Analytical framework for MANET learning and routing

This paper presents an analytical model of a multi-layer (social-communication) network supported with on-the-fly learned and prior social-cognitive features. In this model, routing is formulated in a distributed setting where each node operating with local information only finds the next-hop node, among all social and mobile ad hoc network (MANET) communication link neighbors, whose normalized delay is the smallest. This approach minimizes the end-to-end delay and readily extends to optimize other metrics, e.g., success probability. Delay performance is analytically evaluated as a function of distance to the destination node, which is estimated in social-cognitive learning under two social-cognitive features, frequency of encounters and social pressure metric. Different mobility models (random waypoint and group) are considered for the communication topology and different graph models (uniform random or scale-free) are considered for the underlying social network. Analysis is provided for both infinite and finite node models with different levels of granularity. In addition, real social network data is used to evaluate the analysis. Results show that online learning can closely track the actual distances to the destination and the expected delay with learning is close to that with perfect knowledge of network dynamics, while it saturates with hop distance, implying that the small-world phenomenon is captured.