ABSTRACT
This article reports and summarizes a neural-network-based approach that was used to dynamically configure maps of a satellite communication network, and was incorporated with a state-dependent routing scheme to manage the network traffic. We modified Kohonen’s self-organization paradigm to automate the map configuration task. The modified algorithm consisted of three phases: (i) pattern recognition for selecting an exemplar map which most resembled the input traffic, (ii) a learning phase for fine tuning the chosen exemplar map, and (iii) a decision-maker for replacing the original exemplar map. The intrinsic properties of the proposed unsupervised learning allowed efficient tracking of the random traffic. The proposed traffic management scheme was effective in reducing the block rate of the network, as demonstrated through simulations.
