Wireless Local Area Networks

Abstract

Wireless Local Area Networks (WLANs) have been widely developed during this decade, due to their mobility and exibility. During this period, IEEE 802.11 has become the dominant WLAN protocol. This thesis reports on research into WLANs, especially IEEE 802.11 networks. Since IEEE 802.11 denes rules at the MAC and Physical (PHY) layers, which are introduced in Chapter 1, the rst part (Chapters 2, 3 and 4) of this thesis deals with analytical models for the Distributed Coordination Function (DCF) of the IEEE 802.11 MAC, while the second part (Chapters 5 and 6) focuses on the transmission rates provided by the IEEE 802.11 PHY layer. Analytical models are widely adopted in research into WLANs, especially IEEE 802.11 networks. Despite dierences in details of published analytical models, most of them share common hypotheses. To ensure condence in the predictions made by the analytical models that are based on these common hypotheses, Chapter 2 identies these common hypotheses, and investigates them. By statistically analyzing simulation-based and experimental data, we found the appropriateness of these fundamental hypotheses only exists under some specic limitations. One of the common hypotheses investigated in Chapter 2 is the assumption that the conditional collision probability is constant and independent of the transmission history that is revealed by the back-o stage (the collision-decoupling assumption). Chapter 3 analyzes the relationship between the conditional collision probability and the back-o stage, by building an explorative analytical model without the commonly adopted collision-decoupling assumptions. Thus, Chapter 3 provides an analytical way to the understanding of the collision-decoupling assumption. Another common hypothesis investigated in Chapter 2 is the assumption that the probability of having a non-empty queue after each packet transmission is constant and independent of the transmission history (the queue-decoupling assumption). Although this queue-decoupling assumption is demonstrated to be incorrect in Chapter 2, the analytical models based on this assumption continue to make accurate predictions as reported in some papers [43][45]. To explain this paradox, in Chapter 4, we compare the predictive quantities from models with or without the queue-decoupling assumption. As we found, both models give similar and accurate predictions when the clients in the wireless network are symmetrically loaded. However, when these clients are asymmetrically loaded, the model with the queue-decoupling assumption starts to make errors, while the other model still gives the right answer. Therefore, Chapter 4 proves that the gap between reality and the queue-decoupling assumption can cause errors in model predictions. At the PHY layer, the IEEE 802.11 a/b/g WLAN protocol-suite provides a range of transmission rates determined by distinct physical layer modulation and Forward Error Correction schemes. Based on current channel conditions, a rate control algorithm at each station tries to select the right rate that gives the highest throughput. In the design of the rate control algorithm, it is commonly assumed that higher transmission rates suer more from interference from the noise in any channel conditions (the robustness-to-noise assumption). In Chapter 5, we investigated this assumption with theoretical calculations and experimental measurements. In our observations, there exist some redundant rates that exhibits less robust to the noise than the higher rates. Thus, Chapter 5 identies those redundant rates, and provides a new rate pool that obeys the robustness-to-noise assumptions on the rate control algorithm design. Finally, based on the new rate pool provided by Chapter 5, in Chapter 6, we present `H-RCA', a highly adaptive and collision-aware rate control algorithm. It is designed to minimize the average time each packet spends on the medium including MAC retries, in a fully decentralized fashion with no message exchange. In experiments, H-RCA outperforms both AMRR and SampleRate, which are well-known in the rate control community, in single and multi-client (collisions) scenarios, by providing a higher and more stable throughput.