Statistical Design of Experiments for Analyzing Mobile Ad Hoc Networks

Mobile ad hoc networking performance and scalability may be influenced by many factors, including protocol design at every layer; parameter settings such as retransmission limits and timers; system factors such as network size and traffic load; as well as environmental factors such as channel fading. In this work, we are concerned with understanding the functional relationship between these influential factors and performance in mobile ad hoc networking systems. Oftentimes, the one-factor-at-a-time (OFAT) approach is used as a mechanism to understand factor effects and their relationship to various performance measures. However, the OFAT approach is inefficient and can easily lead to subjective conclusions or conclusions only valid over a small set of conditions and scenarios. In this paper, we show how a systematic statistical design of experiments (DOE) strategy can be used to analyze wireless ad hoc network system and protocol performance, leading to more objective conclusions valid over a wide range of network conditions and environments. Specifically, we use the statistical DOE strategy and a 2 factorial design to quantify the main and interactive effects of five factors (i.e., network density, node mobility, traffic load, network size, and medium access control scheme) on two response metrics (i.e., packet delivery ratio and end-to-end delay). Using these effects measures, we then develop two empirical first-order linear regression models that define the functional relationship between the influential factors and two performance metrics. Keywords—mobile ad hoc networks, statistical design of experiments, performance evaluation