A neural network shortest path algorithm for optimum routing in packet-switched communications networks

The authors consider the application of neural networks to the optimum routing problem in packet-switched communications networks, where the goal is to minimize the network-wide average time delay. Under appropriate assumptions it is shown that the optimum routing algorithm relies heavily on shortest path computations, which have to be carried out in real time. For this purpose an efficient neural network shortest path algorithm based on the Hopfield model is proposed, which is an improved version of previously suggested neural algorithms. The general principles involved in the design of the proposed neural network are discussed. The computational power of the proposed neural model is demonstrated through computer simulations. It is noted that the neural network approach will enable the communications engineer to benefit from the inherent features of neural networks, namely a potential for high computation power and speed, a high degree of robustness and fault tolerance, low power consumption, and real-time operation