1-4244-2575-4/08/$20.00 © 2008 IEEE Interference Aware Multi-path Routing in Wireless Networks Aravind B. Mohanoor, S. Radhakrishnan V. Sarangan School of Computer Science, Computer Science Department University of Oklahoma Oklahoma State University Norman, Oklahoma 73019 Stillwater, Oklahoma 74078 {aravindmc, sridhar}@ou.edu saranga@cs.okstate.edu Abstract We can improve the end-to-end throughput between a sender and receiver in a wireless network using multiple paths which do not interfere with each other. Given that the problem of finding such paths is computationally hard, the paper focuses on finding multiple paths which may have interference between them, but still are able to obtain the maximum possible throughput. It is achieved by observing that the pattern of interference is more important than the number of interfering links. The nature of path sets with non- destructive interference is discussed and based on these observations, combinatorial techniques for finding interference aware disjoint paths in a wireless network are presented. Simulation results indicate that the proposed solutions achieve throughputs that are significantly higher than the established theoretical results. 1. Introduction With the introduction of multi-hop wireless services such as city-wide wide area mesh networks, it has become important to improve the throughput of wireless networks to provide good quality of service. Under the standard radio model, the multihop bandwidth can be at best a third of the single hop bandwidth [6]. Using multiple paths is one way to improve the end-to-end throughput, but interference between these multiple paths causes a significant reduction in the overall throughput [7]. By combining appropriate path selection, and a systematic packet transmission schedule, this throughput reduction can be avoided. While the maximum possible throughput can be achieved using three non-interfering paths [6], the problem of finding such non-interfering paths is the same as the problem of finding a chordless cycle containing a pair of vertices in a graph, which is actually NP-Complete [1]. It is not always necessary to use non-interfering paths, however. By finding vertex disjoint paths between source s and destination t which follow certain patterns of interference, we can achieve throughputs which are optimal or close to optimal. Our contributions are: (a) we demonstrate that it is possible for a set of paths between source ‘s’ and destination ‘t’ with some interference between them to provide high aggregate throughputs provided the interfering edges among the paths follow certain favorable patterns; we present a combinatorial approach for finding such paths in a wireless network (b) we extend our approach to scenarios involving multiple s-t pairs and show that the proposed approach can improve the throughput in such scenarios too (c) our combinatorial approach can also provide a straightforward mechanism for scheduling the transmissions at various links and finally, (d) the computation of the transmission schedule is amenable to a distributed implementation. The paper is organized as follows. Section 2 discusses literature that is relevant to the proposed work. Section 3 presents the ideas underlying the centralized combinatorial approach for finding interference aware multiple s-t paths. Section 4 presents simulation results and Section 5 concludes the discussion. 2. Related Work Many existing algorithms address the issue of improving wireless network throughput by minimizing the impact of interference. Hu et al. [4], Saha et al. [9] and Jones et al. [6] discuss techniques which try to find multiple node disjoint paths between s and t such that there are no edges connecting two vertices belonging to different paths. They do not consider simultaneous multiple s-t transmissions. Jain et al. [5] and Buragohain et al. [2] discuss strategies which use a centralized multi-commodity flow based linear programming (LP) formulation to exhaustively search and determine the maximum achievable throughput with interpath links. However these solutions provide neither the paths nor the schedules for transmission. Nevertheless, they are 516