Central Controller Using Neural Network in a Three-Stage Clos' Packet Switch.

摘要

The paper presents a central controller of a three stage Clos' packet switch with input queueing, which uses neural networks of Hopfield and Tank model to assign an optimal path arrangement. The packet traffic of the switch can be written as a traffic matrix, for every time slot. The optimal path arrangement is to find a series of optimal configuration matrices from the traffic matrix. The neural network performs these operations by computing the minimal energy of neurons. The neural network is nXn (n is number of input/output lines of a middle stage switch) array of inverting amplifiers. The simulations have shown results that the neural network can rapidly find these optimal configuration matrices after 2.5X10(sup -2)mtau second (m is number of switches in middle stage; tau is the time constant of neuron) and in an input queue system, the throughput of this scheme is very close to the throughput of single stage switching.