IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 16, NO. 5, OCTOBER 2008 1147 Performance Optimization of Interference-Limited Multihop Networks Ahmed Bader and Eylem Ekici, Member, IEEE Abstract—The performance of a multihop wireless network is typically affected by the interference caused by transmissions in the same network. In a statistical fading environment, the interference effects become harder to predict. Information sources in a multihop wireless network can improve throughput and delay performance of data streams by implementing interference-aware packet injection mechanisms. Forcing packets to wait at the head of queues and coordinating packet injections among different sources enable effective control of copacket interference. In this paper, throughput and delay performance in interference-limited multihop networks is analyzed. Using nonlinear probabilistic hopping models, waiting times which jointly optimize throughput and delay performances are derived. Optimal coordinated injec- tion strategies are also investigated as functions of the number of information sources and their separations. The resulting analysis demonstrates the interaction of performance constraints and achievable capacity in a wireless multihop network. Index Terms—Hopping dynamics, interference-limited, mul- tihop networks, performance optimization, Rayleigh fading. I. INTRODUCTION I N MULTIHOP wireless networks where all transmitters share the same radio channel, a packet propagating through the network suffers from harmful interference generated by peer packets in the network. The wireless link quality is de- termined by interference, which in turn determines the longest distance a packet can correctly be received at. As the level of mutual interference increases, packets experience shorter hopping distances and slower propagation speeds across the network. Therefore, the network performance is a function of the internally-generated interference. In this work, we aim to analyze the performance of an interference-limited multihop wireless network in terms of information transfer from sources to sinks. The following two performance metrics are considered for this purpose: 1) Throughput (THR): The rate at which packets cross a mea- surement boundary that cuts each flow only once. 2) Head-of-Queue Delay (HQD): The sum of the time a packet spends at the head of the source queue and the multihop transmission time towards its destination. These two metrics are closely related to the transport rate metric which is the product of the throughput and the distance travelled Manuscript received April 25, 2006; revised April 3, 2007. First published March 3, 2008; current version published October 15, 2008. Approved by IEEE/ACM TRANSACTIONS ON NETWORKING Editor S. Das. A. Bader is with VTEL Holdings, Amman, Jordan (e-mail: bader.24@osu. edu). E. Ekici is with the Department of Electrical and Computer Engineering, The Ohio State University, Columbus, OH 43210 USA (e-mail: ekici@ece.osu.edu). Digital Object Identifier 10.1109/TNET.2007.905596 by packets over a single hop [1]. However, transport rate on its own does not capture the end-to-end packet delivery latency as will be shown in Section V-C. This observation motivates our choice to introduce HQD as an additional and necessary performance metric. Given a finite set of source-sink pairs, we are interested in distinct packet flows traversing the network and not necessarily how many packets may coexist in the network. The interference a packet suffers from can be classified as intra-flow and inter-flow. Interference from packets injected from the same source is referred to as intra-flow interference, whereas interference from packets belonging to other flows is referred to as inter-flow interference. The interdependence of THR and HQD in such networks leads to important observa- tions. An increased packet injection rate of information sources (THR) leads to increased numbers of packets propagating in the network, which increases the mutual interference levels. Con- sequently, the progress of packets is slowed down and HQD is adversely affected. Hence, there is an inherent tradeoff between the achievable THR and HQD. This tradeoff can be controlled by managing packet injection processes at the sources, which constitutes the main objective of this study. Information sources can achieve desired tradeoff levels by introducing appropriate waiting times between injection of packets. Forcing packets to wait at the head of the source queue creates a controlled interference environment for the leading packets in the same flow. Moreover, information sources must coordinate their injection processes such that the adverse effects of inter-flow interference are minimized. The analysis presented in this paper is performed for net- works with unlimited node density. We are interested in under- standing the statistical packet flow characteristics which opti- mize the network performance. For this purpose, we first in- troduce nonlinear models describing 1-D and 2-D packet flow dynamics under a probabilistic communication model. We then jointly consider THR and HQD in a multi-objective optimiza- tion problem, where we use nonlinear recursive methods to de- rive the optimal waiting times. We then devise local search tech- niques to derive the optimal number of flows and the optimal co- ordination for multiple-flow packet injection. Obtained results show the interactions between THR and HQD and achievable performance levels with respect to one metric when the other is used as a constraint. II. RELATED WORK In the literature, limitations of interference on the perfor- mance of multihop networks have been analyzed in a number of studies. In [1] and [2], asymptotic bounds on the achiev- able throughput and transport capacities under a deterministic 1063-6692/$25.00 © 2008 IEEE Authorized licensed use limited to: The Ohio State University. Downloaded on February 9, 2009 at 16:13 from IEEE Xplore. Restrictions apply.