Empirical model-based adaptive control of MANETs

The performance of mobile ad hoc networks (MANETs) depends upon a number of dynamic factors that ultimately influence protocol and overall system performance. Adaptive protocols have been proposed that adjust their operation based on the values of factors, such as traffic load, node mobility, and link quality. In this work, however, we are investigating the feasibility of an adaptive model-based self-controller that can manage the values of controllable factors in MANETs. In general, the proposed self-controller should determine a set of factor values that will maximize system performance or satisfy specific performance requirements. The model-based controller adapts or reconfigures system-wide parameters or protocol operation as a function of the dynamically changing network state. In this paper, we describe the proposed self-controller, its design issues, and provide a preliminary case study to demonstrate the effectiveness and tradeoffs of two potential empirical-modeling techniques: regression and artificial neural networks.