Thumbnail Image


Publication or External Link






Mobile Ad Hoc Network is designed and deployed to achieve self-configuring and self-healing. MANET utilizes distributed wireless stations for relaying data packets. Every single station in the MANET can decide routing path for an incoming data packet. MANET has the most unfavorable conditions for routing path discovery due to node mobility and constant topology changes. Large variation of performance due to various environment inputs is a major impediment of implementing existing routing protocols for MANET in the battlefield. Therefore, it is a major challenge to design a routing protocol that can adapt its behavior to environment alteration.

In consideration of adaptability to the environment and flexibility in protocol construction, a novel component based routing protocol methodology is proposed in this paper. Distinguished from conventional investigation of routing protocols as individual entities, this paper will firstly generalize four fundamental components for MANET routing protocols. Then, a significant component diagnosis process is proposed to detect significant component and enhance the overall performance. Finally, preliminary simulation results demonstrate the power of the component based methodology for improving overall performance and reducing performance variation. In conclusion, the evaluation and improvement at the component level is more insightful and effective than that at the protocol level.

The primary contribution of the work is proposing the Component Dependence Network the first time and innovative quantitative methods are proposed to learn the structure and significant component to analyze the impact of component on performance metrics.

Based on conditional independence test, hierarchical structure of Component Dependence Network can be discovered. An Inclusion and Exclusion algorithm is introduced to guarantee the minimal cut set returned for a pair of source and destination nodes. To determine the significant component, a significance indicator will be calculated based on comparing each component's impact by using a backward deriving method. Once the significant component being determined, the parameter of the significant component can be tuned to achieve the best performance. At the end, two real implementations are presented to show the achievement in performance improvement of the component dependence network, structure learning method and significant component indicator.