Self organizing networks : building traffic and environment aware wireless systems

This dissertation investigates how to optimize flow-level performance in interference dominated wireless networks serving dynamic traffic loads. The schemes presented in this dissertation adapt to long-term (hours) spatial load variations, and the main metrics of interest are the file transfer delay or average flow throughput and the mean power expended by the transmitters. The first part presents a system level approach to interference management in an infrastructure based wireless network with full frequency reuse. The key idea is to use loose base station coordination that is tailored to the spatial load distribution and the propagation environment to exploit the diversity in a user population's sensitivity to interference. System architecture and abstractions to enable such coordination are developed for both the downlink and the uplink cases, which present differing interference characteristics. The basis for the approach is clustering and aggregation of traffic loads into classes of users with similar interference sensitivities that enable coarse grained information exchange among base stations with greatly reduced communication overheads. The dissertation explores ways to model and optimize the system under dynamic traffic loads where users come and go resulting in interference induced performance coupling across base stations. Based on extensive system-level simulations, I demonstrate load-dependent reductions in file transfer delay ranging from 20-80% as compared to a simple baseline not unlike systems used in the field today, while simultaneously providing more uniform coverage. Average savings in user power consumption of up to 75% are achieved. Performance results under heterogeneous spatial loads illustrate the importance of being traffic and environment aware. The second part studies the impact of policies to associate users with base stations/access points on flow-level performance in interference limited wireless networks. Most research in this area has used static interference models (i.e., neighboring base stations are always active) and resorted to intuitive objectives such as load balancing. In this dissertation, it is shown that this can be counter productive, and that asymmetries in load can lead to significantly better performance in the presence of dynamic interference which couples the transmission rates experienced by users at various base stations. A methodology that can be used to optimize the performance of a class of coupled systems is proposed, and applied to study the user association problem. It is demonstrated that by properly inducing load asymmetries, substantial performance gains can be achieved relative to a load balancing policy (e.g., 15 times reduction in mean delay). A novel measurement based, interference-aware association policy is presented that infers the degree of interference induced coupling and adapts to it. Systematic simulations establish that both the optimized static and interference-sensitive, adaptive association policies substantially outperform various proposed dynamic policies and that these results are robust to changes in file size distributions, channel parameters, and spatial load distributions.