A mobile ad hoc network (MANET) is a collection of mobile wireless nodes that self-organize without fixed infrastructure or centralized control. MANETs present a system of many factors, ranging from variables in the protocol stack to the movement of nodes. Choosing the appropriate values for internal parameters in response to external changes in conditions, i.e., adapting the parameters to optimize performance while conditions change, is a fundamental problem. Optimizations for fixed conditions are uninformative in scenarios where conditions change rapidly.
Through screening experiments, this dissertation demonstrates that factors in a MANET that contribute to measured responses vary as a function of node speed. At certain speeds one set of factors dictates performance, while at other speeds the set of factors is different. These experiments suggest that a linear model is inappropriate.
To effectively optimize systems of this nature, one must optimize reactively—i.e., in an online manner, while the network runs. Two approaches are explored here. The first is rooted in statistical modelling and process monitoring. The second applies machine learning to the problem.
Exploring a statistical approach develops profile-driven regression, a new modelling methodology. Profile-driven regression alleviates difficulties in modelling large sets of factors that display nonlinearity. Additionally, a cumulative score based model-monitoring scheme is developed. These methods are used to develop, monitor, and improve a model of throughput, as well as optimize throughput under changing node speeds.
The examination of machine learning focusses on artificial neural networks as a means of "learning" throughput conditions. Evaluation of several neural network topologies yields varying results. The best performing topology suggests throughput is affected nonlinearly by some factors, which agrees with the results obtained statistically.
Both examinations yield significant performance improvements over default parameters, though the studies conducted reveal limitations in both techniques. Because models are not explicitly represented in their topology, neural networks, while effective in system optimization, prove a poor choice for model monitoring and optimization. Conversely, while the statistical techniques explored provide useful and robust models of throughput under varying conditions, the potentially prohibitive number of experiments is currently more suitable for simulation than real experimentation.
|School:||Arizona State University|
|School Location:||United States -- Arizona|
|Source:||DAI-B 68/12, Dissertation Abstracts International|
|Subjects:||Industrial engineering, Computer science|
|Keywords:||Ad hoc networks, Machine learning, Mobile ad hoc networks|
Copyright in each Dissertation and Thesis is retained by the author. All Rights Reserved
The supplemental file or files you are about to download were provided to ProQuest by the author as part of a
dissertation or thesis. The supplemental files are provided "AS IS" without warranty. ProQuest is not responsible for the
content, format or impact on the supplemental file(s) on our system. in some cases, the file type may be unknown or
may be a .exe file. We recommend caution as you open such files.
Copyright of the original materials contained in the supplemental file is retained by the author and your access to the
supplemental files is subject to the ProQuest Terms and Conditions of use.
Depending on the size of the file(s) you are downloading, the system may take some time to download them. Please be