Neural channel assignment-the fast way

A new algorithm using a nonpositive neural network model to solve channel assignment problems in arbitrary cellular mobile radio systems is proposed. The channel assignment problems are reduced to node-coloring problems in graph theory: nodes represent cells, and colors represent channels. When trying to solve the channel assignment problem using a neural network, this approach is not feasible any more. Reducing the channel assignment problem to finding a maximum independent set of an interference graph is very promising as there is a certain kind of neural network that is especially useful for finding such sets. Unlike the traditional neural nets used in optimization problems, the proposed network is guaranteed to deliver valid solutions after every update and needs no parameter tuning. The power of the new method to solve this NP-complete problem is shown in a number of simulations. Compared to other approaches, the solution presented here gives a much faster response with only slightly lower assignment rates. This is due to the careful selection of a neural network and a new mapping of channel assignment problem