferers. Our work follows on the lines of previous works done on developing a measurement based model for quantifying the interference limited link capacity [7, 8]. The measurement heuristic described in [7] provides an interference model which not only indicates whether two links interfere, but also quan- tifies the level of interference with each other. The approach requires a O(n 2 ) method to estimate interference between all link pairs in a mesh. However, the above measurement based interference model is not sufficient to estimate the capacity of a link when more than two links are active together. In this work, we particularly focus on capturing the effect of multiple interferers in constructing the interference model. The measurement-based model in [8] models sender-side and receiver-side interference separately, similar to our work, but has two weaknesses. First, the model is based on measured signal power of received packets, which is a very unsta- ble metric, and secondly, the model is difficult to extend when there are multiple interferers. For example, in a net- work with more than two simultaneously active senders, the knowledge of received signal strength between node pairs is not sufficient to determine the transmission rate of each sender. In our approach, we use measurements over a wireless net- work testbed to find the correlation of the impact of inter- ference with various measurable metrics, like the average signal strength of packets received over a link, and the de- livery ratio of a link. Measurements of these parameters can be done in a non-intrusive way by broadcasting beacons pe- riodically, where-in all the nodes listening to these beacons 1