Neural network controller for adaptive routing in survivable communications networks

Abstract

An isolated routing algorithm for adaptive dynamic routing in communications networks that is based on a neural network approach has been developed. A neural network is located within each site of a communications network, and it uses the patterns of incoming traffic to maintain and dynamically update an internal representation of the global topology of the communications network. The information stored in the neural network models is updated using various routing paradigms, including Hebbian learning. The scheme has been implemented in a software simulation running on a Sun workstation